An Introduction to
Marketing Research
Scott M. Smith
James Passey Professor of Marketing
Founder, Qualtrics
Gerald S. Albaum
Emeritus Professor of Marketing
University of New Mexico
Copyright © 2010 by Scott M. Smith and Gerald S. Albaum.
This book is made available electronically to users of Qualtrics without charge through the Qualtrics Survey
University.
All rights reserved. No part of this book may be reproduced except for personal use, or utilized in any form or by any
means, electronic or mechanical, including photocopying, recording, or by any information storage and retrieval
system, without permission in writing from the authors.
Table of Contents
Chapter 1 An Introduction to Marketing Research .....................................................1
Chapter 2 Defining the Research Design and
Controlling Research Errors ......................................................................21
Chapter 3 Secondary Sources of Information.............................................................37
Chapter 4 Conducting Interviews ..............................................................................59
Chapter 5 Modes of Interviewing Personal-Send-Call ..............................................79
Chapter 6 Qualitative Research and Observation.......................................................97
Chapter 7 Sampling Procedures in Research............................................................123
Chapter 8 Experimentation ......................................................................................153
Chapter 9 Measuring Respondent Information:
Attitudes, Satisfaction, Loyalty and Behavior ........................................191
Chapter 10 General Concepts of Measurement and Scaling .....................................219
Chapter 11 Hypothesis Testing and Univariate Analysis ..........................................259
Chapter 12 Bivariate Data Analysis ..........................................................................287
Chapter 13 Multivariate Statistical Analysis I ..........................................................327
Chapter 14 Multivariate Statistical Analysis II ........................................................363
Chapter 15 Preparing the Research Report ................................................................393
Glossary ........................................................................................................ G-1
Appendix 1 ........................................................................................................ A-1
Preface
This book draws its “parentage” from the classic Research for Marketing Decisions by Paul E.
Green, Donald S. Tull, and Gerald Albaum. But, it is not a revision of that book. Rather, it might
best be viewed as a “child” which is targeted to a different audience—primarily senior-level
undergraduate and MBA students who are users of Qualtrics.com.
We believe this book is “novel” in at least three major respects. First, with respect to method, the
unifying concept of this book is that marketing research is a cost-incurring activity whose output
is information of potential value to managers in making decisions. Second, with respect to
technique, this book again departs from tradition in terms of an applied approach to the relatively
large coverage of more sophisticated, yet relatively easily implemented, research techniques. The
entire book focuses on implementation of online marketing research. Question types and
examples are implemented using internet survey provider, Qualtrics.com, so that students can
design, plan and implement an online survey of their own at no charge.
Finally, with respect to analysis, the book is expansive in its coverage, including relative
emphasis on modern analytical tools such as multivariate analysis. In terms of number of
chapters, 30% of the book is devoted to analysis, but, the discussion is at a level that senior-level
undergraduates can understand, and the techniques are explained within the context of computerbased
analysis.
This book is concerned with providing an introduction to marketing research. This means that all
the basic elements of method, techniques, and analysis are covered, including those at a more
sophisticated level. But, the book is NOT a book of only essentials. The methodological scope
regarding research design, data collection techniques, and measurement is broad. For example,
two chapters are devoted to the critical area of measurement and scaling. The book presents its
material from primarily a pragmatic and user-oriented (rather than theoretical research technician)
perspective. User-orientation is based on the premise that users need to know method in order to
evaluate research presented to them.
Because the book is available online, it can be used in a modular fashion at no cost to the student.
For example, if chapters on experimental design or multivariate statistics are beyond the scope of
the instructor’s focus, then they can simply be ignored. Similarly, if the course focuses on survey
research, chapters 9 and 10 could be the focal point, supplemented with chapters 1,2,4,5,6,7 plus
analysis chapters as appropriate. Note that because of the dynamic nature of electronic
publication, chapters may be edited, and additional chapters may be added from time to time.
There is a Glossary of Terms and an appendix that includes some widely-used statistical tables
for analysis. These tables will be useful for analyzing appropriate cases.
Many people helped shape the content and style of this book, but most importantly Professors
Paul E. Green and the late Donald S. Tull have had a profound influence on the authors’ thinking
about research and their book with one of the present authors provided a platform from which the
present book was launched.
SCOTT M. SMITH
GERALD S. ALBAUM
1
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 1
AN INTRODUCTION TO MARKETING RESEARCH
Marketing is a restless, changing, and dynamic business activity. The role of marketing itself has
changed dramatically due to various crises—material and energy shortages, inflation, economic
recessions, high unemployment, dying industries, dying companies, terrorism and war, and
effects due to rapid technological changes in certain industries. Such changes, including the
Internet, have forced today’s marketing executive to becoming more market driven in their
strategic decision-making, requiring a formalized means of acquiring accurate and timely
information about customers, products and the marketplace and the overall environment. The
means to help them do this is marketing research.
WHAT IS RESEARCH?
Research is a systematic and objective investigation of a subject or problem in order to
discover relevant information or principles. It can be considered to be either primarily
fundamental or applied in nature. Fundamental research, frequently called basic or pure
research, seeks to extend the boundaries of knowledge in a given area with no necessary
immediate application to existing problems, for example, the development of a research method
that would be able to predict what people will be like x years in the future. In contrast, applied
research, also known as decisional research, attempts to use existing knowledge to aid in the
solution of some given problem or set of problems.
Marketing research assists in the overall management of the marketing function. A
marketing manager must prioritize the more important and pressing problems selected for
solution, reach the best possible solution based on the information available, implement the
solution, modify the solution when additional information so dictates, and establish policy to act
as a ready-made solution for any recurrence of the problem.
Marketing research often focuses on understanding the “Customer” (purchasers,
consumers, influencers), the “Company” (product design, promotion, pricing, placement, service,
sales), and can also be expanded toward the environment to include “Competitors” (and how
their market offerings interact in the market environment).
Figure 1.1 Marketing Environment (Source: Modified from Perrault and McCarthy, )
2
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Within this “Company-Customer-Competition” environment, many types of marketing
research can be conducted, much of which is focused on using surveys for
• Monitoring customers and markets
• Measuring awareness, attitudes, and image
• Tracking product usage behavior
• Diagnosing immediate business problems
• Supporting strategy development
More specific examples are found in the Qualtrics.com Survey University. This provider
of professional survey software identifies twenty different kinds of surveys that are of use to
marketing researchers. Each focuses on a different aspect of the “Company” and it’s interaction
with the “Customer” and “Competition” in the market environment:
Exhibit 1.1 Twenty Different Types of Marketing Surveys
1 - Market Description Surveys
To determine the size and relative market share of the market. Such studies provide key information about
market growth, competitive positioning and tracking share of market.
2 - Market Profiling-Segmentation Surveys
To identify who the customers are, who they are not, and why they are or are not your customers. This is often a
descriptive market segmentation and market share analysis
3 - Stage in the Purchase Process / Tracking Surveys
Where is the customer in the adoption process? This information shows market Awareness – Knowledge –
Intention – Trial – Purchase – Repurchase of the product.
4 - Customer Intention - Purchase Analysis Surveys
Directed at understanding the current customer. What motivates the customer to move from interest in the
product to actual purchase? This is a key to understanding customer conversion, commitment and loyalty.
5 - Customer Attitudes and Expectations Surveys
Does the product meet customer expectations? What attitudes have customers formed about the product and/or
company. Used to direct advertising and improve customer conversion, commitment and loyalty.
6 - Customer Trust - Loyalty – Retention Analysis Surveys
Especially for high priced consumer goods with long decision and purchase processes (time from need
recognition to purchase), and depth of consumer attitudes formed about the product and/or company.
7 - New Product Concept Analysis Surveys
Concept test studies are appropriate in the initial screening of new product concepts. Likes and dislikes about the
concept and evaluation of acceptability and likelihood of purchase are especially useful measures.
8 - New Product Acceptance and Demand Surveys (Conjoint Analysis)
Primarily for estimating demand for new products that can be described or have been developed in drawing or
concept, but have not yet been developed physically. Develops develop market share estimates of market
potential for the alternative potential products.
9 - Habits and Uses Surveys
Directed at understanding usage situations, including how, when and where the product is used. Habits and uses
studies sometimes include a real or virtual pantry audit.
10 - Product Fulfillment Surveys (Attribute, Features, Promised Benefits)
Evaluation of the product’s promised bundle of benefits (both tangible and image). Are expectations created for
the product by advertising, packaging and the produce appearance fulfilled by the product?
11 - Product Positioning Surveys (Competitive Market Position)
A “Best Practices” study of “How does the market view us relative to the competition?” Competitive positioning
analyses often compare the attributes and benefits that make up the product using multidimensional scaling.
3
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
12 - Brand Equity Analysis Surveys
What is psychological value that a brand holds in the market place? Brand equity is a composite of brand
awareness, brand quality, brand associations and brand loyalty measures.
13 - Advertising Value Identification and Analysis Surveys
Advertising value analysis focuses on mapping the hierarchical attributes, benefits and values that are associated
with and portrayed by an advertisement. Means-end analysis is often part of this type of study.
14 - Advertising Message Effectiveness Surveys (Media and Message)
Message effectiveness testing identifies the impressions, feelings, and effectiveness in moving the respondent to
a desired goal (increased awareness, more product information, trial, repeat purchase).
15 - Sales Force Effectiveness Surveys
A combination of measures that focus on the sales activities, performance and effectiveness in producing the
desired and measurable effect or goal. Often measured as a 360 degree survey completed by the sales person,
the client (evaluating the sales call) and the supervisor responsible for evaluating the sales person.
16 - Sales Lead Generation Surveys
Sales lead generation surveys for (1) assuring timely use and follow-up of sales leads, (2) qualifying sales leads
(thereby saving valuable sales force time) and (3) providing more effective tracking of sales leads.
17 - Customer Service Surveys
Akin to customer satisfaction surveys, but focus in detail on the actual customer service that was received, the
process involved in receiving that service and the evaluation of the participants in the service process.
18 - Customer Service Representative (CSR) Surveys: Attitudes, Burnout, Turnover and Retention:
CSRs hold attitudes that reflect on their job related activities including (1) the allocation of time; (2) solutions to
customer needs; (3) how to improve their job; (4) best practices; (5) How well internal departments help
customers. CSRs often exhibit frustration, burnout and high turnover and surveys focus on CSR retention,
reducing costs and increasing the quality of customer relationships.
19 - Sales Forecasting and Market Tracking Surveys
Sales forecasting and market tracking studies can include expert opinion (experts estimate the market),
judgmental bootstrapping (expert based rules describing how to use available secondary market information),
conjoint analysis (estimation of consumer intentions based on product attributes that are important in the
decision), and intentions evaluations (consumer self reported intentions of future purchases) are to be made.
20 - Price Setting Surveys and Elasticity of Demand Analysis
Price surveys estimate the elasticity of demand and show optimal price points, including prices too low or too
high. Price surveys may estimate the demand for different product or service segments, or different usage
situations.
Source: Twenty Different Types of Marketing Surveys: http://www.qualtrics.com/wiki/index.php/Market_Surveys.
Each of the above surveys focuses on a specific area of research and involves the
development of conceptual models directed at predicting or explaining a specific type of
behavior that is being measured. This level of specificity is desirable for several reasons. Within
the research process, this specificity brings:
1. Clarification. Explication usually results in the clarification of relationships and interactions.
The need for more rigorous definitions of key variables often becomes apparent.
2. Objectivity. The process of explicating the modeled behavior often discloses rationalizations
and unfounded opinions that had not been recognized as such before.
3. Communication. Discussion helps to identify problems and common points of reference
when different people hold alternative implicit models of the same problem situation.
4. Improvement of models. Explicit models can be tested in differing situations to see if the
results are reproducible. The degree and range of adaptability can thus be extended.
4
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
5. Guide to research needs. Formulating models explicitly can better pinpoint information gaps
and, thus, aid in determining the nature of research needs.
While varying information is required for the different types of marketing research
projects, the key to conducting a successful research project lies with the researcher and the
client. They must come to a common understanding of the nature of the exact research problem,
and then agree on the information required to answer this problem. This requires identifying the
appropriate questions, respondents, methodology, analysis and reporting. All studies must
address these same basic issues (see Exhibit 1.2).
EXHIBIT 1.2 Basic Research Issues
As technology advances, marketing researchers are continually looking for ways to adapt new technology
to the practice of research. Both hardware and software are involved in such adaptations. However,
researchers must never forget that research basics cannot be overlooked. Rather, what must be done is to
adapt the new techniques and technologies to these basics. All studies must address the following basic
issues (Anderson, Berdie, & Liestman, 1984):
1. Ask the right questions. This is the essence of project design, and the heart of proper planning. The
research planner must remember that every project is unique, and as such must be tailored to the
user’s needs.
2. Ask the right people. Sample design should be such that only those people who are of interest to the
research user are contacted, and such that those who are contacted are reasonably representative of
the group of interest.
3. Ask questions the right way. It is not enough to be able to ask the right questions; they must be asked in
the right way. This is the essence of questionnaire design. The researcher can use all the aids available
from the new technologies, but if the wording of the questions is not clear to the respondents, the
results will be useless. One basic that is overlooked all too often is pretesting the questionnaire; this is
crucial for ensuring that responses are the ones that are needed to address the problem.
4. Obtain answers to questions. The process of data collection is central to all marketing research.
Techniques used should be selected for how each bears on nonresponse and response alike.
5. Relate answers to the needs of the research user/client. Data seldom speak for themselves.
Proper data analysis is needed if a study is to have any value to the user. Here there is a risk of letting
advanced techniques become the master of the researcher rather than the opposite.
Common sense is a valuable tool for the researcher when considering alternative analysis approaches
for any project.
6. Communicate effectively and in a way that the client understands. Many good projects are ruined in
this stage. The information that is reported to the user should be in a form that is understandable to
the user so that he or she can tell that it is relevant to the issues at hand.
Having considered these general topic-situation issues in conducting research, let’s now
turn to the basic process of conducting a research process.
5
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
THE BASIC RESEARCH PROCESS
How is marketing research actually conducted? What are the general steps in completing
a research project? These questions are answered in the steps of the research process. While the
steps are shown as a linear process, some of the steps may be performed simultaneously, such as
selecting data collection techniques and sample design. There are other times when “later”
decisions influence decisions that are made early in the research planning process. For example,
desired analysis techniques often influence the selection of data collection techniques (e.g.,
measurement) and sample design.
Figure 1.2 The Research Process
It is important to carefully plan the research process and formally recognize the
relationship between the stages. The researcher should write a formal plan for the project,
including the background information and statement of objectives, which then becomes the
master guide for implementing and controlling the research project. Each step in this research
process will now be introduced.
STAGE 1: PROBLEM FORMULATION
In a very real sense, problem formulation is the heart of the research process. As such, it
represents the single most important step to be performed. From the researcher’s point of view,
problem formulation means translating the management problem into a research problem.
As previously discussed, in order to formulate an appropriate research problem, the
researcher must understand the origin and nature of management’s problem and then be able to
rephrase it into meaningful terms from an analytical point of view. This involves timely and clear
communication between manager and researcher.
The end result of problem formulation is a statement of the management problem that is
analytically meaningful and that often points the way to alternative solutions. An accurate
problem formulation specifies the types of information needed to help solve the management
problem. In short, quality thinking about a problem prior to data collection largely determines the
quality of data collection, analysis and problem solving.
6
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 1.2 Examples of Management Problems and Related Research Problems
Management Problems Research Problems
Allocate advertising budget to media Estimate awareness generated by each media type
Decide whether to keep office open Evaluate use of services on Saturday and determine on
Saturday whether customers will shift usage to weekdays
Introduce a new health service Design a concept test and assess acceptance and use
Change the marketing program Design a test-marketing situation such that the effect of
the new program can be estimated
Increase the sales of a product Measure a product’s current image
Closely related to problem formulation is the development of a working hypothesis, or an
assertion about a state of nature. While hypotheses are crucial for basic research because they tell
the researcher what to do, the concept of a hypothesis can also be useful in decisional research to
direct the development of the research problem statement. In most cases, the marketing
researcher will not explicitly state hypotheses for the research. Kerlinger and Lee (2000, Chapter
2) suggest that research problems and hypotheses meet the following criteria:
1. The problem statement expresses a relationship between two or more variables.
2. The problem is stated clearly and unambiguously in question form.
3. The problem statement implies possibilities of empirical testing.
Where properties of good hypotheses include the following:
1. The hypothesis is a statement about the relationship between two or more variables in
declarative statement form.
2. The hypothesis carries clear implications for testing the stated relationship
(i.e., variables must be measurable or potentially measurable).
How to Formulate the Research Problem
Problem formulation is much easier when specific components of the research problem
are defined:
1. Specify the Research Objectives
Objectives guide the researcher in developing good, useful research, and they help the
client evaluate the completed project. Objectives range from the very general, such as profit
maximization, to the highly specific, such as measuring market interest in a new product. It is
rare that the objectives are well explained to the researcher. However, the researcher needs to
take the initiative to develop a clear statement of objectives.
Each study should have a very limited and manageable set of objectives that focus on the
problem being solved. Two or three well targeted objectives is preferable to many that are illconceived.
Fewer the objectives make it easier to keep track of progress toward the objectives,
to ensure that each is properly addressed, and to determine the best methodology. If there are too
many objectives separate studies may be appropriate.
7
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
2. The Environment or Context of the Problem
Consider the problem of deciding whether to introduce a new consumer product. The
marketing researcher must work closely with the client in transforming the client’s problem into
a workable research problem.
The researcher’s efforts should be oriented toward helping the manager decide whether
any investigation is justified based on the potential value of the research findings versus their
cost. The researcher must be aware of, and assist in, the identification of objectives, courses of
action, and environmental variables, insofar as they affect the design of the research
investigation.
If the research is undertaken and if the resulting findings are to be utilized (i.e., have an
influence on the user’s decision making), the manager and researcher must have a productive and
trusting relationship that is based on the researcher’s ability to perform and deliver the research
as promised.
3. The Nature of the Problem
Every research problem may be evaluated on a scale that ranges from very simple to very
complex. The degree of complexity depends on the number of variables that influence the
problem. Understanding the nature of the problem helps a researcher ensure that the right
problem is being investigated and that a marketing plan can be developed to solve the problem.
A thorough preliminary investigation using focus groups of consumers, salespeople, managers,
or others close to the problem may produce much needed insight.
4. Alternative Courses of Action
A course of action specifies a behavioral sequence that occurs over time, such as the
adoption of a new package design, or the introduction of a new product. Such a program of
action becomes a commitment, made in the present, to follow some behavioral pattern in the
future.
It is usually desirable to generate as many alternatives as possible during the problem
formulation stage and state them in the form of research hypotheses to be examined. A
hypothesis often implies a possible course of action with a prediction of the outcome if that
course of action is followed.
Once the nature of the problem has been agreed upon, the course of action must be
specified. This involves:
1. Determining which variables affect the solution to the problem
2. Determining the degree to which each variable can be controlled
3. Determining the functional relationships between the variables and which variables are critical
to the solution of the problem.
The following example shows the results of a failure to follow through with these aspects
of the problem situation model.
8
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 1.3 “New Coke” Versus Original Coke
In the mid-1980s the Coca Cola Company made a decision to introduce a new beverage product (Hartley, 1995,
pp. 129–145). The company had evidence that taste was the single most important cause of Coke’s decline in the market
share in the late 1970s and early 1980s. A new product dubbed “New Coke” was developed that was sweeter than the
original-formula Coke.
Almost 200,000 blind product taste tests were conducted in the United States, and more than one-half of the
participants favored New Coke over both the original formula and Pepsi. The new product was introduced and the original
formula was withdrawn from the market. This turned out to be a big mistake! Eventually, the company reintroduced the
original formula as Coke Classic and tried to market the two products. Ultimately, New Coke was withdrawn from the
market.
What went wrong? Two things stand out. First, there was a flaw in the market research taste tests that were
conducted: They assumed that taste was the deciding factor in consumer purchase behavior. Consumers were not told
that only one product would be marketed. Thus, they were not asked whether they would give up the original formula for
New Coke. Second, no one realized the symbolic value and emotional involvement people had with the original Coke. The
bottom line on this is that relevant variables that would affect the problem solution were not included in the research.
CBS New Coke News Clip: http://www.youtube.com/watch?v=-doEpVWFLsE&NR=1&feature=fvwp
New Coke Commercial: http://www.youtube.com/watch?v=o4YvmN1hvNA
New Coke and Coke Classic Commercial: http://www.youtube.com/watch?v=ky45YGUA3co
5. The Consequences of Alternative Courses of Action
A set of consequences always relate to courses of action and even to the occurrence of events not
under the control of the manager. One of the manager’s primary jobs is to anticipate and
communicate the possible outcomes of various courses of action that may result from following
the research.
6. Degrees of Uncertainty
Most marketing problems are characterized by a situation of uncertainty as to which course of
action is best. Years of experience may allow the decision-making manager to assign various
“likelihoods of occurrence” to the various possible outcomes of specific courses of action.
A carefully formulated problem and statement of research purpose is necessary for competent
research. The statement of purpose involves a translation of the decision maker’s problem into a
research problem and the derivation of a study design from this problem formulation. The
research problem provides relevant information concerning recognized (or newly generated)
alternative solutions to aid in this choice.
STAGE 2: METHOD OF INQUIRY
Market researchers look to the scientific method as the source of their investigative
methods. Even though this method is not the only one used, it is the standard against which other
investigative methods are measured. The scientific method makes great use of existing
knowledge both as a starting point for investigation and as a check on the results of the
investigations (i.e., a test of validity). Its most distinctive characteristic is its total lack of
subjectivity. The scientific method has evolved objective and rigid procedures for verifying
hypotheses or evaluating evidence. It is analytical in its processes and is investigatorindependent.
Thus, the scientific method is for the most part logical and objective, and frequently
makes extensive use of mathematical reasoning and complicated experiments.
9
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The goal of a scientific methodologist, also called an objectivist, is to run a hypothesis
test using publicly stated procedures that are investigator-independent.
• Formulate a problem
• Develop a hypothesis
• Make predictions based on the hypothesis
• Devise a test of the hypothesis
• Conduct the test
• Analyze the results
Even though the terminology used is that associated with basic research, the process
described is analogous to that of decision making. Although the steps are the same, there are
differences in the way in which the steps are performed and in the underlying assumptions about
behavior. For example, the essential difference between the objectivist and the subjectivist is the
latter’s allowance for use of subjective judgments both when collecting data and when analyzing
data (Diesing, 1966).
This objectivist-subjectivist distinction has very practical meaning, particularly when
considering the use of outside research suppliers. There are commercial research firms that tend
to specialize in one or the other method of inquiry. Objectivist-based research is often called
quantitative research, whereas subjectivist-based research is often called qualitative research.
Exhibit 1.4 The Scientific Method
In structure, if not always in application, the scientific method is simple and consists of the following steps:
1. Observation. This is the problem-awareness phase, which involves observing a set of significant
factors that relate to the problem situation.
2. Formulation of hypotheses. In this stage, a hypothesis (i.e., a generalization about reality that permit
prediction) is formed that postulates a connection between seemingly unrelated facts. In a sense, the
hypothesis suggests an explanation of what has been observed.
3. Prediction of the future. After hypotheses are formulated, their logical implications are deduced. This
stage uses the hypotheses to predict what will happen.
4. Testing the hypotheses. This is the evidence collection and evaluation stage. From a research project
perspective this is the design and implementation of the main study. Conclusions are stated based on
the data collected and evaluated.
A simple example will show how the scientific method works. Assume a researcher is performing a
marketing research project for a manufacturer of men’s shirts:
1. Observation: The researcher notices some competitors’ sales are increasing and that many
competitors have shifted to a new plastic wrapping.
2. Formulation of hypotheses: The researcher assumes his client’s products are of similar quality and
that the plastic wrapping is the sole cause of increased competitors’ sales.
3. Prediction of the future: The hypothesis predicts that sales will increase if the manufacturer shifts to
the new wrapping.
4. Testing the hypotheses: The client produces some shirts in the new packaging and market-tests
them.
10
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
STAGE 3: RESEARCH METHOD
Whether a particular method of inquiry is appropriate for a research problem depends in
large part on the nature of the problem itself and the extent or level of existing knowledge. In
addition to selecting a method of inquiry, the research planner must also select a research
method.
Two broad methodologies can be used to answer any research question–experimental
research and non-experimental research. The major advantage of experimental research lies in
the ability to control extraneous variables and manipulate one or more variables by the
intervention of the investigator. In non-experimental research, there is no intervention beyond
that needed for purposes of measurement.
STAGE 4: RESEARCH DESIGN
Research design is defined as the specific methods and procedures for acquiring the
information needed. It is a plan or organizational framework for doing the study and collecting
the data. Research designs are unique to a methodology. We discuss research design in depth
later in this document and in Chapter 3.
STAGE 5: DATA COLLECTION TECHNIQUES
Research design begins to take on detailed focus as the researcher selects the particular
techniques to be used in solving the problem formulated and in carrying out the method selected.
A number of techniques available for collecting data can be used. Some techniques are unique to
a method of inquiry. For example, many of the qualitative research techniques, such as projective
techniques, are used only in subjectivist-type research. In general, data collection uses either
communication or observation.
Communication involves asking questions and receiving responses. This process can be
done in person, by mail, by telephone, by e-mail, and over the Internet. In most instances this
constitutes the broad research technique known as the survey. In contrast to this process, data
may be obtained by observing present or past behavior. Regarding past behavior, data collection
techniques include looking at secondary data such as company records, reviewing studies
published by external sources, and examining physical traces such as erosion and accretion.
In order to collect data from communication or observation there must be a means of
recording responses or behavior. Thus, the process of measurement and the development of
measurement instrument are closely connected to the decision of which data collection
technique(s) should be used. The relationship is two-way. That is, the structure and content of the
measurement instrument can depend on the data collection technique, and measurement
considerations often influence technique selection.
STAGE 6: SAMPLE DESIGN
Rarely will a marketing research project involve examining the entire population that is
relevant to the problem. For the most part, practical considerations (e.g., absolute resources
available, cost vs. value, etc.) dictate that one use a sample, or subset of the relevant population.
In other instances the use of a sample is derived from consideration of the relevant systematic
and variable errors that might arise in a project.
11
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In designing the sample, the researcher must specify three things:
1. Where the sample is to be selected
2. The process of selection
3. The size of the sample
The sample design must be consistent with the relevant population, which is usually
specified in the problem-formulation stage of the research process. This allows the data obtained
from the sample to be used in making inferences about the larger population.
The process of sample selection may be done by probability or non-probability methods.
In probability sampling every element in the population has a known nonzero probability
(chance) of being selected for inclusion in a study. In contrast, a non-probability sample is one
selected on the basis of the judgment of the investigator, convenience, or by some other means
not involving the use of probabilities.
STAGE 7: DATA COLLECTION
Data collection begins after the previous six stages of the research process are complete.
Data collection, whether by communication or observation, requires the use of data collection
personnel which then raises questions regarding managing these people. Because data collection
can be costly, firms often utilize outside limited-service research suppliers, particularly when the
extent of in-house research activity does not warrant the cost of having permanent data collection
personnel. Also, project design may require specialized data collection, which might best be
obtained from an outside supplier.
The working relationship between the data collection agency (a so-called field service) and the
research supplier or client is a major factor affecting the quality of fieldwork and data collection.
A study of marketing research firms found that the major barriers to the communication
of information from clients to research suppliers to field service firms were insufficient
information supplied by the client, the research supplier as an intermediary between client and
field service firm, and lack of client interest in data collection (Segal & Newberry, 1983).
The major suggestion for improving communication is for clients to provide more
information to both suppliers and field service firms. Another way to overcome communication
barriers is for the field service to be consulted on such major issues as scheduling, costs, and
purpose of the study. Finally, it was suggested that two-way communication with suppliers be
established or strengthened. Although this study was conducted more than 20 years ago, these
are enduring problems that exist today.
STAGE 8: ANALYSIS AND INTERPRETATION
Data that are obtained and presented in the same form as originally collected are seldom
useful to anyone. Data must be analyzed. The data must be edited, coded, and tabulated before
performing formal analyses such as statistical tests. The types of analyses that can be properly
performed depend upon the sampling procedures, measurement instruments, and data collection
techniques used. Consequently, it is imperative that the techniques of analysis, associated
descriptive or prescriptive recommendation types, and presentation formats be selected prior to
data collection.
12
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
STAGE 9: THE RESEARCH REPORT
The culmination of the research process is the research report. It includes a clear,
accurate, and honest description of everything that has been done and the results, conclusions,
and— whenever possible—recommendations for courses of action. Two critical attributes of the
report are that it provides all the information readers need using language they understand
(completeness) and that it contains selective information chosen by the researcher (conciseness).
These attributes are often in conflict with each other.
Two approaches can be taken to ensure that this conflict is not a problem. One approach
involves preparing two reports: (1) a technical report that emphasizes the methods used and
underlying assumptions, and presents the findings in a detailed manner; and (2) a popular report
that minimizes technical details and emphasizes simplicity.
The second approach is concerned with how the report is communicated. Because people
vary a great deal in how they are affected by different forms of communication, the ideal
reporting process should try to encompass all major forms. Thus, a written report, by itself, may
be inadequate and only an invitation to inaction. There are simply a lot of people who, for
various reasons, don’t respond to the printed word. There are still more that, although they may
respond, will often misunderstand the meaning of what is written. For these reasons, it is vitally
necessary to get management to sit down with the research manager, or with the researcher and
the outside research firm, in a face-to-face reporting situation.
RESOURCE PLANNING FOR YOUR STUDY
When planning for your research, the resources necessary to complete the study should
also be identified. Resources include personnel, time and money. Resource plans range from very
informal to very formal and may include a list all personnel who will be involved with the project, the
exact assignment of each person, the time to be spent, and the pay for each. Additionally you will need to
prepare a budget and time schedule for the major activities involved in conducting the study. Microsoft
Project or similar software may be helpful in planning and monitoring your research project.
Source: http://office.microsoft.com/en-us/templates/TC012330951033.aspx
13
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
THE MAKE OR BUY DECISION
A decision facing all companies that want to use marketing research is who should do the
research. Alternatives are to have it done in-house, to utilize outside suppliers, or some
combination of the two. In short, sourcing marketing research is a “make or buy” decision. For
some companies, this decision is automatic—the in-house organization will do all research
unless it is beyond their technical expertise. Other companies with in-house capabilities treat the
internal units the same as outside suppliers: they must compete with outside suppliers by
preparing proposals and making bids for the business. Almost all research users will at some
time require the services of outside research suppliers.
Outside suppliers range from a full-service marketing research agency such as M/A/R/C
Research (http://www.marcresearch.com), Burke (http://www.burke.com), and Maritz Marketing
Research, Inc. (http://www.maritz.com), to a specialized survey software company like Qualtrics
(http://www.qualtrics.com) that provides sophisticated, yet easy to use online data collection and
analysis tools, combined with online training, customer support, respondent panels, and
analytical services for the client. Clients can negotiate with full-service companies to perform
only limited services, for example, research design and data collection only, if that is all the
client wants. Thus, there are many variations in the way outside suppliers are used.
When might the use of an outside research supplier be appropriate? There are a number
of situations that may call for the use of such firms:
1. The capabilities or technical expertise of in-house researchers are not adequate.
2. You are not able to hire needed personnel.
3. The outside supplier has the needed facilities for doing the research, such as those needed
for focus groups or laboratory experiments.
4. A research firm has a demonstrated expertise in a specific industry.
5. There is no unused capacity in the in-house research organization.
6. Lack of objectivity on the part of in-house personnel
7. The outside research supplier can do the research quicker.
8. Some aspect such as data collection may be cheaper when done by an outside supplier
(purchasing supermarket-based scanner data may be less costly than collecting it yourself).
9. There is a need for anonymity or confidentiality that may be provided best by an outside
research firm.
10. The results of the research may be used in legal proceedings. If so, the outside research firm
may have more credibility in the eyes of the court or regulatory or legislative body.
Exhibit 1.5 How to Develop a Successful Research-Consulting Relationship
Some rules of thumb for developing a quality relationship with a research client warrant consideration (proposed by
Schmalensee, 2001). For the most part these represent adaptation of more standard techniques and methods to fit a
B2B situation. These suggestions are organized around the typical flow of a research project:
1. Design research to foster customer relationships. This applies to all stages of a project. The research process
should be designed to strengthen relationships with business customers.
2. Lay the groundwork. It is suggested that the researcher allow extra time to talk with the staff, especially those
with customer contact. In B2B situations there may be many people who have customer contact, and their views
may differ enough that it is beneficial to talk with as many as possible.
14
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
3. Select and draw the samples. There may be a choice of respondents within each business organization, including
senior executives, second tier administrators and even customers. In a typical project it is often difficult to decide
which type of respondent to contact. One way to overcome this is to interview all major types identified as
having relevant information for the problem at hand, although the questions asked each type of respondent will
differ.
4. Select the research approach and methodology. Business respondents tend to be busy people, so it is important
to be creative in selecting data collection methodologies. For example a combined telephone and Internet may
lead to better information that a telephone or mail questionnaires. Understanding how your target respondent
can best be contacted can be helpful in selecting the best method for the majority of the sample.
5. Design the questions. Keep the questionnaire as “short and sweet” as possible. This, of course, applies to all
research projects. Business respondents will be more likely to respond if the questions are interesting and allow
them to respond in their own words in a conversational way.
6. Record and analyze the data. Much of the information collected in B2B research is qualitative, making the
analysis crucial.
7. Report the results. A good way to increase credibility and ensure that results lead to action is to personalize
results. This includes use of individual respondent anecdotes and other humanizing details.
8. Plan, communicate, and act. A good way to increase response rates and build relationships with customers is to
share with them what has been learned and what is planned. Communicating with customers allows a company
to involve them in implementing whatever action the research suggests. This, again, is part of relationship
building.
EXHIBIT 1.6 Basic Research Issues
As technology advances, marketing researchers are continually looking for ways to adapt new technology to the
practice of research. Both hardware and software are involved in such adaptations. However, researchers must never
forget that research basics cannot be overlooked. Rather, what must be done is to adapt the new techniques and
technologies to these basics. All studies must address the following basic issues (Anderson, Berdie, & Liestman, 1984):
1. Ask the right questions. This is the essence of project design, and the heart of proper planning. The research
planner must remember that every project is unique, and as such must be tailored to the user’s needs.
2. Ask the right people. Sample design should be such that only those people who are of interest to the research
user are contacted, and such that those who are contacted are reasonably representative of the group of
interest.
3. Ask questions the right way. It is not enough to be able to ask the right questions; they must be asked in the
right way. This is the essence of questionnaire design. The researcher can use all the aids available from the new
technologies, but if the wording of the questions is not clear to the respondents, the results will be useless.
Always pretest the questionnaire to ensure that responses are the ones that are needed to address the problem.
4. Obtain answers to questions. The process of data collection is central to all marketing research. Techniques used
should be selected for how each bears on nonresponse and response alike.
5. Relate answers to the needs of the research user/client. Data seldom speak for themselves. Proper data analysis
is needed if a study is to have any value to the user. Here there is a risk of letting advanced techniques become
the master of the researcher rather than the opposite. Common sense is a valuable tool for the researcher when
considering alternative analysis approaches for any project.
6. Communicate effectively. Many good projects are ruined in this stage. The information that is reported to the
user should be in a form that is understandable to the user so that he or she can tell that it is relevant to the
issues at hand.
15
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Ethical Considerations in Survey Research
There are a number of ethical considerations that arise both in conducting marketing
research projects and in marketing related activities in general. Many of these ethical issues are
the result of marketing activities that are conducted under the guise of surveys. Exhibit 1.7
summarizes the major practices that are considered unethical, these being deceptive and
fraudulent practices, invasion of privacy, and lack of consideration for research subjects and
respondents.
EXHIBIT 1.7 Ethical Considerations in Treatment of Subjects and Respondents
Schneider (1977) enumerated three general areas of ethical concern: deceptive practices, invasion of privacy, and lack of
consideration. An additional concern too frequent interviewing of the respondent.
Deceptive or fraudulent practices include the following:
Unrealized promise of anonymity
Use of disguised questionnaires and interviews
Faked sponsor identification
Implication of required response
Lying about research procedure
Faked testing in experimental research
Promise of undelivered compensation
Sales solicitation disguised as research
Invasions of privacy includes the following examples:
Observation without informed consent
Questions concerning people other than the subject
Projective techniques
Personal classification data
Full disclosure and use of “optional” participation
Lack of consideration for subjects or respondents is exhibited in all of the following practices:
Overuse of public (i.e., unreasonable demands on the time and energy of respondents)
Research in subject areas with a depressing effect on respondents
Subjects of no immediate interest to respondents
Poor interviewers
Contacts at inconvenient times
No mention of procedural aspects
Failure to debrief
Failure to present subject with option to discard results upon completion
Subjects’ rights are an important consideration in the ethical treatment of research
participants. Respondents have rights that should not be ignored or violated. Research should
not be deceptive or coerced. The researcher is often in a position of authority and as such should
assure that the participant does not feel forced to comply, has the ability to choose and make
informed choices, is safe from stress, psychological and physical harm, providing information
detrimental to their self-interest, and have the right to be informed of the purpose of the research.
16
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Furthermore, promises of anonymity must be kept. Exhibit 1.8 identifies the ethical issues
involved in subjects’ rights.
The American Marketing Association has provided a statement identifying principles of
ethical practice of marketing research (Exhibit 1.9). These broad guidelines provide standards
for the protection of the marketer and respondent alike and even extend to researchers and to
marketers who are not engaged in research activities.
Exhibit 1.8 Ethical Questions Regarding Subjects’ Rights
Subjects’ Rights
A. The right to choose
a. Awareness of right
b. Adequate information for an informed choice
c. Opportunity to make a choice
B. The right to be safe
a. Protection of anonymity
b. Subjects’ right to be free from stress
C. The right to be informed
a. Debriefing
b. Dissemination of data
c. Right to not be deceived
Possible Results of Violation of Rights
a. Feelings of forced compliance, biased data
b. May violate the client’s desire for anonymity, may
enable subjects to enact subject role
c. Subjects may avoid environments where this right
is violated
a. Biased data, refusal to participate in future
research
b. Biased data, refusal to participate in future
research
a. Unrelieved stress, feelings of being used, refusal to
participate in future research
b. Subjects may feel that they gain nothing from and
are exploited by participating in research and
consequently may distort their response and
decline to participate in future research
c. Biased data, refusal to participate in future
research
SOURCE: “Ethics in Marketing Research: Their Practical Relevance,” by Tybout, A.M. & Zaltman, G., in Journal of Marketing, 11, p.
359. November, 1974. Published by the American Marketing Association
17
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 1.9 Ethics in Marketing Research
The American Marketing Association on
Ethics in Marketing Research
The American Marketing Association has established principles of ethical practice of marketing research for the
guidance of its members. Marketing management must acknowledge its obligation to protect the public from
misrepresentation and exploitation under the guise of research. Similarly, the research practitioner has an
obligation to the discipline and to those who provide support for it—an obligation to adhere to basic and
commonly accepted standards of scientific investigation as they apply to the domain of marketing research.
FOR RESEARCH USERS, PRACTITIONERS, AND INTERVIEWERS
1. No individual or organization will undertake any activity which is directly or indirectly represented to be
marketing research, but which has as its real purpose the attempted sales of merchandise or services to
some or all of the respondents interviewed in the course of the research.
2. If respondents have been led to believe, directly or indirectly, that they are participating in a marketing
research survey and that their anonymity will be protected, their names shall not be made known to any
one outside the research organization or research department, or used for other than research purposes.
FOR RESEARCH PRACTITIONERS
1. There will be no intentional or deliberate misrepresentation of research methods or results. An adequate
description of methods employed will be made available upon request to the sponsor of the research.
Evidence that fieldwork has been completed according to specifications will, upon request, be made
available to buyers of the research.
2. The identity of the survey sponsor and/or the ultimate client for whom a survey is being done will be held in
confidence at all times, unless this identity is to be revealed as part of the research design. Research
information shall be held in confidence by the research organization or department and not used for
personal gain or made available to any outside party unless the client specifically authorizes such release.
3. A research organization shall not undertake marketing studies for competitive clients when such studies
would jeopardize the confidential nature of client-agency relationships.
FOR USERS OF MARKETING RESEARCH
1. A user of research shall not knowingly disseminate conclusions from a given research project or service that
are inconsistent with or not warranted by the data.
2. To the extent that there is involved in a research project a unique design involving techniques, approaches, or
concepts not commonly available to research practitioners, the prospective user of research shall not solicit
such a design from one practitioner and deliver it to another for execution without the approval of the
design originator.
FOR FIELD INTERVIEWERS
1. Research assignments and materials received, as well as information obtained form respondents, shall be held
in confidence by the interviewer and revealed to no one except the research organization conducting the
marketing study.
2. No information gained through a marketing research activity shall be used, directly or indirectly, for the
personal gain or advantage of the interviewer.
3. Interviews shall be conducted in strict accordance with specifications and instructions received.
4. An interviewer shall not carry out two or more interviewing assignments simultaneously, unless authorized by
all contractors or employers concerned. Members of the American Marketing Association will be expected
to conduct themselves in accordance with the provisions of this code in all of their marketing research
activities.
18
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
SUMMARY
This chapter has introduced the research process planning from the perspective of valuing
research on the basis of how well it has been done (the management of total error). Planning a
research project includes:
• Problem formulation
• Method of inquiry
• Research method
• Research design
• Selection of data collection techniques
• Sample design
• Data collection
• Analysis and interpretation of data
• Research reports
The differences in client requirements give rise to different requirements in research projects.
The value versus cost orientation is an outgrowth of these differences.
In this chapter we explained how managers use marketing research to help decision-making.
Marketing research provides the description and explanation required in order to predict and
evaluate market opportunities. There are several groups who “do” marketing research but the
basic reason for their research is the same—to solve a problem. Whether implicit or explicit, the
best method for decision making is the problem-solving model. Information should be accurate,
current, sufficient, available and, most important, relevant to be meaningful to organizations.
There can be many types of dialogue and challenges between manager and researcher.
The dialogue can encompass objectives, courses of action, and environmental variables affecting
decision outcomes. And there are ethical issues in marketing research. Although there are many
concerns among researchers and others, they can be summarized as deceptive/fraudulent
practices, invasion of privacy, and lack of consideration for subjects/respondents. Ethical
dilemmas arise because of the relationships that exist between a researcher and stakeholders in
the research process. Professional codes of conduct of marketing research were presented, which
are indicative of an industry that is trying to “clean up its own act.”
19
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
American Marketing Association:
http://www.marketingpower.com/AboutAMA/Pages/Statement%20of%20Ethics.aspx?sq=ethics+statement
Anderson, J. F., Berdie, D. R., & Liestman, R. (1984, January 16). Hi-tech techniques OK, but don’t forget research
basics. Marketing News, 18 (Sec. 2), 12.
CBS New Coke News Clip: http://www.youtube.com/watch?v=-doEpVWFLsE&NR=1&feature=fvwp
Diesing, P. (1966, March–June). Objectivism vs. subjectivism in the social sciences. Philosophy of Science, 33,
124–133.
Kerlinger, F. N., & Lee, H. (2000) Foundations of behavioral research (4th ed.). Belmont, CA: Wadsworth
Publishing.
Microsoft Project: http://office.microsoft.com/en-us/templates/TC012330951033.aspx
New Coke Commercial: http://www.youtube.com/watch?v=o4YvmN1hvNA
New Coke and Coke Classic Commercial: http://www.youtube.com/watch?v=ky45YGUA3co
Qualtrics.com (2009), 20 Different Types of Marketing Surveys:
http://www.qualtrics.com/wiki/index.php/Market_Surveys.
Schmalensee, D. (2001, November 19). One researcher’s rules of thumb for B-to-B arena. Marketing News, 35, 17–
19.
Schneider, K. C. (1977, Spring). Subjects and respondent abuse in marketing research. MSU Business Topics, pp.
13–19.
Segal, M. N., & Newberry, C. (1983,Winter). On the field service agency-supplier/client relationships: Problems
and perspectives. Journal of Data Collection, 23, 58–59.
Tybout, A.M. & Zaltman, G. (1974). Ethics in Marketing Research: Their Practical Relevance, in Journal of
Marketing, 11, p. 359. November, 1974. Published by the American Marketing Association
20
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This page is intentionally blank
21
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 2
DEFINING THE RESEARCH DESIGN AND
CONTROLLING RESEARCH ERRORS
A research design specifies the methods and procedures for acquiring the information
needed to structure and solve the research problem. The overall operational design for a research
project stipulates what information is to be collected, from what sources, and by what
procedures. A good research design ensures that the information obtained is relevant to the
research problem, and that it is collected by objective and economical procedures. A research
design might be described as a series of advance decisions that, taken together, form a master
plan or model for conducting a research investigation.
Research designs vary depending on the type of study. Generally designs are associated
with three types of studies, those that focus on providing exploratory research, descriptive
research and causal research. Each will be described in turn.
Exploratory Studies
The major purposes of exploratory studies are for the identification of problems, the
precise formulation of problems (including the identification of relevant variables), and the
formulation of new alternative courses of action.
An exploratory study is often the first in a series of projects. That is, an exploratory study
is often used as an introductory phase of a larger study, and its results are used to bring focus to
the larger study and to develop specific techniques that will be used. Thus flexibility is a key to
designing and conducting exploratory studies.
We can distinguish three separate tasks that are usually included in exploratory studies
and that are typically conducted in the sequence listed:
• A search of secondary information sources
• Interviews with persons knowledgeable about the subject area
• The examination of analogous situations
Search Secondary Sources
Secondary sources of information are the “literature” on the subject. It is the rare research
problem for which there is no relevant information to be found by a relatively quick and
inexpensive search of the literature. Secondary information sources are not limited to external
sources. Searches should also be made of company records.
Interview Knowledgeable Persons
Having searched secondary sources, it is usually desirable to talk with persons who are
well informed in the area being investigated, including company executives, experts, consumers
and mavens, and users outside the organization
A widely used technique in exploratory research is the focus group. In focus group
interviews, a group of knowledgeable people participate in a joint interview that does not use a
structured question-and-answer methodology. The group, usually consisting of 8 to 12 people
(but may have as few as 5 or as many as 20), is purposely selected to include individuals who
22
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
have a common background, or similar buying or use experience, as related to the problem being
researched. The interviewer or moderator of the focus group session works with the client to
develop a general discussion outline that typically includes such topics as usage experience,
problems with use, and how decisions are made. The objective is to foster involvement and
interaction among the group members during the interview that will lead to spontaneous
discussion and the disclosure of attitudes, opinions, and information on present or prospective
buying and use behavior.
Focus groups are used primarily to identify and define problems, provide background
information, and generate hypotheses. Focus groups typically do not provide solutions for
problems. Areas of application include detecting trends in lifestyles, examining new product or
service concepts, generating ideas for improving established products or services, developing
creative concepts for advertising, and determining effective means of marketing products or
services.
If the sole purpose is to create ideas, then individual interviews may be a better
alternative than focus groups. Limited research on this issue conducted more than 20 years ago
suggests that the number and quality of ideas generated may be greater from such interviews
(Fern, 1982).
More specific uses of focus groups include:
1. Identifying and understanding consumer language relating to the product category in question.
What terms do they use? What do they mean?
2. Identifying the range of consumer concerns. How much variability is there among consumers’
perception of the product, and in the considerations leading them to accept or reject the product?
3. Identifying the complexity of consumer concerns. Do a few simple attitudes govern consumer
reaction toward the product, or is the structure complex, involving many contingencies?
4. Identifying specific methodological or logistical problems that are likely to affect either the
cost of the subsequent research, or one’s ability to generate meaningful, actionable findings.
An example of focus group usage might be to determine the reasons for the decline in a
product’s overall rating, as reported in a syndicated research report.
Examine Analogous Situations
It is also logical that a researcher will want to examine analogous situations to determine
what else can be learned about the nature of the problem and its variables. Analogous situations
include case histories and simulations. More discussion of the use of focus groups is given in
Chapter 4.
Descriptive Studies
Much research is concerned with describing market characteristics or functions. For
example, a market potential study may describe the number, distribution, and socioeconomic
characteristics of potential customers of a product. A market-share study finds the share of the
market received by both the company and its major competitors. A sales analysis study describes
sales by territory, type of account, size or model of product, and the like. In marketing,
descriptive studies are also made in the following areas:
23
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Product research: identification and comparison of functional features and specifications of
competitive products
• Promotion research: description of the demographic characteristics of the audience being reached
by the current advertising program
• Distribution research: determining the number and location of retailers handling the company’s
products that are supplied by wholesalers versus those supplied by the company’s distribution
centers
• Pricing research: identifying competitors’ prices by geographic area
These examples of descriptive research cover only a few of the possibilities. Descriptive
designs, often called observational designs by some researchers, provide information on groups
and phenomena that already exist; no new groups are created (Fink, 2003).
One example of a descriptive study is one conducted by a school-employees credit union
in order to gain information useful to provide better service to its members. Management knew
very little about the members, other than that they were school employees, family members of
employees, or former employees. In addition, the credit union knew very little about member
awareness and use of, and attitudes toward individual services available to them. Consequently,
investigators undertook a study to answer the following research questions:
1. What are the demographic and socioeconomic characteristics of primary members?
2. How extensively are existing services being used, and what are members’ attitudes toward such
services?
3. What is the degree of interest in specific new services?
Although associations can be used only to make inferences, and not establish a causal
relationship, they are often useful for predictive purposes. It is not always necessary to
understand causal relations in order to make accurate predictive statements. Descriptive
information often provides a sound basis for the solution of marketing problems, even though it
does not explain the nature of the relationship involved. The basic principle involved is to find
desirable behavior correlates that are measurable when the predictive statement is made.
Causal Studies
Although descriptive information is often useful for predictive purposes, where possible
we would like to know the causes of what we are predicting—the “reasons why.” Further, we
would like to know the relationships of these causal factors to the effects that we are predicting.
If we understand the causes of the effects we want to predict, we invariably improve our ability
both to predict and to control these effects.
Bases for Inferring Causal Relationships
There are three types of evidence that can be used for drawing inferences about causal
relationships:
1. Associative variation
2. Sequence of events
3. Absence of other possible causal factors
24
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In addition, the cause and effect have to be related. That is, there must be logical
implication (or theoretical justification) to imply the specific causal relation.
Associative Variation
Associative variation, or “concomitant variation,” as it is often termed, is a measure of
the extent to which occurrences of two variables are associated. Two types of associative
variation may be distinguished:
1. Association by presence: A measure of the extent to which the presence of one variable
is associated with the presence of the other
2. Association by change: A measure of the extent to which a change in the level of one
variable is associated with a change in the level of the other.
It has been argued that two other conditions may also exist, particularly for continuous variables:
(a) the presence of one variable is associated with a change in the level of other; and (b) a change
in the level of one variable is associated with the presence of the other (Feldman, 1975).
Sequence of Events
A second characteristic of a causal relationship is the requirement that the causal factor
occur first; the cause must precede the result. In order for salesperson retraining to result in
increased sales, the retraining must have taken place prior to the sales increase.
Absence of Other Possible Causal Factors
A final basis for inferring causation is the absence of any other possible causes other than
the one(s) being investigated. If it could be demonstrated, for example, that no other factors
present could have caused the sales increase in the third quarter, we could then logically
conclude that the salesperson training must have been responsible.
Obviously, in an after-the-fact examination of an uncontrolled result such as an increase
in detergent sales, it is impossible to clearly rule out all causal factors other than salesperson
retraining. One could never be completely sure that there were no competitor-, customer-, or
company-initiated causal factors that would account for the sales increase.
Conclusions Concerning Types of Evidence
No single type of evidence, or even the combination of all three types considered, can
ever conclusively demonstrate that a causal relationship exists. Other unknown factors may
exist. However, we can obtain evidence that makes it highly reasonable to conclude that a
particular relationship exists. Exhibit 2.1 shows certain questions that are necessary to answer.
25
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 2.1 Issues in Determining Causation
Several questions arise when determining whether a variable X has causal power over another variable, Y:
1. What is the source of causality—does X cause Y, or does Y cause X?
2. What is the direction of causality—does X positively influence Y, or is the relationship negative?
3. Is X a necessary and sufficient cause—or necessary, but not sufficient cause—of Y? Is X’s causation
deterministic or probabilistic?
4. Which value of the believed cause exerts a causal influence—its presence or absence?
5. Are the causes and effects the states themselves or changes in the states? Is the relationship static or
dynamic?
In the end, the necessary conditions for causality to exist are a physical basis for causality, a cause that temporally
precedes the effect (even for associative variation), and a logical reason to imply the specific causal relation being
examined. (Monroe and Petroshius, n.d.).
SOURCES OF MARKETING INFORMATION
There are five major sources for obtaining marketing information. In this section we
briefly describe each as an introduction to subsequent chapters that describe some of these
sources in more depth.
• Secondary sources
• Respondents
• Natural experiments
• Controlled experiments
• Simulation
Secondary Sources of Information
Secondary information is information that has been collected by persons or agencies for
purposes other than the solution of the problem at hand.
If a furniture manufacturer, for example, needs information on the potential market for furniture
in the Middle Atlantic States, many government and industry sources of secondary information
are available.
The federal government collects and publishes information on the numbers of families, family
formation, income, and the number and sales volume of retail stores, all by geographic area. It
also publishes special reports on the furniture industry. Many state and local governments collect
similar information for their respective areas.
The trade associations in the furniture field collect and publish an extensive amount of
information about the industry. Trade journals are also a valuable source of secondary
information, as are special studies done by other advertising media.
Private research firms collect specialized marketing information on a continuing basis
and sell it to companies. These so-called syndicated services, particularly those for packaged
consumer goods, are becoming more sophisticated as they are increasingly becoming based on
26
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
scanner data. Technology advancements are having a measurable impact on the availability of
secondary data.
Information from Respondents
A second major source of information is obtained from respondents. Asking questions
and observing behavior are primary means of obtaining information whenever people’s actions
are being investigated or predicted.
The term respondent literally means “one who responds or answers.”
Both verbal and behavioral responses should be considered.
In this book we shall consider both the information obtained from asking people
questions, and that provided by observing behavior (or the results of past behavior) to comprise
information from respondents.
Information from Natural and Controlled Experiments
As described earlier, three types of evidence provide the bases for drawing inferences
about causal relationships. While both natural and controlled experimental designs are capable of
providing associative variation and sequence of events, only controlled experiments can provide
reasonably conclusive evidence concerning the third type of evidence, the absence of other
possible producers.
A natural experiment is one in which the investigator intervenes only to the extent
required for measurement. That is, there is no manipulation of an assumed causal variable. The
investigator merely looks at what has happened. As such, the natural experiment is a form of ex
post facto research. In this type of study, the researcher approaches data collection as if a
controlled experimental design were used. The variable of interest has occurred in a natural
setting, and the researcher looks for respondents who have been exposed to it and also, if a
control group is desired, respondents who have not been exposed.
Measurements can then be made on a dependent variable of interest. For example, if the
impact of a television commercial on attitudes were desired, the investigator would conduct a
survey of people after the commercial was shown. Those who saw the commercial would
constitute the experimental group, and those who did not see it would be a type of control group.
Differences in attitudes could be compared as a crude measure of impact. Unfortunately, one can
never be sure whether the obtained relationship is causal or non-causal, since the attitudes may
be affected by the presence of other variables. For a brief discussion of natural experiments, see
Anderson (1971).
In controlled experiments, investigator intervention is required beyond that needed for
measurement purposes. Specifically, two kinds of intervention are required:
1. Manipulation of at least one assumed causal variable
2. Random assignment of subjects to experimental and control groups
The researcher conducts the experiment by assigning the subjects to an experimental
group where the causal variable is manipulated, or to a control group where no manipulation of
the causal variable occurs. The researcher measures the dependent variable in both situations
and then tests for differences between the groups. Differences between the groups, if present, are
attributed to the manipulation variable.
27
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Field experiments are increasingly being completed using online survey instruments. For
example, researchers often use the advanced branching logic, randomization, question block
presentation, question timing, and java scripting capabilities of Qualtrics.com to conduct time
and cost effective field experiments.
Simulation
The expense and time involved in the personal interviews often associated with field
experimentation may preclude it as a source of information for a particular operational situation.
In such cases it may be desirable to construct a model of the operational situation and to
experiment with it instead of venturing into a real-world situation. The manipulation of such
models is called simulation.
Simulation can be defined as a set of techniques for manipulating a model of some realworld
process to find numerical solutions that represent the real process being modeled. Models
that are environmentally rich (that is, that contain many complex interactions and nonlinear
relationships among the variables, probabilistic components, time dependencies, etc.) are usually
too difficult to solve by standard analytical methods such as calculus or other mathematical
programming techniques. Rather, the analyst views a simulation model as a limited imitation of
the process or system under study and attempts to run the system on a computer to see what
would happen if a particular policy were put into effect.
Simulations may be used for research, instruction, decision-making, or some combination
of these applications. During the past 50 or more years, simulations have been developed for
such marketing decision- making applications as marketing systems, marketing-mix elements
(new-product, price advertising, and sales-force decisions), and interviewing costs in marketing
surveys.
TYPES OF ERRORS AFFECTING RESEARCH DESIGNS
The marketing research process (and research design) involves the management of error.
Potential errors can arise at any point from problem formulation through report preparation, and
rarely will a research project be error-free. Consequently, the research designer must adopt a
strategy for managing and minimizing this error. As we shall see in the next section of this
chapter, there are alternative strategies that can be followed.
The objective underlying any research project is to provide information that is as accurate
as possible. Maximizing accuracy requires that total study errors be minimized. Total study error
has two components—sampling error and non-sampling error—and can be expressed as follows:
Total error = Sampling error + Non-sampling error
Total error is usually measured as total error variance, also known as the mean-squared error
(Assael & Keon, 1982):
(Total error)2
= (Sampling error)2
+ (Non-sampling error)2
Sampling error refers to the variable error resulting from the chance specification of
population from elements according to the sampling plan. Since this introduces random
variability into the precision with which a sample statistic is calculated, it is often called random
sampling error. Exhibit 2.2 gives an illustration of how total error is assessed.
28
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 2.2 How Errors Add Up
It is important to know all the sources of error that contribute to inaccuracy, and to assess the
impact of each. As an example, consider the figure below, which shows components of error in a
study designed to estimate the size of the personal computer market (Lilien, Brown, & Searls, 1991).
When estimating the market, adjustments are made for each source of error. The components are
then combined mathematically to create the total error. For purposes of simplicity, total error is
shown here as the sum of the component errors. In actuality, total error would be smaller, as it is
usually based on the square roots of summed squares of component errors. Assessing the individual
components of total error is highly judgmental and subjective, but it is worth the effort.
Non-sampling error consists of all other errors associated with a research project. Such
errors are diverse in nature and are often thought of as resulting in some sort of bias, which
implies systematic error. Bias can be defined simply as the difference between the true value of
that which is being measured and the researchers’ estimate of the true value. However, there can
be a random component of non-sampling error. For example, misrecording a response during
data collection would represent a random error, whereas using a loaded question would be a
systematic error. Non-sampling errors have both nonresponse and response based origins.
29
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
To a large extent these major error components are inversely related. Increasing the
sample size to reduce sampling error can increase non-sampling error in that, for example, there
are more instances where such things as recording errors can occur, and the impact of biased
(i.e., nonobjective) questions and other systematic errors will be greater. Thus, this inverse
relationship lies at the heart of our concern for total error.
Ideally, efforts should be made to minimize each component. Considering time and cost
limitations this can rarely be done. The researcher must make a decision that involves a tradeoff
between sampling and non-sampling errors. Unfortunately, very little is known empirically about
the relative size of the two error components, although there is some evidence that non-sampling
error tends to be the larger of the two. In a study comparing several research designs and data
collection methods, Assael and Keon (1982) concluded that non-sampling error far outweighs
random sampling error in contributing to total survey error.
As an introduction, Exhibit 2.3 briefly defines eight major types of errors that can influence
research results.
EXHIBIT 2.3 Types of Errors in the Research Process
Different types of errors can influence research results:
• Population specification: noncorrespondence of the required population to the population selected by
the researcher
• Sampling: noncorrespondence of the sample selected by probability means and the representative
sample sought by the researcher
• Selection: noncorrespondence of the sample selected by nonprobability means and the sought
representative sample
• Frame: noncorrespondence of the sought sample to the required sample
• Nonresponse: noncorrespondence of the achieved (or obtained) sample to the selected sample
• Surrogate information: noncorrespondence of the information being sought by the researcher and that
required to solve the problem
• Measurement: noncorrespondence of the information obtained by the measurement process and the
information sought by the researcher
• Experimental: noncorrespondence of the true (or actual) impact of, and the impact attributed to, the
independent variable(s)
Population Specification Error
This type of error occurs when the researcher selects an inappropriate population or
universe from which to obtain data.
Examples: Cessna Aircraft conducts an online survey to learn what features should be added to a
proposed corporate jet. They consider conducting a survey of purchasing agents from major
corporations presently owning such aircraft. However, they believe that that this would be an
inappropriate research universe; since pilots are most likely play a key role in the purchase decision.
Similarly, packaged goods manufacturers often conduct surveys of housewives, because they are
easier to contact, and it is assumed they decide what is to be purchased and also do the actual
purchasing. In this situation there often is population specification error. The husband may purchase
a significant share of the packaged goods, and have significant direct and indirect influence over what
is bought.
30
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Sampling Error
Sampling error occurs when a probability sampling method is used to select a sample, but
the resulting sample is not representative of the population concern.
Example: Suppose that we collected a random sample of 500 people from the general adult
population and upon analysis found it to be composed only of people aged 35 to 55. This sample
would not be representative of the general adult population. Sampling error is affected by the
homogeneity of the population being studied and sampled from and by the size of the sample.
In general, the more homogeneous the population (meaning smaller variance on any
given characteristic of interest), the smaller the sampling error; as sample size increases,
sampling error decreases. If a census were conducted (i.e., all elements of the population were
included) there would be no sampling error.
Selection Error
Selection error is the sampling error for a sample selected by a nonprobability method.
Example: Interviewers conducting a mall intercept study have a natural tendency to select those
respondents who are the most accessible and agreeable whenever there is latitude to do so. Such
samples often comprise friends and associates who bear some degree of resemblance in
characteristics to those of the desired population.
Selection error often reflects people, who are most easily reached, better dressed, and
have better kept homes or more pleasant personalities. Samples of these types rarely are
representative of the desired population.
Frame Error
A sampling frame is the source for sampling that represents all the members of the
population. It is usually a listing of the prospective respondents to be sampled.
Example: Consider the sample frame for a shopper intercept study at a shopping mall. This sample
frame includes all shoppers in the mall during the period of data collection. A commonly used frame
for consumer research is the telephone directory. This frame introduces error because many elements
of the population are not included in the directory (unlisted phone numbers, new arrivals), some
elements are listed more than once, and nonpopulation elements are also included (businesses, people
who have left the area).
A perfect frame identifies each member of the population once, but only once, and does
not include members not in the population of interest.
Nonresponse Error
Nonresponse error can exist when an obtained sample differs from the original selected
sample. There are two ways in which nonresponse can occur: (a) noncontact (the inability to
contact all members of the sample); and (b) refusal (nonresponse to some or all items on the
measurement instrument). Errors arise in virtually every survey from the inability to reach
respondents.
31
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Example: In telephone surveys, some respondents are inaccessible because they are not at home
(NAH) for the initial call or call-backs. Others have moved or are away from home for the period of
the survey. Not-at-home respondents are typically younger with no small children, and have a much
higher proportion of working wives than households with someone at home. People who have moved
or are away for the survey period have a higher geographic mobility than the average of the
population. Thus, most surveys can anticipate errors from non-contact of respondents.
Refusals may be by item or for the entire interview. Income, religion, sex, and politics are
topics that may elicit item refusals. Some respondents refuse to participate at all because of time
requirements, health issues, past experiences in which an “interviewer” turned out to be a
telemarketer, or other reasons. Refusals can also be specific to the method of data collection, as
in nonresponse to a mail and email questionnaires or using caller ID to screen and avoid
telephone surveys. Nonresponse to mail and email questionnaires sometimes runs as high as 90
percent of the initial mailing, even after several successive mailings.
The amount of effort involved in data collection is another possible way to affect
nonresponse error. However, little research has been done to examine the impact of effort.
Example: In a national telephone survey, a so-called five-day “standard” survey was compared to a
“rigorous” survey conducted over an eight-week period (Keeter, Miller, Kohut, Groves, & Presser,
2000). Response rates were significantly different; the rigorous survey generated about two-thirds
greater response. But the two surveys produced similar results. Most of the statistically significant
differences were for demographic items. Very few differences were found on substantive variables.
Nonresponse is also a potential problem in business-to-business and within organization
research situations. Although specific respondents are individuals, organizations are not, as they
are differentiated and hierarchical. These characteristics may affect organizational response to
survey requests.
Tomaskovic-Devey, Leiter, and Thompson (1994) in a study of organizational response,
stated the likelihood that an organizational respondent will respond is a function of three
characteristics of the respondent:
1. Authority to respond: The degree to which a designated respondent has the formal or informal
authority to respond to a survey request
2. Capacity to respond: Organizational practices and the division of labor and information affect the
assembly of relevant knowledge to reply adequately
3. Motive to respond: Both individual and organizational motivations to provide information (or not
provide information) about the organization.
Surrogate Information Error
In many research situations, it is necessary to obtain information that acts as a surrogate
for that which is required. The necessity to accept substitute information arises from either the
inability or unwillingness of respondents to provide the information requested.
Decision-oriented behavioral research is always concerned with the prediction of
behavior. This limits most marketing research projects to using proxy information, since one
32
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
cannot observe future behavior. Typically, researchers obtain one or more kinds of surrogate
information believed to be useful in predicting behavior.
Examples: One may obtain information on past behavior because it is believed that there is sufficient
stability in the underlying behavior pattern to give it reasonably high predictive validity. One may ask
about intended behavior as a means of prediction. Or one may obtain information about attitudes,
level of knowledge, or socioeconomic characteristics of the respondent in the belief that, individually
or collectively, they have a high degree of association with future behavior.
Since the type of information required is identified during the problem-formulation stage
of the research process, minimizing this error requires an accurate problem definition.
Measurement Error
Measurement error is generated by the measurement process itself, and represents the
difference between the information generated and the information wanted by the researcher.
Such error can potentially arise at any stage of the measurement process, from the development
of an instrument through the analysis of the findings. To illustrate, Figure 2.1 depicts the stages
at which errors in eliciting information may arise when interviewing respondents for a survey.
FIGURE 2.1 Potential Sources of Measurement Error in a Survey
In the transmittal stage, errors may be due to the faulty wording of questions or
preparation of nonverbal materials, unintentional interviewer modification of the question’s
wording, or the way in which a respondent interprets the question. In the response phase, errors
may occur because the respondent gives incorrect information, the interviewer interprets it
incorrectly, or recording errors occur. One aspect of this regards form; form-related errors
concern psychological orientation toward responding differently to different item formats and
include:
1. Leniency: the tendency to rate something too high or too low
2. Central tendency: reluctance to give extreme scores
3. Proximity: giving similar responses to items that occur close to one another
(Yu, Albaum, & Swenson, 2003, p. 217)
33
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In the analysis stage, errors of incorrect editing and coding, descriptive summarization,
and inference can contribute substantially to measurement error. Measurement error is
particularly troublesome for the researcher, since it can arise from many different sources and
take on many different forms.
Experimental Error
When an experiment is conducted, the researcher attempts to measure the impact of one
or more manipulated independent variables on some dependent variable of interest, while
controlling for the influence of all other (i.e., extraneous) variables. Unfortunately, control over
all possible extraneous variables is rarely possible. Consequently, what may be measured is not
the effect of the independent variables but the effect of the experimental situation itself.
METHODS FOR DEALING WITH POTENTIAL ERRORS
For any research design, recognizing that potential errors exist is one thing, but doing
something about them is another matter. There are two basic approaches for handling potential
errors:
1. Minimize errors through precision in the research design
2. Measure or estimate the error or its impact
Minimize Error
Two different approaches can be taken to minimize total error. The first uses the research
design to minimize errors that may result from each of the individual error components. Much of
the material in Chapters 3 through 9 of this book discusses effective research methods, and as
such, involves techniques designed to minimize individual errors. This is consistent with our
view that research design innately involves error management. However, this approach is often
limited by the budget allotted to a project.
The second approach recognizes that individual error components are not necessarily
independent of each other. Thus, attempts to minimize one component may lead to an increase in
another. Reducing sampling error by increasing sample size, for example, leads to potentially
greater non-sampling error. This means that the research designer must trade off errors when
developing a research design that minimizes total error. For a fixed project budget, therefore, it
may be prudent for the research designer to choose a smaller sample size (which will increase
sampling error) if the cost savings by doing this can develop techniques that will reduce
nonresponse and/or improve the measurement process. If the reduction in these nonsampling
errors exceeds the increase in sampling error, there will be a reduction in total error.
Estimate or Measure Error
Estimating or measuring individual components and total error is not easy, primarily due
to the nature of non-sampling errors. There is a body of accepted sampling theory that allows the
researcher to estimate sampling error for a probability sample, but nothing comparable exists for
non-sampling errors. Consequently, subjective or judgmental estimates must be made.
For individual error components, many diverse procedures can be used to estimate and
measure their impact as illustrated in Table 2.1. These are discussed where appropriate in
subsequent chapters.
34
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 2.1 Selected Methods for Handling Non-Sampling Errors
Type of Error Design to Avoid Measure Estimate
Surrogate Information Strive for realism No method of direct
measurement, as event has
not yet occurred
Use track record of studies,
Use surrogate variables
Measurement
1. Instrument
induced
Pretest, alternative wording,
alternative positions, etc.
Experiment by using
alternative positioning, etc.
in a subsample
Estimate will likely be for no
bias but some variable error
2. Interviewing-
associated
(e.g., bias,
recording, cheating
in telephone and
personal
interviews)
Select and train interviewers
correctly
Use same editor of all
interviews by one
interviewer
Use cheater questions
Use computer program to
analyze for patterns of
responses by interviewer
Re-interview subsample
using expert interviewer
Analysis of variance
Use cheater questions
Use computer program to
analyze for patterns
Use interpenetrating sample
Estimate will be for both bias
and variable error
3. Response Randomize response
technique
Ask for verification checks
Cross-check questions
Use mail-back technique
Compare with known data Have interviewer evaluate
respondent
Estimate will be for both bias
and variable error
4. Editing Prepare editing manual
Train editors
Require daily return of data
Use master editor to edit
subsample
Estimate will be for limited
bias, some variable error
5. Coding
(text and manually
entered data)
Pre-code variables
Use coding manual
User computer program to
clean data
Use master coder to validate
subsample
Some bias and variable error
6. Tabulation Use verification for data
entry
Recheck sample of forms Variable error
7. Analysis No remedy except
competence
Use more competent analyst
Frame Use multiple frames Take subsample of excluded
segments
Use compensating weights
Use past data
Selection Make sample element and
sample unit the same
Use probability sample
Compare with known
population
Use compensating weights
Nonresponse Use callbacks
Call at appropriate time
Use trained interviewers
Take subsample of
nonrespondents
Use Politz-Simmons method
Use wave analysis
As a final note, even though the researcher has designed a project to minimize error, it is
almost never completely eliminated. Consequently, the error that exists for every project must be
estimated or measured. This is recognized for sampling error when probability samples are used,
though non-sampling errors typically are ignored. Although estimating or measuring errors is
better than ignoring them, there may be times when ignoring non-sampling error may not be that
bad. For example, if non-sampling error is viewed as a multiple of sampling error, ignoring nonsampling
errors up to an amount equal to one-half of sampling error reduces a .95 confidence
level only to .92 (Tull & Albaum, 1973). However, ignoring a non-sampling error equal in
amount to sampling error reduces the .95 level to .83.
35
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
CHOOSING A RESEARCH DESIGN
The overview of research designs and sources of error just presented should make it
apparent that, given a specified problem, many competing designs can provide relevant
information. Each design will have an associated expected value of information and incurred
cost.
Suppose, for example, that a researcher is assigned to determine the market share of the
ten leading brands of energy drinks. There are many possible ways of measuring market share of
energy drink brands, including questioning a sample of respondents, observing purchases at a
sample of retail outlets, obtaining sales figures from a sample of wholesalers, obtaining sales
figures from a sample of retailers and vending machine operators, obtaining tax data, subscribing
to a national consumer panel, subscribing to a national panel of retail stores, and, possibly,
obtaining data directly from trade association reports or a recent study by some other
investigative agency. Though lengthy, this listing is not exhaustive.
The selection of the best design from the alternatives is no different in principle from
choosing among the alternatives in making any decision. The associated expected value and cost
of information must be determined for each competing design option. If the design is such that
the project will yield information for solving more than one problem, the expected value should
be determined for all applicable problems and summed. The design with the highest, positive, net
expected payoff of research should be selected.
SUMMARY
In this chapter we dealt with a subject of single most importance to the research project:
the research design. We described what a research design is, discussed the classes of designs, and
examined major sources of marketing information that various designs employ. Finally, we
considered the errors that affect research designs.
Presenting these topics as an introduction and overview, we deal with the topics in more
depth in the next several chapters. These chapters deal with major sources of marketing
information—respondents and experimentation—and the means of obtaining and analyzing
research information.
36
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
Anderson, B. F. (1971). The psychology experiment: An introduction to the scientific method
(2nd ed.). Belmont, CA: Brooks/Cole.
Assael, H., & Keon, J. (1982, Spring). Non-sampling vs. sampling errors in survey research.
Journal of Marketing, 46, 114–123.
Feldman, J. (1975). Considerations in the use of causal-correlational technique in applied
psychology. Journal of Applied Psychology, 60, 663–670.
Fern, E. F. (1982). The use of focus groups for idea generation: The effects of group size,
acquaintanceship, and moderator on response quantity and quality. Journal of Marketing
Research, 19 (February), 1–13.
Fink, A. (2003). How to design surveys (2nd ed.). Thousand Oaks, CA: Sage.
Keeter, S., Miller, C., Kohut, A., Groves, R. M., & Presser, S. (2000). Consequences of reducing
non-response in a national telephone survey. Public Opinion Quarterly, 64, 125–148.
Lilien, G., Brown, R., & Searls, K. (1991, January 7). How errors add up. Marketing News, 25,
20–21.
Monroe, K. B., & Petroshius, S. M. (n.d.). Developing causal priorities. Unpublished working
paper, College of Business, Virginia Polytechnic Institute and State University.
Tomaskovic-Devey, D., Leiter, J., & Thompson, S. (1994). Organizational survey response.
Administrative Science Quarterly, 39, 439–457.
Tull, D. S., & Albaum, G. S. (1973). Survey research: A decisional approach. New York: Intext
Educational Publishers.
Yu, J., Albaum, G., & Swenson, M. (2003). Is a central tendency error inherent in the use of
semantic differential scales in different cultures? International Journal of Market Research, 45
(2), 213–228.
37
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 3
SECONDARY SOURCES OF INFORMATION
“A few months of field work collecting data can save several hours in the library”
Secondary data (or secondary information) is information that has been collected by
persons or agencies for purposes other than the solution of the marketing research problem at
hand. These data may have been collected from sources within the researcher’s firm or from
sources outside the firm. The key point is that the data were collected for some other project, or
reason, than the current one.
In contrast, primary data is data collected for the researcher’s current research project.
Primary data is often collected from a respondent, an individual who provides information either
passively through the observation of his or her behavior, or actively through verbal response.
Researchers using primary data must be concerned with information obtained by asking
questions, by observing behavior or by examining the results of past behavior.
In addition to primary and secondary data, there exists commercial data sold in the form
of syndicated services. These data are collected by commercial marketing research firms or
industry associations and, as such, have characteristics of both primary and secondary data. Since
these data relate to ongoing concerns of a marketer they can be viewed as primary data.
However, the commercial agency did not design its service solely to provide information for one
company’s specific project. Thus, there are elements of secondary data. It should be clear that
distinctions between primary and secondary commercial data may be minimal.
In this chapter, we discuss the reasons for obtaining secondary information, types of
secondary information, sources of external secondary data, and syndicated services that provide
commercial data. Data, in all their forms, are the heart of research. Secondary research can help
provide a clearer picture of a problem so that researchers and managers can make the necessary
critical decisions.
REASONS FOR OBTAINING SECONDARY INFORMATION
As a general rule, no research project should be conducted without a search of secondary
information sources. This search should be conducted early in the problem investigation stage
and prior to any organized collection of information from primary sources. There are several
reasons for this.
Secondary Information May Solve the Problem
If adequate data are available from secondary sources, primary data collection will not be
required. For example, Campbell Soup Co. based a long-running advertising campaign on the
theme “soup is good food.” This theme emerged from federal government data pertaining to
eating habits, nutritional health, and related topics collected over a period of 15 years. Campbells
(1959: youtube.com/watch?v=qMJtLP8jMWQ , 1985 youtube.com/watch?v=4PtZYlcyXfI )
Secondary Information Search Costs Substantially Less
A comprehensive search of secondary sources can almost always be made in a fraction of
the time and cost required for the collection of primary information. This is particularly true
today with online access to research publications and databases. Searching for secondary
38
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
research helps you avoid duplicating primary research and optimizes research expenditures by
acquiring only information that cannot be found elsewhere. Many marketing problems do not
warrant expenditures for primary information collection, but are worth the time and cost of
secondary information.
Secondary Information Has Important Supplementary Uses
When secondary information cannot solve the research problem, it can provide invaluable
supplemental uses:
1. Defining the problem and formulating hypotheses about its solution. The analysis of available
secondary data will almost always provide a better understanding of the problem and its context,
and will frequently suggest solutions not considered previously.
2. Planning the collection of primary data. An examination of the methods and techniques
employed by other investigators in similar studies may be useful in planning the present one. It
may also be of value in establishing classifications that are compatible with past studies so that
trends may be more readily analyzed.
3. Defining the population and selecting the sample. Past information and samples may help
establish classifications for current primary information collection.
The researcher must be careful when using only secondary data. To be useful, secondary
data must be available, relevant to the information needs (which includes being timely), accurate,
and sufficient to meet data requirements for the problem at hand. Often, little is known about the
reliability of secondary research studies. It is important that the researcher know how the
secondary data being considered for use were collected, if the data is reliable, and if the right
techniques were used.
Example: One company wants to do a segmentation analysis on foreign markets with a particular
emphasis on examining demographics (Albaum, Duerr, & Strandskov, 2005, Chap. 5). The
company is considering using the official government census of the population. However, they
become aware that the data are not available from all markets in equal quantity, aggregation, and
detail, and the reliability of data is not the same.
What one gets from a census depends on what was on the census form in the first place—
typically a mixture of traditional questions and new items of interest to public policy makers and
civil servants at the time. Some countries publish information about noncitizens, and others
collect data on religion—both of these topics are ignored in U.S. Censuses. Income is one of the
major dimensions of U.S. segmentation research, but many highly developed nations ignore the
income question in their censuses.
In short, one cannot always expect to find the same range of data topics that you are
interested in. Moreover, the data may not use the same categories when showing relevant
distributions of demographic variables, such as age. In short, comparability and equivalence
issues arise, and these can hinder your effectiveness in using secondary data.
39
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
TYPES OF SECONDARY INFORMATION
Secondary information falls into two categories, data that is available within the company
(internal data) and that which must be obtained from outside sources (external data).
Internal Secondary Information
All companies collect information in the everyday course of conducting business. Orders
are received and filled, costs are recorded, warranty cards are returned, salespeople’s reports are
submitted, engineering reports are made—all are collected for other purposes, but may be useful
to the researcher (Andreasen, 1988, pp. 77–89). The key, of course is knowing where they are
and how to access them. In order to do this efficiently, the firm must have an effective marketing
information system.
Example: Spectra Physics Lasers Division (producing laser grocery store scanners) regularly
performs customer satisfaction studies. Whereas these studies are primary research to the Retail
Systems organization, they are internal secondary information to other divisions that may want
to look at them. Also, they can be secondary data to Retail Systems should they be used at a
much later date for aiding in decision-making or for purposes other than those originally
intended when the studies were done.
External Secondary Information
External secondary information is available in staggering assortments and volumes. It
also is applicable to all of the major types of marketing research projects and is mainly
concerned with the noncontrollable aspects of the problem:
• Total market size
• Market characteristics
• Competitor products, prices, promotional efforts, and distribution methods
As an example, a consumer goods company is considering whether it should establish a
direct selling operation. Direct selling is defined as personal contact between a salesperson and a
consumer away from a fixed business location such as a retail store. The Direct Selling
Association (DSA) provides secondary information in the form of a regular survey of the
industry. Some types of information on industry statistics and sales force demographics are
available on a regular basis. Some of the types of industry statistics are the following:
• Estimated U.S. sales
• Estimated U.S. salespeople
• Percent of sales by major product groups
• Location of sales
• Percent of sales by census region
• Sales strategy
• Compensation structure by percent of firms
• Compensation structure by percent of sales dollars
• Compensation structure by percent of salespeople
40
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The following are some of the sales force demographics available:
• Gender
• Age
• Education
• Independent contractor/employee status
• Hours per week dedicated to direct selling
• Average time spent on direct selling tasks
• Main reasons for becoming a direct sales representative
• Percent of salespeople by distributorship type
Age, education, average time spent on direct selling tasks, and main reason for becoming
a direct sales representative data are from DSA’s National Sales force Survey. (Direct Selling
Association, 2003; http://www.dsa.org/pubs/).
SOURCES OF EXTERNAL SECONDARY DATA
The major original sources of external secondary information are:
1. Government (supranational, federal, state, and local)
2. Trade associations and trade press
3. Periodicals and professional journals
4. Institutions (e.g., universities)
5. Commercial services
Government Data Sources
The federal government is by far the largest single source of this type of data. Both
governmental and trade sources are so important that the experienced researcher will be
thoroughly familiar with them in his or her field of specialization. Periodicals and research
publications of universities and research institutes frequently provide valuable information.
Commercial services of many types are available that are highly useful for specific research
problems.
Market performance studies on consumer products, for example, will normally provide
such demographic information as the number of consumers (or consuming units) by age group,
income class, gender, and geographic area. Such data are usually available on a reasonably
recent basis from censuses conducted by federal, state, local, and, when needed, supranational
governments.
Often, a good first source is the Statistical Abstract of the United States, available online
from the Bureau of the Census http://www.census.gov/compendia/statab/. This reference abstracts
data from original reports and gives some useful material on social, political, and economic
matters. The source is a good reference to the more detailed data in the original sources.
The State and Metropolitan Area Data Book is a publication of the Bureau of the Census
that is available online in PDF format. It provides detailed comparative data on states,
metropolitan areas and their component counties, and central cities. It covers information about
numerous topics relevant for both B2C and B2B marketing, including population, income, labor
force, commercial office space, banking, health care, housing, and so forth.
http://www.census.gov/statab/www/smadb.html
41
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The Census of Population http://www.census.gov/population/www/ and the Census of
Housing http://www.census.gov/hhes/www/housing.html taken by the U.S. Department of
Commerce every 10 years are the most comprehensive of such censuses. Updates of various
census measurements based on smaller yearly surveys are available in Current Population
Reports http://www.census.gov/main/www/cprs.html and Current Construction and Housing
Reports http://www.census.gov/prod/www/abs/cons-hou.html. Many other up-to-date estimates are
made periodically by governmental and non–governmental agencies.
Data from the U.S. Census Bureau is available online for custom data analysis, on CDROM,
and in report form as downloadable PDF files. There are, however, private companies that
make such data available—for a fee—in more processed form, which, in effect, adds value to the
Census Bureau data. The company previously mentioned, GeoLytics (http://geolytics.com/),
markets a line of census data products and a variety of custom data retrieval services.
• Demographic reports (and maps)
– Custom data sets and reports
– Area segmentation
– Area-to-area correspondence files
– Banking and realtor tract level data and maps
• Services
– Geocoding (GPS) addresses
– Custom-built databases
– Normalized data for across census comparisons
Other companies include census data in mapping software that is used for geographic
market analysis. This type of software is potentially useful for such applications as retail site
analysis, real estate site reports, direct marketing, database creation, and so forth. One supplier is
Scan/US, Inc. (http://www.scanus.com ), whose software product Scan/US Streets and Data
U.S.A. includes maps for the entire United States that include all types of demographics.
Private Data Sources
Private organizations are another source of demographic information useful to marketers.
To illustrate, SRDS ( www.srds.com/frontMatter/ips/lifestyle/index.html ) publishes The Lifestyle
Market Analyst. This annual provides demographic and lifestyle information for 210 Designated
Market Areas (DMAs) in the United States. As shown in Table 3.1 and Figure 3.1, this market
data can be accessed in graphic and tabular formats for demographic and lifestyle variables:
• Demographic categories for each DMA: Start with a specific demographic segment, such as dualincome
households, and identify lifestyles and geographic locations.
• Most popular lifestyles for each DMA: Specify a lifestyle and then identify what other interests
frequently appeal to those consumers and what demographic information corresponds to that
profile.
42
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 3.1 Demographics and Lifestyles, Seattle-Tacoma, Washington
43
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
44
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 3.1 Map of a Selected Life Style Incidence Across the U.S.
Source: http://www.srds.com/frontMatter/ips/lifestyle/images/LMAA_overview-with-map.jpg
Nielsen’s Claritas division, a provider of solutions for geographic, lifestyle and behavioral target
marketing, has developed a demographic widget that is available as a free download for personal
electronics http://www.claritas.com/target-marketing/nielsen-claritas-demographic-widget.jsp .
Market size studies (e.g., size in sales dollars or units) often are conducted by trade
associations, media, firms in the industry, and private research organizations. These studies are
published and made available to interested parties. Industry-type studies may be concerned with
such types of information as total market size, market characteristics, market segments and their
size and characteristics, and similar types of information.
Example: Mediamark Research, Inc. conducts a single-source continuing survey, primarily
aimed at the advertising industry that provides demographics, lifestyles, product usage, and
exposure to all advertising media data. One part of this study is a series of studies on specific
products/services that is published as syndicated reports.
http://www.mediamark.com
Information on new products and processes is available from such sources as patent
disclosures, trade journals, competitors’ catalogs, testing agencies, and the reports of
governmental agencies, such as the Food and Drug Administration, the Department of
Agriculture, and the National Bureau of Standards.
An extensive amount of information is available concerning advertising.
• Publishers Information Bureau provides a compilation of expenditures by medium for each
competitor. http://www.magazine.org/advertising/revenue/index.aspx
• Audit Bureau of Circulations provides data on the numbers of magazine copies sold under
specified conditions. http://www.accessabc.com/
• Standard Rate and Data Service provides complete information on the rates and specifications for
buying advertising space and time.
http://www.srds.com/portal/main?action=LinkHit&frameset=yes&link=ips
• Mediamark Research, Inc. publishes data on multiple major local media markets, relating detailed
media behavior to demographic characteristics of readers/viewers/listeners.
http://www.mediamark.com
45
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Arbitron Radio and Television Reports (http://www.arbitron.com/home/content.stm ), the Nielsen
Radio-Television Index http://ca.acnielsen.com/products/product_list.shtml,
Starch Advertising Readership Service http://www.starchresearch.com/services.html,
all measure of audience exposure to specific advertisements or programs.
EXHIBIT 3.2 Starch Readership Reports
The best way to create print ads for the future, and for the long term, is to get feedback on a
constant basis in order to find out what works and what doesn’t.
Each year, Starch measures over 25,000 ads in over 400 magazine issues. On the most basic level you get
raw readership scores—the percent of readers who saw the ad and read the copy. Then the data are put
into a context: The ad is ranked not only against other ads in the issue but also against other ads in its
product category over the past two years. These norms are a fast and easy way to judge the
performance of your ad over time and against the competition.
The Benefits of Starch Ad Readership
In-Depth Analysis
• Campaign analyses inform clients not only about the scores of the ads but also why they
performed as they did and what can be done to improve the ads. Moreover Starch also is unique
in its ability to tell clients about the best positions in various publications (e.g., whether farforward
positioning is superior to ads in the back of the book).
Extra Questions
• To give you information on advertising likeability, persuasiveness and intent to purchase
• Many times, if you ask a publisher to Starch an issue your ads will appear in, they will assume
the cost and pass on the data to you for free.
The Starch Ad Readership Program
Through-the-Book, Recognition Method
• One-to-one in-person interviews
• Generally, 100–200 sample, but can be more if client desires
• Sample approximates readership of publication, but is not representative
• Reports present data on:
o Noted: percent who saw any part of the ad
o Associated: percent who saw advertiser’s name
o Read Some: percent who read any of the copy
o Read Most: percent who read more than half the copy
• Most reports also offer indexed scores, based on ads of the same size, color, product category
SOURCE: Adapted from Roper, 2002.
46
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Internet Databases
The Internet has become the staple of research and provides access to most commercial
electronic databases. Thousands of such databases are available from numerous subscription
systems, such as DIALOG (http://www.dialog.com ), LexisNexis (http://www.lexisnexis.com/), or
Dow Jones News/Retrieval
http://www.dowjones.com/Products_Services/ElectronicPublishing/EnterpriseMedia.htm.
In general, there are five categories of commercial databases:
1. Bibliographic databases that index publications
2. Financial databases with detailed information about companies
3. Statistical databases of demographic, econometric, and other numeric data for forecasting and
doing projections
4. Directories and encyclopedias offering factual information about people, companies, and
organizations
5. Full-text databases from which an entire document can be printed out.
The advantages of such current databases are obvious. All that is needed is personal
computer with internet access or a CD-ROM/DVD.
Computerized databases have led to an expanded focus on database marketing. Database
marketing has been defined as an extension and refinement of traditional direct marketing, but
uses databases to target direct response advertising efforts and tracks response and/or
transactions. In database marketing, the marketer identifies behavioral, demographic,
psychographic, sociological, attitudinal, and other information on individual
consumers/households that are already served or that are potential customers. Data may come
from secondary and/or primary sources. Qualtrics clients are increasingly APIs (Application
Programming Interface) to link and integrate customer databases with survey data and
respondent panels. APIs can be used to link and integrate data from multiple sources in real
time. Thus, information in database profiles is augmented by new contact and survey data, and
can be viewed in dashboards that report current information and can be used to better target and
predict market response. Databases can be used to estimate market size, find segments and
niches for specialized offerings, and even view current customer use and spending (Morgan,
2001). In short, it helps the marketer develop more specific, effective, and efficient marketing
programs.
Today, data mining is in high demand as a research focus. Data mining involves sifting
through large volumes of data to discover patterns and relationships involving a company’s
products and customers. Viewed as a complement to more traditional statistical techniques of
analysis, two of the more powerful data mining techniques are neural networks and decision
trees (Garver, 2002). Further discussion of data mining techniques is beyond the scope of this
text, but good discussions are found in Berry & Linoff (1997, 2002) and Dehmater & Hancock
(2001).
47
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 3.3 Attaining Market Knowledge From Online Sources
Elkind and Kassel (1995) provided essential guidelines for attaining market knowledge from online
sources:
• Develop an online research plan. The plan will outline all the key areas of inquiry and will provide
a systematic pathway to search, retrieve, and arrive at the desired data, information, and
knowledge.
• Clearly define your information needs, knowledge gaps, and issue to be resolved. One of the best
ways is to do a knowledge inventory or review to determine what you already know or have in
your possession, both primary and secondary research.
• Focus the search. Start by applying the learning from your knowledge inventory and specify the
new areas that are critical to your project. The focus can be further enhanced by specifying key
hypotheses regarding possible findings, information categories relevant to the issue, and other
criteria such as product categories, consumer targets, market areas, time frames, and so on.
• Search across multiple sources. Don’t expect to find what you need in single pieces of data or
sources of information. You only rarely will find what you need in one place.
• Integrate information from the multiple sources. Use techniques of trend analysis to identify
patterns that emerge when various information elements are combined; for example, content
analysis, stakeholder analysis, paradigm shift, trendlines, critical path analysis, sector analysis
(technological, social/cultural, occupational, political, economic), or other analytic techniques
that facilitate integration of diverse data and information and identification of underlying
patterns.
• Search for databases that contain analyses rather than limiting the search to just data or
information. Many of the professional online database producers and vendors offer thousands
of full-text articles and resources that contain analyses. You may be able to find material that
already provides some interpretation that may be helpful.
• Enhance the robustness of your data or information through multiple-source validation. You can
increase confidence in the validity of the findings of your secondary searches by looking for
redundant patterns that cut across different sources and studies.
SYNDICATED SERVICES
Some of the aforementioned commercial services are examples of what are called
syndicated services. Research organizations providing such services collect and tabulate
specialized types of marketing information on a continuing basis for purposes of sale to a large
number of firms. In general, syndicated data are made available to all who wish to subscribe.
Reports are made available on a regular basis (for example, weekly, monthly, quarterly). Since
these data are not collected for a particular firm for use in a specific research problem situation,
they can properly be viewed as secondary data. Syndicated services are widely used in such areas
as movement of consumer products through retail outlets, direct measures of consumer
purchases, social trends and lifestyles, and media readership, viewing, and listening.
The syndicated Survey of American Consumers (based on surveying more than 25,000
adults) by Mediamark Research, Inc. provides an illustration of syndicated services. This survey
provides data useful for detecting a marketer’s best prospects by providing answers to such
questions as the following:
48
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• How many customers are there for the products or services we market? Is the size of the market growing?
Stabilizing? Or shrinking?
• Who are the customers? How old are they? What do they earn? Where do they live?
• How do customers differ in terms of how often and how much they buy? Who are the heaviest purchasers
of the product?
• What brands are customers buying? How have shares of the market changed? What differences are there
among brand buyers?
• What’s the best way of reaching prospects? Which media vehicles and formats are most efficient in
delivering the message to the customer?
Mediamark is able to profile American consumers on the basis of more than 60
demographic characteristics and covers usage of some 500 product categories and services and
6,000 brands.
Types of Syndicated Services
Syndicated data may be obtained by personal interviews, direct observation, selfreporting
and observation, or use of certain types of mechanical reporting or measuring devices.
One of the most widely used approaches is the continuous panel, which refers to a sample of
individuals, households, or firms from whom information is obtained at successive time periods.
Continuous panels are commonly used for the following purposes:
1. As consumer purchase panels, which record purchases in a consumer diary and submit them
periodically.
2. As advertising audience panels, which record programs viewed, programs listened to, and
publications read.
3. As dealer panels, which are used to provide information on levels of inventory, sales, and prices.
Such panels have been established by many different organizations, including the federal
government, various universities, newspapers, manufacturers, and marketing research firms.
These types of panels furnish information on at regular intervals on continuing purchases of the
products covered.
For example, typical consumer panels might report the type of product purchased by brand,
weight or quantity of unit, number of units, the kind of package or container, price per unit,
whether a special promotion was in effect, store name, and date and day of week of purchase.
Data are recorded in diaries, either online or are mailed in each month.
One of the largest consumer panels is maintained by NPD Research
(http://www.npdor.com). This panel comprises 13,000 families and is national in coverage. NPD
also maintains self-contained panels in 29 local markets. Other well-known national consumer
panels are maintained by Synovate (synovate.com), NFO (mysurveys.com) ACNielsen
(http://en-us.nielsen.com), and IRI (infores.com).
Advertising audience panels are undoubtedly more widely publicized than other panels. It
is from these panels that television and radio program ratings are derived. These panels are
operated by independent research agencies rather than the media—both for reasons of economy
and to avoid any question of partisanship.
49
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
For example, ACNielsen uses a metering device that provides information on what TV shows are
being watched, how many households are watching, and which family members are watching.
The type of activity is recorded automatically; household members merely have to indicate their
presence by pressing a button. The sample is about 5,000 households. In local markets, the
sample may be 300 to 400 households.
Single source data tracks TV viewing and product purchase information from the same
household. Mediamark’s national survey and IRI’s BehaviorScan are examples of such singlesource
data. The single-source concept was developed for manufacturers who wanted
comprehensive information about brand sales and share, retail prices, consumer and trade
promotion activity, TV viewing, and household purchases.
The information obtained from the types of syndicated services described previously has
many applications. The changes in level of sales to consumers may be analyzed directly without
the problem of determining changes in inventory levels in the distribution channel. Trends and
shifts in market composition may be analyzed both by type of consumer and by geographic
areas. A continuing analysis of brand position may be made for all brands of the product class.
Analyses of trends of sales by package or container types may be made. The relative importance
of types of retail outlets may be determined. Trends in competitor pricing and special promotions
and their effects can be analyzed along with the effects of the manufacturer’s own price and
promotional changes. Heavy purchasers may be identified and their associated characteristics
determined. Similarly, innovative buyers may be identified for new products and an analysis of
their characteristics made to aid in the prediction of the growth of sales. Brand-switching and
brand-loyalty studies may be made on a continuing basis. One reported use of this syndicated
service has been to design products for specific segments.
The products from syndicated services are continually changing with client needs and
new technological opportunities.
SUMMARY
This chapter has been concerned with secondary information and sources of such
information. We started with some reasons why secondary information is essential to most
marketing research projects. Then, various sources and types of secondary information—internal
and external—were discussed in some depth. Also given more than cursory treatment was
syndicated data, a major type of service provided by commercial agencies.
50
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
ACNielsen: http://en-us.nielsen.com
ACNielsen – Claritas Demographic Widget, http://www.claritas.com/target-marketing/nielsen-
claritas-demographic-widget.jsp
Albaum, G., Duerr, E., & Strandskov, J. (2005). International marketing and export management
(5th ed.). Harlow, UK: Pearson Education.
Andreasen, A. (1988). Cheap but good marketing research. Homewood, IL: Dow-Jones Irwin.
Berry, M., & Linoff, G. (1997). Data mining techniques: For marketing, sales, and customer
support. New York: Wiley.
Berry, M., & Linoff, G. (2002). Mining the Web: Transforming customer data. New York:
Wiley.
Campbells 1959 attribute commercial http://www.youtube.com/watch?v=qMJtLP8jMWQ
Campbells 1985 Soup is good food commercial http://www.youtube.com/watch?v=4PtZYlcyXfI
Dehmater, R., & Hancock, M. (2001). Data mining explained: A manager’s guide to customer
centric business intelligence. Burlington, MA: Digital Press
DIALOG http://www.dialog.com
Direct Selling Association, 2003 http://www.dsa.org/pubs/
Dow Jones News/Retrieval
http://www.dowjones.com/Products_Services/ElectronicPublishing/EnterpriseMedia.htm
Elkind, F., & Kassel, A. (1995, July 31). A marketer’s guide for navigating the information
superhighway. Marketing News, 29, 2ff.
Garver, M. S. (2002, September 16). Try new data mining techniques. Marketing News, 36, 31–
33.
IRI: infores.com
LexisNexis http://www.lexisnexis.com
Morgan, M. S. (2001, October 8). Research boosts database’s power. Marketing News, 35, 16.
NFO: mysurveys.com
NPD Research http://www.npd.com
51
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
NPD Research, Inc. http://www.npdor.com
RoperASW. (2002). Available at RoperASW.com. Accessed January 9, 2002.
SRDS Life Style Market Analyst, http://www.srds.com/frontMatter/ips/lifestyle/index.html
SRDS Graphical Reports,
http://www.srds.com/frontMatter/ips/lifestyle/images/LMAA_overview-with-map.jpg
Synovate: synovate.com
U.S. Census Bureau, Statistical Abstract, http://www.census.gov/compendia/statab/
U.S. Census of Population, http://www.census.gov/population/www/
U.S. Census of Housing, http://www.census.gov/hhes/www/housing.html
U.S. Current Population Reports, http://www.census.gov/main/www/cprs.html
U.S. Current Construction and Housing Reports, http://www.census.gov/prod/www/abs/cons-
hou.html
52
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
APPENDIX 3.1
Selected Secondary Sources
Marketing/Advertising
Directories
American Marketing Association. Green Book: International Directory of Marketing Research Houses and
Services. Annual. www.greenbook.org
The Green Book lists and describes services offered by major marketing firms. Foreign as well as U.S. research
firms are entered alphabetically in a single list.
Broadcasting/Cablecasting Yearbook. Annual. http://www.bowker.com/index.php/component/content/article/2/33
In addition to its radio section, this source includes sections that survey the broadcasting industry, profile television
and cable stations, and focus on programming, professional services, technology, advertising and marketing, and
satellites.
The Direct Marketing Market Place. Annual. http://www.dirmktgplace.com/
Of possible interest to direct marketers, this directory lists the following: direct marketers of products and services
(catalog and retail sales, financial and investment services, fund raising, etc.); service firms and suppliers to direct
marketing companies (printers, list brokers and managers, computer services, etc.); creative and consulting services
(agencies, artists, copywriters, etc.); and lists of organizations, courses, awards, and so on. For each it includes a
description of the firm’s product or service, names of key executives, sometimes number of employees, gross sales
or billings, advertising budget and expenditures spent in direct marketing media, and countries in which business is
conducted.
Rand McNally. Commercial Atlas and Marketing Guide. Annual. http://www.randmcnally.com/
Brings together current economic, demographic, and geographic information. In addition to a series of area maps
highlighting major military installations, trading areas, retail sales, and manufacturing centers, it includes statistics
on business and manufacturers, retail trade, sales, and the largest corporations in the United States.
Standard Directory of Advertising Agencies. Annual. http://www.redbooks.com/Nonsub/index.asp
Also known as the Agency Red Book, this directory contains current information about 4,500 American advertising
agencies and their branches, including their names and addresses, specialization, major accounts, and key staff
members.
Standard Directory of Advertisers. Annual.
A companion directory to the aforementioned Agency Red Book. It focuses on companies and other organizations
that advertise and includes trade names, the types of media used for advertising, the advertising agencies employed,
and, frequently, annual advertising budgets.
Indexes
ABI Inform CD Rom.. http://www.proquest.com/en-US/catalogs/databases/detail/abi_inform.shtml
Among the fields covered are marketing and advertising. Over 800 journals are indexed and each article is fully
annotated. A full list can be printed from the CD-ROM. Current five years on disc
Business Index. Monthly.
This is a microfilm index that is a cumulative, three-year index to over 800 business periodicals and newspapers.
Business Periodicals Index. Monthly. http://www.ovid.com/site/catalog/DataBase/173.jsp
Indexes over 300 business periodicals. Most are general business periodicals; trade journals are not well represented.
Gale Group F & S Index United States. Weekly.
http://www.gale.cengage.com/servlet/BrowseSeriesServlet?region=9&imprint=000&cf=ps&titleCode=FSUS
Excellent for the latest news on product and industry developments or for information on specific companies,
particularly those that are not well known. Indexes over 750 periodicals, reports, and trade journals.
53
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Public Affairs Information Source Bulletin. Semimonthly.
The index covers periodical articles, books, government documents, conference proceeds, and other publications
that touch on topics of interest to public policy. Over 1,400 journals are indexed. Also on CD-ROM.
Consumer Index. Monthly.
Topicator. Monthly.
Other Sources
New York Times Index. http://www.nytimes.com/ref/membercenter/nytarchive.html
Statistical Indexes
1. American Statistics Index. Monthly. http://academic.lexisnexis.com/cis/subject-
area.aspx?pid=65&type=&parentid=58
The ASI is a master guide to statistical publications of the federal government. It is published in two parts, an Index
Section and an Abstract Section. It is arranged by subject with references to microfiche reproduction of the
publication found there.
2. Statistical Reference Index. Monthly.
Same format as aforementioned ASI except it indexes and abstracts statistics contained in publications not issued by
the federal government. It includes statistics published by private organizations, corporations, commercial
publishers, and independent and university-affiliated research organizations.
Government Publications
1. Monthly Catalog of United States Government Publications.
Records and indexes documents received by the Government Printing Office from all areas of government,
legislative, judicial, and executive branches, and independent and regulatory agencies as well.
2. Congressional Information Service. CIS/Index. Monthly.
This is a comprehensive subject index and abstract to the working papers of Congress comprising committee
hearings, reports, and prints, as well as publications of joint committees and subcommittees, executive documents,
and special publications.
Bibliographies
1. Special Librarian Association. What’s New in Advertising and Marketing. Monthly.
List by subject new books, periodicals, and trade publications in marketing and advertising. It includes many titles
that are free, especially summaries of market research.
Special Marketing, Advertising, and Consumer Reference Sources
Editor and Publisher Market Guide. Annual. http://www.editorandpublisher.com/eandp/resources/market_guide.jsp
Market data for over 1,600 U.S. and Canadian newspaper cities covering facts and figures about location,
transportation, population, households, banks, autos, gas and electric meters, disposable income, income per capita
and per household, and many others.
Leading National Advertiser. Ad. $ Summary. Annual.
Lists brands alphabetically and shows total 9-media expenditures, media used, and parent company. Also included
in this report are industry class totals and rankings of the 1,000 leading companies by total 9-media spending.
Sales and Marketing Management. Publishes an annual statistical issue and a loose-leaf data source.
1. Survey of Buying Power. Annual http://www.surveyofbuyingpower.com/sbponline/index.jsp
Contains geographically oriented demographic, income, and retail sales statistics for metropolitan areas and
counties. Also lists areas by population, number of households, total retail sales, and much more.
2. Survey of Buying Power Data Service. Annual.
Has more detailed and additional information than the Survey of Buying Power.
54
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Simmons Study of Media and Markets. http://www.libs.uga.edu/ref/simmons/index.html
Forty volumes devoted to the determination of use of various products and services.
Standard Rate and Data Service. http://www.srds.com/portal/main?action=LinkHit&frameset=yes&link=ips
This service offers separate directories giving advertising rates, specifications, and circulation for publications,
broadcast stations, and so forth in the following media: business publications (monthly), community publications
(semiannual), consumer magazines and agri-media (monthly), newspapers (monthly), print media production
(quarterly), spot radio (monthly), spot television (monthly), and weekly newspapers (monthly). This service also
publishes an annual Newspaper Circulation Analysis covering newspaper circulation and metro area, TV market,
and county penetration. Entries include daily and Sunday circulation figures, by county, for each newspaper.
Company and Industry Data
Dun and Bradstreet. Million Dollar Directory. Annual.
Lists over 160,000 American companies having an indicated net worth of $500,000 or more.
Standard and Poor’s Register of Corporations, Directors and Executives. Annual.
Lists over 45,000 U.S. businesses.
Directory of Corporate Affiliations. (Annual) and America’s Corporate Families: The Billion Dollar Directory
(Annual).
Each directory features lists of divisions and subsidiaries of parent companies.
MOODY’s Investor Service. Various call numbers. Seven annual volumes plus twice-a-week supplements.
Includes Bank and Finance, Industrial, Municipal and Government, Over-the-Counter, Public Utility,
Transportation, and International.
Standard and Poor’s Corporation Records.
Value Line Investment Survey. Weekly.
Walker’s Manual of Western Corporations and Services. Annual.
Standard and Poor’s Corporation. Industry Surveys.
This loose-leaf service includes for 22 industry categories detailed analysis of each category and of the industries
that comprise it.
Predicasts Forecasts. Quarterly.
Is useful both as a source of forecasts and projections of industries, products, and services and as a finding aid to
other sources.
U.S. Industrial Outlook. Annual.
Contains short-term forecasts for specific industries.
International Business
Many sources listed previously also are relevant to foreign markets and international marketing activities.
Indexes and Guides
Predicasts Funk and Scott Indexes.
1. F & S Europe. Monthly.
Covers articles or data in articles on foreign companies, products and industries.
2. F & S International Index. Monthly.
Covers articles or data in articles on foreign companies, products, and industries.
55
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Economic Trends
Asia Yearbook. Annual.
Contains information on individual countries plus finance, investment, economics, trade, and aid, and so on.
Europa Yearbook. Annual.
Includes history, economic affairs, economic statistics, constitution, government, and so on of individual countries.
Foreign Economic Trends and Their Implications for the United States. Semiannual.
Series of brief reports on more than a hundred countries that are prepared by U.S. embassies and consulates and
provide current data on gross national product, foreign trade, unemployment figures, wage and price index, and so
on.
Organization for Economic Cooperation and Development Economic Surveys of the O.E.C.D.
Annual.
Economic surveys of each of the 24 OECD member countries that contain information on current trends in demand
and output, price and wages, foreign trade and payments, economic policies, and future economic prospects.
Overseas Business Reports. Irregular.
Provides current and detailed marketing information for businesses evaluating the export market. Includes trade
outlooks, statistics, advertising and market research, distribution and sales channels, regulations, and market
profiles.
Price Waterhouse Guide Series. Continual update.
A series of guides on various aspects of doing business where Price Waterhouse has offices or has business contacts.
Topics covered include investments, corporate information, business regulations, accounting, taxes, and so on. Over
75 countries are represented in this series.
World Economic Survey. Annual.
A comprehensive picture of the economic situation and prospects for the world as a whole and for major world
regions. Analysis of inflation, rates of interest, exchange rates, trade balances, commodity prices, and indebtedness
are included.
Directories
Bradford’s Directory of Marketing Research Agencies and Management Consultants in the United States and the
World.
Directory of American Firms Operating in Foreign Countries.
Gives alphabetical index of U.S. Corporations and international geographic distribution; alphabetized by country,
then firm.
Directory of Foreign Firms Operating in the U.S.
Has:
1. Alphabetical listing by country—shows American Company and parent company.
2. Alphabetical listing of foreign parent companies and corresponding American subsidiaries or affiliates.
3. Alphabetical listing of American subsidiaries, branches, or affiliates of foreign companies.
Dun and Bradstreet. Principal International Businesses. Annual.
Complete legal name, parent company, address, cable or telex, sales volume, number of employees, SIC number,
description of activities, and chief executives are given for over 50,000 businesses.
Foreign Commerce Handbook. Irregular.
Information on all phases of international business is included in this reference source, including foreign-trade
services, daily language of foreign commerce, bibliography, and lists of organizations.
International Directory of Corporate Affiliations. Annual.
56
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
A comprehensive reference source of foreign companies with U.S. holdings and U.S. companies with foreign
holdings.
Worldwide Chamber of Commerce Directory. Annual.
Listing of U.S. Chambers of Commerce, American Chambers of Commerce abroad and chief executive officer, and
Foreign Chambers of Commerce in principal cities of the world.
Special Services
African Research Bulletin: Economic, Financial and Technical Series. Monthly.
Asian Recorder. Weekly.
Business International.
1. Business International Loose-Leaf Services. Weekly.
This is a newsletter service that offers short articles on capital sources, economy, industry, exporting, foreign trade,
management, and marketing. These newsletters cover the following areas: Asia, China, Eastern Europe, Europe,
International, and Latin America.
2. Financing Foreign Operations. Irregular.
Current guide to sources of capital and credit in 34 major markets.
3. Investing, Licensing and Trading Conditions Abroad. Annual.
Covers African-Middle East, Europe, Asia, North America, and Latin America.
Includes information on state role in industry, rules of competition, price controls, corporate taxes, personal taxes,
incentives, labor, and foreign trade.
4. Research Reports. Irregular.
These are in-depth reports prepared by the B.I. service on various subjects and countries. “Marketing in China,”
“Andean Common Market,” and so on.
5. Worldwide Economic Indicators. Annual
Includes key economic indicators for over 130 countries. Includes G.D.P., demographic and labor force data, foreign
trade, and production and consumption data.
Commerce Clearing House. Common Market Reporter. Weekly.
Ernst and Whinney. International Business Series. Irregular.
Very much like Price Waterhouse Guides described previously but not as extensive in coverage.
Federation of International Trade Associations. Really Useful (Web) Sites for International Trade Professionals. Biweekly
(http://fita.org, newsletter@fita.org)
Moody’s International Manual. Twice weekly.
Provides financial and business information on more than 5,000 major foreign corporations and national and
transnational institutions in 100 countries.
Statistics
Demographic Yearbook. Annual.
Comprehensive collection of international demographic statistics. Includes population, demographic and social
characteristics, geographical, educational, and economic information.
U.S. Statistical Office. Statistical Yearbook. Annual.
Kept up to date by the Monthly Bulletin of Statistics. Comprehensive compendium of international comparable data
for the analysis of socioeconomic development at the world, regional, and national levels.
International Labor Office. Yearbook of Labor Statistics.
57
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Includes total and economically active population, employment, unemployment, hours of work, wages, and so on.
U.N. Department of Economic and Social Affairs. International Trade Statistics Yearbook. Annual.
Kept up to date by Commodity Trade Statistics and Direction of Trade.
U.N. Statistical Office. National Accounts Statistics. Annual.
UNESCO. Statistical Yearbook. Annual.
Contains tables grouped according to various subjects: population, education, libraries and museums, book
production, newspapers and broadcasting, television, and cultural expenditure. Over 200 countries or territories
represented.
U.N. Statistical Office. Yearbook of Industrial Statistics. Annual.
Covers general industrial statistics data for each country and commodity production data. The United Nations has
separate Economic and Social Commissions for each geographical region. Each publishes an annual report that
surveys the economic and social trends of that area, including economic prospects, foreign trade, investments, oil
industry, and agricultural. The titles of these annual reports are listed next, followed by the title of the periodical and
supplements that update them if one exists.
1. International Financial Statistics. Monthly.
Data on exchange rates, international reserves, money and banking, trade, prices, and production for all IMF
member countries.
2. International Monetary Fund. Balance of Payments Yearbook. Annual.
Five-year detailed balance of payments statistics for about 100 countries. Includes statistics for goods, services,
capital, SDRs, and so on. Pay special attention to notes that accompany each table.
This appendix was developed from information provided by Rodney Christensen, former
reference librarian, Knight Library, University of Oregon.
58
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This page is intentionally blank
59
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 4
CONDUCTING INTERVIEWS
THE INTERVIEW
The research interview is a means of obtaining information from respondents. We can
define an interview as a form of dyadic (person-to-person) communication that involves the
asking and answering of questions.
The dyadic nature of the interview is such that bias and error potentially will be present
as a result of interviewer and the interviewee background characteristics (e.g., age, education,
socioeconomic status, race, religion, gender, etc.), psychological attributes (e.g., perceptions,
attitudes, expectations, motives), and behavior (e.g., errors in asking questions, probing,
motivating, recording responses). Interviewer and respondent are perceive and react to the
observable background characteristics and specific behaviors of each other. These factors have a
direct influence when an interactive interview is conducted and are implied in self-report
interviews (e.g., mail, e-mail).
Intervening variables are also part of the didactic interaction. The respondent’s role
expectations, the interviewer’s task behavior, differences in social desirability of response
alternatives, extent of topic threat, and salience of topic all affect the responses received.
The interviewer can likewise control the style of interviewing and thereby affect the
quality of information obtained. For example, styles can range from socio-emotional
(maintaining a warm, sympathetic, and understanding relationship with the respondent) to formal
(where the person-oriented actions of the interviewer are held to a socially acceptable minimum).
The researcher must assess the desirability of one style over the other in a given situation.
EXHIBIT 4.1 What’s in an Interview?
Universal dimensions underlie the relationships that are shaped as part of every Interview.
• Involvement encompasses the degree to which each party wants to take part in the
interview, including the degree of commitment of each to making it a success.
• Control refers to the degree of power the interviewer or interviewee has to affect the
interview process and its outcome.
• Relationship is the degree of warmth or friendship between the interview parties.
A number of elements also define the environment in which each Interview takes place:
1. Context. The total situation in which an interview takes place, including location,
physical arrangements, the people present, and those absent. This also includes status
differences between parties, temperature, privacy, and time.
2. Content. What the parties talk about during the interview. It involves topic selection
and treatment, arguments, supporting materials, language, and questions and answers.
Information From Respondents
3. Structure. Includes the interviewer’s or interviewee’s basic organizational patterns,
sequences of topics and questions, and the means used to open and close interviews.
60
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
4. Disclosure. The willingness on the part of both parties to reveal their “true” selves to
one another.
5. Feedback. The continuous stream of verbal and nonverbal signals (e.g., smiles, puzzled
expressions, raised eyebrows, moans) sent between interview parties that reveal
feelings, belief or disbelief, approval or disapproval, understanding or
misunderstanding, interest or disinterest, and awareness or
6. Cooperation. The degree to which the interview parties are willing and able to reduce
the competition inherent in most interview situations and work together for their
mutual benefit.
7. Conflict. The potential or actual struggle between parties because of incompatible or
opposing needs, desires, demands, and perceptions.
8. Trust. Belief in the good, worth, ethics, believability, and reliability of the other party.
Involvement, Control and Relationships have some effect upon each of the elements. These
dimensions and elements of relationships are present in each interview but are not of equal
importance. Although they are independent of each other, they have strong
interdependence as well.
SOURCE: From Stewart, C. J. and Cash, W. B., Interviewing Principles and Practices, 4/e. © 1985
William C. Brown, Publishers. Reprinted with permission of The McGraw-Hill Companies, pp. 9–13
Structure of the Interview
Interviews in marketing research and the behavioral sciences typically involve information
gathering and are usually classified by two major characteristics. An interview is either
structured or unstructured, depending on whether a formal questionnaire has been formulated
and the questions asked in a prearranged order. An interview is also categorized as either direct
or indirect, reflecting whether the purposes of the questions are intentionally disguised. Crossclassifying
these two characteristics helps us to identify four different types of interviews
Objective Interviews (a) structured and direct
(b) unstructured and direct
Subjective Interviews (c) structured and indirect
(d) unstructured and indirect.
Types a and b are basically objectivist; types c and d, subjectivist. We discuss each type
of interview in turn (although the discussion of the two indirect types of interviews is combined).
We then discuss the media through which interviews may be conducted.
Structured-Direct Interviews
Structured-direct interviews are the usual type of consumer survey to “Get the Facts” and
obtain descriptive information. A formal questionnaire is used consisting of nondisguised
questions.
Example: A marketing research manager of a bedroom-furniture manufacturer wants to find out
how many and what kinds of people prefer various styles of headboards and dressers. The
question sequence is fixed only those questions are asked. The resulting interview is structureddirect
in nature.
61
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The structured-direct interview has many desirable features. Since the questions are
formulated in advance, all the required information can be obtained in an orderly and systematic
fashion. The exact wording and phrasing of the questions can be worked out carefully to reduce
the likelihood of misunderstandings or influencing the answer. Pretests can (and should) be made
on the questionnaire to discover any problems in the wording or ordering of questions before the
questionnaire is finalized.
In the structured-direct interview, the questionnaire is, in effect, the dominant factor in
the interview. The interviewer’s role is simply to ask questions. The same questions are asked of
62
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
all respondents in the same order. This provides maximum control of the interviewing process
and reduces the variability in results caused by differences in interviewer characteristics. This
type of interview is less demanding insofar as the abilities of the interviewer are concerned,
permitting the use of less-skilled interviewers and resulting in a lower cost per interview. The
standardized, direct questions allow for uniform recording of answers, thereby reducing errors in
editing, tabulating, and analysis of the information.
The major problems associated with this type of interview involve wording questions
properly and the difficulties encountered in getting unbiased and complete answers to questions
concerning personal and motivational factors. The structured-direct interview is by far the most
commonly used type of interview in marketing research. An alternative approach is suggested in
Exhibit 4.2.
EXHIBIT 4.2 Structuring Conversational Interviewing
Despite pretesting, every survey question contains terms that have the potential to be
understood differently than the survey designer intends. For example, in a study of physicians conducted
for a pharmaceutical company, the interviewer might ask, “During the past two weeks, did you prescribe
any anti-inflammatory drug?” A respondent might answer, “Well, that depends. What exactly do you
mean by anti-inflammatory?” The interviewer is now faced with a choice. Should he use his knowledge to
answer the respondent’s question, or leave the interpretation of “anti-inflammatory” up to the
respondent?
The normal (standardization approach) way of handling this situation would be to leave the
interpretation of the question up to the respondent. Interviewers must read exactly the same question
and never interpret the question in any way (Fowler, 1991; Fowler & Mangione, 1990). When a
respondent asks for help, the interviewer should use so-called neutral probing techniques, like repeating
the question, presenting response alternatives, and so forth.
Another school of thought holds that the interviewer in the example above should help the
respondent and define “anti-inflammatory.” This group argues that response validity can be undermined
if respondents interpret questions idiosyncratically. An approach suggested is that interviewers be
allowed to use conversationally flexible interviewing techniques. This means that interviewers should
engage respondents in a manner similar to ordinary conversation, deviating from the standardized script
to ensure that respondents interpret questions consistently and correctly.
Online surveys have the ability to provide standard definitions through context sensitive help.
The respondent uses the mouse cursor to touch a context laden word (underlined or otherwise
designated) and a popup definition will appear. To the degree that respondents needing help would
receive it, response validity would increase, but the interaction with the respondent would be
standardized.
Unstructured-Direct Interviews
Unstructured-direct interviews are most often used in exploratory studies, and in
qualitative research (see Chapter 5). In the unstructured-direct method of interviewing, the
interviewer is given only general instructions to obtain the type of information desired. He or she
is left free to ask the necessary direct questions to obtain this information, using the wording and
order that seems most appropriate in the context of each interview.
Many research projects go through an exploratory phase in which researchers contact
respondents and hold unstructured interviews. These interviews are useful for obtaining a clearer
understanding of the problem, and determining what areas to investigate. This type of interview
is also often useful for obtaining information on motives. Following the exploratory interviews, a
formal questionnaire is developed for the final interviews.
63
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
To use the bedroom furniture example again, if the owner of a bedroom set is asked the
free-answer question, “Why did you buy your bedroom set?” the answer is almost certain
to be incomplete, provide proximate causes and may be worthless.
If the interviewer were seeking motivations, consider answers such as “Because
we needed a bed,” “Our old bed was worn out,” or “Because it was on sale.” When
motivations are given, such as “We enjoy a comfortable mattress that gives us a good
night’s sleep,” they are rarely complete.
The added enjoyment may be because the mattress is firmer, because of the pillow
top, because of the prestige the owner attaches to having a carved oak bedroom set, or
some combination of these and other factors. In addition, it is probable that motives other
than “enjoyment” influenced the purchase.
When used to establish motives, the unstructured-direct interview is known as a depth
interview. The interviewer will continue to ask probing questions: “What did you mean by that
statement?” “Why do you feel this way?” “What other reasons do you have?” The interviewer
continues with similar questions until satisfied that all the information that can be obtained has
been obtained, considering time limitations, problem requirements, and the willingness and
ability of the respondents to verbalize motives.
The unstructured interview is free of the restrictions imposed by a formal list of
questions. The interview may be conducted in a seemingly casual, informal manner in which the
flow of the conversation determines which questions are asked and the order in which they are
raised. The level of vocabulary used can be adapted to that of the respondent to ensure that
questions are fully understood and rapport is developed and maintained. The flexibility inherent
in this type of interview, when coupled with the greater informality that results when it is
skillfully used, often results in the disclosure of information that would not be obtained in a
structured-direct interview.
In the unstructured interview, the interviewer must both formulate and ask questions. The
unstructured interview can therefore be only as effective in obtaining complete, objective, and
unbiased information as the interviewer is skilled in formulating and asking questions.
Accordingly, the major problem in unstructured direct interviews is ensuring that competent
interviewers are used. Higher per-interview costs result, both as a result of this requirement and
the fact that unstructured interviews generally are longer than those that use a questionnaire. In
addition, editing and tabulating problems are more complicated as a result of the varied order of
asking questions and recording answers.
Structured-Indirect and Unstructured-Indirect Interviews
A number of techniques have been devised to obtain information from respondents by
indirect means. Both structured and unstructured approaches can be used. Many of these
techniques employ the principle of projection, in which a respondent is given a non-personal,
ambiguous situation and asked to describe it. It is assumed that the respondent will tend to
interpret the situation in terms of his or her own needs, motives, and values. The description,
therefore, involves a projection of personality characteristics to the situation described. These
techniques are discussed in more depth in Chapter 5.
64
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REDUCING RESPONSE AND NONRESPONSE BIAS
A major concern of the research planner when choosing which interview medium to use
is the potential systematic error (i.e., bias) that might arise. In Chapter 2, we discussed total error
and looked at its major components. At this point, it is useful to explore how to reduce the nonsampling
based error that occurs during the interview process. In communication, error may be
due to the nature of response given (inaccuracy and ambiguity) or due to the fact that the
sample member has not responded. The following sections will discuss these topics.
Inaccuracy
Inaccuracy refers to intentional and unintentional errors made by the respondent when
they provide information. There are two types of inaccuracies: predictive and concurrent.
Predictive inaccuracy is a special case or response error caused by inaccurate intentions:
Suppose a respondent indicates that he intends to buy a new sports car within 6 months;
and then does not. Or, alternatively, he does not now intend to buy, he answers “No” to
the question, and then buys a car within the six-month period. In each case, the
respondent intention was clear, but was not followed. This situation is a predictive
inaccuracy.
A similar type of predictive inaccuracy can occur when marketing researchers try to
predict actual market response to a price by asking consumers, “How much are you willing to
pay for Product X?” Differences between predicted and actual price acceptability may occur
because the true range of acceptable prices may change between the time of data collection and
the point of purchase for a number of reasons, such as budget constraints or windfalls, the price
of substitutes at point of purchase, search costs, and purchase urgency.
Concurrent inaccuracy occurs when the respondent intentionally does not provide
accurate information. Everyday experiences and empirical evidence suggest that inaccurate
information results from the respondent’s inability or unwillingness to provide the desired
information.
For our car purchase example, suppose the respondent answers “Yes”, but really has no
intention of buying a sports car within this period or, conversely, answers “No”, but does
intend to buy a sports car. In this case, we may say that there is concurrent inaccuracy in
his statement. The intention is to provide inaccurate information.
Concurrent inaccuracies are a major concern for many kinds of information obtained
from respondents (information on past behavior, socioeconomic characteristics, level of
knowledge, and opinion-attitude). Concurrent inaccuracies may also apply to instances where
observation is used; the observer charged with reporting the events or behavior may be unable or
unwilling to provide the desired information.
It is clear from this brief introduction to predictive and concurrent inaccuracies that
inability and unwillingness to respond are major contributors to response bias and warrant more
detailed attention to understand how they can be controlled.
65
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Inability to Respond
Even such a simple and straightforward question as “What is the model year of your
family car?” may result in an information-formulation problem, particularly if the car is several
years old. If respondents were asked, “What brand or brands of tires do you now have on your
car?” most would have even more difficulty in providing an accurate answer without looking at
the tires. Finally, if respondents were asked, “What reasons did you have for buying Brand “A”
tires instead of some other brand?” most respondents would have even more difficulty in
providing an accurate answer. Semon (2000a, 2000b) suggests that inaccuracies due to inability
to respond stem from three major conditions:
• Memory error: A respondent gives the wrong factual information because he or she
simply does not remember the details of a specific event. Often time since an event occurred
(purchase) are underestimated or overestimated. While better questionnaire and survey design
can help reduce this error, such proven techniques are not used because they add to the length of
the survey. For instance, in a personal or telephone interview survey, follow-up calls are often
not made to confirm the answers given.
• Ignorance error: This refers to the respondent’s lack of understanding, awareness or
perception of irrelevance for a question and is due to a poor research design in terms of question
content and sampling. A question (or even an entire questionnaire) may be unrealistic, deficient,
or directed to the wrong persons.
• Misunderstanding: This can be a matter of careless question design. Poorly defined terms
or words with different meanings can lead to inaccurate, or even deliberately falsified responses.
Proper question design would avoid words with multiple meanings and definitions, or should
clearly define the context in which the word is being used in the questionnaire.
In addition, the following items also create inaccuracies through the inability to respond
accurately:
Telescoping
Questions that ask respondents to reconstruct past experiences run a high risk of response
bias. More specific is the possibility of a respondent telescoping, or misremembering when an
event occurred during a short recent time period. In a study of durable-goods purchases in the
United States, respondents on average displayed forward-telescoping biases (reporting something
happened more recently than it did), and the magnitude of this reporting increased (Morwitz,
1997). Overall, the tendency to make forward-telescoping errors may differ by the demographic
of the respondent and by the event being studied— recall of purchase, reading of an ad, or some
other event will have an effect on the nature of any telescoping errors.
Telescoping can be reduced by using bounded recall procedures, which involve asking
questions about the events of concern in previous time periods as well as the time period of
research interest (Sudman, Finn, & Lannam, 1984). Other approaches include asking respondents
to use finer time intervals, and to use a landmark event such as New Year’s Day or Easter, or an
individual event landmark such as the date of a child’s wedding.
66
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 4.3 Response Error and Questionnaire Design
One of the keys to minimizing concurrent errors is for researchers to better select questions that fulfill
the clients’ information needs.
Important/Urgent questions are a first priority and easy to identify. Unimportant/Non-urgent questions
are likewise easy to identify and exclude. It is the other two categories that can cause most of the
problems.
Remember that Urgent/Unimportant questions may be best answered by a judgment-call than by
extensive research, and that the Important/Non-urgent questions are the ones that often need to be
addressed. In the short run, a company will carry on without answers to Important/Non-urgent
questions. But these answers may be essential to the long-term future direction of the company.
Unwillingness to Respond
When we move to the problem of unwillingness of respondents to provide accurate
information, the topic is more complex. Here we are dealing with the motivations of people: why
they are not willing to accurately provide the information desired.
Except in those instances where the respondent provides information by being observed
in a natural situation, there are always costs (negative utilities) attached to his or her formulating
and sharing information.
No fully accepted general theory to explain this behavior, but we again apply everyday
experiences to this problem and add some research findings to suggest why people may not be
willing to make accurate information accessible.
Investigator Expectations
A complex source of inaccuracy in response stems from the respondents’ appraisal of the
investigator and the opinions and expectations imputed to him or her.
A classic example is a cosmetics study that showed an unexpectedly high reported usage
of luxury cosmetics among women from low-income families. In this case, one
exceptionally well-dressed, carefully groomed, competent interviewer conducted all of
the interviews. The study was repeated with a matronly woman, in dress similar to the
women interviewed, calling on the same respondents on the following days. The reported
brands of cosmetics used were much less expensive, in this series of interviews.
Investigator Unwillingness
Sometimes, what appears to be respondent unwillingness to provide accurate data is
actually a case of interviewer cheating where the investigator is unwilling to obtain accurate
information. This happens when an interviewer finds a particular question too embarrassing to
ask; when the interview finds it easier to self complete the survey forms rather than conduct the
interviews; or when interviewers have friends complete the survey. Interviewers may also
complete some questions legitimately and then make an estimate or inference of other questions
such as age, income, and certain attitudes or behaviors of respondents.
Interviewer cheating can be kept to a low level of incidence but not eliminated
completely. Careful selection, training, and supervision of interviewers will eliminate much of
67
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
the problem. In addition, control procedures can and should be established to reduce it even
more.
The simplest control procedure is to call-back a subsample of respondents. If the
information on an initial interview is found to disagree significantly with that on the call-back
interview, additional call-backs may be made on respondents originally interviewed by the same
person. The fear of being caught will discourage cheating.
Other control procedures include the use of “cheater” questions and the analysis of
response patterns. Cheater questions are informational questions that will disclose fabricated
answers with a reasonably high probability of success. Likewise, the analysis of patterns of
responses for interviewer differences will disclose interviewer cheating when significant
variations from expected norms occur. Such analyses can be made at very little additional cost.
Time Costs
Perhaps the most common reason for respondent unwillingness to provide accurate
information, or any information for that matter, is the result of the time required to make the
information available. Respondents often give hasty, ill-considered, or incomplete answers and
resist probing for more accurate information. When possible to do so, a respondent will tend to
act in a manner that will reduce time costs. Such behavior often results in inaccurate or missing
information.
When conducting telephone and personal interviews the interviewer might ask “Is this a
good time to answer some questions, or would you rather set a time when I could contact you
again?” Experience has shown this latter technique only slightly lowers response rates.
Perceived Losses of Prestige
When information attributing prestige to the respondent is sought, there is always a
tendency to receive higher-prestige responses. All researchers experience difficulty both in
recognizing the items that demand prestige content, and in measuring the resulting amount of
inaccuracy. Information that affects prestige is often sensitive information, including
socioeconomic (age, income, educational level, and occupation), place of birth or residence.
An example of a still more subtle prestige association occurred in a study on nationally known
brands of beer. One of the questions asked was, “Do you prefer light or regular beer?” The
response was overwhelmingly in favor of light beer. Since sales data indicated a strong
preference for regular beer, it was evident that the information was inaccurate. Subsequent
investigation revealed that the respondents viewed people who drank light beer as being more
discriminating in taste. They had, therefore, given answers that, in their view, were associated
with a higher level of prestige.
Measuring the amount of inaccuracy is a difficult task. One solution to this problem is to
ask for the information in two different ways. For example, when obtaining information on
respondents’ ages, it is a common practice to ask early in the interview, “What is your present
age?” and later “In what year did you graduate high school?”
In one study, when respondents were asked, “Are you afraid to fly?” Very few people
indicated any fear of flying. In a follow-up study, when they were asked, “Do you think your
neighbor is afraid to fly?” (a technique known as the third-person technique), most of the
neighbors turned out to have severe anxieties about flying.
68
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Invasion of Privacy
Clearly, some topics on which information is sought are considered to be private matters.
When such is the case, both nonresponse and inaccuracy in the responses obtained can be
anticipated. Matters about which respondents resent questions include money matters or finance,
family, life, personal hygiene, political beliefs, religious beliefs, and even job or occupation. It
should be recognized however, that invasion of privacy is an individual matter. Thus,
information that one person considers sensitive may not be viewed that way by others.
The investigator should attempt to determine sensitivity if it is suspected to be a problem.
One way of handling this is to add questions in the pretest stage which ask about the extent of
sensitivity to topics and specific questions. A comprehensive treatment of sensitive information
and how to ask questions about it is given by Bradburn and Sudman (1979).
Ambiguity
Ambiguity includes errors made in interpreting spoken or written words or behavior.
Ambiguity, therefore, occurs in the transmission of information, through either communication
or observation.
Ambiguity in Communication
Ambiguity is present in all languages. Unambiguous communication in research requires
that the question asked and the answers given each mean the same thing to the questioner and the
respondent.
The first step in this process is the controlling one. If the question is not clearly
understood by the respondent, frequently the answer will not be clearly understood by the
questioner. To illustrate this point, after pretesting in an actual research project on tomato juice,
the following question change occurred after pretesting.

Even a careful reading of these two questions may not disclose any real difference in
their meaning. The analyst who drew up the question assumed that “like” refers to taste. In
pretesting, however, it was discovered that some housewives answered “Yes” with other referent
in mind. They “like” the amount of Vitamin C their children get when they drink tomato juice,
they “liked” the tenderizing effect that tomato juice has when used in cooking of meat dishes,
and so on. If the wording of the question had not been changed, there would have been a
complete misunderstanding in some cases of the simple, one-word answer “Yes.”
A related issue is one where a shortened form of a sentence is used. Examples are, “How
come?”, “What?”, and “How?” This “elliptical sentence” requires the respondent to first consider
the context of the sentence and then add the missing parts. When the mental process of
transformation is different for the researcher and respondent, communication is lost and
interpretation of a person’s response is faulty and ambiguous.
69
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The understanding of questions is an issue that goes beyond ambiguity. All too often a
respondent may not understand a question, but may have no opportunity to request clarification.
Most personal and telephone interviewing uses standardized interviewing, meaning that the
interpretation of questions is left up to the respondent. As discussed earlier in Exhibit 4.2, one
interesting approach taken in online surveys by Qualtrics.com is to use context-sensitive help
that provides standardized clarification or instruction for a given question.
Procedures for Recognizing and Reducing Ambiguity in Communication
Every research design that uses communication to obtain information should have as
many safeguards against ambiguity as possible. Procedures should be employed to recognize
where ambiguity may be present and to reduce it to the lowest practicable level.
Three procedural steps are useful for these purposes and should be considered in every
project:
1. Alternative question wording
2. Pretesting
3. Verification by observation
1. Alternative question wording. We have already seen that the present state-of-the-art
question formulation cannot guarantee unambiguous questions. In questions where ambiguity is
suspected, it is advisable to consider alternative wordings and forms of questions to be asked of
sub-samples of respondents.
The use of this simple experimental technique costs no more for online surveys
(randomly assign respondents to different blocks of questions). In personal and telephone
interviewing situations the interviewers can likewise be instructed to change the order for onehalf
the interviews. Where significant differences in response are discovered, it will be even
more worthwhile as a warning in interpreting the information.
2. Pretesting. Pretesting of questionnaires is a virtual necessity (Converse & Presser,
1986, pp. 51–75). The only way to gain real assurance that questions are unambiguous is to try
them. Pretesting is usually done initially by asking proposed questions of associates. To be truly
effective, however, pretesting of questions should be conducted by asking them of a group of
respondents who are similar to those to be interviewed in the final sample.
A typical way to assess problems with individual questions included in the questionnaire
is to ask those participating whether they had any trouble with each of the questions. Be sure to
ask about the exact nature of the problem.
If the pretest is done by an interviewer, each respondent can be asked about each question
and probing can get more depth in the response. It is the rule, rather than the exception, that
questions will be revised as a result of pretesting. Several versions of a question may need to be
considered as a result of pretesting before the final version is decided upon.
3. Verification by observation. Information obtained through communication should be
verified by observation whenever cost, time, and the type of information desired permit. Clearly,
verification by observation is not always possible or practical. For example, a housewife may
object to a pantry audit to verify that brands she indicated as preferred are on hand.
70
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Ambiguity in Observation
Although it has been suggested that, where practical to do so, information obtained by
communication should be verified by observation, the implication should not be drawn that
observation is free of ambiguity. In making observations we each select, organize, and interpret
visual stimuli into a picture that is as meaningful and as coherent to us as we can make it. Which
stimuli are selected and how they are organized and interpreted are highly dependent on both the
expertise, background and frame of reference of the observer.
As an illustration, a cereal manufacturer ran a promotional campaign involving a drawing
contest for children. Each child who entered was required to submit (along with a box top) a
picture he or she had drawn that depicted Brand X cereal being eaten. The contest was run, the
prizes awarded on the basis of artistic merit, and the brand manager turned his attention to other
matters. Later, a psychologist who worked for the company happened to see the pictures and was
permitted to study them. He found that a sizable proportion of them showed a child eating cereal
alone, often with no other dishes on the table. This suggested to him that cereal is often eaten by
children as a between-meal snack. Later studies by the company’s marketing research
department showed that cereals are eaten between meals by children in greater amounts than are
eaten for breakfast. The advertising program of the company was subsequently changed to stress
the benefits of its cereals as between-meal snacks.
Nonresponse Error
A nonresponse error occurs when an individual is included in the sample but, for any of
many possible reasons, is not reached or does not complete the survey. In most consumer
surveys this is a source of a potentially sizable error.
Non-response errors differ in nature depending on the mode of survey administration.
For example, when using telephone or personal interview methodologies, families who cannot be
reached generally have different characteristics than those who can be reached. They may be
away from home during the day and differ from those in which at least one member can usually
be found at home with respect to age, number of small children, and the proportion of time in
which the wife is employed. Similarly, fathers who are unwed, poor, and live in large cities; busy
executives and professionals, or occupational group such as hospital purchasing agents
responsible for chemical agents provide examples of hard-to-reach-populations that are difficult
to locate and interview (Teitler, Reichman, & Sprachman, 2003).
Internet surveys have the potential of increasing contact because the survey invitation
appears in the inbox awaiting the potential respondent’s reply. However other respondent
differences (time pressure, occupation, or lack of interest) and technological issues (spam filters)
may similarly increase non-response.
The seriousness of nonresponse error is magnified by the fact that the direction of the
error is often unknown. Hansen and Smith (2009) recently showed an 8% increase in response
rates by using a two stage presentation of a highly interesting question (a highly interesting
question was asked at the beginning of the survey and the respondent was told that the second
part of the question would appear at the end of the survey). These additional respondents were
shown to provide more centrally distributed responses, thus producing no additional error.
Researchers believe that major reasons for refusal include the public’s concerns about
data privacy and personal protection; a negative association with telemarketing efforts of all
types; consumers’ natural aversion to telephone surveys combined with a lack of survey choices
71
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
for consumers; low salaries of interviewers; and the fact that financial remuneration is not widely
used in surveys to compensate consumers for their time.
Evangelista, Albaum and Poon, (1999) suggest that four general motivations drive survey
response. Exhibit 4.4 suggests that response rates may be increased (and non-response bias
decreased) by using specific motivational techniques and inducements.
EXHIBIT 4.4 Theories of Survey Response
Why do people participate as respondents in a survey? The question is often asked by marketing
researchers, perhaps all too often implicitly, and seldom is an answer provided other than in terms of
specific techniques (including inducements) that have been used to increase participation. The following
theories are among those proposed (and studied to varying degrees) as answers to this question
(Evangelista, Albaum and Poon, 1999).
Exchange
The process of using survey techniques to obtain information from potential respondents can be viewed
as a special case of social exchange. Very simply, social exchange theory asserts that the actions of
individuals are motivated by the return (or rewards) these actions are expected to, or usually do, bring
from others. Whether a given behavior occurs is a function of the perceived costs of engaging in that
activity and the rewards (not necessarily monetary) one expects the other participant to provide at a later
date. In order that survey response be maximized by this theory, three conditions must be present:
1. The costs for responding must be minimized.
2. The rewards must be maximized.
3. There must be a belief that such rewards will, in fact, be provided.
Cognitive Dissonance
Cognitive dissonance theory appears to provide a mechanism for integrating, within a single model, much
of the empirical research that has been done on inducement techniques for survey response. As used to
explain survey response, the theory postulates that reducing dissonance is an important component of
the “respond/not respond” decision by potential survey respondents.
The process is triggered by receipt of a questionnaire and invitation requesting participation. Assuming
that failure to respond might be inconsistent with a person’s self-perception of being a helpful person, or
perhaps at least one who honors reasonable requests, failure to respond will produce a state of
dissonance that the potential respondent seeks to reduce by becoming a survey respondent. Since the
decision process involves a series of decisions for some people, delaying the ultimate decision may be a
way to avoid completing the questionnaire without having to reject the request outright (and thus
experience dissonance). Delaying a decision, therefore, may in itself be a dissonance-reducing response.
Self-Perception
Self-perception theory asserts that people infer attitudes and knowledge of themselves through
interpretations made about the causes of their behavior. Interpretations are made on the basis of selfobservation.
To the extent that a person’s behavior is attributed to internal causes and is not perceived as
due to circumstantial pressures, a positive attitude toward the behavior develops. These attitudes (selfperception)
then affect subsequent behavior.
The self-perception paradigm has been extended to the broad issue of survey response. To increase the
precision of this paradigm, the concepts of salience (behaviors one has attended to), favorability (the
affect or feeling generated by a given behavioral experience), and availability (information in memory) are
utilized. In addition, to enhance the effects, researchers should create labels. Labeling involves classifying
people on the basis of their behavior such that they will later act in a manner consistent with the
72
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
characterization. Self-perception would predict that labeling one’s behavior would cause that person to
view himself or herself as the kind of person who engages in such behavior; therefore, the likelihood of
later label consistent behavior is increased.
Commitment and Involvement
Of concern here is the range of allegiance an individual may be said to have for any system of which he or
she is a member. Consistent behavior is a central theme, including the following characteristics:
1. Persists over some period of time
2. Leads to the pursuit of at least one common goal
3. Rejects other acts of behavior
Consequently, the major elements of commitment are viewed as including the following:
1. The individual is in a position in which his or her decision regarding particular behavior has
consequences for other interests and activities not necessarily related to it.
2. The person is in that position by his or her own prior behavior.
3. The committed person must recognize the interest created by one’s prior action, and realize it
as being necessary.
A person who is highly committed to some activity is less likely to terminate the activity than one who is
uncommitted.
The theory of commitment (or involvement) can be extended to explain survey response behavior. To do
this requires recognition that commitment can be attached to many different aspects of a survey, such as
the source or the sponsor, the researcher, the topic and issues being studied, and/or the research process
itself. To a large extent, commitment is manifested by interest in what is being asked of the potential
respondent. The following hypotheses (untested) can be proposed:
1. The less favorable the attitude toward a survey’s sponsor, topic, and so forth, the less
involvement with, and thus commitment to, anything related to that study.
2. The less the extent of involvement, the more behavior productive of disorder (e.g.,
nonresponse, deliberate reporting of false information, etc.) is perceived as legitimate.
3. The more behavior productive of disorder is perceived as legitimate, the less favorable the
attitude toward the survey.
REDUCING INTERNET SURVEY ERROR
Conducting online surveys has become not only accepted, but the dominant form of
conducting structured/direct interviews. This shift to online research is due largely to reduced
cost, the availability of dynamic surveys using advanced survey flow logic, the ability to display
visually interesting and even interactive graphics, the ease of survey creation and administration,
and the ability to eliminate errors associated with data entry, coding and transcription.
In this light, we will focus our attention on online surveys to discuss how we can further
reduce or manage four other major sources of error in survey interviewing:
• Coverage error
• Sampling error
• Nonresponse error
• Measurement error
These same sources of error must be addressed regardless of the mode of survey data collection.
73
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Coverage Error
Coverage error occurs when the sample frame (the group from which the sample is
drawn) does not represent the population as a whole. For example, a random sample of Apple
Mac users would be a mismatch for the adult population of the United States. In more traditional
research methods such as mail or telephone methodologies, samples are drawn from sources such
as telephone directories, driver’s license records, rolls of property owners, credit reports, and so
forth. However such sampling frames are very information specific and often do not contain
email addresses.
E-mail list brokers offer panels and e-mail address lists that may be targeted to reduce
coverage error. Respondent lists can be selected by many variables, including gender, interests
(computers, electronics, family, finance, Internet, medical, and travel), and online purchasing.
These lists are typically double opt-in, meaning that the users have specifically indicated their
agreement to receive surveys or other promotional materials.
When the researcher requires a more detailed set of sample criteria, the cost of reducing
coverage error increases. Targeted specialty lists such as physicians of a given specialty are
expensive, costing as much as $100 per completed response. While this amount seems large, the
cost is much less than other methods of data collection. E-mail name brokers make a practice of
not providing the list, but of sending the survey invitation out, thereby controlling their list and
avoiding survey abuse of the potential respondents on the list.
Online sampling frames rarely include all elements of the target population. Therefore
coverage error will continue to be the greatest source of inaccuracy for online surveys for many
years to come. While this same problem is often encountered in the use of mail and phone lists, it
is not as severe as with online e-mail lists, which are often based on lists from online websites,
including magazines that have specialized hobby and interest affiliations. Carefully selecting lists
from well constructed probability panels or panels having millions of members will help to
reduce coverage error.
Sampling Error
Sampling error occurs when a non-representative sample is drawn from the sampling
frame. The estimation of sampling error requires that probability sampling methods be used,
where every element of the frame population has a known nonzero probability of being selected,
which may be made the same (i.e., equal) for all. However when the relationship between the
sample frame and the target population is unknown, statistical inferences to the target population
using confidence intervals may be inaccurate or entirely misleading. In online surveys the degree
of sampling error is generally unknown unless the sample is drawn from an online panel or other
frame with known size and characteristics. This information is rarely found in consumer research
and is rarely estimated.
Online surveys are therefore subject to certain amounts of sampling error. Sampling error
may be reduced in part by increasing the sample size. This is an easy task, especially when using
an online panel.
Nonresponse Error
Internet researchers are confronted with many non-respondent problems that have
elements both unique and common to those faced in telephone surveys. Spam filters, like caller
ID monitoring, prevent many survey requests from reaching the “In Box.” Internet users often
have limited discretionary time resulting in decreased willingness to participate in surveys. This
74
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
self-selection bias is manifest as potential respondents consider the appeal of the survey topic,
survey length, and incentives to complete the survey. The net impact is that without adequate
survey response and sample representativeness, non-response error will reduce validity and
accuracy of results (Shaffer and Dillman, 1998).
Increasing response rates and reducing nonresponse error most often includes the use of
multiple notifications and requests, and the use of personalization in the contact email requesting
completion of the interview. Sometimes even these techniques to not produced the desired
results. When a population of interest is not adequately represented online or is particularly
difficult to interview, a mixed-mode survey strategy should be considered to reduce nonresponse
error: a combination of e-mail and telephone, mail, or mall intercept techniques.
For example, many airline passengers making a connection in Cincinnati during July,
2009 encountered an interviewer in the terminal who was giving travelers a business card
with instructions and an online survey code. She requested that the traveler complete the
airline satisfaction survey when they returned home or to their office. The contact was
quick, novel and non-intrusive. Most travelers kindly accepted the card.
The single most important factor in reducing survey nonresponse is the number of
attempts to make contact with each prospective respondent. While many studies have confirmed
this fact, one of the more rigorous studies compared response rates for mail and e-mail surveys
(Shaffer and Dillman, 1998). In this field study, respondents in the mail and e-mail treatment
groups were contacted four times through (1) pre-notifications, (2) letters and surveys, (3) thankyou/reminder
notes, and (4) replacement surveys. Results showed no statistically significant
difference between the 57.5 percent response rate for the mail group, and the 58.0 percent
response rate for the e-mail group.
EXHIBIT 4.5 Increasing Response Rates
The variation in response rates for surveys is enormous, especially when interest and incentives are
considered. Ryan Smith, director of sales and marketing at Qualtrics.com, relates his experience with
three client surveys that differed greatly in their respective response rates (Smith, 2007). These three very
different surveys provide insight into the types of variables that influence response rate:
1. The first survey consisted of a short 10-question survey entitled “What Do Women Want . . .
For Valentine’s Day?” This somewhat whimsical survey was sent using a single e-mail blast
(with no second communication) to a “random sample” of Internet users through an e-mail list
broker. Recipients of the survey were offered the chance to win $500 cash in a random drawing
and in addition were promised a copy of the results. This combination of incentives plus a short,
interesting survey produced an amazing 43 percent response rate.
2. A second e-mail survey, a very long academic survey of more than 100 questions, focused on
developing a demographic, psychographic, and technological expertise profile of the online
shopper. This survey measuring attitudes and behaviors was sent through the same broker to a
random sample of “Internet shoppers.” Respondents were promised the chance to win $500
cash in one of seven random drawings. The university sponsorship of the survey was identified
in the cover letter that contained the professor’s name, contact information, and link to the
survey. The response rate was 11 percent.
A parallel paper and pencil survey was conducted for comparison purposes using a national
sample provided by Experian, a provider of credit rating reports. This mail survey was
75
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
implemented using three separate mailings (1) a pre-notification, (2) the survey, and (3) a
follow-up reminder. The mail version produced a 20 percent response rate. Comparison of the
mail and online survey results showed that demographic profiles were very different. This
difference was attributed to the difference in sampling frames. Respondents to the mail sample
were older, had different family structures and were more financially secure. However,
demographic differences aside, the psychographic profiles related to online shopping were
nearly identical.
3. A third survey sent to brokers by a leading investment firm resulted in a .002% response rate
after two mail outs. Further follow up revealed that this fast paced group of potential respondents
was too busy to be bothered with a survey.
Smith believes that five actions will greatly increase your online survey response rates:
(1) make your survey as short as possible by removing marginal questions
(2) make your survey interesting to the respondent
(3) include an offer of incentives
(4) use group affiliations whenever possible
(5) use requests that focus on altruistic self perception appeals (I need your help)
It should be noted that although Dillman’s response rates for a university faculty
population were considerably higher than would be expected for a consumer survey, the
similarity across survey modes stands as a solid finding. Perhaps most noteworthy is the finding
that when compared with the mail survey, the survey administered by e-mail produced 12.8
percent more respondents who completed 95 percent or more of the questions. Individual item
response rates and item completion rates were also higher. For the e-mail based open-ended text
responses, the same 12% increase in completion rates was observed, but in addition, responses
were longer, averaging 40 words versus 10 words for the paper-and-pencil survey.
Non-response is also important at the question level. Response rates can be improved by
using forced responses (the respondent cannot continue until all questions are answered).
Albaum et. al. (2010) show that contrary to expectations, using forced response increases both
the total response rate and the data quality improve.
It is clear that response rates are improved through the use of multiple contacts to secure
cooperation, sending reminders to complete the survey, and forcing a response. This applies not
only in traditional mail surveys but also in e-mail surveys. Yet, as shown in Exhibit 4.5 great
response rate variations can exist.
Management of the data collection process through state-of-the-art online survey
technology offers many new capabilities like survey tracking and personalization of invitations
and survey questions. The integration of panel information with the survey (using embedded
codes and data) facilitates the identification and tracking of survey respondents and
nonrespondents. Also with this integration, personalized follow-up mailings and reminders can
be sent to nonrespondents, and survey questions can be personalized to further increase response
rates.
Additional technologies enable the careful tracking of respondents. Statistics can be
compiled about the data collection status, including the number of surveys e-mailed, the number
received by potential respondents, the number of e-mails opened, the number of surveys viewed
(link clicked on), and the number of surveys completed. While technological advances help the
researcher to reduce non-response rates, it is clear that multiple factors are responsible for
76
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
nonresponse rates, many of which are not addressable through the administration and handling of
the survey.
Measurement Error
Measurement error is a result of the measurement process itself and represents the
difference between the information generated on the measurement scale and the true value of the
information. Measurement error may be due to such factors as faulty wording of questions, poor
preparation of graphical images, respondent misinterpretation of the question, or incorrect
answers provided by the respondent.
Measurement error is troublesome to the researcher because it can arise from many
different sources and can take on many different forms. For telephone and personal interviews,
measurement error will often occur when the interviewer misinterprets responses, makes errors
recording responses, or makes incorrect inferences in reporting the data.
Technical issues may similarly create measurement error in online surveys. The size and
resolution of the monitor, browser, operating system (Mac, Microsoft Windows, Linux), and
even web page color pallet may change the appearance of the survey. Additionally, skins or
templates may affect the actual survey’s appearance by adjusting the spacing between questions,
the appearance of horizontal lines separating questions or sections, the use of horizontal versus
vertical scales, drop-down boxes versus checkboxes or radio buttons, and even font
characteristics including size, typeface, the use of boldface and italics, and spacing between scale
items.
Researchers have for decades compared measurement error differences for the various
modes of data collection. While differences do exist, online surveys compare favorably with
traditional paper-and-pencil and telephone surveys. Standard surveys follow the structured/direct
approach for data collection and change little when transitioning from paper and pencil to radio
button or checkbox formats. However, as will be discussed later, the differences between the less
personal online surveys and the in-person qualitative research are far more extreme.
Many of the traditional measurement errors associated with transcription and recording of
data are eliminated with electronic real-time entry of the data. With Internet surveys, the survey
as well as the analysis of results can be conducted in real-time and posted to a secure Web site in
hours. In one recent survey of programmers and software developers conducted by Qualtrics.com
for Microsoft, 6,000 invitations were sent out with the promise of a $20 Amazon.com gift
certificate. Nine hundred completed surveys were received within 48 hours as results were
monitored online.
Online studies may be completed in 24 hours, as compared to the four to ten weeks
required for paper-and-pencil methodologies. Mail surveys must be prepared, printed, mailed,
followed up with mail reminders, manually coded, typed or scanned into the database, analyzed
and then compiled into a managerial report. These many steps involve many participants with
varying levels of expertise, and each may introduce error. Internet based surveys eliminate many
of these steps and combine other steps to complete the research much more quickly and easily.
Efficiencies aside, no matter which mode is used for survey completion, error control must be
addressed to assure quality results.
77
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
SUMMARY
This chapter first examined the various types of information that can be obtained from
respondents. It then considered communication as a means to obtain information from
respondents. The types of respondent interviews—structured-direct, unstructured-direct, and
structured- and unstructured-indirect—were discussed.
The next section introduced the concepts of inaccuracy and ambiguity as the major
sources of response and non-response bias. Predictive and concurrent sources of inaccuracy
were discussed in the context of respondent inability or unwillingness to respond. Methods of
reducing non-response error were then discussed in the context of theories of survey response.
Finally our discussion focused on how to reduce coverage, sampling, non-response and
measurement errors in online surveys.
The objective of marketing research is to understand the consumer and apply information
and knowledge for mutual benefit. Technological advances in online marketing research provide
the ability to monitor customer knowledge, perceptions, and decisions to dynamically generate
solutions tailored to customer needs. In this chapter we have stressed the need to improve the
research process by reducing errors in the research process. Perhaps the biggest mistake the
market researcher can make is to view research options as time- and cost-saving tradeoffs across
the data collection options. New technologies continue to be developed, but each must be tested
for applicability under specific research conditions, and refined so that marketers are able to
better identify and measure the constructs being investigated.
REFERENCES
Gerald Albaum, Catherine A. Roster, James Wiley, John Rossiter, and Scott M. Smith (2010).
Designing Web Surveys in Marketing Research: Does Use of Forced Answering Affect
Completion Rates?, Working paper.
Evangelista, F., Albaum, G., & Poon, P. (1999, April). An empirical test of alternative theories
of survey response behavior. Journal of the Market Research Society, 41, 2, 227–244
Bradburn, N., & Sudman, S. (1979). Improving interview method and questionnaire design. San
Francisco: Jossey-Bass.
Stewart, C. J. and Cash, W. B., Interviewing Principles and Practices, 4/e. © 1985, William C.
Brown, Publishers. Reprinted with permission of The McGraw-Hill Companies, pp. 9–13
Converse, J. M., & Presser, S. (1986). Survey questions: Handcrafting the standardized
questionnaire. Beverly Hills, CA: Sage.
Fowler, F. J. (1991). Reducing interviewer-related error through interviewer training,
supervision, and other means. In P. P. Bremer, R. M. Groves, L. E. Lyberg, N. A. Mathiowetz, &
S. Sudman, Measurement errors in surveys (pp. 259–278). New York: Wiley.
Fowler, F. J., & Mangione, T. W. (1990). Standardized survey interviewing: Minimizing
interviewer-related errors. Newbury Park, CA:
78
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Hansen, Jared M., Scott M. Smith, and Michael D. Geurts (2009), “Improving Survey
Completion Rates and Sample Representativeness Using Highly-Interesting Questions: A
National Panel Experiment Comparing One and Two Stage Questions”
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1313265
Morwitz, V. G. (1997). It seems like only yesterday: The nature and consequences of telescoping
errors in marketing research. Journal of Consumer Psychology, 6(1), 1–29.
Semon, T. T. (2000a, August 14). Better questions means more honesty. Marketing News, 34,
10.
Semon, T. T. (2000b, January 17). If you think a question is stupid—it is. Marketing News, 34, 7.
Schaffer, D. R., & Dillman, D. A. (1998). Development of a standard e-mail methodology:
Results of an experiment. Public Opinion Quarterly, 62, 378–397.
Smith, R. (2007). Personal communication [interview].
Stewart, C. J. and Cash, W. B., Interviewing Principles and Practices, 4/e. © 1985. William C.
Brown, Publishers. Reprinted with permission of The McGraw-Hill Companies, pp. 9–13
Sudman, S., Finn, A., & Lannam, L. (1984, Summer). The use of bounded recall procedures in
single interviews. Public Opinion Quarterly, 48, 520–524.
Teitler, J. D., Reichman, N. E., & Sprachman, S. (2003). Cost and benefits of improving
response rates for a hard-to-reach population. Public Opinion Quarterly, 67, 126–138.
79
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 5
INTERVIEWING MODES: PERSONAL-CALL-SEND
Survey information is obtained from respondents through communication in several
alternative media modes. Respondents may be interviewed in person, by telephone, or they may
be sent a questionnaire. Mail, FAX and the Internet surveys are similar in that non-personal self
reporting is involved.
PERSONAL INTERVIEWS
As the name implies, the personal interview consists of an interviewer asking questions of one or
more respondents in a face-to-face situation. The interviewer’s role is to get in touch with the
respondent(s), ask the desired questions, and record the answers obtained. Recording of the
information may be done either during or after the interview. In either case, it is the interviewer’s
responsibility to ensure that the content of the answers is clear, unambiguous and that
information has been recorded correctly.
While it is significantly more expensive on a per-completed-interview basis, the personal
interview, as a collection medium, has several advantages relative to telephone interviews and
mail questionnaires. It provides the opportunity to obtain a better sample, since virtually all the
sample units can be reached and, with proper controls and well-trained interviewers, nonresponse
to the survey or to individual questions can be held to a minimum. It also gives the opportunity
to obtain more information, as a personal interview can be of substantially greater length than
either a telephone interview or mail questionnaire. Finally, it permits greater flexibility. More
freedom is provided for adapting and interpreting questions as the situation requires, especially
in the case of unstructured personal interviews where visual, auditory, or olfactory aids are used.
The limitations of the personal interview include time, cost, and the response bias that
may be induced by poorly trained or improperly selected interviewers. Problems with personal
interviews arise from its very nature in that it is a social interaction between strangers, often on
the respondent’s territory, initiated by an interviewer who may have little in common with the
respondent.
In addition to the home and workplace, many studies conduct consumer interviews in
malls, where the so-called mall-intercept method is used. This method avoids the logistic,
financial and time costs of travel to meet with respondents. The mall intercept method involves
having interviewers stationed at selected places in a mall who request interviews from people
passing by. Presumably the people are chosen on the basis of a predetermined sampling plan. At
times, monetary incentives may have positive effects (Wiseman, Schafer, & Schafer, 1983).
The mall-intercept method is a widely used method of data collection in marketing
research. Many malls within the United States have permanent research facilities located within
them. These facilities may be equipped with videotape equipment, private interviewing
compartments, food preparation facilities for taste tests, and a variety of other research
equipment. Soundproof rooms free from distractions and equipped with proper lighting and
materials can contribute to reliable data collection. Researchers can observe an interviewer’s
technique and check the completed work immediately.
Overall quality of data (completeness, depth) appears to be about equal to that of other
methods, since mall-intercept respondents are more frequent users of shopping centers, and they
80
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
may be better able to provide more brand and store-oriented information than respondents
contacted by other means.
EXHIBIT 5.1 Mall Intercepts Are Widely Used
According to Katherine Smith (1989), mall intercepts have the following advantages:
1. They allow researchers to conduct visual, auditory and taste tests of ads, products and other physical stimuli.
2. They offer an opportunity to obtain immediate response.
3. They potentially provide more depth of response than non-face-to-face interviews.
4. Researchers can use equipment to analyze responses (for example, voice-pitch or eye movement tracking
analysis).
5. A large number of respondents from a wide geographic area can be interviewed in a limited time.
6. Researchers can control the interviewing environment and supervise the interviewer.
Mall intercept studies are less expensive than door-to-door interviewing, because travel time and the “not-athome
problem” are eliminated. However, it is becoming increasingly more difficult to locate people at home, and
even more people are hesitant to let strangers inside.
Using the mall intercept, interviewing often takes place where members of the population of interest are doing
something related to what is being measured. For studying certain types of products or behaviors, the mall is a
more realistic setting when a respondent is being asked to make choices. Finally, using certain sampling methods,
the mall-intercept procedure may give a better distribution of respondents.
Despite all these virtues, mall intercepts have limitations:
1. The mall customer may not reflect the general population.
2. The intercept is not well-suited to probability sampling.
3. Shoppers in a hurry may respond carelessly.
4. The interview time constraint is more severe with mall intercepts than with other personal interviewing
methods.
A variation of the mall intercept that is often used in business is to interview at
conferences, sales meetings, or other gatherings representing the population of interest.
In more general terms, research is better conducted on site whenever the topic is about
the business, when the purchase decision is made on the premises, or when the population of
interest is represented. Respondents are most likely to recall and discuss their experiences
during the experience, not days later during a survey. David Kay (1997), a partner in Research
Dimensions International, suggests there are five types of interviews for on-site research:
1. Stream of consciousness interview. This is a conversation with questions designed to
elicit what the respondent is experiencing at every moment of shopping.
2. Spontaneous reaction interview. This asks for spontaneous, minimally prompted
reactions of customers to their environment.
3. Directed general-response interview. Useful to assess effectiveness of strategy, this
method asks general questions directed to the strategy.
4. Directed specific-response interview. This is useful to determine why consumers feel
as they do, as indicated by answers to other questions.
81
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
5. Prompted reaction to execution elements. This is designed to elicit response to
specific elements. For example, an in-store taste test might include the question “What
do you think about the taste of China Sea brand Spring Rolls?”
The obvious advantages of on-site interviews are that the respondent is usually in a
proper state of mind and has better task or product recall. In addition it is easier to contact the
actual target group, making the response rates are higher. On-site interviews seem to produce
more robust information.
Paying people to participate in surveys, in the form of prepaid incentives, tends to
increase overall response rates for personal interviews, as well as for other types of interviews.
But, do not seem to influence the quality of the data collected (Davern, Rockwood, Sherrod, and
Campbell, 2003).
THE TELEPHONE INTERVIEW
Telephone interviews are often used in lieu of personal interviews, especially when
personal contact is desired, when the information must be collected quickly and inexpensively,
and when the amount of information required is relatively limited. Compared to e-mail or mail
surveys, telephone interviews often are more costly in terms of total costs of data collection.
However, when cost is figured on a per completed questionnaire basis, telephone interviews may
be less costly than mail, but more costly than e-mail. In addition, telephone surveys offer the
opportunity to probe for clarification or further information.
It is generally recognized that for business to business and consumer research, telephone
interviewing is as effective as personal interviewing for scope and depth of information obtained.
In addition, when a telephone survey is conducted from a call center, they can be better
supervised than personal interviews.
Virtually all telephone interviews are structured direct interviews. However, when the
population to be studied is business decision makers, some research practitioners believe that
more information may be obtained using the telephone than by conducting focus groups
(Eisenfeld, 2003). For business people and consumers alike, it is frequently easier to get 10
minutes of telephone cooperation, than a longer personal interview or attendance at a focus
group.
With a detailed database to use as a sample frame, interviews of business people, current
customers, former customers, and prospects all can be contacted relatively easily. Furthermore,
pre-notification letters can also be sent. A recent study of political telephone surveys concluded
that advance pre-notification letters can significantly increase response rates (Goldstein &
Jennings, 2002).
The likelihood of the potential respondent refusing to be interviewed is always present
when starting a telephone interview. Telephone surveys are unique in that they allow the
interviewer to respond to the potential respondent and attempt to turn a refusal into a completed
interview. In his classic treatise on telephone surveys, Dillman (1978) identifies common reasons
people give for refusals and suggests some possible responses the interviewer can give. These
responses can help the researcher handle objections and refine their interviewing skills. These are
shown in Table 5.1.
82
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
TABLE 5.1 Possible Answers to Reasons for Refusals
Reasons for
Refusing Possible Responses
Too busy This should only take a few minutes. Sorry to have caught you at a bad time, I would be
happy to call back. When would be a good time for me to call in the next day or two?
Bad health I’m sorry to hear that. Have you been sick long? I would be happy to call back in a day or
two. Would that be okay? (If lengthy or serious illness, substitute another family member.
If that isn’t possible, excuse yourself and indicate they will not be called again.)
Too old Older people’s opinions are just as important in this particular survey as anyone else’s.
In order for the results to be representative, we have to be sure that older people have as
much chance to give their opinion as anyone else. We really do want your opinion.
Feel inadequate:
Don’t know
enough to answer
The questions are not at all difficult. They mostly concern your attitudes about local
recreation areas and activities, rather than how much you know about certain things.
Some of the people we have already interviewed had the same concern you have, but
once we got started they didn’t have any difficulty answering the questions. Maybe I
could read just a few questions to you and you can see what they are like.
Not interested It’s awfully important that we get the opinions of everyone in the sample; otherwise the
results won’t be very useful. So, I’d really like to talk with you.
No one else’s
business what I
think
I can certainly understand, that’s why all of our interviews are confidential. Protecting
people’s privacy is one of our major concerns, and to do it people’s names are separated
from the answers just as soon as the interview is over. And, all the results are released in
a way that no single individual can ever be identified.
Objects to surveys We think this particular survey is very important because the questions are ones that
people in parks and recreation want to know answers to, so they would really like to have
your opinion.
Objects to
telephone surveys
We have just recently started doing our surveys by telephone, because this way is so
much faster and it costs a lot less, especially when the survey is not very long, like this
survey.
SOURCE: Reprinted from Mail and Telephone Surveys: The Total Design Method by Dillman, D. Copyright
© 1978. This material is used by permission of John Wiley & Sons, Inc.
As with all modes of surveying, telephone surveys benefit from the use of inducements or
incentives—monetary or nonmonetary—to encourage potential respondents to participate.
Incentives may be promised or sent in advance with a preliminary letter when the mailing
address of the potential respondent is known, or they may also be offered when the initial request
for participation is a refusal. When used this way it is known as a refusal conversion incentive.
The main purpose of such incentives is to generate a greater response rate with the effect
of reducing nonresponse error. But the use of incentives has implications as well. First, total cost
will increase, although cost per response may decrease depending on how effective the incentive
is. Second, data quality may be affected, leading to a change in response bias, which may be a
positive or negative change. Third, sample composition may be affected, again with a positive or
negative effect. Fourth, expectations of interviewer and respondent may be changed. Finally,
interviewer effort may be affected.
The telephone survey may be a good approach to reach specific market segments,
particularly when door-to-door interviews are not possible or might lead to serious distortions in
response. It is obvious that there must be sufficiently high telephone penetration in the segment
for this mode of data collection to be advantageous. For example, the use of surname sorts makes
telephone surveys the most efficient way to locate, contact and survey ethnic groups in the
United States.
83
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The basic limitations of telephone interviews are the relatively limited amounts of
information that can be obtained (at least compared with alternative methods) and the bias that
exists in any sample of home telephone subscribers. More than 25 percent nationally and more
than 50 percent in large cities are not listed in a published directory, either because they have an
unlisted number or as a result of moving (www.busreslab.com/articles/article3.htm). A
technique for including unlisted telephone numbers in the sample frame is called random digit
dialing (RDD).
Another additional problem for telephone researchers is that home telephone subscribers
are disappearing. Currently about 1 in 5 homes do not have a “landline” telephone, but rely
instead on cell phones or computer based phone services.
http://www.azcentral.com/business/articles/2009/07/20/20090720biz-cellnation0721side.html
Additional problems associated with telephone interviewing are those of sample control
and interviewer performance. Often this is manifested by inadequate efforts to complete
interviews with some of the harder-to-reach respondents. Adding another sample is no substitute
for dealing properly with the original sample.
Another aspect of interviewer performance that can influence response rates and data
quality is actually something beyond any given interviewer’s control. This is the interviewer’s
accent. There is evidence that accent can influence respondents’ participation. Linguistic experts
have found that listeners form mental impressions of people who speak with an accent different
from theirs, impressions that may lead to a refusal or bias the responses. In the United States
there are many region-specific accents (New England, the Deep South) and also those that are
cultural (Hispanic, Asian, Indian). When faced with an unfamiliar accent, people may have
trouble communicating. When communication becomes difficult, refusals will increase.
Alternatively, some accents increase curiosity (British, Irish) and can actually increase response
rates. In addition, respondents who identify an interviewer’s accent may apply preconceived
biases to the interviewer and to the survey. Accent-free interviewing eliminates one potential
source of nonresponse and bias. At the very least, if a study is regional in nature, then having
interviewers from that region will also reduce nonresponse and bias.
The so-called “caller-id” telephone technology problem has emerged for telephone
surveys as the number of households having caller-id and answering machines has increased.
One way people use answering machines is to screen incoming calls. The second, of course, is to
allow those calling to leave a message when members of the household are not at home. One
might be tempted to assume that screening and leaving messages would allow potential
respondents to choose to not participate—a form of refusal. Interestingly, some research
conducted on this issue has found that households with answering machines were more likely to
complete the interview and less likely to refuse to participate compared to households where
there was no answer on the initial call attempt (Xu, Bates, & Schweitzer, 1993). This suggests
that where answering machines are operating, the call can represent a form of pre-notification. It
is generally believed that pre-notification in any type of survey increases participation rates. But
this phenomenon also has the potential to generate bias in the sample (Oldendick & Link, 1994).
In the United States, major legislation affecting telemarketing went into effect in 2003.
This was the creation of the National Do Not Call Registry. Research practitioners are exempt
from this law. Of primary concern to the research community is the use by telemarketers of
selling-under-the-guise-of-research techniques. For those interested, the European Society for
Opinion and Marketing Research (ESOMAR) has responded by issuing guidelines for
84
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
distinguishing telephone research from telemarketing (ESOMAR, 1989). A good overview of
telephone surveys is given by Bourque and Fielder (2003a).
THE MAIL INTERVIEW
Mail interviews have in the past been widely used. Mail questions provide great
versatility at relatively low cost and are particularly cost effective when included as part of a
scheduled mailing, such as a monthly correspondence or billing. A questionnaire may be
prepared and mailed to people in any location at the same cost per person: the cost of preparing
the questionnaire, addressing the letter or card sent, and the postage involved. Respondents
remain anonymous unless a name is requested, the questionnaire is openly coded, or some
ethically-questionable practice is employed.
Timeliness of responses is critical in mail surveys. If the time given is reasonable, say
one or two weeks, stating a deadline should not adversely affect the response rate. Stating such a
deadline may encourage the potential respondent not to postpone the task indefinitely.
The overall process of data collection from a mail survey is summarized below as a
sequence of contact activities for an optimal mail survey. With minor modifications this general
sequence is applicable to personal interview, telephone, and e-mail surveys. When designing a
survey, the researcher must consider issues that can affect response rate and data quality,
including those shown in Table 5.2.
TABLE 5.2 Selected Dimensions of a Mail Survey and Alternatives for Choice
Dimension Alternatives
Preliminary notification Letter, postcard, telephone call, e-mail, none
Reminder Letter, postcard, telephone call, e-mail, none
Cover letter Separate item, included as first page of
questionnaire Personalized, non-personalized
Color of ink in signature (black, blue)
Length of questionnaire Number of pages
Format of questionnaire Print front, print front and back, individual stapled
pages, booklet
Type of outgoing postage First-class stamp, first-class metered, bulk, nonprofit
(where appropriate)
Return envelope postage First-class stamp, metered, first-class permit, none
Inducements Monetary (amount), nonmonetary (pen, silver
jewelry, trinkets of all types), contribution to charity,
none when given (prepaid, promise to pay)
Coding with a number Yes (on questionnaire, on return envelope), none
Anonymity/Confidentiality Yes, no
Endorsement Yes, no
Increasing Response Rates
Perhaps the most serious problem with mail surveys is that of nonresponse. Typically,
people indifferent to the topic being researched will not respond. It is usually necessary to send
additional mailings (i.e., follow-ups) to increase response. But even with added mailings,
response to mail questionnaires is generally a small percentage of those sent; the modal response
rate is often only 20 to 40 percent. On the front end of the surveying effort, response rates are
85
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
increased through preliminary contact by letter or telephone call, cover letters, and monetary or
non-monetary (a gift) inducements.
Other experimental research has evaluated response effects of questionnaire format and
length, survey sponsorship, endorsement, type of postage, personalization, type of cover letter,
anonymity and confidentiality, deadline date premiums and rewards, perceived time for task, and
the use of a follow-up reminder,. Reported nonresponse rates and accuracy of data for
experiments involving these techniques vary, and there appears to be no strong empirical
evidence that any one is universally better than the others, except that it is better to use a followup
and use monetary or nonmonetary incentives.
Cover letters are included as the first page of a questionnaire is shown in Exhibit 5.2.
This letter could have been a separate page if more space was needed for the questionnaire. Note
that email invitations should be much shorter, perhaps limited to only one short paragraph. There
is no evidence that any alternative is universally better than another within each dimension. The
best rule of thumb is to use common sense. Further discussion of these will be found in the many
review articles and studies published in such sources as the Journal of Marketing Research and
Public Opinion Quarterly, in the book by Bourque and Fielder (2003b), and in the classic works
of Dillman (1978, 2000).
EXHIBIT 5.2 Example of a Cover Letter
My colleague, Dr. David Boush, and I are engaged in a study of consumers’ use of financial services. The broad
objective of this study is to gain an understanding of how people use banks and similar financial organizations, and
what characteristics influence their behavior with such companies. The Bank of Anytown has agreed to cooperate
with us in this endeavor by assisting us in data collection.
The enclosed questionnaire is being sent to a large number of the customers of the Bank of Anytown, each of
whom has been selected by a random process. I would greatly appreciate your completing the questionnaire and
returning it in the envelope provided. Please note that you do not have to add postage to this envelope.
All individual replies will be kept in strictest confidence. No person associated with The Bank of Anytown will see
any questionnaire. Only aggregate results will be shown in our write-up of the results. No person other than Dr.
Boush, myself, and our research assistant will ever see a completed questionnaire. If you do not wish to participate
in this survey simply discard the questionnaire. Completing and returning the questionnaire constitutes your
consent to participate.
The code number at the top of the questionnaire will be used only for identifying those people who have not
responded so that he or she will not be burdened by receiving a follow-up mailing. After the second mailing has
been made, all records that match a number with a person’s name will be destroyed.
The success of this project depends upon the assistance of persons such as yourself. If you have any questions,
please call me at 503-346-4423.
Sincerely,
Gerald Albaum
Professor of Marketing
86
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Endorsements are an intriguing dimension of survey research. An endorsement is an
identifying sponsorship that provides “approval and support for a survey from an individual or
organization.” An endorsement can be included as a company logo or a person under who’s
signature the letter is sent. Unknowingly, an endorsement may have a positive, neutral, or
negative effect, depending on how the endorser is perceived by a potential respondent.
Rochford and Venable (1995) found that significantly higher response rates were
observed when there was endorsement by an external party associated with the targeted audience
than when there was no such endorsement. In addition, endorsements by locally known
individuals produced higher response rates than endorsements by highly placed but less wellknown
individuals from a national headquarters office.
Since people responding to a mail questionnaire tend to do so because they have stronger
feelings about the subject than the nonrespondents, biased results are to be expected. To measure
this bias, it is necessary to contact a sample of the nonrespondents by other means, usually
telephone interviews. This is a type of nonresponse validation. The low level of response, when
combined with the additional mailings and telephone (or personal) interviews of nonrespondents,
results in substantial increases in the per-interview cost. The initial low cost per mailing may
therefore be illusory. On the other hand, the nonresponse validation may indicate that populate
subgroups have not been omitted and that results may not be biased.
Variations on Mail Interviews
Many variations of the mail interview are frequently used. These include the warranty
card, hand delivered surveys, newspaper/magazine surveys, the fax back survey, survey on the
back of checks, website polls (a one question survey) and of course, the email survey.
Warranty cards often ask for information about where the item was purchased, what kind
of store or outlet sold it, when it was purchased, and other variables such as demographics and
life style. Although warranty cards do not provide an extensive amount of information, response
rates are substantially higher than for the usual mail questionnaire and the cards are useful for
creating a customer database.
Another variation is the questionnaire that is either printed or inserted in a newspaper or
magazine. Potential respondents are requested to mail or fax this back to a designated
address/fax number. Among the many problems with this approach is the lack of any formalized
control over the sample. Yet, despite this major limitation, the approach does have a possibility
of better hitting the target population.
Another modification of including a questionnaire in a printed publication asks
respondents to return the questionnaire by fax. One advantage of this approach is that those who
do respond will be truly committed to the project. But, again there is little control over the
sample. One major Airlines routinely printed questionnaires to be returned by fax in its in-flight
magazine.
Potential advantages of using fax include quick contact, rapid response, retention of the
original document format and visual images used, modest cost, and automated faxing that works
directly with the computer (Baker, Hozier, & Rogers, 1999).
One hybrid approach, used by McDonalds, had patrons to take a brief in-store survey on
a mark sense sheet, as shown in Figure 4.3. When completed, the store faxed the sheets for
processing using a technique that converted the image directly to data without being printed to
paper on the receiving end. The results were then available online in real time.
87
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
WEB AND E-MAIL INTERVIEWS
As computer coverage in home markets increase, the use of electronic surveys has
increased. Web and e-mail surveys are fulfilling their promise to be a driving force in marketing
research. Currently (2009) 73 percent of all U.S. households have Internet access in their home.
http://www.marketingcharts.com/interactive/home-internet-access-in-us-still-room-for-growth-8280/
The Internet has experienced a growth rate that has exceeded any other modern
technology, including the telephone, VCR, and even TV. The Internet has diffused from a
highly educated, white-collar, upper-income, male dominated core. At the opposite end of the
spectrum, the approximately 25% who are non-adopters include a disproportionate number of
elderly, single mothers, African Americans and Hispanics, and lower-income individuals. For
some studies, this may be a serious limitation. However even today, most groups like company
employees, students, or association members have nearly 100 percent Internet access and check
e-mail on a daily basis.
FIGURE 5.1 Example of Questionnaire Processed by Fax
88
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Electronics driven lifestyles that include online social networks, massive use of texting
and pervasive internet connections are no doubt responsible in part for seemingly responsible for
attitude and behavioral changes in the way we view our increasingly virtual world. Strong
upward trends are observed in the percentage of Internet purchases for airline tickets, CDs,
DVDs, books, computer software, hardware and systems. These online customers provide
excellent access for research purposes.
Advocates of online surveying quickly point to the elimination of mailing and
interviewing costs, elimination of data transcription costs, and reduced turnaround time as the
answer to client demand for lower cost, timelier, and more efficient surveys. As a result, online
marketing research has become so widely accepted that online research has been optimistically
projected to account for as much as half of all marketing research revenue, topping $3 billion.
While these numbers appear to be overly optimistic, it is clear that online research is growing
and that researchers operate in a much faster-paced environment than ever before. This pace will
continue to increase as new modalities for research open: wireless PDAs, Internet-capable
mobile phones, Internet TVs, and other Internet-based appliances yet to be announced. Each is
an acceptable venue for interacting with the marketplace and conducting online research.
Substantial benefits accrue from the various approaches to computer-assisted data
collection in surveys as shown below:
• Respondents need few computer related skills
• Respondent choose their own schedule for completing survey
• Can easily incorporate complex branching into survey
• Can easily pipe and use respondent generated words in questions throughout the survey
• Can accurately measure response times of respondents to key questions
• Can easily display a variety of graphics and directly relate them to questions
• Eliminates need to encode data from paper surveys
• Errors in data less likely, compared to equivalent manual method
• Speedier data collection and encoding compared to equivalent manual method.
ACNielsen (Miller, 2001) reported the results of 75 parallel tests comparing online and
traditional mall intercept methods. Researchers noted high correlations in aggregate purchase
intentions. While online measures may yield somewhat lower score values, recalibration of
averages against appropriate norms produced accurate sales forecasts. Wilkie further reported
that while responses gathered using different survey modes may be similar, the demographic
profiles of online and traditional respondents groups do differ.
Given that the current percentage of households online is approximately 75 percent,
statistical weighting of cases could be used to adjust demographic differences of online groups to
match mall intercept or telephone populations. However, the possibility of weighting actually
raises the question of whether it is better to model phone or mall intercept behavior (which are
also inaccurate) or to attempt to independently model the actual behavior of the respondents.
89
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
TABLE 5.3 Comparative Evaluation of Alternative Survey Methods of Data Collection
Criteria
Telephone
CATI
In-Home
Interviews
Mall-
Intercept
Interviews CAPI
Mail
Surveys
Mail
Panels
Internet/
Web
Flexibility of data
collection
Moderate
to high High High
Moderate
to high Low Low
Moderate
to high
Diversity of
questions Low High High High Moderate Moderate
Moderate
to high
Use of physical
stimuli Low
Moderate
to high High High Moderate Moderate Moderate
Sample Control
Moderate
to high
Potentially
high Moderate Moderate Low
Moderate
to high
Low to
moderate
Control of data
collection
environment Moderate
Moderate
to high High High Low Low Low
Control of field
force Moderate Low Moderate Moderate High High High
Quantity of data Low High Moderate Moderate Moderate High Moderate
Response rate Moderate High High High Low Moderate Very low
Perceived
anonymity of
respondent Moderate Low Low Low High High High
Social desirability Moderate High High
Moderate
to high Low Low Low
Obtaining
sensitive
information High Low Low
Low to
moderate High
Moderate
to high High
Potential for
interviewer bias Moderate High High Low None None None
Speed High Moderate
Moderate
to high
Moderate
to high Low
Low to
moderate Very high
Cost Moderate High
Moderate
to high
Moderate
to high Low
Low to
moderate Low
SOURCE: From Malhotra, N., Marketing Research: An Applied Orientation, 4th edition, Copyright © 2004.
Reprinted with permission of Pearson Education, Inc., Upper Saddle River, NJ.
PROBABILITY AND NONPROBABILITY SURVEY APPROACHES
A variety of approaches to presentation of surveys and recruitment of respondents are
used on the Web. Surveys based on probability samples, if done properly, provide a bias-free
method of selecting sample units and permit the measurement of sampling error. However the
majority of online research can be typified as a one-shot mailout, the objective of which is to
obtain a sufficient number of completed responses. With these types of studies, there is rarely
much thought given to the issues of representativeness of the sample, or validity and accuracy of
the results. The question is, then, what is required for effective online research?
Online Nonprobability Surveys
Nonprobability samples offer neither of these features. Nonprobability-sample–based
surveys, generally for entertainment or to create interest in a Web site, are self-selected by the
respondent from survey Web sites either for interest or compensation, or are provided to
members of volunteer panels.
As an example of the Web site interest variety of survey, Figure 5.2 shows a CNN.com
“Quick Vote” survey, which includes a link to the online results page.
90
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
FIGURE 5.2 Web Site Interest Survey
Source: http://edition.cnn.com/
The National Geographic Society Web site offers surveys that focus on a variety of
educational, social and environmental issues at www.nationalgeographic.com/geosurvey/.
Surveys include lengthy inventories covering demographics, Internet usage, and attitudes about
such topics as geographic literacy, conservation and endangered species, culture, and a variety of
other topics. The 2006 Global Geographic Literacy Survey was conducted jointly with Roper
Research to assess the geographic knowledge of people ages 18 to 24 across the United States.
Specific questions focused on benchmarking attitudes towards the importance of geography and
how aware young adults are of geography in the context of current events.
Other well-recognized nonprobability surveys include the ACNielsen BASES (test
marketing) and Harris-Black panels. Although nonprobability surveys, these panels are
continually redefined to match the demographic characteristics of telephone and mall intercept
surveys. The parallel telephone and mall intercept studies provide weighting to proportionately
adjust online samples to reduce selection bias.
Online Probability Surveys
Probability-based surveys allow the researcher to estimate the effects of sampling error
and thereby provide inferences about the target population through hypothesis testing. Coverage
errors, nonresponse errors, and measurement errors still apply and may reduce the
generalizability of the data. Online probability samples generally result where e-mail surveys are
sent to comprehensive lists that represent the target population. When the target population is
large, random samples from the list will be used. For smaller populations such as employees of a
company, the survey may be sent to the entire population, thus representing a census.
Where the target population of interest is visitors to a given Web site, pop-up surveys
may be presented randomly to visitors during their visit to the site. In this case, the target
population is well defined and the sample element has a known nonzero probability. The
Qualtrics.com Site Intercept tool allows the market researcher to control all pop-up and popunder
content without the assistance of the IT department. Surveys, white papers, and messages
can be distributed based on a variety of conditions, including multiple pages visited that contain
specific key words. For example alternative surveys about kitchen appliances could be
administered if the visitor viewed pages dealing with major appliances (stoves and refrigerators)
rather than counter top appliances (mixers and waffle irons).
91
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Pre-recruited online panels, when recruited using probability-based sampling methods
such as random-digit telephone dialing, also produce probability surveys. In this case, random
digit dialing would be used to contact the prospective panel members who would be qualified as
Internet worthy before being recruited for the panel. The initial interview may also include a
variety of demographic, lifestyle, and computer usage questions that would help in weighting the
panel, thereby reducing selection bias.
Mixed Mode Studies
Mixed-mode designs provide another alternative for the researcher, presenting
respondents with a choice of responding via online survey or via another mode. Respondents
contacted by mall intercept, telephone, mail, or other probability-based sampling mechanism, are
given the opportunity to respond in several modes, including online. It is common for businesses
or individuals to prefer the online survey format.
Wisconsin cheese producers respond annually to an industry group survey that reports
production by the type of cheese. This more-than-90-page survey details the desired information
for a separate type of cheese product on each page. When asked if they would prefer a paperand-pencil
or online survey, more than 50 percent favored the online mode. While the online
methodology may be preferred, access to the survey must be provided to all cheese producers,
even those without Internet access. A mixed-mode survey design is the obvious choice.
When online samples are used to make inferences about the general population, we must
account for the multiple factors that distinguish online samples from the general population.
These factors include nonsampling errors unique to the Internet methodology: for example,
fewer households have adopted the Internet than have telephone or mail and researchers lack
control of the respondent’s computer configuration (browser, operating system, fonts, and
resolution). In addition, refusals, partial completions, and all other nonsampling factors that bias
traditional survey measurement and results still apply to online surveys. (Sources of nonsampling
error were discussed in Chapter 2.)
Online survey techniques are also subject to many of the other errors that affect telephone
and mail surveys. Marketing researchers, both professional and casual, often neglect to consider
the implications that nonprobability sampling and surveys have on the ability to make inferences
regarding the target population. While this brief review has done little more than identify the
topic areas to be considered, much research on the topic has been completed for both traditional
and online surveys. In Chapter 10, we build upon this discussion of general survey and sampling
methodology.
STRATEGIES OF DATA COLLECTION
When designing a survey, our concern should be for the total package of survey
procedures rather than any single technique (Dillman, 1978, 2000). The total package concept
underlying survey design strategy goes beyond contacts with respondents. All aspects of a study
must be considered when comparing alternative strategies. This is the essence of a total design.
Pretesting and conducting pilot surveys are part of this package.
Pretesting and Pilot Survey
A distinction should be made between a pretest and a pilot survey. Pretesting is an
activity related to the development of the questionnaire or measurement instrument to be used in
92
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
a survey or experiment. In contrast, a pilot survey is a small-scale test of what the survey is to be,
including all activities that will go into the final survey. Pretesting a questionnaire answers two
broad questions:
1. Are we asking “good” questions?
2. Does the questionnaire flow smoothly, and is the question sequence is logical?
Pretesting does not, however, ensure that the questionnaire (or even the survey) will be
valid, particularly in its content. A general rule of thumb for most surveys is that a pretest of
about 30 to100 interviews is adequate, provided this covers all subgroups in the main survey
population. Ideally, the sample for the pretest should mirror in composition that of the main
survey.
The pilot study is designed to ascertain whether all the elements in the survey fit together.
Thus, questionnaire pretesting may be part of the pilot study but normally should not be. One
aspect of the pilot survey is that it can help researchers decide the size of the original sample for
the main survey. Response to the pilot can be used, together with the desired sample size, to
determine the size of the required sample.
Both pretesting and pilot surveys can provide information helpful to manage some of the
sources of potential research error. Moreover, in the long run they can both make a survey more
efficient and effective.
LONGITUDINAL DATA COLLECTION WITH PANELS
Panels are widely used in marketing research. In the preceding chapter we discussed the
continuous panel as used by syndicated services. In this chapter we have discussed methods for
reducing response error along with some general characteristics of panels. Although the panel
concept has been used in business-to-business marketing research, its greatest application has
been in studying consumer purchase, consumption and behavioral patterns.
For example, panels have been effectively used to develop early forecasts of long-run
sales of new products. There are major commercial consumer panel organizations, and
many consumer product companies that maintain their own panels or create short-term ad
hoc panels as the need arises to test new products. In addition, several universities
maintain consumer panels to obtain research data and generate revenues by providing
data to others. Moreover, the application of Internet technology has helped many
corporations to reduce data costs while increasing customer feedback and research.
The distinguishing feature of a panel is the ability to repeat data collection and collect
longitudinal data from a sample of respondents. The repeated collection of data from panels
creates both opportunities and problems. Panel studies offer at least three advantages over onetime
surveys:
1. Deeper analysis of the data is possible so that the researcher can determine if overall
change is attributable primarily to a unidirectional shift of the whole sample or only
reflects changes in subgroups.
2. Additional measurement precision is gained from matching response from one interview
data collection point to another.
93
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
3. Panel studies offer flexibility that allows later inquiries to explain earlier findings.
When responses are obtained at two or more times, the researcher assumes that an event
happens or can happen (i.e., changes may occur) during the time interval of interest. In fact, it is
just such changes, analyzed in the form of a turnover table, that provide the heart of panel
analyses.
Assume that we have changed the package in one market for a brand of paper towels
called Wipe, and that we run a survey of 200 people purchasing the product two weeks
before the change (T1) and a similar measure for the week after (T2). The results are
shown in Table 4.4. Both (A) and (B) tell us that the gross increase in sales of Wipe over
X (this represents all other brands) is 20 units (or 10 percent). However, only the turnover
table from the panel in (B) can tell us that 20 former buyers of Wipe switched to X and
that 40 former buyers of X switched to Wipe. In those instances where there is
experimental manipulation, such as the introduction of a new product or the use of splitcable
advertising, the manipulation is presumed to cause changes between time x (when
the change is made) and time x + 1.
TABLE 5.4 Changes in Sales of Wipe Paper Towels between T1 and T2 (Hypothetical)
(A) Cross-Sectional T1 T2
Bought Wipe
Bought X
Number of Purchasers
100
100
200
120
80
200
(B) Panel At T1 Bought Wipe At T1 Bought X Total
At T2 Bought Wipe
At T2 Bought X
Total
80
20
100
40
60
100
120
80
N=200
Panel studies are a special case of longitudinal research, where respondents are typically
conscious of their ongoing part in responding to similar questions over a period of time. This
consciousness of continuing participation can lead to panel conditioning, which may bias
responses relative to what would be obtained through a cross-sectional study. As in any effort at
scientific measurement, the researcher should be concerned with threats to internal validity, since
internal validity is a precondition for establishing, with some degree of confidence, the causal
relationship between variables. Another issue of concern is panel attrition, the extent of
nonresponse that occurs in later waves of study interviewing. Some persons who were
interviewed at the first time may be unwilling or unable to be interviewed later on.
There are many distinguishing characteristic of panel types. We have already mentioned
different types of sponsoring organizations (such as commercial), permanence (continuous or ad
hoc), and research design (non-experiment). Panels can also be characterized by geographic
coverage (ranging from national to local), whether a diary is used, data collection method (all
types are used), sampling method employed for a given study (probability or not), and type of
respondent.
94
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
A unique type of panel is the scanning diary panel. This panel involves recruiting
shoppers in the target market area to participate in the panel, and each person typically is
compensated for participation. An identification card (similar to a credit card) is given to each
member of the panel household. At the end of a normal shopping trip in a cooperating store, the
card is presented at the start of the checkout process. This identifies the respondent for input of
purchases into the computer data bank. The types of information available from this sort of panel
are similar to those discussed in the preceding chapter for scanner-based syndicated services. An
added advantage here, of course, is that there is a carefully designed sample providing purchase
data.
One last comment about panels is that they are often used for a cross-sectional study.
When used this way and only one measurement is made, the panel is merely the source of a
sample (the sample frame).
SUMMARY
This chapter first examined the media through which interviews may be conducted. The
personal interview, the telephone interview, the mail interview and online e-mail interview were
discussed, including the merits and limitations of each. Variations of these basic methods,
including electronic-based variations, were briefly described. Finally, the use of panels was
presented in a general context.
The objective of marketing research is to understand the consumer and apply information
and knowledge for mutual benefit. Technological advances in online marketing research provide
the ability to monitor customer knowledge, perceptions, and decisions to dynamically generate
solutions tailored to customer needs. In this chapter we have stressed the advantages as well as
the caveats associated with online research. Perhaps the biggest mistake the market researcher
could make would be to view online research as simply a time- and cost-saving extension of
traditional modes of data collection. As new technologies continue to be developed, they are
tested for applicability in marketing research settings, and refined so that marketers are able to
better identify the needs and wants of today’s consumers.
95
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
www.azcentral.com/business/articles/2009/07/20/20090720biz-cellnation0721side.html
Baker, K. G., Hozier, G. C., Jr., & Rogers, R. D. (1999, February). E-Mail, mail and fax survey
research: Response pattern comparisons. Paper presented at the 6th Annual Meeting of the
American Society of Business and Behavioral Sciences, Las Vegas, NV.
Becker, E. E. (1995, January 2). Automated interviewing has advantages. Marketing News, 29, 9.
Bourque, L. B., & Fielder, E. P. (2003a). How to conduct telephone surveys. Thousand Oaks,
CA: Sage.
Bourque, L. B., & Fielder, E. P. (2003b). How to conduct self-administered and mail surveys.
Thousand Oaks, CA: Sage.
www.busreslab.com/articles/article3.htm
www.cnn.com/
Davern, M., Rockford, T. H., Sherrod, R., & Campbell, S. (2003). Prepaid monetary incentives
and data quality in face-to-face interviews. Public Opinion Quarterly, 67, 139–147.
Dillman, D. A. (1978). Mail and telephone surveys: The total design method. New York: Wiley-
Interscience.
Dillman, D. A. (2000). Mail and internet surveys: The tailored design method (2nd ed.). New
York:Wiley.
Eisenfeld, B. (2003, March 3). Phone interviews may garner more data. Marketing News, 37, 57.
European Society for Opinion and Marketing Research. (1989). Distinguishing telephone
research from telemarketing: ESOMAR guidelines. Amsterdam: ESOMAR.
Goldstein, K. M., & Jennings, M. K. (2002). The effect of advance letters on cooperation in a list
sample telephone survey. Public Opinion Quarterly, 66, 608–617.
Kay, D. (1997, January 6). Go where the consumers are and talk to them. Marketing News, 31,
14.
Malhotra, N. K. (2004). Marketing research: An applied orientation (4th ed.). Upper Saddle
River, NJ: Pearson Education.
www.marketingcharts.com/interactive/home-internet-access-in-us-still-room-for-growth-8280/
Miller, T. W. (2001, September 24). Make the call: Online results are mixed bag. Marketing
News, 30.
96
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
www.nationalgeographic.com/geosurvey/
Oldendick, R. W. (1994). The answering machine generation. Public Opinion Quarterly, 58,
264–273.
Rochford, L., & Venable, C. F. (1995, Spring). Surveying a targeted population segment: The
effects of endorsement on mail questionnaire response rate. Journal of Marketing Theory and
Practice, pp. 86–97.
Smith, K. T. (1989, September 11). Most research firms use mall intercepts. Marketing News, 23,
16.
Wiseman, F., Schafer, M., & Schafer, R. (1983). An experimental test of the effects of a
monetary incentive on cooperation rates and data collection costs in central-location
interviewing. Journal of Marketing Research, 20, 439–442.
Xu, M., Bates, B. J., & Schweitzer, J. C. (1993). The impact of messages on survey participation
in answering machine households. Public Opinion Quarterly, 57, 232–237.
97
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 6
QUALITATIVE RESEARCH AND OBSERVATION
In Chapter 4 we discussed communication as a means of obtaining information from
respondents. Respondents provide information by answering questions (via an interview) or by
having their behavior observed. It will be recalled that at the extremes, an interview will be
either structured or unstructured in organization and that it may be either direct or indirect in
questioning. In this chapter we consider the use of indirect interviews of all kinds, special types
of unstructured-direct interviews, and observation as means of obtaining information from
respondents. We conclude the chapter with an assessment of direct and indirect research
techniques.
FOCUS GROUP INTERVIEWS
Perhaps the best-known and most widely used type of indirect interview is that conducted
with a focus group. As we mentioned in Chapter 2, a focus group interview involves group of
people jointly participating in an interview that does not use a structured question and- answer
method to obtain information from these people. A trained moderator conducts the interview
with a group of, ideally, 8 to 12 (but increasingly only 6 to 8) willingly recruited participants.
The composition of the group varies according to the needs of the client, especially the problem
under study. Although the technique is widely used in exploratory research, it also is useful in
nonexploratory research. Such applications include gaining greater understanding of consumer
wants, thoughts and needs; idea generation (e.g., problem, unmet needs, ideas for new products),
concept development and screening, tests for comprehension of promotion and communication
materials, and establishment of “opinion leader” panels.
It seems to be a well-accepted fact that focus groups work, especially when used with
other techniques. Practitioner researchers accept the idea that qualitative techniques like focus
groups, explore, define and describe. In contrast, quantitative methodologies measure, estimate
and quantify. These techniques are complements, not substitutes. Typically, the qualitative
research precedes quantitative research, but, when a company already has a substantial amount
of existing research data to prepare a quantitative survey questionnaire, the focus group can add
meaning to the survey findings (Garee & Schori, 1996).
Example: The Federal Duck Stamp Office of the Department of Interior, U.S. Fish and Wildlife
Service (Rydholm, 2000) focus group study. Migratory Bird Hunting and Conservation Stamps,
known commonly as duck stamps, are required for hunters of ducks. They are also popular
among stamp collectors. Funds from the sale of these stamps are used to help fund the
preservation of wetlands in the United States. Because the number of hunters has been declining,
there was a need to find a way to reach a new audience and to broaden the market for the stamps.
Focus groups were used during the development of a marketing campaign in which a certificate
bearing a duck stamp and stating that the recipient has helped preserve one-tenth of an acre of
wetlands could be purchased for $30. The objective was to appeal to bird watchers, hikers, and
other naturalists. Focus groups showed that the stamp itself was not enough to make the sale.
The certificate idea came directly from the focus group research. Focus groups were conducted
throughout the country, two groups per city—one with environmentally active people, the other
with people who were not predisposed against environmental issues. Once a range of appeals
98
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
was identified in the groups, the questioning centered on isolating the elements that made people
change their minds.
Example: is the Jacksonville, Florida, symphony orchestra’s use of focus groups to identify
lifestyle marketing issues to explore entertainment alternatives, and to provide some ideas about
what future audiences would want and expect from the orchestra (LaFlamme, 1988).
Raymond Johnson (1988) has identified four distinctive categories of focus groups on the
basis of examining tapes from the project files of several research companies. Johnson, a
practitioner, has defined each type of focus group by the adaptation of an interviewing technique
to answer one of four basic research questions. The focus group types are as follows:
• Exploratory studies of consumer lifestyles and probing to “just find out what’s on the
consumers’ minds these days.”
• Concept testing studies of how a group, without prompting, interprets a deliberately
sketchy idea for a new product or service. Potential users are able to react to a concept
still in its formative or experimental stage.
• Habits and usage studies deal with the real world of actual consumers. The topic is
framed by the moderator’s instructions to describe, usually by situation-specific
narratives, the details of personal experiences in using a particular product or service.
• Media testing in which participants are asked to interpret the message covered in media
usually seen in rough form is the fourth type. All types of media may be covered. Group
members talk about their understanding of the message and evaluate the extent to which
they find it credible, interesting, and emotionally involving.
The use of focus groups is not limited to consumer products and services. This technique
can provide a relatively easy and cost-effective way to interact with business consumers in
industries ranging from pharmaceuticals to computer software. The ways in which focus groups
are structured and conducted are similar for consumer-based and business-to-business groups,
except as identified by Fedder (1990).
A natural question, of course, is, “Why do focus groups work?” One view is that clients
are provided with a gut-level grasp of their customers. This means that a sense of what is unique
about customers is gained —their self-perceptions, desires, and needs that affect everything they
do. For more detailed discussions, see Bloor, Frankland, Thomas, and Robson (2001), and Fern
(1981). Qualitative research offers not just an intellectual comprehension of consumers but a
vivid, visceral recognition that affects, on a very deep level, how clients see, feel about, and deal
with their customers from then on. Some guidelines and questions to assist clients in observing
focus groups more effectively are discussed briefly in Exhibit 6.1.
One critical aspect of a focus group’s success is the moderator (Exhibit 6.2). The
moderator’s job is to focus the group’s discussion on the topics of interest. To accomplish this,
the moderator needs to explain to the group the operational procedures to be used, the group’s
purpose, and why the group is being recorded. Rapport with the group must be established and
each topic introduced. The respondents need to feel relaxed early on, and often moderator
humor helps this to happen. Once the respondents are comfortable, the moderator needs to keep
the discussion on track while not influencing how the discussion proceeds. A moderator has
99
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
done a good job if, after the topic has been introduced, the discussion proceeds with little
direction from the moderator. There must be synergy among the group members. Thus, a
critical measure of success is that the group members have talked with each other, not with the
moderator.
EXHIBIT 6.1 Observation of Focus Groups by Clients
As an observer, a client should be cognizant of certain things as he or she observes the dynamics of a
focus group in the “back room” from behind a one-way mirror. According to Judith Langer (2001), a
client should consider the following:
1. Determine your overall impression of the people in the group by looking at their sophistication level,
appearance, and the way they express themselves.
2. Do the respondents’ reactions support your assumptions? A way to assess whether the people are
atypical or less than fully honest is to have outside data.
3. Are there segments that seem to exist in the focus groups, perhaps based on psychographics?
4. Are there patterns that emerge after several groups? Watch out for making conclusions after just one
session. Look for variance! Do not count numbers.
5. A single comment by a respondent may be quite insightful.
6. Look at how people say things, not just what they say. Nonverbal communication can be valuable in
interpreting the verbal responses.
7. If a new product or product concept is involved, are the respondents enthusiastic or are they neutral
about it?
8. Although certain statements by respondents may appear to be complimentary, they may not really
be. This is known as a false positive and is something to be avoided.
9. Be aware of any contradictions that arise between what respondents say and what they report as
behavior.
10. Are respondents open to changing their minds, given other information?
11. After the session, talk with the moderator and ask him or her to put the responses in perspective.
The last suggestion may be difficult for a client to accept. Never take what is said as being personal.
Forget about ego and company politics.
12. Joining the participants of the focus group (after the focus group script is ended). Seeing and hearing
consumers, up close, has an impact that no set of data and no written report alone can have. It
makes the abstract real because it is human and individual.
When conducting focus groups with professionals, the practical aspects of focus group
use are somewhat unique. Special attention must be given to recruitment, type of compensation
or gratuity, convenience of the facility to be used, and the moderator. When professionals are
involved, the moderator’s interaction must be such that he or she is presented as an authority on
research in the field of the professional. At least a working knowledge of current technical
terminology is necessary.
100
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 6.2 The Make or Buy Decision for Focus Groups
Some marketing managers believe that almost one-half the cost of the research can be saved by having
internal marketing research people conduct the groups. As the costs of doing focus groups increase,
some companies consider hiring an “outsider” to conduct groups. In general, a managers’ argument for
having the groups done internally is based on the following:
1. Money can be saved as the cost of the moderator is eliminated.
2. Company personnel are more familiar with the product and will be able to ask the right
questions, recognize the nuances, and know when the group is not telling the truth.
3. The groups can be scheduled as the managers’ schedules permit.
4. There have been bad experiences with the outside moderators used in the past.
To an extent, these points are valid and could represent an advantage for doing it internally. However,
Thomas Greenbaum (1991) believes that when the total situation is considered, it is preferable to hire the
right outsider to conduct the focus groups. He lists the following reasons:
1. An outside moderator can be more objective than an insider, and is less likely to lead the group
in a specific direction.
2. Focus group moderation is a learned skill that is developed over time.
3. An outside moderator can be helpful in designing the groups and in developing an effective
moderator guide.
4. The lack of detailed product knowledge often is an advantage, not a disadvantage. Since
moderators pretend not to know much about the category being discussed, it is possible to ask
the seemingly dumb question or seek “help” from the participants that will generate information
to achieve the goals of the research.
5. There is less chance that the participants will refrain from showing concern or reacting negatively
to ideas when an outside moderator is used.
6. An outsider will be more objective in interpreting results than will an insider.
7. Clients work better by doing what they need to do from the back room, behind the mirror, rather
than by conducting the groups themselves.
It seems clear that much of the argument for using outsiders is based on their greater experience and
objectivity. Of course, there may be insiders who possess the experience and can, indeed, be
objective in doing focus group research. Greenbaum (1999) presents a practical perspective and a
more detailed guide of moderators.
In addition to in-person group interaction, a focus group can be conducted over the
telephone by use of a conference call. Respondents are recruited from across the country and are
told to call a toll-free number at a certain time to participate. Groups have included doctors, car
dealers, accountants, travel agents, and others for projects relating to product development,
promotion feedback, reasons why a product was not selling, and similar issues. Simon (1988)
listed some of the advantages of this approach:
1. Groups can have tremendous geographic diversity.
2. Travel costs can be virtually eliminated.
3. Recruitment is easier because you do not ask a respondent to spend an evening
traveling to, sitting in, and returning from a facility.
101
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
4. Mixed groups are no problem.
5. Bad weather generally has no effect on the carrying out of a group session.
6. The information from a telephone group is clean, concise, and to the point.
7. Overbearing respondents can be better handled without disrupting the group.
8. Concept testing is easy.
9. Researchers and clients do not have to go all over the country to put together a
sufficient number of representative groups from among the smaller sub-specialties.
One disadvantage of the telephone focus group is that the researcher loses the ability to
evaluate verbal response within the context of nonverbal communication. More recent
technology has allowed clients to view focus groups live without traveling to the geographic
areas where they are held. One company has developed video software capable of broadcasting
live focus group interviews from a nationwide network of independently-owned focus group
facilities to a conference room in a company’s own office. Clients view all of the action on a
large monitor and control two cameras that allow a full group view, close-up, zoom, or pan.
They can maintain audio contact with the moderator, video the complete sessions, and hold
“open-mike” post-group debriefings. Major advantages of this system include saving time and
money. Such video-system focus-group facilities have expanded to include global options.
Technology has had a dramatic impact on focus group structure and method.
Videoconferencing is being used to reduce costs. Related to this is conducting focus groups at
conventions, which has application for business-to-business situations and to situations where the
target market consists of professionals. With the right convention, a large concentration of the
target market can be accessed (Greenbaum, 1993). Focus groups conducted via the Internet can
replace face-to-face research. This so-called virtual research has the appeal of groups being run
more quickly and at lower cost. But, there is a cost! This cost is that the researcher loses the
ability to relate nonverbal communication by respondents to the verbal aspects. That is, body
language, voice inflection, facial expression, and interaction between people cannot be observed
(Miller, 1994).
INDIRECT INTERVIEWS AND QUALITATIVE RESEARCH
A number of techniques have been devised to obtain information by indirect means.
Most of these techniques employ the principle of projection. That is, the subject is given a
nonpersonal, ambiguous situation and asked to describe it, expand on it, or build a structure
around it. The person giving the description will tend to interpret the situation in terms of his or
her own needs, motives, and values. The description therefore involves a projection of the
respondent’s own personal characteristics to the situation described.
Projection techniques include word association tasks, sentence completion tests, and
interpretation of pictures and pictorial representations that have been developed as a means of
inducing people to project their feelings (see Table 6.1). These techniques have been most
widely used for studies of consumer products that are similar in quality, performance, and
price—notably for such products as automobiles, soaps and detergents, gasoline, cigarettes, food
products, beverages, and drug sundries. Projection techniques can stimulate a relaxed free flow
of associations that tap and identify deep, unacknowledged feelings to a degree not usually
possible by other research techniques. Because projective techniques are designed to bypass
people’s built-in censoring mechanisms, they are useful in eliciting information about sensitive
or threatening topics and products.
102
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 6.1 Classification of Projective Techniques
Technique Response Requested of Subjects
Construction To create a story based on a stimuli presented
Ordering
Item preference test
To order stimulus items by preference
Expressive techniques Play a role; draw a picture of a person doing something;
describe a character in a simulated situation
Association
Word-association test
To reply to a stimulus with the first word, image, or percept
that comes to mind
Completion
Sentence-completion test
To complete incomplete expressions, images, or situations
Most indirect interviews are at least partially structured in that they are conducted using a
predefined set of words, statements, cartoons, pictures, or other representation to which the
subject is asked to respond. However, the interviewer is usually allowed considerable freedom
in questioning the respondent in order to ensure a full response. Indirect interviews, therefore,
are commonly neither fully structured nor fully unstructured; ordinarily they utilize both types of
question. Within the marketing research community these techniques constitute qualitative
research techniques.
The Third-Person Technique
The simplest way of obtaining information through indirect questioning of a respondent
is to ask for the view of a neighbor, an (unnamed) associate, or some other person whose views
on the subject at hand might reasonably be known. This permits the respondent to project his or
her own views with no feeling of social pressure to give an “acceptable” answer.
An early study using a variation of this technique that has come to be regarded as a
classic is the study by Mason Haire (1950) on instant coffee. This study was conducted when
instant coffee was first being introduced. The purpose of the study was to determine the
motivations of consumers toward instant coffee in general and Nescafe, a brand of instant coffee,
in particular. Interviews of consumers had been conducted using a questionnaire employing
direct questions. Among the questions asked were “Do you use instant coffee?” and (if “No”)
“What do you dislike about it?” The majority of the unfavorable responses were of the general
content “I don’t like the flavor.” This answer was suspected to be a stereotype rather than
revealing the true reasons. An indirect approach was therefore chosen.
Two shopping lists were prepared that were identical in every respect except that one
contained “Nescafe instant coffee” and the other “Maxwell House coffee (drip grind).” These
shopping lists were shown alternately to a sample of 100 respondents, each being unaware of the
other list. Each subject was given the following instructions:
Read the shopping list below. Try to project yourself into the situation as far as possible until
you can more or less characterize the woman who bought the groceries. Then write a brief
description of her personality and character. Wherever possible indicate what factors
influenced your judgment.
The results were quite revealing. The descriptions given were summarized as follows (Haire,
1950, p. 652):
103
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Forty-eight percent of the people described the woman who bought Nescafe as lazy;
four percent described the woman who bought Maxwell House as lazy.
• Forty-eight percent of the people described the woman who bought Nescafe as failing
to plan household purchases and schedules well; 12 percent described the woman
who bought Maxwell House this way.
• Four percent described the Nescafe woman as thrifty; 16 percent described the
Maxwell House woman as thrifty; 12 percent described the Nescafe woman as
spendthrift; 0 percent described the Maxwell House woman this way.
• Sixteen percent described the Nescafe woman as not a good wife; 0 percent described
the Maxwell House woman this way; 4 percent described the Nescafe woman as a
good wife; 16 percent described the Maxwell House woman as a good wife.
The implications of these findings seem clear. The woman using the instant coffee was
characterized as being lazier, less well organized, more of a spendthrift, and not as good a wife
as the one using the conventional coffee. These imputed characteristics must have been the
result of the respondents’ projecting their own feelings toward instant coffee in their descriptions
of the woman using it.
This study has been replicated a number of times. The general acceptance of instant
coffee and the change in dietary habits since the original study have resulted in different findings
in the more recent studies. Nevertheless, the original study remains as a classic application of
the third-person technique of research.
Word Association Tests
Word association tests consist of presenting a series of stimulus words to a respondent
who is asked to answer quickly with the first word that comes to mind after hearing each. The
respondent, by answering quickly, presumably gives the word that he or she associates most
closely with the stimulus word.
Word association tests are simple and easy to use, and offer powerful insights into the
perceptions and associations related to the concepts being tested.
Sentence Completion Tests
Sentence completion tests are similar to word association tests, both in concept and in
use. A sentence stem (the beginning phrase of a sentence) is read to the respondent, who is
asked to complete the sentence quickly and with the first thought that occurs to him or her.
Recognizing that people may react in more than one way to a sentence stem, participants are
asked to fill in the sentence several times rather than once. This increases the likelihood of
uncovering all major feelings on a topic.
Sentence completion tests provide a top-of-mind association between the respondent and
the topic/product/subject being investigated. This data is easy to collect, but is difficult to
analyze so that an accurate perspective is obtained of the differences between groups and the
meanings of the comments.
Example: A sentence completion was used in a study of automobile buying to probe the
motivations of automobile buyers and thereby provide a sounder basis for advertising.
Analysis of selected responses of men and women to two of the sentence stems illustrates
how inferences of motivational influences can be drawn through the use of this technique
(Newman, 1957, pp. 227–228).
104
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Sentence stem: When you first get a car . . .
Women’s responses:
. . . you can’t wait till you drive.
. . . you would go for a ride.
. . . you would take rides in it, naturally.
. . . you would put gas in it and go places.
Men’s responses:
. . . you take good care of it.
. . . I want to make darn sure it has a good coat of wax.
. . . check the engine.
. . . how soon can I start polishing it.
Sentence stem: A car of your own . . .
Women’s responses:
. . . is a pleasant convenience.
. . . is fine to have.
. . . is nice to have.
Men’s responses:
. . . I would take care of it.
. . . is a good thing.
. . . absolutely a necessity.
Women’s responses indicated that for them a car is something to use and that pride of
ownership stresses being seen in the car. For men a car was something for which they should be
protective and responsible. Their emphasis was on examining the car and doing things to it.
Men appeared to feel closer to their car and regarded it as more of a necessity than did women.
The Depth Interview
The unstructured informal interview in marketing research is referred to as a depth
interview. It is used to explore the underlying predispositions, needs, desires, feelings, and
emotions of the consumer toward products and services. Depth interviews, if conducted in
sufficient detail, produce accurate and understandable qualitative information about the research
problem (exploratory studies), or for concept or habits and uses studies.
Insofar as obtaining information on motivations is concerned, the concept of “depth”
refers to the level at which underlying motivations are uncovered. When conducting depth
interviews, the key is to ask questions that probe for more depth in underlying relationships and
motivations. Mariampolski (1988) created a useful list of different probes that can be used:
1. The silent probe (use eye contact and body language)
2. Request elaboration
3. Request definition
4. Request word association
5. Request context or situation of use, occurrence or activity
6. Shift the context or situation
7. Request a clarification
8. Request a comparison
9. Request classification or typology
10. Compare and contrast to a previous statement
11. Challenge veracity
12. Challenge completeness
13. Confrontational probe
14. Echo probe
15. Interpretive probe
16. Summary probe
17. Purposive misunderstanding
18. Playing naive
19. Projective probe
105
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
These alternative techniques of probing have varied effectiveness, but are all useful tools
in interviewing. By following leads and cues provided by respondents, phrasing questions to
continue the flow and pattern of the conversation and to maintain the rapport established, the
competent interviewer can explore and probe the underlying motivations of the respondent.
Dissatisfaction with the group influence and the high cost of focus groups, together with
certain evolving factors in the marketing environment, have led to recent increased use of the
individual depth interview. The depth interview is ideal for obtaining from consumers anecdotes
of times they used a product or service, and such “stories” provide the marketer with a good view
of what products mean to consumers. When used this way, the result of the interview is usually
presented verbatim. Telephone interviewing has proved to be effective in obtaining such
consumer stories. The depth interview offers insights with depth, whereas focus groups offer
information with breadth. Thus, each has its place in the total set of methodologies from which
marketing researchers can draw.
There are three distinct research situations where depth interviews can be used (Kates, 2000):
1. Depth interviews are useful in exploratory research to obtain background information to
support the development of a quantitative survey instrument.
2. Depth interviews may be the sole research method used such as hard-to-reach groups, or
when focus groups are not feasible. Obviously, results normally cannot be statistically
significant, but results can be projected if the size of the sample interviewed is large
enough (at least 60 percent of the population, indicating, of course, a relatively small
population).
3. Depth interviews can obtain information on a subject without being biased by the group
dynamic that often occurs in a focus group. For instance, in the case of a rollout of a new
product the marketer may not want respondents to be influenced by the views of others.
Another approach, labeled ZMET, for Zaltman Metaphor Elicitation Technique was
developed by Zaltman (2003). This approach uses a rigorous in-depth interviewing procedure to
understand the deep meaning customers attach to product decisions. Founded in the assumption
that 95% of human thoughts and feelings are unconscious, we process current information
through the context of past experiences to provide their meaning and interpretation to us.
ZMET identifies the metaphors that shape decision making and behavior. For example,
we can identify the surface metaphors (statements like “He invests his time wisely”), thematic
metaphors (themes: “time is like money”) and deep metaphors (“time is a resource”) that help us
understand and evaluate the thinking and emotions that precede our behavior (what we hear, feel,
think, say and do).
The ZMET interview focuses on photos or drawings that subjects are asked to bring with
them, representing their feelings about the topic of discussion for example, donating to a charity
or university, or purchasing a specific product or brand). These representations are discussed in
depth to identify the deep metaphors associated with the topic.
To better understand this result, Olson, Waltersdorff and Forr (2008) provide, in Figure
6.1, a tabular explanation of how a ZMET analysis can benefit the marketing process.
106
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 6.1 Applications of ZMET in Product Marketing
Source: Olson, Waltersdorff and Forr (2008)
Table 6.2 shows examples of these deep metaphors that each of us in expressing thoughts
and emotions. Interviewees are asked to explain their true feelings about objects of concern with
images, and not just words.
Table 6.2 Deep Metaphors and Expressions in Conversation
DEEP
METAPHORS EXPRESSIONS IN THE VIEW
Physicality References to bodily functions and senses such as taste it; feel it; pick up, ingest, see my point;
hurts me
Orientation References to spatial orientation--up/down, higher/lower; bigger/smaller; upright/lie down;
front/back.
Pleasure/pain References to the positive (or negative); to things that give one pleasure; feeling good or bad;
References to enjoyment, fun, happiness, euphoria, well-being ... or to he opposite (fear, disgust,
pain); Feeling good versus hurting, physically or emotionally.
Entity Considering an intangible idea, concept as a thing, a physical object, an entity; e.g.; "I can't get my
breath"
Balance References to equilibrium, stability, equalize or compensate; Including both sides; Images of
scales, teeter-totter, balance beam; References to reciprocity--give and take; References to
'stable" emotional states such as calm, relaxed, serene; Feeling 'right" with the world.
Connection or
Linkage
References to connecting to things or people; Making an association; References to linking or
attaching; To be a part of; to not be isolated from; Liking or loving someone or something;
References to getting in touch with yourself; find your true self.
Resource References to having/getting the requisite knowledge, energy, tools, or materials to accomplish
some task; Having or getting help and assistance from others (we are a team, I need support);
107
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Container References to being in (or out) of a place (house, room); References to keeping or storing;
References to "in" and "out;" Keeping things out as well as in; Being wrapped up ... or out in open.
Motion or
Movement
References to moving (flowing, traveling, running or walking); References to movement actions
(hurrying, getting going); Keep moving; Keep it going.
Journey References to taking a trip; Following a path, choosing a direction; Getting there; Journey of life.
Transformation References to changing from one state to another-physical or emotional; Becoming something or
someone else; References to evolving, maturing, growing;
Time References to the passage of time; Reference to past events or historical perspectives; Images of
clocks and watches; References to old memories, remembering past events.
Nature References to nature, outdoors, natural world; Specific images of nature--rain forest, desert,
woods; References to pure, unadulterated, pristine, uncontaminated and to wild, untamed,
chaotic, stormy. References to breeding, evolving, growing.
Force References to power, a powerful presence, or a source of energy; References to the
consequences of force (getting hit; slammed, impact)
Fight vs. Flight References to war; fights, battle, attacking, aggression; Choose your battles; References to
weapons; Avoid a fight; Don't get involved; Running away or hiding from something; Ignoring an
issue.
Knowing and
knowledge
References to knowing, understanding, learning. Gaining knowledge and understanding through
study; Knowledge and wisdom; Comprehension; insight. In contrast, the unknown, ignorance,
stupidity; Inability to comprehend and understand something.
Contrast or
Paradox
References to looking at both/two sides; Comparing opposites; Juxtaposition of opposites;
References to paradox--being one thing and its opposite at same time;
Personal
expression
Reference to things that express one's personal goals, values, points of view ... to one's self or to
others.
The Ideal Reference to the ideal object, situation, feeling. Statements about one's ideal self. References to
perfection, the perfect one.
System References to machine metaphors (wheels and gears, well-oiled machine); or a constructed
process or approach for solving a problem; A set of rules or procedures to accomplish a task;
Following a ritual
Sacred
(Profane)
References to divine or spiritual qualities; Symbolic cleanliness, purity, vs. dirty, impure, devilish,
malevolent
The Mother;
Care giving
References to love, fondness; warm relationships with family and friends; Caring for or nurturing
others, animals, plants; Being a caring resource for others; taking responsibility for and supporting
and helping others
Masculine/
Feminine
References to gender--male or female; Having masculine or feminine traits; Attributions of gender
to nonliving things, ideas, concepts.
Birth/Death References to beginning or end of life, of an idea, or concept; Being rejuvenated or brought back
to life (reborn); References to decline, dying, or death.
Quantify/
Measure
References to quantification; Keeping track; Counting; Yardsticks, scales, sextons; Computing
amounts and quantities
Holism/
Completeness
References to being whole, entire, complete, not lacking anything, not having weaknesses that
compromise any other part of the whole.
Source: Jerry C. Olson, Olson Zaltman Associates (www.olsonzaltman.com).
As a final note for conducting in-depth interviews, it is important to use empathy in
understanding and appreciating someone else’s beliefs, attitudes, feelings, and behaviors, as well
as the social factors that influence their behavior. Many standard qualitative techniques neglect
empathy. Specific guidelines for conducting empathic interviews include the following (Lawless,
1999):
108
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
1. The researcher needs to imagine himself/herself in the respondent’s situation and must
listen to the respondent fully.
2. Do not be hindered by the discussion guide; react and improvise as needed.
3. Ask open-ended, non-leading questions that start with how, what, and why.
4. Avoid self-referencing by setting aside thoughts, preconceptions, and interpretations.
5. Challenge generalizations by asking for specific examples.
6. Probe non-judgmentally to understand the person’s beliefs, feelings, and behaviors.
7. Let the respondent reveal himself/herself through personal studies.
The next section demonstrates a depth interviewing technique called means-end analysis,
which focuses on discovering the product attributes/features, benefits and values that motivate
the use or purchase of a product.
MEANS-END ANALYSIS
Means-end analysis, also known as Laddering and Means-End Chain, is an in-depth, oneon-one
interviewing technique that is directed at identifying the linkages people make between
product attributes (the means), the benefits derived from those attributes (the consequences), and
the values or end state (the ends) that underlie the desire to achieve the consequences.
The premise of means-end analysis is that consumers make attributed based decisions, selecting
those attributes that lead to their desired benefits or consequences and that are consistent with
their achieving, fulfilling, or identifying with specific personal values.
Laddering interviews, as an in-depth interviewing technique, employ structured dialogues
that identify the most important attributes, the benefits (consequences) derived from those
attributes and the values linked to the consequences in a given usage situation (Figure 6.2). The
interviewer moves up and down the means-end chain, identifying the hierarchical structure of the
product attribute / product benefit / personal benefit / personal value components and linkages.
The interviewer usually asks questions like “Why is that important to you?” or “What do you
gain from having that characteristic?” Interviews typically last between 45 and 90 minutes and
are recorded for analysis and preparation of the resulting laddering maps of the components and
linkages.
Figure 6.2 Understanding Consumer Decision Making
109
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Laddering focuses on both the positive and negative linkages between attributes and
consequences that are important in choosing a brand. This laddering of the reasons that underlie
a decision provides a much deeper understanding of the consumer than does a traditional “ratings
survey” of product attributes, but does not penetrate as deeply as does the ZMET analysis.
The first task of a laddering exercise is to elicit the attributes that are important in
distinguishing between brands. Exhibit 6.3 identifies a series of approaches that might be used
to elicit the attributes that are most important in distinguishing between brands. In practice,
several different methods may be used to capture a full range of meaningful distinctions between
a brand and its competitors. For example, a laddering interview might start with top-of-mind
imaging to understand general product-category beliefs, then increase in brand-related specificity
by asking about the usage context (contextual environment), and finally about alternative usage
occasions. Other tools, such as a worksheet (Exhibit 6.4) might also be used.
EXHIBIT 6.3 Methods for Eliciting Brand Attitudes
A variety of methods can be used in marketing research to elicit brand attitudes:
• Top-of-mind imaging. The respondent gives positive and negative associations for the brand or product
category, along with reasons why the characteristic is viewed that way. This line of questions uncovers
the attributes and consequences that distinguish the characteristic.
• Grouping similar brands. Grouping identifies similar and dissimilar brand groupings within a product
category and the reasons for this perceived similarity or dissimilarity. The primary reasons, most
important attributes, and most representative brands are identified and attributes and consequences are
laddered.
• Contextual environment. The usage context for a brand or product can be described either as physical
occasions (place, time, people), or need state occasions (relaxing, rejuvenating, building relationships,
feeling powerful, reducing stress, and getting organized). A brand or product is associated with a usage
context.
• Preference, usage, similarity and dissimilarity differences. Comparing brands based on personal
preference or usage is commonly used to distinguish between brands. Similarity and dissimilarity
groupings also provide a direct method of distinguishing between brands. Questions of why Brand A was
grouped differently or ranked higher than Brand B produce elicitations of attributes and consequences.
• Timing of purchase or consumption. Timing issues are often related to product or brand choice and usage.
For example, a respondent might be asked to identify products used for relief of a stuffy nose into several
stages like onset, full-blown, and on-the-mend, or daytime and nighttime. Then the respondent would
relate which brands were preferred for each time-related stage.
• Usage trends. Dialogues about past and expected future usage of a brand help to elicit attributes and
consequences that lead to different usage patterns. For example, respondents may be asked, “Will this
brand be used more often, less often, or about the same as you have used it in the past?” Then, reasons
for increased, decreased, or unchanged usage are discussed.
• Product or brand substitution. Product and brand substitution methods elicit the degree of similarity of
perceived attributes and consequences associated with usage. When questions are asked about the
degree of substitutability, attributes and consequences are discovered that inhibit or promote
substitution (attributes or consequences that need to be added or removed for substitution or trial to
occur. The respondent can first sample or be given a brand descriptions, followed by questions like: how
110
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
likely would you be to substitute (name of the new brand) for your current brand for this occasion—why
is that?
• Alternative usage occasions. Alternative uses are presented to determine if and why the brand is present
or absent from the choice set. Questions might be phrased to ask: why would you consider using Brand A
for this occasion, or what is keeping you from using Brand A for this occasion now? Both positive reasons
why a brand fits a new occasion and negative reasons why it does not fit can be elicited and laddered. .
Adapted from by Reynolds, Dethloff, and Westberg, 2001, and Reynolds and Whitlark, 1995
One study using the Reynolds and Gutman (1988) approach to analyzing and quantifying
the results of laddering exercises, attempted to understand the development of a new bank credit
card.
Using the following types of probing questions, the study identified nine critical
consumer attributes for a new card: no annual fee, status, low interest rate, added value features,
acceptance, credit limit, ability to carry a balance, location of the sponsoring bank, and
availability.
• Why do you prefer these cards?
• How do they benefit you?
• How are they different from others?
• What is your least favorite banking or credit card and why?
• What specific things did you dislike about this card?
• Have you had negative experiences that affect your current preferences?
• Under what conditions would you consult these rejected sources for a card?
• How does your preferred card affect your life?
• How does this card make you feel as a consumer?
• What does this preferred card do for you that the rejected cards don’t?
• During what period of your life are you most likely to need these services?
• During which period has it most benefited you? Why?
• Describe the best personal experience you have ever had concerning credit cards.
• How has that experience changed your life?
• Do you seek out similar experiences today?
• How would you describe a card that could help you achieve this?
These attributes were found to be linked to 12 benefits (consequences) that were
perceived as part of card usage: not feeling cheated, independence, convenience, dependability,
saving money, financial responsibility, freedom, establishing a credit history, power, security,
supporting the community (local card), and confidence.
Finally the consequences were linked to four personal values: family concerns, improved
self-esteem, improved quality of life, and peace of mind. This analysis is actually mapped to
show the hierarchy of attribute relationships.
In Figure 6.3, four distinct sub-hierarchies are shown that focus respectively on the four
values. Means-end analysis provides a valuable platform upon which media can be developed.
In this case, it is easy to imagine four separate commercials, each stressing a different set of
attributes and consequences with overall personal value appeals to “Providing for the Family,”
“Superior Quality of Life,” “Peace of Mind,” and “Heightened Self-Esteem.” Reflection upon
the most effective credit card commercials shows that they are well targeted toward these very
same appeals.
The advantage of the benefit chain is that it is relatively easy to administer without a
highly sophisticated interviewer. The only key decision is when to stop probing.
111
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 6.3 Hierarchical Value Map
SOURCE: Gibby, Holland, Lee, Lundell, Merrill, and Robison, 1995.
OTHER TECHNIQUES
Other variations of projective techniques have been used that are similar in nature but
different in form from those already described. The repertory grid, for example, is a partially
structured technique that requires the respondent to compare objects along dimensions that he or
she selects. The usual procedure is to present the interviewee with a pack of cards on which
brand names or other stimuli are printed. The respondent is asked to cull unfamiliar brands
(stimuli) from the pack, and three cards with familiar brand names (stimuli) are selected at
random. Following this, the respondent is asked to describe a way in which any two of the
familiar brands (stimuli) are like each other and different from the third. The respondent is then
asked to rate all the brands (stimuli) with respect to this dimension. The response may be in the
form of a paired comparison, a ranking, or a numerical rating on a scale. This process is
repeated using three different brands (stimuli) until the dimensions of the respondent are
exhausted. Additional respondents are interviewed until no new dimensions are given. On the
average, 40 interviews are required to identify most of the relevant dimensions.
An example of the use of the repertory grid is a study done for a major producer of frozen
foods who was interested in introducing a new type of toaster item that could be made to taste
like a variety of freshly baked products. The research director was interested in the attributes of
bakery items that consumers use to distinguish one product from another. The stimuli were 20
products as shown in Table 6.3. The study resulted in a list of 22 attributes that were scaled as
112
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
bipolar scales (see Chapter 7) for further analysis, as shown in Table 6.4. The stimuli and
attributes can be set up as a grid whereby respondents can then rate each stimulus (i.e., bakery
item) on each attribute.
Table 6.3 List of Food-Item Stimuli Used in Repertory-Grid Study
Stimulus Stimulus
1. Toast pop-up
2. Buttered toast (white)
3. English muffin and margarine
4. Jelly donut
5. Cinnamon toast
6. Blueberry muffin and margarine
7. Hard rolls and butter
8. Toast and marmalade
9. Buttered toast and jelly
10. Toast and margarine (white)
11. Cinnamon bun
12. Danish pastry
13. Buttered toast (rye)
14. Chocolate chip cookie
15. Glazed donut
16. Coffee cake
17. Apple strudel
18. Toasted pound cake
19. Corn muffin and butter
20. Bagel and cream cheese
SOURCE: Green, Tull, and Albaum, 1988, p. 712.
Table 6.4 Twenty-Two Bipolar scales Found in a Repertory Grid
SOURCE: Green, Tull, and Albaum, 1988, p. 713.
113
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 6.4 discusses the use of protocols (Ericsson & Simon, 1984). This technique allows
respondents to respond freely without intervention of an interviewer.
EXHIBIT 6.4 Protocols for the Qualitative Research Tool Kit
A protocol is a record of a respondent’s verbalized thought processes while performing a decision task or
while problem solving. This record is obtained by asking the respondent to “think out loud” or talk about
anything going through his or her head while performing the task. Protocols can be collected either in a
laboratory situation while the respondent is making a simulated purchase or in the field while an actual
purchase decision is being made.
Protocols can be recorded concurrent with the task or retrospective when the verbalizing aloud is done
just after the task has been finished.
In contrast to traditional survey methods, protocol methodology allows a person to respond freely in his
or her own terms in relation to the actual choice task or decision situation. The form and particular
stimuli to which the research subject should respond is not defined or specified by the researcher.
Protocols can be useful in studying brand choice, product categorization, product usage patterns and
attitudes, and the impact of shopping environment and situational variables on behavior.
Example: Comcast, a major provider of cable TV, telephone and Internet services offered an unsuccessful
self install kit that often resulted in customers calling for a technician to complete the installation. A
protocol analysis was conducted when individuals in a laboratory situation were give a self-install kit and
asked to complete an install while narrating their activities. Comcast identified the key problem areas in
the install process and was able to revise installation instructions and provide an 800 support line, thereby
making significant reduction in install service calls.
Story completion, an extension of the sentence completion technique, consists of
presenting the beginning of a situational narrative to a respondent, who is asked to complete it.
The general underlying principle is that the person will project his or her own psychological
interpretation of the situation into the response. For example, the situation could be formulated
as follows: “Last weekend my partner and I were deciding which jewelry store to visit for a
purchase. When I mentioned XYZ, my partner remembered the last visit there. Now you
complete the story.”
OBSERVATION
The remaining major method of collecting qualitative information is through observation.
Observation is used to obtain information on both current and past behavior of people. Rather
than asking respondents about their current behavior, it is often less costly and/or more accurate
if the behavior is observed. We clearly cannot observe past behavior, but the results of such
behavior are often observable through an approach known as the case study or customer case
research. This exploratory qualitative methodology traces the stories of people, circumstances,
decisions and events leading to actual purchase decisions through one-on-one interviews
(Berstell & Nitterhouse, 2001). The case study approach allows for determining any underlying
patterns and may uncover unforeseen problems and unexpected opportunities. Some key
characteristics of this approach are (Berstell, 1992):
• Case studies uncover motivations through demonstrated actions, not through opinions.
• Studies are conducted where a product is bought or used.
114
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Observation and documentation are used to stimulate questions and corroborate
responses.
• Case studies can access multiple decision makers because they are done on site.
• Case studies require researchers who are essentially “market detectives” rather than
“census takers.” Such detectives must have the skills necessary to continue asking “why”
questions until answers emerge that show and explain motivations.
Observation may be used as the sole means of collecting data or, as is frequently the case,
it may be used in conjunction with other means. It is a method that should always be considered
in designing marketing research investigations that call for information on past or current
behavior. In some circumstances, observation is the only means of collecting the data desired.
Respondents often cannot and sometimes will not report information accurately. Brands usage
reports of well-established brands generally show a “halo effect,” an upward bias reflecting the
prestige the respondent associates with the use of the brand. Many companies, for example, have
found that respondent reports of brand purchases of products vary widely from the actual brand
of product that the consumer has on hand.
Other examples include a food retailer who tested a new type of shelving for canned
goods by observing shoppers as they used the new shelves, and a toy manufacturer who, in a
laboratory setting, observes children playing with product prototypes. In all these cases other
research techniques could have been used, but observation was preferable.
There are some problems in using observation. One concern is with selective perception.
Since human perception is quite selective, what people observe depends upon their backgrounds.
Thus, what is observed, and reported as such by an observer, depends upon who is doing the
observing. The second potential problem is that the behavior being observed is not
representative. It may be that what has been observed is the “exception” rather than the rule.
That is, the behavior observed may be a unique incident. Of particular concern is whether those
being observed know they are being observed. If a person is being observed, his or her behavior
may not be a “true” behavior. Also, the presence of others (e.g., in a crowded retail store) may
influence behavior at the time. Thus, the situation and setting are critical to the observation
experience.
The Audit
Audits of both distributor inventories and consumer purchases are widely conducted to
understand purchase patterns. The distributor audit is the more widely known of the two. The
commercially available Nielsen Retail Index, an audit of retail stores performed regularly, was
described in Chapter 2. As indicated there, data from this and audits available through other
research agencies provide estimates of market size, market share, geographic pattern of the
market, seasonal purchasing patterns, and results of promotional and pricing changes.
The pantry audit of consumer homes is the second type of audit that is sometimes
performed. In this type of audit, the field worker takes an inventory of the brands, quantities,
and package sizes that the consumer has on hand. When this type of audit is performed on a
recurring basis, inconspicuous labels may be attached to the package showing the date the item
was first included in the inventory. When the audit is combined with questioning of the
consumer, an estimate of usage may be made. The pantry audit is relatively expensive in terms
of data obtained, compared with a self-reporting consumer panel, however. Its use has declined
as the use of consumer panels has increased.
115
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Recording Devices
A number of electromechanical devices for “observing” the behavior of respondents are
in use in marketing research. Some of these devices are used primarily in laboratory-type
investigations and others are used to record behavior in its natural setting. Types of recording
instruments used in laboratory studies are the eye/pupilometric camera, and the
psychogalvanometer. Three of the devices used in noncontrived situations are the video camera,
and the Audimeter.
The “observing” of respondent behavior in a laboratory situation with the aid of
recording devices has been largely confined to the pretesting of advertising. Eye/pupilometric
cameras, for example, are specially designed cameras that record the pupil size changes and eye
movements to specific materials on a page. Subjects are fitted with headgear and software is
trained to the individual user. Subjects are then given an advertisement or other print/text media.
The computer records and maps their line of eye travel, the ad sections that attract attention
initially and sequentially, the relative amounts of time and points of focus in looking at images,
and which portions of the copy are actually read, and so on.
Pupil size fluctuations are also recorded. The dilation and restriction of the pupil of the
eye has been found to correlate with the degree of interest aroused by the visual stimulus.
Interesting or arousing stimuli result in the dilation of the pupil. An advertisement or a product
that is of interest to the subject will be evidenced by greater dilation of the pupil, indicating
degree of interest.
The psycho galvanometer measures the extent of the subject “responds” to the
advertisement. The principle involved is that the perspiration rate of the body is increased by the
amount of stimulation provided by an advertisement. This relationship is measured as the
change in electrical resistance in the palms of the subject’s hands.
Other devices are also used for “observing” behavior under laboratory conditions. In
general, all such devices have the advantage of permitting careful and detailed observations of
behavior that could not be made otherwise. They have the added advantage of providing
permanent records of the behavior observed. In using these devices, however, one should always
keep in mind two important questions:
1. Is the behavior we are observing a valid predictor of the behavior we want to predict?
2. Are the subjects behaving as they would in a natural situation?
The answer to the second question can clearly be in the affirmative if the observation is
made outside the laboratory and in the natural situation, such as in ethnographic research.
Hidden video cameras, for example, are used in many situations to record respondent behavior.
Direct Observation
Direct observation of people and how they behave in situations of interest is a commonly
used method of collecting information. Many studies have been made of shopping behavior to
determine the relative effects of such variables as displays, availability, and reliance on
salesperson advice on the brand selected and the quantity purchased. Supermarkets and
department store managers continually rely on observation of traffic flows and length of waiting
lines to determine the proper location of the various lines of products and the number and
location of salespeople and cash registers. An important consideration in the location of banks,
retail stores, and entire shopping centers is the amount and pattern of traffic at alternative sites.
116
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Information obtained from direct observation of purchasers can be highly useful in
helping to answer such questions as
• Who actually buys the product?
• Do they appear to be influenced by an accompanying person?
• To what extent do brand choices appear to have been made earlier versus at the point of
purchase?
• What proportion of shoppers appears to check prices?
• What proportion of shoppers studies the package before purchase?
Unobtrusive Measures
Observation is the method of data collection underlying a set of methods known as
unobtrusive measures. By their very nature these are nonreactive measures. Included are all the
types of data collection mentioned in this section.
One such application was a study performed for a manufacturer of frozen juice
concentrates who was considering changing the design and amount of information given
on the label. Before this change was made, information was needed on the extent to
which consumers actually read information on labels. Hidden cameras were stationed in
a sample of supermarkets in front of the frozen food cases, and pictures were taken of
consumers selecting frozen juice concentrates. An analysis of these pictures indicated
that far more time was spent in the selection and more careful attention given to the label
than had previously been believed to be the case. It is not necessary for the camera to be
hidden in order for it to be a useful device for recording behavior.
In addition, other types of unobtrusive measures are traces and archives. Regarding
traces, studies of garbage can tell much about consumers. Garbage data can be used to examine
many aspects of consumption behavior (Reilly, 1984):
1. Brand and product type switching patterns
2. Market share estimation
3. Lifestyle patterns
4. Ethnic and social group differences
5. Media usage patterns
6. Free samples, deal packs, trial sizes, coupons
As an example, consider a “classic” study where alcoholic beverage purchases are
tracked in a “dry” community. This unobtrusive research would show which types and brands
are purchased and consumed by examining residents’ garbage. In this type of unobtrusive trace
research, there are some biases that can arise. While the chose example and explanation focuses
on “garbage research”, the biases and advantages outlined below apply more generally to all
unobtrusive research. In general, biases may include the following:
Missing evidence. Compost piles, pets, garbage disposals, and recycling all affect
estimates of product usage and waste and can lead to inaccurate estimates.
Incomplete evidence. Discard of packaging material may not correspond to usage.
117
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Despite these limitations, it appears they are minor relative to the advantages, particularly
when the refuse analysis is part of an accepted project. Reilly (1984, 127–128) listed the
following advantages of garbage analysis:
• Unobtrusiveness. Under the current practices used (random sampling of household
refuse within selected neighborhoods), individuals are not aware that their refuse is being
analyzed.
• Nonreactivity. Because they are not aware that their consumption is being monitored,
consumers are unlikely to alter their behavior to appear more rational, more socially
acceptable, or more economical.
• Nonresponse. There is no selective bias in participation. Problems in estimating the
effects of an imperfect sample are not evident.
• Interviewer effects. Garbage is coded according to objective standards. There is limited
possibility for the recorder to consciously or unconsciously bias the outcome of the
analysis.
• Response effects. As a result of the unobtrusiveness of the procedure, respondents are
not capable of misrepresenting their behavior, either because they can’t remember
accurately or because they wish to create some type of favorable impression.
• Longitudinal analysis. The behavior of the same household can be observed over time.
Patterns of brand/type/product switching can be accurately observed.
• Satisfaction. Refuse provides an accurate measure of the waste of the product, which is
a good indicator of the consumer’s liking for the product.
• Completeness. Garbage provides information on products that are difficult to monitor
through traditional means. Illegal behaviors, purchases from unscanned stores, outlier
behaviors (such as beer consumption at parties), and socially sensitive aspects of
behavior are all amenable to quantitative analysis using refuse evidence.
• Consumption. Garbage analysis includes accurate measures of when products were
used, as opposed to measures of when the products were purchased. This makes
measurement of stock-up effects and cross-consumption possible.
For a more detailed discussion of unobtrusive methods, we refer the reader to the writings
of Webb, Campbell, Schwartz, and Sechrest (1966); Sechrest (1979); and Bouchard (1976). In
many cases it is desirable to use these methods in conjunction with other more traditional ones.
This is the process known as triangulation.
DIRECT VERSUS INDIRECT RESEARCH TECHNIQUES—AN ASSESSMENT
Opinion has been divided among practitioners about the role and relative merits of
indirect research techniques in marketing research. This division reflects in marketing research
the objectivist-subjectivist debate in the behavioral sciences in general. The controversy has
largely centered on three areas:
• The applicability of the techniques
• Sample selection and sizes employed
• Accuracy of utilizing disguised modes of obtaining such information
118
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
We will discuss each of these areas.
Applicability of Indirect Research Techniques
The basic premises leading to the use of indirect research techniques are as follows:
1. The criteria employed and the evaluations made in most buying and use decisions
have subconscious thoughts and emotions.
2. This subconscious content is an important determinant of what we hear, feel, think,
say and do when performing a choice behavior such as buying a product.
3. Such content is not adequately or accurately verbalized by the respondent through
direct communicative techniques.
4. Such content is adequately and accurately verbalized by the respondent through
indirect communicative techniques.
How valid are these premises? We have already seen that they are valid for some
problems. While correct, this is not a very satisfactory answer. It is more useful to review
situations in which indirect research information might reasonably be sought from respondents.
Four situational categories can be distinguished in which information might be sought from
respondents.
Category one is where the information desired is known to the respondent and he or she
will give it if asked. Direct questioning will therefore provide all of the needed information in
this situation. If the reason a consumer does not buy brand X tires is because he believes they do
not wear as well as they should, he will willingly say so given the opportunity.
Category two is where the information desired is known to the respondent, but he or she
does not want to divulge it. Matters that are considered to be private in nature, that are believed
to be prestige- or status-bearing, or that are perceived as presenting a potential respondentinvestigator
opinion conflict may not be answered accurately. That otherwise quiet and retiring
people sometimes buy powerful cars because it gives them a feeling of superiority on the
highway are not reasons that will likely be expressed openly and candidly in response to direct
questions. When underlying motivations of this general nature are believed to exist, indirect
techniques are well suited to elicit such information.
Third, the information desired is obtainable from the respondent, but he or she is unable
to verbalize it directly. When respondents have reasons they are unaware of, such as the
association of the use of instant coffee with lack of planning and spendthrift purchasing, properly
designed and administered indirect techniques can be highly useful for uncovering such
motivations.
Fourth, the information desired is obtainable from the respondent only through inference
from observation. In some cases motivations of respondents are so deep-seated that neither
direct nor indirect methods of questioning will bring them to the surface.
An experiment in which the same detergent in three different-colored boxes resulted in the
opinion of housewives using them that the detergent in the blue box left clothes dingy, that the
one in the yellow box was too harsh, and that the one in the blue-and-yellow box was both gentle
and effective in cleaning is an illustration of color association and its effect on assessment of
119
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
product quality that very likely would not have been discovered through either direct or indirect
questioning.
In another experiment, orange-scented nylons were placed on a counter in a department store next
to identical, but unscented, hosiery. The scented hosiery was bought by approximately 90% of
the women making purchases. Questioning of the women who bought the scented hose as to why
they preferred the hose they bought resulted in answers such as “of better quality,” “sheerer,” and
the like.
Of these four informational categories, only the second and third lend themselves to the
use of indirect research techniques. It remains, of course, for the analyst to decide in which one
or more of these categories the information he or she requires will fall.
While neither the universally applicable methodology nor the panacea that some
proponents have claimed, indirect research techniques can provide information on some types of
marketing problems that is not now obtainable by other means.
Sample Selection and Sizes in Qualitative Research
The subject of sampling is considered in detail in the next chapter. However, it is
desirable to examine here the typical sampling procedures and practices that have been used in
qualitative research studies, as this has been an area of considerable controversy.
Sample selection in qualitative research studies has tended to be done on nonprobabilistic
(purposive) bases rather than by probabilistic methods. Typically, selection has been on a
judgment or quota basis. Serious sampling errors can result from purposive sampling and the
extent of the sampling error is unknown. There are times, however, when the sample is not
statistically representative of the target market, but is close enough to be used as a basis for
judgment.
A second area of controversy over the samples typically taken in qualitative research
studies relates to their size. Generally, samples have been small, often ranging from 20 to 50 in
size. The use of a small sample in a qualitative research study suggests that the population of
psychological attributes and motivations being sampled is sufficiently homogeneous that only a
limited sample is required to provide an adequate representation of the population. However, the
bulk of the evidence amassed by psychologists suggests that motivations are myriad and varied
in their effect on behavior. To assume that the motivations of a very small group of people
adequately represent those of the population at large is to ignore the high degree of variability
that empirical studies have substantiated.
The Validity of the Findings
What about the validity of indirect research findings? How has their performance in these
respects compared with that of the more conventional research methods? The question of validity
of findings is, of course, the heart of the issue here, as it is in the general objectivist-subjectivist
controversy. Unfortunately, to raise the question is to beg it; no definitive answer can be given.
As has already been indicated, the answer is necessarily conditional on the nature of the problem
being investigated.
An observation does need to be made, however, about the differences in judging validity
by the “clients” of basic research versus those of decisional research projects. The client of the
basic research project is the professional in the field. Judgments of a study’s validity are a highly
impersonal process and one that is seldom urgent. When the purpose of the project is either to
120
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
make the best estimate of a population parameter or to conduct the best test of a hypothesis
within the constraints of available resources, the rules of evidence for a basic research study
require that the procedures be public, the results investigator-independent, and the project
replicable.
Since indirect research methods violate each of these requirements to some extent, there
has been reluctance on the part of some basic researchers to give even tentative acceptance to
unvalidated findings of studies that employ indirect methods. They tend to look upon indirect
research methods as a means of generating hypotheses for testing by objectivist methods rather
than as a source of valid findings. In the absence of data that can be used for direct validation,
the basic research project is judged tentatively on the basis of method.
The client for a decisional research project has a different set of requirements. Rather
than wanting to be assured that the best estimate of a parameter or the most definitive test of a
hypothesis has been made, the client needs information that will assist him or her in making the
best decision possible in the circumstances. The procedures of the investigation need not be
public, and there is seldom a need for replication. The client works directly with the researcher
and is able to raise any questions he or she has about the project. The client usually will have
had the opportunity to judge the validity of the findings of past research projects conducted by
either the researcher or the organization for which he or she works. An assessment of validity of
the findings must be made now; for a decision to await the outcome to determine if the findings
are valid would obviate the very purpose for which the research was conducted. Judgment of
degree of validity therefore turns out to be a much more subjective process in decisional than in
basic research.
Indirect techniques serve several useful purposes in marketing research. They can be
used to obtain information from respondents unwilling or unable to provide it by direct methods,
to check the validity of responses to direct techniques, and to provide supplemental information.
Included in the supplemental information that is of value is that which suggests hypotheses that
can be tested by direct methods.
SUMMARY
In this chapter we first examined the various types of indirect interviews and qualitative
research techniques that can be used to obtain information from respondents. In the indirect
types of interviews, we described the more commonly used projective techniques, including the
third person technique (“what does your neighbor think of ___?”), word association, sentence
completion and depth interviews. Also discussed were focus groups and in-depth interviews,
including ZMET and means ends laddering interviews.
We then considered the means of obtaining information through observation of people.
The use of audits, recording devices, and direct observation were described and their applications
discussed. Of particular interest is the use of trace analysis (analysis of post-consumption
garbage) to study behavior.
Finally, an assessment was made of direct (discussed in Chapters 4) versus indirect
research techniques from the standpoints of applicability to marketing problems, sample
selection and sizes, and validity of findings.
121
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
Berstell, G. (1992, January 6). Study what people do, not what they say. Marketing News, 26, 15ff.
Berstell, G., & Nitterhouse, D. (2001, February 26). Getting whole story can reveal surprise opportunities.
Marketing News, 35, 45.
Bloor, M., Frankland, J., Thomas, M., & Robson, K. (2001). Focus groups in social research. Thousand Oaks, CA:
Sage.
Bouchard, T. J., Jr. (1976, February). Unobtrusive measures: An inventory of uses. Sociological Methods &
Research, 4, 267–300.
Ericsson, K. A., & Simon, H. A. (1984). Protocol analysis: Verbal reports of data. Cambridge: MIT Press.
Fedder, C. J. (1990, January 8). Some differences between consumer-based and business-to-business groups require
a special touch. Marketing News, 24, 46.
Fern, E. (1981). Why do focus groups work? A review and integration of small group process theories. Proceedings
of the Association for Consumer Research Annual Conference, pp. 441–451.
Garee, M. L., & Schori, T. R. (1996, September 23). Focus groups illuminate quantitative research. Marketing News,
30, 41.
Gibby, K., Holland, D., Lee, F., Lundell, C., Merrill, S., & Robison, J. (1995). Positioning a credit union Visa.
(Field study conducted under the supervision of Professor Scott M. Smith.) Provo, UT: Brigham Young University.
Green, P. E., Tull, D. S., & Albaum, G. (1988). Research for marketing decisions (5th ed.). Englewood Cliffs, NJ:
Prentice Hall.
Greenbaum, T. (1993, January 4). Convention focus groups are a way to increase geographic representation.
Marketing News, 27, FG-2.
Haire, M. (1950, April). Projective techniques in marketing research. Journal of Marketing, 14, 649–656.
Johnson, R. R. (1988, October 24). Focus groups are divided into 4 distinctive categories. Marketing News, 22, 21.
LaFlamme, T. A. (1988, August 29). Symphony strikes a note for research as it prepares to launch a new season.
Marketing News, 22, 12.
Lawless, P. (1999, January 4). Empathic interviewing for insightful results. Marketing News, 33, 20
Mariampolski, H. (1988, October 24). Probing correctly uncovers truth behind answers in focus group. Marketing
News, 22, 22ff
Newman, J. W. (1957). Motivation research and marketing management. Cambridge, MA: Harvard University
Graduate School of Business Administration.
Olson, Jerry C., Kristin Waltersdorff and James Forr (2008), Incorporating Deep Customer Insights in the
Innovation Process, in Hans H. Hinterhuber and K. Matzler (eds.), Kundenorientierte Unternehmensführung, 6th
edition, Gabler Verlag: Wiesbaden, 2008.
http://www.olsonzaltman.com/downloads/2008 Insights in the Innovation Process.pdf
Jerry C. Olson, Deep Metaphors and Expressions in Conversation, Olson Zaltman Associates
(www.olsonzaltman.com).
122
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Reilly, M. D. (1984, September/October). Household refuse analysis and market research. American Behavioral
Science, 28, 115–128.
Reynolds, T. J., & Gutman, J. (1988, February/ March). Laddering theory, method, analysis, and interpretation.
Journal of Advertising Research, 28, 11–31.
Reynolds, T. J., Dethloff, C., & Westberg, S. (2001). Advancements in laddering. In T. J. Reynolds & J. C. Olson,
Eds., Understanding Consumer Decision Making: A Means-End Approach to Marketing and Advertising Strategy.
Mahwah, NJ: Lawrence Erlbaum.
Reynolds, T. J., & Whitlark, D. (1995, July/August). Applying laddering data to communications strategy and
advertising practice. Journal of Advertising Research, 35(4), 9–17.
Rydholm, J. (2000, March). Preserving the preservationists. Quirk’s Marketing Research Review, pp. 8–19ff.
Sechrest, L., Ed. (1979). Unobtrusive measurement today. San Francisco: Jossey-Bass
Simon, M. (1988, August 29). Focus groups by phone: Better way to research health care. Marketing News, 22, 47–
48.
Webb, E., Campbell, D., Schwartz, R., & Sechrest, L. (1966). Unobtrusive methods: Nonreactive research in the
social sciences. Chicago: Rand McNally.
Zaltman, G. (2003). How consumers think. Boston: Harvard Business School Press.
123
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 7
SAMPLING PROCEDURES IN RESEARCH
Researchers must answer many questions when developing a project to collect
information about a population of interest. Consider the following questions involving sampling:
• Should we take a census (complete canvas) or a sample?
• What kind of sample should be taken?
• What size should the sample be?
Answering these questions depends upon the application of statistical inference. This
chapter first considers the selection of sample types and the process of sample planning. It then
describes different kinds of samples. Finally, related to design, but a unique decision area in its
own right, is the determination of the size of the sample.
PLANNING THE SAMPLE
Two broad objectives are fundamental to the use of samples in survey research projects:
• Estimation of information about a population based on a sample from the population,
• Hypothesis testing to compare the relationships between data items for some selected
population groups.
Each involves making inferences about a population on the basis of information from a
sample. The precision and accuracy of project results are affected by the manner in which the
sample has been chosen. However, as Exhibit 7.1 illustrates, precision is only a reflection of
sampling error and confidence limits and has nothing to do with accuracy.
Exhibit 7.1 Precision versus Accuracy in Sampling
There often is confusion between precision and accuracy in marketing research. When a
researcher speaks of sampling error and the size of the confidence limits placed on an estimate, that
researcher is speaking of precision and not accuracy.
Accuracy is affected by nonresponse bias, memory error, misunderstanding of questions,
problematic definition of terms, and processing errors, and not by sample size (Semon, 2000). The
researcher must constantly ask: “Have we asked the right questions?” and “Have we measured
what we want to measure properly and correctly?”
The difference between precision and accuracy is analogous to the difference between
errors of degree and errors of kind. Accuracy, like errors of kind, are clearly more serious. If
a researcher has made the correct measurements, precision will show the degree of error;
but if the right measurements have not been taken, precision is probably meaningless
(Semon, 2000). The quality of a sample depends upon the quality of design and execution of
a research project at every stage of the process.
124
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Consequently, strict attention must be paid to the planning of the sample. It must also be
recognized that sample planning is only one part of planning the total research project. The
process of selecting a sample follows the well-defined progression of steps shown in Figure 7.1,
and will be discussed in turn.
Figure 7.1 Steps in Sample Planning
Defining the Population
The first step in sample panning is to define the population to be investigated. A
population, also known as a universe, is defined as the totality of all units or elements
(individuals, households, organizations, etc.) to which one desires to generalize study results.
While seemingly an easy task, an imprecise research problem definition often leads to an
imprecise population definition.
Specifying a population involves identifying which elements (in terms of kind) are
included, as well as where and when. For example, a group medical practice that is considering
expanding into sports medicine might acquire information from any or all of the distinct
population groups listed in Table 7.1. The population element is the unit of analysis, and may be
defined as an individual, household, institution, patient visit, and so on.
Table 7.1 Possible Population Choices for a Research Study by a Medical Practice
Group Members Where When
All patients Designated group practice Last 12 months
Patients who have had orthopedic work Designated group practices Last 12 months
All people Specified geographic area Last 12 months
All people who have had orthopedic work Specified geographic area Last 12 months
The second and third columns of Table 7.1 define population choices in terms of its
location and timeframe. One useful approach is to first define the ideal population to meet study
objectives, and then apply practical constraints to ultimately limit and define the study
population. However over defining the population should be avoided unless it is absolutely
necessary. Over defining can limit the extent to which findings can be generalized, and greatly
increase the operational cost and difficulty of finding population elements (Sudman, 1976).
125
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Census, or Sample?
Once the population has been defined, the investigator must decide whether to conduct
the survey among all members of the population, or only a sample subset of the population. The
desirability and advantages of using a sample rather than a census depend on a variety of factors
such as geographic location, the absolute size of the population, and the sample size required for
results sufficiently accurate and precise to achieve the required purposes.
Two major advantages of using a sample rather than a census are speed and timeliness. A
survey based on a sample takes much less time to complete than one based on a census.
Frequently, the use of a sample results is a notable economy of time, money and effort,
especially when a census requires hiring, training and supervising many people.
In other situations, a sample is necessary because of the destructive nature of the
measurement, such as in product testing. There is a related problem over surveying human
populations when many different surveys need to be conducted on the same population within a
relatively short period of time. Nonprobability sampling techniques explicitly protect against this
problem.
In still other situations, a sample may control non-sampling errors. Samples (a smaller
number of interviews compared to a census) may result in better interviewing, higher response
rates through more call backs, and better measurement in general. The total amount of sampling
and non-sampling error of a sample may actually be than the non-sampling error alone would be
for a census. In Chapter 2, we emphasized the importance of minimizing total error.
However, under certain conditions a census may be preferable to a sample. When the
population is small, the variance in the characteristic being measured is high, the cost of error is
high, or the fixed costs of sampling are high, sampling may not be useful. In addition, if the
characteristic or attribute of interest occurs rarely in the population, then a census of a tightly
defined population might be desirable (for example, people with a rare genetic disorder). In this
case, it would be necessary to sample a relatively large proportion of the general population to
provide statistically reliable information. Obviously, the practicality of this depends upon the
absolute size of the population and the occurrence rate for the characteristic of interest.
Sample Design
Sample design is “the theoretical basis and the practical means by which data are collected so that the
characteristics of a population can be inferred with known estimates of error” (Survey Sampling, Inc.).
Operationally, sample design is at the heart of sample planning. Sample design
specification, including the method of selecting individual sample members, involves both
theoretical and practical considerations (such as cost, time, labor involved, organization). The
following checklist is suggested to obtain a sample that represents the target population (Fink,
2003):
1. Are the survey objectives stated precisely?
2. Are the eligibility criteria for survey respondents or experimental subjects clear
and definite? Exclusion criteria rule out certain people.
3. Are rigorous sampling methods chosen? This involves selecting an appropriate
probability or nonprobability sampling method.
4. Further questions to be answered in this section include:
126
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• What type of sample should be used?
• What is the appropriate sampling unit?
• What is the appropriate frame (that is, list of sampling units from which the sample is
to be drawn) for the particular design and unit decided upon?
• How are refusals and nonresponse to be handled?
Type of Sample
Much of the sampling in the behavioral sciences and in marketing research, by its nature,
involves samples are selected by the judgment of the investigator, convenience, or other
nonprobabilistic (nonrandom) processes. In contrast, probability samples offer the promise of
bias-free selection of sample units and permits the measurement of sampling error.
Nonprobability samples offer neither of these features. In nonprobability sampling one must rely
on the expertise of the person taking the sample has selected respondents in an unbiased manner,
whereas in probability sampling the sampling of respondents is independent of the investigator.
Example: A dog food manufacturer tested consumer reactions to a new dog food by giving
product samples to employees who own dogs and eliciting their responses about a week later. The
employees’ dogs liked the food and the pet food manufacturer actually introduced the new dog
food product. However when it hit the market the product was a flop… dogs simply would not
eat it. As managers scrambled to find out what went wrong, research showed that employees
were so loyal to the company’s products that their dogs had little variety and would eat anything
for a change. In the broader market dogs were used to a greater variety of dog foods including
table scraps and just did not like the taste of the new dog food. In this case, a biased sample was
erroneously assumed to conform to the general population of dogs and dog owners.
A researcher choosing between probability and nonprobability sampling implicitly
chooses the probability sample’s relative size of sampling error against the nonprobability
sample’s combined sampling error and selection bias. For a given cost, one can usually select a
larger nonprobability sample than probability sample, meaning that the sampling error should be
lower in the nonprobability sample, but that the nonrandom process used for selecting the sample
may have introduced a selection bias.
The Sampling Unit
The sampling unit is the unit of the population to actually be chosen during the sampling
process. The sampling unit may contain one or more elements describing the population. For
instance, a group medical practice may be interested in surveying past patient behavior of the
male wage earner or his entire household. In either case, it may be preferable to select a sample
of households as sampling units.
The Sampling Frame
A sampling frame is a means of identifying, assessing and selecting the elements in the
population. The sampling frame usually is a physical listing of the population elements. In those
instances where such a listing is not available, the frame is a procedure producing a result
equivalent to a physical listing. For instance, in a consumer survey where personal interviews are
conducted with a mall intercept method is used to obtain data, the sample frame includes all
those people who enter the mall during the study period.
127
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Ideally, the sample frame should identify each population element once, but only once,
and not include elements not in the defined population. Such a perfect frame is seldom available
for marketing research purposes. As shown in Figure 7.2, a sampling frame may be incomplete,
too comprehensive, or a combination of both. In addition, the frame may include individual
population elements more than once. Any of these situations can lead to coverage error.
Figure 7.2 Sampling Frame – Population Relationship
Perhaps the most widely used frame in survey research for sampling human populations
is the telephone directory. Use of such a frame, however, may lead to frame error from the
exclusion of non-subscribers (no home phone or cell phone only), voluntarily unlisted
subscribers, and involuntarily unlisted subscribers. For example, more than one in five American
homes (20.2%) has only wireless telephones (National Center for Health Statistics, 2009).
Combined with the fact that in many major cities more than 50% of home phones are unlisted,
the potential for frame errors is huge. Similar frame errors can arise from such other sampling
frames as city directories, maps, trade association membership lists, or any other incomplete or
dated listing or representation of a population. The National Do Not Call Registry provisions do
not cover calls from political organizations, charities, telephone surveyors, or companies with
which a consumer has an existing business relationship, nonetheless, resident push back leads to
further frame error.
Even though today’s panel management software makes compiling and managing sample
lists relatively easy, sampling is the single most important problem in e-mail and Internet-based
surveys. Dillman (2000) suggests asking the following five questions about any potential
sampling list:
• Does the list contain everyone in the survey population?
• Does the list include names of people who are not in the study population?
• How is the list maintained and updated?
• Are the same sample units included on the list more than once?
• Does the list contain other respondent information that can be used to customize the
survey or direct survey logic?
128
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Whether the researcher uses their own list, or one purchased from a commercial source,
these questions are equally applicable.
Sample Size
Sample size, as part of sample planning is related to precision. Four traditional
approaches can determine the appropriate sample size for any given research project:
1. Arbitrarily or judgmentally determined
2. Minimum cell size needed for analysis
3. Budget-based
4. Specifying desired precision in advance of sampling and then applying the
appropriate standard error formula to determine the sample size.
For approaches 1-3, the precision could be measured after compiling the data and then
applying the appropriate standard error formula or formulas if a probability design is being used.
Costs of Sampling
The sample plan must account for the estimated costs of sampling. Such costs include
overhead costs, which are relatively fixed for a sampling procedure, and variable costs, which
depend on the number of respondents contacted and interviewed in the study. In reality, costs
from all aspects of data collection are usually considered together. The dashed line shown in
Figure 7.1 when the sample design and sample size are considered, the estimated costs may be so
great that other sample designs and/or smaller-sized samples may be considered.
Execution of the Sampling Process
The last step in sample planning is to execute the sampling process. A sample is chose
that is thought to be both representative and to adequately mirror the various patterns and
subclasses of the population (Exhibit 7.2). The sample should be of sufficient size to provide
confidence in the stability of its characteristics.
Exhibit 7.2 Samples Are Not Always an Exact Match
Many researchers feel that the best way to assess the validity of a sample is to compare its
demographic profile (i.e., distributions of the key demographic characteristics) with a national or
otherwise known profile. This alone does not necessarily guarantee a good sample. At the same
time, a poor fit does not necessarily mean that the obtained sample is bad.
When the fit is not good, some researchers tend to develop a weighting scheme. However many
studies have shown that large differences in demographic characteristics may to translate into
small differences in the variable of interest, whether it is a behavior or an attitudes, interests, or
opinions (AIO) measure. Of course, there could be serious distortion if specific segments are of
concern. But the differences generally have to be much greater than we would think for a
significant effect.
This, in turn, requires a measure of precision, which necessitates using probability-based
design. From this discussion, one might conclude that an ideal sample is obtained with a
probability process. In general, this is preferable. However, recognize that it is more important to
129
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
avoid distorted samples than to be able to measure sampling error. There may be a tendency to
ignore the existence of potential bias when using probability designs.
This chapter is primarily concerned with probability sampling, but before discussing this
topic it is useful to describe some of the procedures by which nonprobability samples are taken
in marketing research.
NONPROBABILITY SAMPLING PROCEDURES
Nonprobability sampling is distinguished from probability sampling in that
nonprobability sample elements do not have a known, nonzero chance of being selected for the
sample. As such, sampling error is generally not able to be measured. Nonprobability samples
are widely used in exploratory research, but are also valuable for non-exploratory research.
The Quota Sample
The quota sample is the most commonly employed nonprobability sampling procedure in
marketing research. Quota samples are collected to reflect proportions in the various subclasses
(or strata) of the population of interest. This might be, for example, the proportion of the adult
population who fall into various age-by-gender-by-education groupings.
To prepare a quota sample, the subclass proportions are first estimated from some outside
source, such as census data. Next, if an interviewer has a total number of, say, 600 interviews to
obtain, the age-gender-education proportions in the population are applied to the 600 total
interviews to determine the appropriate quotas.
However, in quota sampling the interviewer is not required to randomly select the
respondents necessary to fill each quota category. The lack of random selection is the major
distinction between quota sampling and stratified random sampling.
In quota sampling, the interviewer’s judgment is relied upon to select actual respondents
within each quota. Therefore, many sources of selection bias are potentially present. For
example, the interviewer may not bother to call back if the first call results in a “not-at-home.”
Interviewers may go to selected areas where the chances are good that a particular type of
respondent is available. Certain houses may be skipped because the interviewer does not like the
appearance of the property, and certain people may be allowed to pass in a mall-intercept
approach because they do not “look right.” Still other interviewer habits and biases exist that can
influence their selection of respondents within the quota.
The advantages of quota sampling are the lower costs and greater convenience provided
to the interviewer when selecting respondents to fill each quota. Quota sampling is quite close to
traditional probability sampling when selection is tightly controlled (Sudman, 1976).
Quota sampling is discussed rather extensively in Stephan and McCarthy (1958). A
critique of the method is offered by Deming (1960). A comparison of quota and probability
sampling was the subject of a seminar held in 1994 at the Survey Methods Centre in the United
Kingdom (Survey Methods Centre, 1995).
The Judgment Sample
A somewhat representative sample may be provided through the use of purposive, or
judgment, sampling. Sound judgment or expertise and an appropriate strategy, leads one to
carefully and consciously choose the elements so as to develop a suitable sample. The intent is to
select respondents representative of the population in such a way that errors of judgment in the
130
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
selection will cancel each other out. The relative advantages of judgment sampling are that it is
inexpensive, convenient to use, less time-consuming, and provides results as good as probability
sampling. One weakness of this approach is that without an objective basis for making the
judgments, there is no way of knowing whether the so-called typical cases are, in fact, typical.
The Convenience Sample
Convenience sampling is a generic term that covers a wide variety of ad hoc procedures
for selecting respondents. Convenience sampling means that the sampling units are accessible,
convenient and easy to measure, cooperative, or articulate. An illustration of convenience
sampling is “intercept” interviews among shopping-mall customers or in other areas where large
numbers of consumers may congregate. In this case, the researcher needs to select respondents
with care. People frequent malls for different reasons and stay in malls different lengths of time,
which leads to a biasing effect if a probability sample is attempted. A sample is said to be length
biased if the probability of observing an individual at a particular site is dependent on the
individual’s length of stay at the site.
Firms may also authorize samples to be taken from such intact groups as Parent-Teacher
Associations, church groups, philanthropic organizations, and so on. Again, the purpose is to
obtain a relatively large number of interviews quickly from a cooperating group of respondents.
Usually the sponsoring organization receives a donation from the interviewing firm for the help
and cooperation of its members. However depending on the purpose of the research, many
potential sources of selection bias may exist in that only certain members may respond, who may
be disproportionately different on one of many demographic, attitudinal or behavioral
dimensions.
Snowball Sampling
Snowball sampling (also known as multiplicity or chain-referral sampling) is the rather
colorful name given to a procedure in which initial respondents are selected randomly, but where
additional respondents are then obtained from referrals or other information provided by the
initial respondents. One major advantage of snowball sampling is the ability to estimate various
characteristics that are rare in the total population. A more complete discussion of this technique
is given by Zinkhan, Burton, and Wallendorf (1983).
Example: A study of international tourism, required researchers to interview respondents in the
United Kingdom, France, and Germany who visited the United States in its bicentennial year.
Given the likelihood of finding a qualified adult respondent was less than two percent, stratified
probability methods were used to select initial respondents. A referral procedure (up to two
referrals per qualified respondent) was then used to obtain a second group of qualified
respondents. (However, this particular study did not obtain subsequent referrals from this second
group of respondents.)
In other types of snowball sampling, referrals from referrals are obtained, and so on, thus
leading to the “snowballing” effect. Even though a probability-based procedure may be used to
select the initial group of respondents, the overall sample is a nonprobability sample.
Major advantages of this type of sampling over conventional methods are the
substantially increased probability of finding the desired characteristic in the population, and
lower sampling variance and costs.
131
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
PROBABILITY SAMPLING PROCEDURES
The best-known type of probability sample is no doubt the simple random sample.
However, many occasions in marketing research require more specialized sampling procedures
than those of simple random-sampling methods. Statisticians have developed a variety of
specialized probability-sampling designs that, although derived from simple random-sampling
principles, can be used to reduce sampling error and cost. Five major modifications can be made
to the basic selection process, as shown in Table 7.2.
These techniques are discussed in turn, following a review of simple random sampling.
Our purpose is to describe the major characteristics of each technique, rather than to present a
detailed mathematical exposition of its procedures. Many excellent statistics and research books
review the mathematical aspects of these sampling techniques.
The Simple Random Sample
In a simple random sample, each sample element has a known and equal probability of
selection, and each possible sample of n elements has a known and equal probability of being the
sample actually selected. It is drawn by a random procedure from a sample frame—a list
containing an exclusive and exhaustive enumeration of all sample elements. One widely used
process for generating a simple random sample is: upload the elements of the sample frame to a
spreadsheet, number them, and then use the random-number generator function to select the
sample members.
Simple random samples are not widely used in consumer research, for two reasons. First,
it is often difficult to obtain a sampling frame that will permit a simple random sample to be
drawn and second, one may not want to give all sample units an equal probability of being
selected. Consumer research usually requires people, households, stores, or areas to be the basic
sampling units. While a complete representation of areas is available through maps, there
normally is no complete listing of persons, the households in which they live, or the stores
available. When persons, households, or stores are to be sampled, some other sample design
must be used.
In business-to-business (B2B) research, there is a greater opportunity to apply simple
random sampling. In this case purchasing agents, companies, or areas are the usual sampling
units and the population under study is often relatively small. One is therefore in a better
position to develop a complete list of respondents or sample frame.
Table 7.2 Selection Methods for Probability Samples
Probability Samples Nonprobability Samples
I. Equal probability for all elements
a. Equal probabilities at all stages
b. Equal overall probabilities for all
elements obtained through
compensating unequal probabilities
at several stages
Unequal probabilities for different elements; ordinarily
compensated with inverse weights
a. Caused by irregularities in selection frames and
procedures
b. Disproportionate allocation designed for
optimum allocation
II. Element Sampling: single stage, sampling Cluster Sampling: sampling units are clusters of elements
132
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
unit contains only one element a. One-stage cluster sampling
b. Sub-sampling or multistage sampling
c. Equal clusters
d. Unequal clusters
III. Unstratified Selection: sampling units
selected from entire population
Stratified Sampling: separated selections from partitions,
or strata, of population
IV. Random Selection of individual sampling
units from entire stratum or population
Systematic Selection of sampling units with selection
interval applied to list
V. One-Phase Sampling: final sample selected
directly from entire population
Two-Phase (or Double) Sampling: final sample selected
from first-phase sample, which obtains information for
stratification or estimation
SOURCE: Adapted from Kish, 1965, p. 20.
The Systematic Sample
Systematic sampling involves only a slight variation from simple random sampling. In a
systematic sample, each sample element has a known and equal probability of selection. The
permissible samples of size n that are possible to draw each have a known and equal probability
of selection, while the remaining samples of size n have zero probability of being selected.
The mechanics of taking a systematic sample are rather simple. If the population contains
N ordered elements and a sample size n is desired, one merely finds the ratio of N/n and rounds
to the nearest integer to obtain the sampling interval. For example, if there are 600 members of
the population and one desires a sample of 60, the sampling interval is 10. A random number is
then selected between 1 and 10, inclusively; suppose the number turns out to be 4. The analyst
then takes as the sample elements 4, 14, 24, and so on.
Essentially, systematic sampling assumes that population elements are ordered in some
fashion—names in a telephone directory, a card index file, or the like. Some types of ordering,
such as an alphabetical listing, will usually be uncorrelated with the characteristic (say, income
level) being investigated. In other instances, the ordering may be related to the characteristic
under study, as when a customer list is arranged in decreasing order of annual purchase volume.
If the arrangement of the elements of the sample is itself random with regard to the
characteristic under study, systematic sampling will tend to give results close to those provided
by simple random sampling. We say “close” because in systematic sampling, all combinations of
the characteristic do not have the same chance of being included. For example, it is clear that in
the preceding example, the fifth, sixth, and so on items have zero chance of being chosen in the
particular sample after the first item has been determined.
Systematic sampling may increase the sample’s representativeness when items are
ordered with regard to the characteristic of interest. For example, if the analyst is sampling a
customer group ordered by decreasing purchase volume, a systematic sample will be sure to
contain both high- and low-volume customers. On the other hand, the simple random sample
may yield, say, only low-volume customers, and may thus be unrepresentative of the population
being sampled if the characteristic of interest is related to purchase volume.
It is also possible that systematic sampling may decrease the representativeness of the
sample in instances where the items are ordered in a cyclical pattern. For example, a sample
interval of 7 for a systematic sampling of daily retail-store sales figures would reflect the same
day of data for each week and would not reveal the day-of-the-week variations in sales.
133
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The Stratified Sample
It is sometimes desirable to break the population into different strata based on one or
more characteristics, such as the frequency of purchase of a product, type of purchase (e.g.,
credit card versus non-credit card), or the industry in which a company competes. In such cases,
a separate sample is then taken from each stratum. Technically, a stratified random sample is one
in which a simple random sample is taken from each stratum of interest in the population. In
practice, however, systematic and other types of random samples are sometimes taken from each
of the strata. In this case, the resulting design is still referred to as a stratified sample.
Stratified samples are generally conducted according to the following procedure:
• The entire population is first divided into an exclusive and exhaustive set of strata, using
some external source, such as census data, to form the strata.
• A separate random sample is selected within each stratum.
• From each separate sample, some statistic (such as a mean) is computed and properly
weighted to form an overall estimated mean for the whole population.
• Sample variances are also computed within each separate stratum and appropriately
weighted to yield a combined estimate for the whole population.
The two basic varieties of stratified samples are proportionate and disproportionate. In
proportionate stratified sampling, the sample drawn from each stratum is made proportionate in
size to the relative size of that stratum in the total population. In disproportionate stratified
sampling, one departs from the preceding proportionality by taking other circumstances, such as
the relative size of stratum variances, into account.
Example: A company is interested in estimating the average purchases of consumers of hot
cereal. The researcher may be willing to assume that, although average consumption would vary
markedly by family size, the variances around the means of the strata would be more or less equal
among family sizes. If so, the researcher would make use of proportionate stratified sampling.
More generally, however, both means and variances will differ among strata. If this is the
case, the researcher would make use of disproportionate stratified sampling. In this instance, the
number of families included in each stratum would be proportionate to (the product of) the
relative size of the different family-sized strata in the population and the standard deviation of
each family class. This requires, of course, that the researcher be able to estimate (from past
studies) the within-group standard deviation around the average purchase quantity of each
purchasing stratum. Formulas for computing sampling errors in stratified samples are briefly
discussed later in this chapter and can be found in standard texts on sampling.
As intuition would suggest, the increased efficiency of stratified sampling over simple
random sampling depends on how different the means (or some other statistic) really are among
strata, relative to the within-stratum variability. The greater the within-stratum homogeneity and
among-stratum heterogeneity, the more efficient the stratified sampling is relative to simple
random sampling.
Two final considerations need to be addressed regarding stratified sampling. First, a
sample size needs to be calculated for each stratum or subgroup. Second, the process of sample
selection can be time-consuming and costly to carry out if many subgroups are to be used.
134
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Nevertheless, this method continues to be widely used due to the segmentation of markets that
companies routinely engage in.
The Cluster Sample
The researcher will ordinarily be interested in the characteristics of some elementary
element in the population such as individual family attitudes toward a new product. However,
when larger primary sampling units are desired, cluster sampling may be used. For example, the
researcher may choose to sample city blocks and interview all the individual families residing
therein. The blocks, not the individual families, would be selected at random. Each block
consists of a cluster of respondents. The main advantage of a cluster sample relative to simple
random sampling is in lower interviewing costs rather than in greater reliability.
The Area Sample: Single Stage and Multistage
As the name suggests, area sampling pertains to primary sampling of geographical
areas—for example, counties, townships, blocks and other area descriptions. A single-stage area
sample occurs when only one level of sampling takes place (such as a sampling of blocks) before
the basic elements are sampled (the households). If a hierarchy of samples within the larger area
is taken before settling on the final clusters, the resulting design is usually referred to as a
multistage area sample.
Example: Consider the sample design used by the Gallup Organization for taking a nationwide
poll. Gallup draws a random sample of locations as the first stage of the sampling process. Blocks
or geographic segments are then randomly sampled from each of these locations in a second
stage, followed by a systematic sampling of households within the blocks or segments. A total of
about 1,500 persons are usually interviewed in the typical Gallup poll.
METHODS FOR DETERMINING SAMPLE SIZE
There are several ways to classify techniques for determining sample size. Two that are
of primary importance are the following:
• Whether the technique deals with fixed or sequential sampling
• Whether its logic is based on traditional or Bayesian inferential methods
Other than the brief discussion of sequential sampling that follows, this chapter is concerned
with the determination of a fixed sample size with emphasis on traditional inference, such as
Neyman-Pearson, rather than Bayesian inference.1
Although the discussion will focus on the statistical aspects of setting sample size, it
should be recognized that nonstatistical dimensions can affect the value of a research project.
Such things as study objectives, the length of a questionnaire, budget, time schedule, and the
requirements of data analysis procedures, all have a direct effect on sample size decisions.
Fixed Versus Sequential Sampling
As the name implies, in fixed-size sampling the number of items sampled is decided in
advance. The size of the sample is chosen to achieve a balance between sample reliability and
sample cost. In general, all sample observations are taken before the data are analyzed.
In sequential sampling, however, the number of items is not preselected. Rather, the
135
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
analyst sets up in advance a decision rule that includes not only the alternative of stopping the
sampling process (and taking appropriate action, based on the sample evidence already in hand)
but also the possibility of collecting more information before making a final decision.
Observations may be taken either singly or in groups, the chief novelty being that the data are
analyzed as they are assembled and sample size is not predetermined.
In general, sequential sampling will lead to smaller sample sizes, on average, than those
associated with fixed-size samples of a given reliability. The mathematics underlying sequential
sampling are, however, more complex and time consuming. In addition, the problem may be
such that it is less expensive to select and analyze a sample of many items at one time than to
draw items one at a time (or in small groups) and analyze each item before selecting the next.
Sampling Basics: Terminology
Intuitively we would expect that when we increase the size of the sample, our estimate of
the population parameter should get closer to the true value. Also, we would expect that the less
dispersed the population’s characteristics are, the closer our sample estimates should be to the
true parameter. After all, the reason why we sample in the first place is to make some inference
about the population. These inferences should be more reliable when the sample is larger and
when there is less variability in the population variables measured. To see this in action,
consider the example in Exhibit 7.3.
When you read about samples in newspapers or other documents, the researcher often
reports the margin of error or confidence interval for the statistical findings reported in the study.
The “margin of error” or “confidence interval” is the plus-or-minus figure that represents the
accuracy of the reported. Consider another example:
A Canadian national sample showed “Who Canadians spend their money on for Mother’s Day.”
Eighty-two percent of Canadians expect to buy gifts for their mom, compared to 20 percent for
their wife and 15 percent for their mother-in-law. In terms of spending, Canadians expect to
spend $93 on their wife this Mother’s Day versus $58 on their mother. The national findings are
accurate, plus or minus 2.75 percent, 19 times out of 20.
In this example, if 82% of your sample indicates they will "buy a gift for mom" and you
use a confidence interval of 2.75%, you can be "95% confident that for ALL CANADIANS,
somewhere between 79.25% (82%-2.75%) and 84.75% (82%+2.75%) would have picked that
answer.
The “confidence level” tell you how confident you are of this result. It is expressed as a
percentage of times that different samples (if repeated samples were drawn) would produce this
result. The 95% confidence level means that if 20 different samples were drawn, 19 times out of
20, the results would fall in this - + confidence interval. A 99% confidence level would mean
that 99 out of 100 times, the results would fall into the stated -+ confidence interval. The 95%
confidence level is the most commonly used.
When you put the confidence level and the confidence interval together, you can say that
you are 95% (19 out of 20 times) sure that the true percentage of the Canadian population that
will "buy a gift for mom" is between 79.25% and 84.75%.
Wider confidence intervals increase the certainty that the true answer is within the range
specified. These wider confidence intervals are associated with smaller sample sizes and of
course produce larger sampling errors. When the costs incurred from making an error are
136
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
extremely high (you are betting your company, or a multi-million dollar decision is being made)
the confidence interval should be kept small. This can be done by increasing the sample size to
reduce the sampling error.
Exhibit 7.3 How Do Election Polls Work?
The following is an Edited version of "Inside the paper's election polls", an article by Elsa
McDowell that appeared in The Charleston Post and Courier:
The beauty of... election polls is that they are straightforward. They use statistical
formulae to estimate how many people will vote one way and how many will vote another. No
spin. No qualifying clauses to muddy the picture. The difficulty of.. election polls is that they are
not always straightforward. How else could you explain that a poll done by one candidate shows
him in the lead and that a poll done by his opponent shows her in the lead? Statisticians say
there are ways to twist questions or interpret answers to give one candidate an advantage over
another.
One reader took issue with a recent poll results run in The Post and Courier. He
questioned whether the methodology was described in enough detail, whether the sample size
was adequate. He was right about one point. The story did not make clear who was polled. It
said "voters" and failed to elaborate. It should have indicated that the people polled were
registered and likely to vote in the November elections. His next point is debatable. He said the
sample size of 625 likely voters was insufficient for a state with nearly 4 million residents and
suggested at least 800 should have been polled.
Brad Coker, the researcher responsible for the study responded that "the standard
sample size used by polling groups nationally is 625. It produces, as the story stated, a margin of
error of plus-or-minus 4 percent. Increasing the sample size to 800 would have produced a
margin of error of plus-or-minus 3.5 - more accurate, but not so much more accurate to justify
the additional cost."
"Many people do not understand how sample sizes work. They believe that, the larger
the pool, the larger the sample size needs to be. It's not like that. You can take a drop of blood
from a 400-pound person and it will contain the same data you would get if you took it from a
100-pound person," he said.
The reader's next concern was that the margin of error of plus-or-minus 4 applies only
to the group viewed in its entirety. "If 'minorities' constituted 27 percent of the total sample,
then only 169 were sampled. The margin of error then skyrockets into double digits." Coker said
the reader is right and wrong. The margin of error (also known as a confidence interval) does
jump for subgroups, but does not reach double digits. In this case, it moves from plus-or-minus 4
to plus-or-minus 6 to 8.
Two days before The Post and Courier ran their poll, another short story was run about
a poll commissioned by MSNBC. That poll indicated incumbent Gov. Jim Hodges with a 45-43
percent lead. (Our) poll indicated challenger Mark Sanford was ahead 45 to 41 percent. When
the margin of error is considered, both polls show the race is still a toss-up.
137
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Controlling the Size of the Confidence Interval
Sampling theory teaches us that the accuracy of a sample estimate is dependent on such
factors as the dispersion and skewness of the population’s responses, the sample size, and the
size of the population. Controlling these variables contributes to the incidence (and elimination)
of sampling error. Note that "non-sampling" errors, such as bad question design or selection of a
"bad" sample frame are not controlled by sample size.
Sample Size
Larger sample sizes generally produce a more accurate picture of the true characteristics
of the population. Larger samples tighten the size of the confidence interval, making your
estimate much more accurate. This relationship is not linear as shown in Table 7.3. Increasing
sample size from 500 to 1000 reduces the confidence interval from +- 4.38 to +- 3.1.
Dispersion
The accuracy of an estimate also depends on the dispersion and skewness of the
population on the question being asked. A sample of individuals registered for the republican
political party would likely give a less dispersed evaluation of former president George W. Bush
than would a sample of democrats. Likewise, a sample of Catholic priests would have less
variability on the issue of abortion than would a survey of the general population. Accuracy of
the sample estimate increases as the dispersion in the population decreases. Depending on the
method of sample size calculation, dispersion is expressed as sample variance or as a proportion
holding alternative positions (favorable or unfavorable toward an issue).
When using a proportion to compute sample size or estimate confidence intervals, it is
easy to cover all eventualities by assuming maximum variability (50-50 proportions). Likewise,
once your data has been collected, the observed proportion and final sample size can be used to
obtain a more accurate estimate of the actual confidence interval.
Population Size
The size of the population also influences the size of the confidence interval, but not as
much as you might expect. For a sample of 1000 respondents from a population of 100,000, the
confidence interval is +- 3.08%. However, if the population were instead 1 million, the
confidence interval widens to only +- 3.1%. The confidence interval is far more sensitive to
changes in the sample size than to the size of the total population.
Non-sampling errors cannot be compensated for by increased sample size. Often, larger
samples accentuate non-sampling errors rather than reduce them. Non-sampling errors come
from samples that are not truly random, bad scales, misleading questions, incomplete surveys,
etc.
138
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 7.3 Proportion Based Confidence Intervals Computed at the 95% Confidence Level
Sample
Size
Variability Proportions
50/50% 40/60% 30/70% 20/80% 90/10% 95/5%
25 20 19.6 18.3 16 12 8.7
50 14.2 13.9 13 11.4 8.5 6.2
75 11.5 11.3 10.5 9.2 6.9 5
100 10 9.8 9.2 8 6 4.4
150 8.2 8 7.5 6.6 4.9 3.6
200 7.1 7 6.5 5.7 4.3 3.1
250 6.3 6.2 5.8 5 3.8 2.7
300 5.8 5.7 5.3 4.6 3.5 2.5
400 5 4.9 4.6 4 3 2.2
500 4.5 4.4 *4.1 3.6 2.7 2
600 4.1 4 3.8 3.3 2.5 1.8
800 3.5 3.4 3.2 2.8 2.1 1.5
1000 3.1 3.0 2.8 2.5 1.9 1.4
1500 2.5 2.5 2.3 2.0 1.5 1.1
2000 2.2 2.2 2.0 1.6 1.2 0.96
2500 2 1.9 1.8 1.6 1.2 0.85
5000 1.4 1.4 1.3 1.1 .83 0.6
*Example Interpretation: In a product usage study where the expected product
usage incidence rate is 30%, a sample of 500 will yield a precision of +/- 4.1
percentage points at the 95% confidence level.
This table is compute using the following formula:
(Number of Standard Errors)
2
* ((proportion)*(1-proportion)) / (Accuracy)
----------------------------------------------------------------------------------------------------
(1+((Number of Standard Errors)
2
* ((proportion)*(1-proportion)) / (Accuracy)-1) / (the population size)
This formula is easily entered into a spreadsheet, to compute a sample size
determination table.
Sampling Basics: Sampling Distributions and Standard Errors
The reader will recall from elementary statistics the concept of a sampling distribution. For a
specified sample statistic (e.g., the sample mean) the sampling distribution is the probability
distribution for all possible random samples of a given size n drawn from the specified
population. The standard error of the statistic is the standard deviation of the specified sampling
distribution. We shall use the following symbols in our brief review of the elementary formulas
for calculating the standard error of the mean and proportion (under simple random sampling):
µ = population mean
π = population proportion regarding some attribute
σ = standard deviation of the population
s = standard deviation of the sample, adjusted to serve as an estimate of the standard
deviation of the population
X = arithmetic mean of a sample
p = sample proportion
n = number of items in the sample
139
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
We here identify eight important properties associated with sampling distributions:
1. The arithmetic mean of the sampling distribution of the mean ( X ) or of the proportion (p)
for any given size sample equals the corresponding parameter values µ and π , respectively.
2. The sampling distribution of the means of random samples will tend toward the normal
distribution as sample size n increases, regardless of the original form of the population being
sampled.
3. For large samples (e.g., 100≥n and for π fairly close to 0.5) the normal distribution also
represents a reasonable approximation of the binomial distribution for sample proportions.
4. In the case of finite universes, where the sample size n is some appreciable fraction of the
total number of items in the universe, N, the standard error formulas should be adjusted by
multiplication by the following finite multiplier:
1−
−
N
nN
For practical purposes, however, use of the finite multiplier is not required unless the sample
contains an appreciable fraction, say 10% or more, of the population being sampled. At a
10% sampling fraction, taking it into account will reduce the random sampling error by 5%.
If the sampling fraction is 5%, 2%, or 1%, not ignoring it will reduce error very slightly—
2.5%, 1.0%, and 0.5%, respectively.
5. Probabilities of normally distributed variables depend on the distance (expressed in multiples
of the standard deviation) of the value of the variable from the distribution’s mean. If we
subtract a given population mean µ from a normally distributed variable Xi and divide this
result by the original standard deviation σ , we get a standardized variable Zi that is also
normally distributed but with zero mean and unit standard deviation. In symbols this is as
follows:
σ
µ−
= i
i
X
Z
Table A.1 in Appendix A at the end of this book presents the standardized normal distribution
in tabular form. Note further that the original variate may be a sample mean, X . If so, the
denominator is the standard error (i.e., standard deviation of the sampling distribution). We
can then define Z as some number of standard errors away from the mean of the sampling
distribution
X
X
Z
σ
µ−
=
where X
σ denotes the standard error of the mean. (A similar idea is involved in the case of
the standard error of the proportion.)
6. The formulas for the standard error of the mean and proportion of simple random samples
are, respectively, the following: Mean Proportion
nX
σ
σ =
nP
)1( ππ
σ
−
=
140
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
7. If the population standard deviation σ is not known, which is often the case, we can estimate
it from the sample observations by use of the following formula:
1
)(
1
2
1
−
−
=
∑=
n
XX
s
n
i
We can consider s to be an estimator of the population standard deviation, σ . In small
samples (e.g., less than 30), the t distribution of Table A.2 in Appendix A is appropriate for
finding probability points. However, if the sample size exceeds 30 or so, the standardized
normal distribution of Table A.1 is a good approximation of the t distribution.
In cases where σ is estimated by s, the standard error of the mean becomes
n
s
est X
=σ.
where X
est σ. denotes the fact that σ is estimated from s, as defined in the preceding
equation.
8. Analogously, in the case of the standard error of the proportion, we can use the
sample proportion p as an estimate of π to obtain
n
pp
est P
)1(
.
−
=σ
as an estimated standard error of the proportion. Strictly speaking, n–1 should appear in the
denominator. However, if n exceeds about 100 (which is typical of the samples obtained in
marketing research), this adjustment makes little difference in the results.
Methods of Estimating Sample Size
In our discussion of sample planning we pointed out that there are four traditional
approaches to determining sample size.
• First, the analyst can simply select a size either arbitrarily or on the basis of some
judgmentally based criterion. Similarly, there may be instances where the size of sample
represents all that was available at the time—such as when a sample is composed of
members of some organization and data collection occurs during a meeting of the
organization.
• Second, analysis considerations may be involved and the sample size is determined from
the minimum cell size needed. For example, if the critical aspect of the analysis required
a cross-tabulation on three categorical variables that created 12 cells (2 categories x 3
categories x 2 categories = 12 cells), and it was felt that there should be at least 30
observations in a cell, then the absolute minimum sample size needed would be 360.
• Third, the budget may determine the sample size. If, for example, the research design for
an online survey of physicians, the cost of each interview was estimated to be $50, and
the budget allotted to data collection was $10,000, then the sample size would be 200.
It may appear that these methods are for nonprobability samples. While this certainly is
true, these methods are also applicable to probability samples and have occasionally been used
141
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
for such samples. For probability samples, the precision must be determined after the fact.
• In a fourth approach to sample size determination, we specify the desired precision in
advance and then applying the appropriate standard error formula to calculate the sample
size. This approach using traditional inference (Neyman-Pearson), relies on either of two
major classes of procedures available for estimating sample sizes. The first and better
known of these is based on the idea of constructing confidence intervals around sample
means or proportions. This can be called the confidence-interval approach. The second
approach makes use of both type I (rejecting the true null hypothesis) and type II
(accepting a false null hypothesis) error risks and can be called the hypothesis-testing
approach. We discuss each of these approaches in turn.
Before doing this, however, two points must be made. First, as with the other approaches,
the analyst must still calculate the standard error after data collection in order to know what it is
for the actual sample that provided data. Second, the size of sample that results from traditional
inference refers to the completed (or resulting) sample. Depending on the data collection method
used, the original survey delivery may have to be much larger. For example, suppose that the
size of the desired sample was 582. A mail survey is used for data collection and past experience
has shown that the response rate would be around 25%. The original sample size in this case
would have to be 2,328 so that 582 responses would be obtained.
The Confidence Interval Approach
It is not unusual to construct a confidence interval around some sample-based mean or
proportion. The standard error formulas are employed for this purpose. For example, suppose a
researcher sampled 100 student consumers and noted that their average per capita consumption
of specialty fruit/energy drinks was 2.6 pints per week. Past studies indicate that the population
standard deviation σ can be assumed to be 0.3 pint.
With this information, we can find a range around the sample mean level of 2.6 pints for
which some prespecified probability statement can be made about the process underlying the
construction of such confidence intervals.
For example, suppose that we wished to set up a 95% confidence interval around the
sample mean of 2.6 pints. We would proceed by first computing the standard error of the mean:
03.0
100
3.0
===
nX
σ
σ
From Table A.1 in Appendix A we find that the central 95% of the normal distribution
lies within ±1.96 Z variates (2.5% of the total area is in each tail of the normal curve).
With this information we can then set up the 95% confidence interval as
)03.0(96.16.296.1 ±=± X
X σ
and we note that the 95% confidence interval ranges from 2.54 to 2.66 pints.
Thus, the preassigned chance of finding the true population mean to be within 2.54 and 2.66
pints is 95%.
This basic idea can be adapted for finding the appropriate sample size that will lead to a
142
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
certain desired confidence interval. To illustrate, let us now suppose that a researcher is
interested in estimating annual per capita consumption of specialty fruit/energy drinks for adults
living in a particular area of the United States. The researcher knows that it is possible to take a
random sample of respondents in the area and compute a sample mean. However, what the
researcher really wants to do is be able to state with, say, 95% confidence that the population
mean falls within some allowable interval computed about the sample mean. The researcher
wants to find a sample size that will permit this kind of statement.
The Case of the Sample Mean
Let us first assume that the allowable error is 0.5 gallon of energy drinks per capita and
the level of confidence is 95%. With this in mind, we go through the following checklist:
1. Specify the amount of error (E) that can be allowed. This is the maximum allowable
difference between the sample mean and the population mean. EX ± , therefore,
defines the interval within which µ will lie with some prespecified level of
confidence. In our example, the allowable error is set at E, or 0.5 gallon per year.
2. Specify the desired level of confidence. In our illustrative problem involving specialty
fruit/energy drink consumption, the confidence level is set at 95%
3. Determine the number of standard errors (Z) associated with the confidence level.
This is accomplished by use of a table of probabilities for a normal distribution. For a
95% confidence level, reference to Table A.1 indicates that the Z value that allows a
0.025 probability that the population mean will fall outside one end of the interval is
Z = 1.96. Since we can allow a total probability of 0.05 that the population mean will
lie outside either end of the interval, Z = 1.96 is the correct value for a 95%
confidence level.
4. Estimate the standard deviation of the population. The standard deviation can be
estimated by (a) judgment; (b) reference to other studies; or (c) by the use of a pilot
sample. Suppose that the standard deviation of the area’s population for specialty
fruit/energy drink consumption is assumed to be 4.0 gallons per capita per year.
5. Calculate the sample size using the formula for the standard error of the mean. One
standard error of the mean is to be set equal to the allowable error (E = 0.5) divided
by the appropriate Z value of 1.96.
255.0
96.1
5.0
===
Z
E
X
σ
This will assure us that the interval to be computed around the to-be-found sample mean will
have a 95% preassigned chance of being ±0.5 gallon away from the population mean.
6. Neglecting the finite multiplier, we then solve for n in the formula
nZ
E
X
σ
σ == or
nX
0.4
255.0 ==σ
Hence, n ≅ 246 (rounded)
143
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
7. In general, we can find n directly from the following formula:
246
)5.0(
)96.1(16
2
2
2
22
≅==
E
Z
n
σ
If the resulting sample size represents a significant proportion of the population, say 10% or
more, the finite population multiplier is required and the sample size must be recalculated
using the following formula, where N is the size of the population:
222
)22
(
ZNE
ZN
n
σ
σ
+
=
The Case of the Sample Proportion
Suppose that, in addition to estimating the mean number of gallons of specialty
fruit/energy drinks consumed per capita per year, the researcher is also concerned with
estimating the proportion of respondents using one or more specialty fruit/energy drinks in the
past year. How should the sample size be determined in this case?
The procedures for determining sample size for interval estimates of proportions are very
similar to those for interval estimates of means. In this case the following checklist would be
used:
1. Specify the amount of error that can be allowed. Suppose that the desired reliability is such
that an allowable interval of 05.0±=−πp is set; that is, the allowable error E is 0.05, or 5
percentage points.
2. Specify the desired level of confidence. Suppose that the level of confidence here, as in the
preceding problem, is set at 95%.
3. Determine the number of standard errors Z associated with the confidence level. This will be
the same as for the preceding estimation; Z = 1.96.
4. Estimate the population proportion, )(π . The population proportion can again be estimated
by judgment, by reference to other studies, or by the results of a pilot sample. Suppose that
π is assumed to be 0.4 in this case; that is, the researcher assumes that 40% of the population
used one or more specialty fruit/energy drinks last year.
5. Calculate the sample size using the formula for the standard error of the proportion. One
standard error of the proportion is to be set equal to the allowable error (E = 0.05) divided by
the appropriate Z value of 1.96.
0255.0
96.1
05.0
===
Z
E
P
σ
6. Neglecting the finite multiplier, we then solve for n in the following formula:
nnZ
E
P
)6.0(4.0
0255.0
)1(
==
−
==
ππ
σ
Hence, n ≅ 369 (rounded)
7. In general, we can find n directly from the formula
369
)05.0(
)96.1)(4.01(4.0)1(
2
2
2
2
≅
−
=
−
=
E
Z
n
ππ
Once again, if the resulting n is 10% or more of the population size, the finite population
multiplier is required and the sample size can be computed from the following:
144
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
22
2
)1(
)1(
ZNE
ZN
n
ππ
ππ
−+
−
=
Determining Sample Size When More Than One Interval Estimate Is to Be Made
from the Same Sample
The usual case when collecting sample data for estimation of various parameters is that
more than one estimate is to be made. The sample size for each of the estimates will usually be
different. Since only one sample is to be chosen, what size should it be?
A strict adherence to the allowable error and the confidence levels specified in the
calculation of the sample sizes for the individual estimation problems leaves no choice but to
take the largest sample size calculated. This will give more precision for the other estimates than
was specified but will meet the specification for the estimate for which the size of sample was
calculated. In the specialty fruit/energy drink consumption problem, for example, the sample size
would be 369 (the sample size calculated for estimating the proportion of users) rather than 246
(the sample size calculated for estimating the mean amount used). Remember, the sample sizes
determined in this manner are for obtained samples. In order to determine the size of the original
sample the researcher must estimate the rate of response expected. For example, if a mail survey
was to be conducted and it was believed that only a 20% response would be obtained, a desired
obtained sample of 250 would need an original sample of 1,250.
Devices for Calculating Sample Size
In practice, manual devices (paper nomographs) and online sample size calculators can
be used to easily compute a sample size helpful in rough-guide situations where the researcher is
not sure of either allowable error levels or population standard deviations. One such calculator is
located at http://marketing.byu.edu/samplesizecalculator.html .
The Hypothesis-Testing Approach
As indicated earlier, sample sizes can also be determined (within the apparatus of
traditional statistical inference) by the hypothesis-testing approach. In this case the procedures
are more are more elaborate. We shall need both an assumed probability of making a type I
error—called the alpha risk—and an assumed probability of making a type II error—called the
beta risk. These risks are, in turn, based on H0: the null hypothesis, and H1: the alternate
hypothesis.
In hypothesis testing the sample results sometimes lead us to reject H0 when it is true.
This is a type I error. On other occasions the sample findings may lead us to accept H0 when it is
false. This is a type II error. The nature of these errors is shown in Table 7.4.
A numerical example should make this approach clearer. We first consider the case for
means and then the case for proportions. Before doing this, however, a few words on the
relationship between the Type I and Type II errors are in order. The relationship between these
two errors is an inverse. The ability of a sample to protect against the type II error is called
statistical power.
145
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 7.3 Sample Size When Estimating Population Mean and Proportion (Selected Samples)
SOURCE: Tatham, 1979, p.b.
Table 7.4 Types of Error in Making a Wrong Decision
Action Ho is true Ho is false
Accept Ho
Reject Ho
No error
Type I error (α)
Type II error (β)
No error
When the hypothesis is one of difference, a type II error occurs when what is really
chance variation is accepted as a real difference. Taking statistical power into account often
indicates that larger (and, thus, more costly) samples are needed. Sample size is affected by the
effect size—the needed or expected difference. When it is large, say more than 15 percentage
points, statistical power is usually not a problem. Very small effect sizes (e.g., two percentage
points) require such large sample sizes as to be impractical; surveys cannot really reliably
measure such small differences or changes. As an illustration, if a researcher wants to discern an
effect size of, say, 10 points at a 95 % confidence level and a desired power to detect the
superiority of one alternative over another of 80%, the approximate sample size needed would be
310. If the desired power is raised to 95%, the sample size needed would be about 540 (Semon,
1994).
The researcher and the manager as well, must consider the relative business risks when
setting appropriate levels of protection against the two types of errors (Semon, 1990).
146
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The Case Involving Means
As an illustrative example, let us assume that a store test of a new bleaching agent is to be
conducted. It has been determined earlier that if the (population) sales per store average only 7
cases per week, the new product should not be marketed. On the other hand, a mean sales level
of 10 cases per week would justify marketing the new product nationally. Using methods of
traditional inference, how should the number of sample stores for the market test be determined?
The procedures are similar to those for interval estimation problems but are somewhat more
complicated. Specifically, we go through the following checklist:
1. Specify the values for the null (H0) and the alternate (H1) hypotheses to be tested in
terms of population means 0µ and 1µ , respectively. (By convention, the null
hypothesis is the one that would result in no change being made, if accepted.) In the
bleach market-introduction problem, the values are set at H0: 0µ = 7 cases per week,
and H1: 1µ = 10 cases per week.
2. Specify the allowable probabilities (α and β, respectively) of type I and type II errors.
The type I error is the error of rejecting a true null hypothesis. The type II error is
made when the alternate hypothesis is rejected when it is true. α and β are the
allowable probabilities of making those two types of errors, respectively. They are
shown graphically in Figure 7.3, where we assume that in the bleach-introduction
problem the allowable probabilities of error are assigned as α = 0.05 and β = 0.01.
Figure 7.3 Alpha and Beta Risks in the Hypothesis-Testing Approach
3. Determine the number of standard errors associated with each of the error
probabilities α and β. For a one-tailed test the Z values for the 0.05 and 0.01 risks,
respectively, are found from Table A.1 in Appendix A to be Z α = 1.64 and Z β = 2.33.
These are shown in figure 7.3. Note that in the figure we affix a minus sign to the
value of Z β since the critical values lies to the left of 1µ = 10.
147
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
4. Estimate the population standard deviation σ . In the case of the new bleach the
standard deviation of cases sold per store per week is assumed to be five cases.
5. Calculate the sample size that will meet the α and β error requirements. Because two
sampling distributions are involved, a simultaneous solution of two equations is
required to determine the sample size and critical value that will satisfy both
equations. These equations are the following:
critical value =
n
Z
σ
µ α+0
critical value =
n
Z
σ
µ βα−1
6. Setting the right-hand side of these two equations equal and solving for n gives
2
01
22
)(
)(
µµ
σβα
−
+
=
ZZ
n
In the bleach problem the desired sample size is
≅
−
+
= 2
22
)710(
5)33.264.1(
n 44 stores (rounded)
Having solved for n, the sample size, we can then go on to solve for the critical value for the
mean number of cases by means of the following substitution:
critical value
n
Z
σ
µ β−= 1
= 10 – (2.33)
44
5 = 8.24 cases
Alternatively, we could find the critical value from the first of the two equations:
critical value = 7 + (1.64)
44
5 = 8.24
The decision rule then becomes the following: Take a sample of 44 stores for the
controlled store test. If the mean number of cases of the new bleach sold per week in the
sample stores is less than or equal to 8.24 cases, do not introduce the product. If the mean
number of cases of bleach sold per week is greater than 8.24 cases, introduce the product.
The Case Involving Proportions
For sample-size determination involving proportions, the following analogous steps are
required:
1. Specify the values of the null (H0) and the alternate (H1) hypotheses to be tested in terms of
population proportions 0π and 1π , respectively.
2. Specify the allowable probabilities (α and β, respectively) of type I and type II errors.
3. Determine the number of standard errors associated with each of these error probabilities (Zα
and Zβ).
4. Calculate the desired sample size n from the formula:
148
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
2
01
1100 )1()1(








−
−+−
=
ππ
ππππ βα ZZ
n
This formula is appropriate for relatively large samples (n≥100), where the normal distribution
is a good approximation to the binomial. To illustrate its application, suppose that a researcher is
interested in the true proportion of residents in a large city who would be willing to pay over
$400 for a portable refrigerator-bar combination if it were commercialized.
Assume that the marketing researcher would recommend commercialization of the firm’s
refrigerator-bar combination if the true proportion of consumers who would pay over $400 for
this class of goods is 70%. If the proportion is only 60%, the researcher would not recommend
commercialization. The hypotheses are:
H0: 0π = 0.6
H1: 1π = 0.7
The alpha risk associated with the null (status quo) hypothesis is selected by the researcher to be
0.05 if the true proportion π is equal to 0.6. Moreover, the researcher is willing to assume a beta
risk of 0.1 if the true proportion is equal to 0.7. With these assumptions it is possible to obtain
the approximate sample size by using the preceding formula:
2
01
1100 )1()1(








−
−+−
=
ππ
ππππ βα ZZ
n
where Zα = Z0.05 = 1.64, Zβ = Z0.1 = 1.28, 0π = 0.6, and 1π = 0.7. The solution follows:
193
6.07.0
)3.0(7.028.1)4.0(6.064.1
2
≅








−
+
=n (rounded)
Accordingly, in this example the sample size to take is 193. The critical value can be found
analogously as follows:
critical value =
n
Z
)1( 11
1
ππ
π β
−
−
=
193
)3.0(7.0
)28.1(7.0 − = 0.658
In this case the decision rule is the following: Take a sample of 193 residents. If the sample
proportion who would pay over $400 is less than or equal to 0.658, do not commercialize the
refrigerator-bar combination. If the sample proportion exceeds 0.658, commercialize the product.
DETERMINING SAMPLE SIZE FOR OTHER PROBABILITY–SAMPLE
DESIGNS
Thus far we have discussed only the determination of simple random-sample sizes using
the methods of traditional statistical inference. How are the sizes for other types of randomsample
designs—systematic, stratified, cluster, area, and multistage—determined?
149
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The answer to this question is that the same general procedures are used to determine the
overall sample size, but the formulas for the standard errors differ. The formulas become more
complex and difficult to estimate as one considers stratified sampling, cluster sampling, or the
other more elaborate designs. This is because the standard error for these designs is partially a
function of the standard deviation (or proportion) of each stratum or cluster included in the
design. For a multistage sample consisting of several strata in one stage followed by clusters in
another and systematic sampling in a third, the standard error formula can become very complex
indeed. And once the overall sample size is determined; it must be apportioned among the strata
and clusters, which also adds to the complexity.
Appropriate formulas for estimating standard errors and sample sizes for other randomsample
designs are available elsewhere (Sudman, 1976; Kish, 1965). In general, as compared
with the size of simple random samples, systematic samples may be the same (for purposes of
calculating the standard error, the assumption is typically made that the systematic sample is a
simple random sample). Stratified samples are usually smaller, and cluster samples will usually
be larger in size to provide the same reliability as a simple random sample. If used properly,
stratification usually results in a smaller sampling error than is given by a comparable-size
simple random sample. Consequently, the advantages of a stratified sample design over a simple
random sampling design are as follows (Sangren, 2000, p. 67):
• For the same level of precision, one would need a smaller sample size in total, and
this leads to a lower cost
• For the same total sample size, one would gain a greater precision for any
estimates that are to be made
EVALUATION OF THE TRADITIONAL APPROACH
If one were to devise the ideal method of determining sample size, at a minimum one
would want it to meet the criteria of being:
1. Logically complete
2. Adaptable to a wide range of sampling situations
3. Simple to use
If the traditional (Neyman-Pearson) approach to sample-size determination were to be
rated on these criteria, the rating would be low for logical completeness and high for both
adaptability and simplicity.
The traditional approach is logically incomplete because sample size is specified as being
a function only of the conditional probabilities of making errors. The conditional costs of wrong
decisions, prior probabilities, nonsampling errors, and the cost of sampling are not considered in
the model. More advanced texts, however, do consider traditional sample-size determination by
means of formulas that include the costs of sampling (see Cochran, 1963, Chapter 4).
The fact that these variables are excluded implies that somehow they must be taken into
account outside the model. However, the only way that accommodation can be made is through
adjustment of either the specified confidence level or the assigned alpha and beta risks.
150
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
SUMMARY
One of the most difficult problems in research design is the one concerned with the size
of sample to take. We discussed the determination of sample size from the standpoint of
traditional inferential methods and all aspects of sample size determination were covered.
Estimation and hypothesis-testing applications were discussed. We concluded the chapter with
an evaluation of this traditional approach to determining sample size.
REFERENCES
Cochran, W. G. (1963). Sampling techniques (2nd ed.). New York: Wiley.
Deming, W. E. (1960). Sample designs in business research. New York: Wiley.
Dillman, D. A. (2000). Mail and internet surveys: The tailored design method (2nd ed.). New
York:Wiley.
McDowell, Elisa (2002), "Inside the paper's election polls", The Charleston Post and Courier
Accessed August 21, 2009 http://www.accessmylibrary.com/coms2/summary_0286-
2181489_ITM
Kish, L. (1965). Survey sampling. New York: Wiley.
National Center for Health Statistics (2009), Data: Cell Phone-Only Households State-by-State,
Associated Press, as reported on Fox News, March 11, 2009
http://www.foxnews.com/story/0,2933,508926,00.html;
http://www.cdc.gov/nchs/data/nhis/earlyrelease/wireless200905.htm
http://marketing.byu.edu/samplesizecalculator.html
http://www.qualtrics.com/wiki
Sangren, S. (2000, April). Survey and sampling in an imperfect world: More second-best
solutions. Quirk’s Marketing Research Review, pp. 16ff.
Semon, T. T. (1990, January 8). Don’t panic when sample isn’t quite what you expected.
Marketing News, 24, 31.
Semon, T. T. (1994, January 3). Save a few bucks on sample size, risk millions in opportunity
loss. Marketing News, 28, 19.
Semon, T. T. (2000, June 9). Quality sampling, research means avoiding sloppiness. Marketing
News, 34, 11.
Sudman, S. (1976). Applied sampling. New York:Academic Press.
151
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Survey Methods Centre (1995). Quota versus probability sampling. Survey Methods Centre
Newsletter, 15(1).
Tatham, E. (1979). A report on sample size selection. Shawnee Mission, KS: Customer
Satisfaction Institute.
Unlisted phone numbers on the rise (1989)
http://findarticles.com/p/articles/mi_m3065/is_n8_v18/ai_7812479/
Zinkhan, G., Burton, S., & Wallendorf, M. (1983). Marketing applications for snowball
sampling. In W.R. Darden et al. (Eds.), Research methods and causal models in marketing.
Chicago: American Marketing Association.
152
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This page is intentionally blank
153
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
CHAPTER 8
EXPERIMENTATION
Experimentation is widely used in marketing research. Marketing experiments have been
conducted in such diverse activities as evaluating new products, selecting advertising copy
themes, determining the frequency of salespeople's calls, and evaluating all aspects of a movie
(including ending, pacing, music, and even the story line). For example, as shown in Exhibit 8.1
the ending of the very successful movie Fatal Attraction was changed because test audiences did
not like the original ending.
This chapter discusses the objectives of experimentation illustrates techniques for designing and
analyzing marketing experiments including:
• the nature of experimentation
• ingredients of a marketing experiment
• sources of invalidity
• models of experimental design
• panels and experimental design
• difficulties in field experiments in marketing
This chapter further discusses experimental designs and logic for advanced surveys using the
capabilities of Qualtrics.com, including the following:
• piping of text, graphics and experimental treatments
• simple branching and compound branching logic, looping and piping of answers
• question and treatment blocks
• quota fulfillment
• randomization of answer choices, questions, treatment blocks and alternative
questionnaire forms.
The Nature of Experimentation
Two general types of experimental designs exist—natural and controlled. A natural
experiment is one in which the investigator intervenes only to the extent required for
measurement, and there is no deliberate manipulation of an assumed causal variable. “Nature”
produces the changes. In contrast, in a controlled experiment two kinds of intervention are
needed:
1. manipulation of at least one assumed causal variable
2. random assignment of subjects to experimental and control groups.
True experiments have both types of intervention, while quasi-experiments manipulate
the variables but do not randomly assign the subjects. All true experiments have certain things in
common—treatments (i.e., assumed causal variables), an outcome measure, units of assignment,
and some comparison from which change can be inferred and, it is hoped, attributed to the
treatment. Quasi-experiments, on the other hand, have treatment, outcome measures, and
experimental units but do not randomly assign subjects to treatments. Rather, subjects already
154
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
belong to groups that differ from each other in ways other than the presence of a treatment whose
effects are being tested (Cook and Campbell, 1990).
Exhibit 8-1 Test Audiences have Profound Effect on Movies
What if E.T. hadn't made it home, or Richard Gere had failed to come back for his pretty woman,
Julia Roberts? Believe it or not, in the original versions of these films, E.T. died on American soil and it
was Roberts who rejected Gere at the end of "Pretty Woman."
So who changed these potential misses into hits? It wasn't the producers or the writers who
prevailed for change – it was the moviegoers. In what may be Hollywood's last and most closely guarded
secret, the test audience is having a profound effect on the movies you watch.
Scientific Findings?
In the original version of the hit "My Best Friend's Wedding," Rupert Everett had a minor role as
Julia Roberts' gay best friend. But test audiences wanted more. So the ending was scrapped, the set
rebuilt, and Everett's character came back for one final appearance. That's what can happen if test
audiences love you. But what if they loathe you? In the 1987 thriller Fatal Attraction test audiences so
despised Glenn Close's character that they became responsible for having her killed off in the end.
Director Ron Howard and his partner, producer Brian Grazer, are responsible for hits like Far and
Away, Ransom, Apollo 13, and A Beautiful Mind. Howard says, "What I would hate to do is put the
movie out there, find out that the audience is confused about something or upset about something that
you could have fixed and realize I had no idea they'd respond that way." Grazer and Howard
were dealt one surprise when they tested the 1989 film Parenthood. "The audience told us there was
too much vulgarity in the movie," Grasier says. "We took out a lot of the vulgarity. The scores went up,
made us feel better, the movie played better. It didn't offend anybody."
Source: Adapted from Bay (1998)
Objectives
The term experimentation is used in a variety of ways and for a variety of objectives. In
the discussion of this chapter we shall use the term “experimentation” to describe an experiment
may be conducted for the purpose of identifying relevant variables as well as the functional form
of the model that links these variables with the criterion variable under study.
Perhaps the characteristic that best distinguishes experimentation from observational studies
(which are also employed in measurement and estimation) is that experimentation denotes some
researcher intervention and control over the factors affecting the response variable of interest to
the researcher.
Experimentation permits the establishment of causal relationships. In contrast,
correlation analysis (a useful technique in observational studies) permits the analyst to measure
the degree to which changes in two or more variables are associated with each other.
An Industry Example
A national producer of packaged candies was interested in children’s preference for
various formulations of one of its well-known candy bars. Type of chocolate, quantity of
peanuts, and amount of caramel were independently varied in a factorial design (described later
in the chapter) of 2 types of chocolate by 3 quantities of peanuts by 3 amounts of caramel. Paired
155
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
combinations (i.e., two combinations at a time) comparisons involving the 18 combinations were
made up and evaluated by various school children between 8 and 12 years of age. Interestingly
enough, the company found that preferences for type of chocolate varied with the amount of
caramel. In addition, while children preferred more peanuts to fewer peanuts, the intermediate
level of caramel was the most preferred. The company modified its formulation to match the
most preferred test combination.
Many other experiments have been carried out involving taste testing, package design,
advertising type and quality, price sensitivity, and other marketing variables.
Ingredients of a Marketing Experiment
An experiment involves a series of interrelated steps, as shown in Figure 8-1 (note the
similarity to Figure 2-1). Our concern in this chapter is primarily with defining variables,
designing the experimental procedure, and conducting the experiment. The other steps are
discussed elsewhere in this book, and for experimentation they do not differ from general
concepts.
Figure 8-1. Components of an Experiment
All experiments involve three types of variables. First, there is the “treatment” variable
whose effect upon some other variable the experiment is designed to measure. This is the
variable that is manipulated, and presumed to be the cause. It is often referred to as the
independent variable. Marketing experiments often involve more than one treatment variable.
When this is the case, the researcher may be interested in observing the effects of combinations
of treatment variables, in addition to the effect of each one individually. In short, there may be
interaction effects. Interaction refers to the situation like the candy bar example cited above,
where the response to changes in the levels of one treatment variable (type of chocolate) is
dependent on the level of some other treatment variable(s) (amount of caramel) in the
experiment.
The second broad type of variable in an experiment is the outcome or dependent variable.
In the preceding candy bar example, the dependent variable was product preference.
Survey literature
Survey theories
STATE THE PROBLEM
FORMULATE HYPOTHESES
DEFINE VARIABLES
DESIGN EXPERIMENTAL PROCEDURE
CONDUCT THE EXPERIMENT
ANALYZE THE RESULTS
WRITE UP REPORT
Independent
Dependent
Others
I Measurement
II Subjects
III Manipulation
V Experimental design
VI Experimental Procedures
(a) Instructions
(b) Instruments
(c) Apparatus
156
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The last category of variables consists of those other than the manipulated independent
variables that could influence the observed effects (i.e., dependent variable). These are known as
extraneous variables, and unless controlled adequately they are the source of errors in an
experiment. These will be discussed in a later section of this chapter.
Test Objects
The terms test units, test objects, and subjects are used interchangeably in discussing
experimentation. All are used to refer to the units whose responses to the experimental treatment
are being studied. In marketing research the experimenter has a choice of three possible
universes of test units—people, stores, and market areas. Which is most appropriate depends on
the problem forming the basis of the experiment (Banks, 1965, pg. 13-15).
The experimenter must contend with differences among the inherent properties of the test
objects. For example, if a researcher is interested in the effect of shelf height on the sales of a
packaged consumer product, it is to be expected that stores will vary in their amount of shopping
traffic, placement of shelving units, and the like. If the experimenter is interested in the effect of
various shelf heights on product sales over a variety of store sizes, several stores will have to be
used in the analysis. If so, he or she may use a technique called covariance analysis, in which
responses to the controlled variables (shelf height) are adjusted for inherent differences in the test
objects (stores) through measurement of these characteristics before (or during) the experiment.
Measurement, Manipulation, and Experimental Procedures
A critical aspect of all experiments, indeed of all marketing research, is measurement.
Our concern at this point is with the operational problems of measurement. The concepts, levels,
and techniques of measurement, and scaling are covered in Chapter 9.
In a marketing experiment, it is the outcome or dependent variable that is measured.
Generally, the operational measures used can be classified as physiological or psychological
measures. Physiological measures include those obtained from devices that measure such things
as eye movements or electrical conductivity of the skin (psychogalvanometer) during advertising
readership. Such devices are used in laboratory experiments. Psychological measures include
verbal measures include spoken and written responses, including responses provided
interactively with a personal computer. They may additionally include direct measures such as
dollar amounts or units of a product that are sold or consumed, and actual behavior (or assumed
actual behavior) of people under the conditions of the experiment. Such measures are obtained
by self-report, observation, or electronic means.
Turning now to manipulation, an experimental treatment must be capable of variation.
There are at least three ways in which variation in the independent variable can be achieved.
First, the type of variable can be manipulated. For example, a company interested in the effect of
image on sales could conduct an experiment by running a series of advertisements, each of which
was designed to convey a different image of its product. Variation is generated in the type of
image conveyed.
Second, there is the presence versus absence technique. For instance, one group of people could
be shown a new advertisement and their responses to an attitude measurement could be
compared with the response from a group that did not see the advertisement.
Finally, the amount of a variable can be manipulated; different amounts are administered to
different groups. This technique is used in such experiments as those where different prices for a
product are “tested” and the outcome measured as “units sold.”
157
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
As with any approach to marketing research, all phases of an experiment should be
carefully planned in advance. After decisions have been made concerning measurement,
research subjects, experimental design, control techniques, and manipulation, there is the need to
plan everything that will take place in the actual experiment itself through to the end of data
collection. This includes the setting of the experiment, physical arrangements, apparatus that
will be used, data collection forms, instructions, recording of the dependent variable, and so
forth.
Difficulty of Control
In any experiment, as previously mentioned, extraneous factors could potentially affect
the dependent variable (response). Those factors that probably would not influence the
dependent variable can be ignored. On the other hand, those extraneous variables that might
reasonably influence the dependent variable must be controlled in some way in order that error,
and thus threats to valid inference, can be minimized.
In any experiment, control over all possible variables affecting the response is rarely
possible. Even in the laboratory it is not possible to control all variables that could conceivably
affect the outcome. But compared with the laboratory situation, the researcher who is working in
the marketplace has a really difficult control job to do. The marketing researcher must try to
design the experiment so that the effects of uncontrolled variables do not obscure and bias the
nature of the response to the treatment variables that are being controlled.
There are five general techniques for controlling extraneous variables.
1. Elimination of extraneous variables, including controlling the situation in which an
experiment is being conducted so as to keep out extraneous forces. Used in this way,
control is much easier to achieve in a laboratory environment than in a field setting.
2. Constancy of conditions, including the ability to determine which subjects or test units
receive a particular treatment at a particular time. Control of the treatment variable helps
separate out the effects attributable to irrelevancies that are correlated with a treatment.
3. Matching (sometimes called balancing). Marketing researchers most often match by
equating subjects or by holding variables constant. In equating subjects, each group is
controlled for all variables except the treatment variable. For example, if it were felt that
gender and age of subject would influence taste test results, each group would have the
same gender-and-age proportions. Matching by holding variables constant involves
creating constancy for all groups. Again referring to a taste-testing experiment, the
gender variable could be controlled by using only all male or all female subjects. If time
of day is an important extraneous variable that could affect one’s “taste buds,” then the
experiment should be conducted at approximately the same time on successive days. Or,
if time of day affects subjects’ purchasing behavior (purchase occurs after a certain time),
then the experiment should not be conducted before that time; McDonald’s tested pizza
for the dinner menu and required that tests be conducted after 4:00 pm. Similarly, using
the same facilitator in all versions of the study may control experimenter effects.
4. Counterbalancing is a technique used to control confounding, or the tangling effects of
two or more levels of a treatment variable (for two or more treatment variables), used in
158
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
experimentation. An illustration should make this point clearer. Suppose that a
marketing researcher is interested in conducting a series of taste-testing experiments for a
new soft drink. Subjective interpretations of, say, “sweetness” may well vary from
subject to subject. A control procedure called counterbalancing would have each subject
taste each of two drinks on the assumption that ratings will be expressed in terms of
differences in sweetness over each subject. To avoid “ordering” effects on responses, the
new and the control drinking would be presented in randomized order or one-half of the
group would follow the sequence “established-new” while the other one-half would use
the sequence “new-established.” To reduce carryover tendencies, the subject would be
asked to take a sip of water between testing trials.
5. Randomization provides assurance that known and unknown extraneous factors will not
cause systematic bias. Statisticians have made a major contribution to experimental
design in the development of statistical models that feature randomization over
uncontrolled variables. Consequently, it is assumed that the effects of the extraneous
variables will affect all groups in an experiment to the same extent. In theory,
randomization is supposed to accomplish the conversion of all irrelevant sources of
possibly systematic variability into unsystematic variability, i.e, into random error
(Brown and Melamed, 1990, p. 3).
The above discussed techniques for control of extraneous variables are a key part of all
experiments, however the fact remains that confounding effects can never be entirely eliminated.
Sources of Invalidity
In Chapter 3, we briefly introduced experimental errors as the extraneous forces that can
affect the outcome of an experiment. Each extraneous force potentially has a bearing on the
validity of an experiment and, consequently may threaten the validity of the results.
In the context of experimentation, the term validity refers to the extent to which we really
observe (or measure) what we say we observe (or measure). Four distinct types of validity have
been identified (Cook and Campbell, 1990, chap. 2):
• Statistical conclusion
• Internal
• Construct
• External
A necessary condition for inferring causation is that there be covariation between the
independent variables. Statistical conclusion validity involves the specific question as to
whether the presumed independent variable, X, and the presumed dependent variable, Y, are
indeed related (Rosenthal and Rosnow, 1991, chap. 3). After it has been determined that the
variables covary, the question arises as to whether they are causally related. This is the essence
of internal validity. A given experiment is internally valid when the observed effect is due
solely to the experimental treatments and not due to some extraneous variables. In short, internal
validity is concerned with how good the experiment is, as an experiment. The third type of
validity, construct validity, is essentially a measurement issue. The issue revolves around the
159
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
extent to which generalizations can be made about higher-order constructs from research
operations and is applicable to causes and effects. Because construct validity is concerned with
generalization, it is a special aspect of external validity. External validity however, refers to the
ability to generalize a relationship beyond the circumstances under which it is observed. That is,
external validity is concerned with how good an experiment is: to what degree can conclusions
of an experiment be applied to and across populations of persons, settings, times, and so on.
To a large extent, the four kinds of validity are not independent of each other. That is,
ways of increasing one kind may decrease another kind. Consequently, in planning an
experiment it is essential that validity types be prioritized, and this varies with the kind of
research being done. For applied marketing research, it has been observed that the priority
ordering is internal, external, construct of the effect, statistical conclusion, and construct of the
cause (Cook and Campbell, 1990). Accordingly, we now examine those extraneous factors that
affect internal and external validity. Construct validity is discussed in Chapter 9, and factors
affecting statistical conclusion validity are covered throughout the later sections of this book.
Internal Validity
Internal validity is concerned with whether or not the observed effect is due solely to the
experimental treatments or due to some other extraneous variables. The kind of evidence that is
required to support the inference that independent variables other than the one(s) used in an
experiment could have caused the observed effect(s) varies, depending on the independent
variable(s) being investigated. However, there are some general classes of variables affecting
designs that deserve mention. The following factors affect internal validity:
1. History. An extraneous event that takes place between the pre-measurement and
post-measurement of the dependent variable has an impact on the results.
2. Maturation. The results of an experiment are contaminated by changes within the
participants with the passage of time.
3. Testing. A prior measurement can have an effect on a later measurement.
4. Instrumentation. Changes in the measuring instrument or process, including
interviewers’ instructions, over time, affects the results.
5. Selection. Different kinds of research subjects have been selected for at least one
experimental group than have been selected for other groups.
6. Mortality. Different types of persons drop out from experimental groups during
the course of an experiment.
7. Statistical Regression. This error may arise when experimental groups have been
selected on the basis of extreme pretest scores or correlates of pretest scores.
8. Interactions with Selection. Independent effects of the interaction of selection
with history and maturation have an impact on experimental results.
Example: To illustrate each of these factors, suppose that a controlled experiment is
setup for salesperson retraining. Assume one group of salespeople had taken a
retraining course during a three-month period, while another group had not been
retrained. The brand manager of a particular brand of detergent wants to determine if
retraining is a producer (i.e., a cause) of sales performance. During the three-month
period after retraining, sales of the detergent showed an unusually large increase.
160
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
History deals with events outside the design that affect the dependent variable. History is
therefore comprised of the producers that are extraneous to the design. For salespersonretraining,
the level of competitive promotion and advertising, the overall level of demand, or
any one of many other producers may have changed substantially in some territories. Clearly,
the longer the time period involved, the greater the probability that history will significantly
affect the results.
Maturation is concerned with the changes that occur with the passage of time in the people
involved in the design. For example, as time passes, salespeople gain more experience in selling
and hence know their customers better, and the customers become better acquainted with the
product. Similarly, some biological and psychological changes within salespeople and
customers over time can affect the performance of either.
Testing effect has to do with the effect of a first measurement on the scores of a second
measurement. Familiarity with a measurement (i.e., a test) can sometimes enhance performance
because items and error responses are more likely to be remembered at later measurement
sessions. Suppose that the attitudes of all salespeople toward their job, performance, the
company, and so on, had been measured prior to and after the retraining program. There is the
possibility that some responses obtained after the retraining program were due to the salespeople
remembering responses given on the first measurement.
Instrument effect refers to the changes in the measuring instrument or process that may affect
the measurements obtained. This can occur in many ways, including changes in cost-value (a
drop in home prices) over time, the learning process on the part of the investigators, a change in
investigators, or simply boredom or fatigue may affect the measurements and thus the
interpretation of results. A related issue arises when the research subjects themselves self report
or record responses. In this case, there exists the possibility of a confounding of the instrument
with the experimental treatment because the subjects’ understanding or awareness of the variable
being measured may have changed.
Selection is concerned with the effect of the selection procedure for the test and control
groups on the results of the study. If the selection procedure is randomized, the effect will be a
measurable random variation. However, if self-selection or some other nonrandom (purposive)
procedure occurs, the results will be affected in a non-measurable manner. Sizable systematic
errors may well result. The concern, of course, is that the resulting groups may differ on
important characteristics, and these differences may be influencing the dependent variable.
Statistical regression dictates that under conditions where measures are unreliable, high
pretest scorers will score relatively lower at the posttest and low pretest scorers will score higher.
There is a tendency with repeated measures for scores to regress to the population mean of the
group (Cook and Campbell, 1990).
Some of the foregoing error sources affecting internal validity can interact with selection
to produce forces that might appear to be treatment effects. Selection-maturation results when
experimental groups mature at different speeds. Selection-history can occur when the
experimental groups come from different settings. For example, the salespeople receiving the
retraining all come from one region of the country, whereas those not retrained come from some
other region. In this situation, each group may have a unique local history that might affect
outcome variables.
161
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
External Validity
External validity is concerned with whether or not conclusions are externally valid, that
is, whether conclusion apply to and across populations of persons, settings, times, and so on.
The following are sources of error in external validity:
1. Reactive Effect of Testing (Interaction). Pre-measurement may have an impact on the
experimental subject’s sensitivity or responsiveness to the treatment variable.
2. Reactive Effects of Experimental Situation. Experimental subjects react to the
experimental situation (such as setting, arrangements, and experimenter).
3. Interaction of History and Treatment. Measuring the dependent variable at a point in
time that does not reflect the actual effect of the independent variables can skew
results.
4. Interaction of Selection and Treatment. The method of selection affects the extent to
which the measured effect can be generalized to the population of interest.
A frequent problem in research design is that a “before” measurement is desired, but it is
recognized that making such a measurement may alert the subjects that they are participating in a
study. If an experiment is being conducted on the impact of advertising, a pre-measurement may
sensitize an experimental subject to pay particular attention to a company’s advertising,
product(s), or both. Although this potential source of error may seem the same as that from
testing, it is distinctly different that it is an interaction of testing and treatment. Despite the
plausibility, the empirical evidence of a pretest treatment interaction bearing on attitude change
research is, in general, meager.
Experimental situations refers to “all aspects of the experiment which cause the subject to
perceive, interpret, and act upon what he believes is expected or desired of him by the
experimenter (Sawyer, 1975, p. 20). This includes the setting within which the experiment is
conducted, the arrangements made for the experiment such as the apparatus used, and the
presence and behavior of an experimenter. These reactive effects involve both experimenter and
subject effects (Christensen, 2004, Chap. 7).
To a large extent, the effects of demand characteristic depend on the roles and
expectations adopted by the experimenter and the subjects, particularly when subjects become
aware of, or believe they know, the experimental hypothesis. Ideally, the investigator would
seek subjects who are faithful or naïve (i.e., do not know the hypothesis).
Unfortunately, you can never be absolutely certain that demand characteristics will not be
present in the chosen experimental procedures. Research design, measurement of dependent
variables, and use of procedures (e.g., deception and natural environments) will reduce such
demand characteristics (Sawyer, 1975, pp. 25-8). In addition, manipulation and confounding
checks can be used to investigate the plausibility of demand characteristics (Perdue and
Summers, 1986, p. 38).
History-treatment interaction can affect external validity when the dependent variable
measurement and the treatment variable measurement are not time generalized. For example, if
an experiment were conducted on the effectiveness of advertising and this experiment was run at
a time close to any holiday, then a question would remain as to whether the same cause-effect
relationship would exist at some other time during the year. Moreover, with some independent
variables their effects may be long-run in nature, but typically dependent variable measurement
occurs at one point in time.
162
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Selection-treatment interaction asks if the observed effects can be generalized beyond the
groups used to establish the relationships? Even when research subjects belong to the target
group of interest, the recruitment approach used may limit any generalizations to only those who
participated in a given experiment. Experiments, including laboratory and Internet based
experiments, rely heavily upon volunteers, paid or otherwise. There is some evidence that
volunteer experimental subjects differ from non-volunteer subjects in more or less systematic
ways: they are better educated, higher social class, higher need for social approval, more
sociable, more intelligent (Rosenthal and Rosnow, 1975, pp. 88-90).
Models of Experimental Design
A number of experimental designs have been developed to overcome and reduce the
various sources of invalidity. Experimental designs can be categorized into two broad groups—
classical and statistical. Classical designs consider the impact of only one dependent variable at
a time, whereas statistical designs allow for examining the impact of two or more independent
variables.
Classical Designs
The major types of classical designs were mentioned earlier in this chapter. Preexperimental
designs are so called because there is such a total absence of control and are of
value in exploratory research, but are of minimal value in establishing causality. Quasiexperimental
designs involve control but lack random assignment of subjects, as required for true
experiments. Where any given design fits in this classification spectrum will depend on whether
the treatment variable has been deliberately manipulated, the nature of control, and whether there
has been random assignment of subjects to experimental groups.
The following notational system will be used in the discussion of classical experimental
designs:
• X represents the exposure of test groups to an experimental treatment of a producer or
event whose effect is to be observed and/or measured.
• O refers to a measurement or observation taken.
• R indicates that individuals have been assigned at random to differing treatments.
• Movement from left to right indicates a sequence of events. When O’s and X’s are
found in a given row, they are to be interpreted as having occurred in sequence and to
the same specific individual or group. Vertical arrangement of symbols is to be
interpreted as the simultaneous occurrence of the events that they denote.
Three classes of designs seem to fit for pre- and quasi-experiments:
• Time-series and trend designs
• Cross-sectional designs
• Combinations of the two previous classes
Time-Series and Trend Designs
Time-series and trend designs are similar in concept, yet their differences in
implementation and analytic procedures warrant a brief discussion. A time-series design
involves obtaining data from the same sample (or population) for successive points in time. The
common method of gathering primary data of this kind is to collect current data at successive
intervals through the use of a continuous panel. One may, however, collect current and
163
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
retrospective data from respondents during a single interview. If the latter technique issued,
respondent recall must be relied on to reconstruct quasi-historical data. An alternative method of
obtaining data for past periods is to use secondary sources, when available.
Trend data differ from time-series data in that they are obtained from statistically
matched samples drawn from the same population over time. Current data are gathered from
each successive sample.
Both time-series and trend data are used to investigate the existence and nature of causal
relationships based on associative-variation and sequence-of-events types of evidence. While
individuals or households are the most commonly used sample units, data are also obtained from
retail stores, wholesalers, manufacturers, and other units.
Time-series and trend designs involve at least one treatment and a subsequent
measurement and can involve a large number of measurements with several interspersed
treatments over a long period of time. We now describe and discuss four types of time-series
and trend designs.
After-Only without Control Group
This design is often termed a “tryout” or “one-shot case study.” It is the simplest of all designs,
as it involves only one nonrandomly selected group, one treatment, and one measurement.
Symbolically, it may be diagrammed as follows:
X O (1)
The many weaknesses of this design may be illustrated by applying it to the salespersonretraining
problem. Assume that no prior measurement of sales volume of the salespeople to be
retrained had been made. A group of salespeople are selected by a nonrandom method and
retrained (X); a measurement (O) is made after the retraining.
Since no prior measurement of sales volume was made for each of the retrained
salespeople, there is no method, short of making assumptions as to what would have happened in
the absence of retraining, of estimating what the effect of retraining was. The effects of history,
maturation, and selection are all potentially substantial and non-measurable. Because of these
limitations the use of this design is to be avoided if at all possible.
Before-After without Control Group
This design is the same as (1) with the addition of a “before” measurement. In its simplest form
it is shown as
O1 X O2 (2)
More advanced forms of this design include multiple measurements (O’s) before and after the
single measurement.
Although design (2) is relatively weak, it is frequently used. It is a decided improvement
over (1) in that the apparent effect of the treatment, O2 – O1, is measured. In terms of the
salesperson-retraining illustration, a measurement (O1) of sales volume of the retrained
salesperson is made for the quarter of the preceding year.
Multiple Time Series
In using a time-series design the possibility of establishing a control group should always be
investigated. It may be possible to find a comparable, if not equivalent, group to serve as a
control against which to compare the results of the group that underwent the treatment involved.
This design may be diagrammed as
164
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
O1 O2 O3 O4 X O5 O6 O7 O8 (3)
O′1 O′2 O′3 O′4 O′5 O′6 O′7 O′8
where the primed O′s represent measurement of the control group. Note that the individuals
constituting the groups were not selected at random. It may be possible, however, to select at
random the group that will receive the treatment.
This design can easily be adapted to the sales-retraining evaluation problem. If it is
assumed that the sales volume of each of the salespeople is measured during each period as a
matter of course anyway, a group could be selected for retraining after period 4. Either a
comparable group or all the rest of the salespeople could be selected as the control group. After
the sales of both groups in the periods after the training had been measured, the apparent effect
of the training would be shown by comparing the differences in average sales volume for the two
groups before and after treatment.
Cross-Sectional Designs
Cross-sectional designs involve measuring the dependent variable of interest for several
groups at the same time, the groups having been exposed to different levels of treatments of the
treatment variable whose effect is being studied. Cross-sectional designs may be viewed
diagrammatically as follows:
X1 O1
X2 O2
X3 O3 (4)
. .
Xn On
Examples of frequent applications of this design are studies of the effect of such variables as
price, package design, or level of advertising on sales in different geographic areas. This design
can be used when direct manipulation of the producer involved is not possible or practical.
When this is the case, the design is being used in a natural experiment. The effect of the
different levels of treatment is measured by determining the degree of association between
producer and product. The techniques that can be employed are discussed in later chapters.
A variation of this design is the static-group comparison. This is a design in which a
group exposed to a treatment is compared with one that was not:
X O1 (4a)
O2
History may play a critically important role in cross-sectional designs. There could be a
sizable differential effect of extraneous variables between the groups being measured.
Combination Cross-Sectional, Time-Series Designs
A number of designs employing a combination of time-series and cross-sectional
treatment and measurement may be used in observational studies. The multiple-time-series
design (3) could be considered a combination of the two types, as it involves measurements of a
product for different groups at the same time as well as for the same group over time.
Combination designs are well adapted for use with consumer-panel data. One commonly
used design is the ex post facto test-control group. In this design the test and control group are
not known until after the treatment has been administered. This design is illustrated as follows:
165
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
O1 X O3 (5)
O2 O4
This design is widely used in connection with testing the sales effectiveness of price
changes, “deals,” and advertising. Data on the sales of the brand of interest are reported
regularly by the members of a continuous consumer panel. After a given advertisement is run
(X), panel members may be questioned to determine whether or not they saw it. Those who saw
it are part of the test group, since they have had exposure to the treatment involved. Those who
did not see the advertisement become a part of the control group. The apparent effect is
determined by comparing the difference in test and control-group purchases before with that
after the advertising was run.
True Experimental Designs
As previously mentioned, for true experiments two kinds of investigator intervention are
required:
• Manipulation of at least one assumed causal variable
• Random assignment of subjects to experimental and control groups
Through use of a random selection procedure, systematic errors due to selection are eliminated
and the effects of the many extraneous variables tend to be equalized between the experimental
and the control groups as the size of these groups increases. Random selection permits the use of
inferential statistical techniques for analyzing the experimental results.
Three single-variable experimental designs are described below.
After-Only with Control Group
The simplest of all experimental designs is the after-only with control group. It requires only
one treatment and an “after” measurement of both the experimental and the control group. Yet it
has the essential requirements of the true experiment: manipulation of at least one variable and
randomly selected test and control groups. It is illustrated as follows:
R X O1 (6)
R O2
The absence of a "before" measurement (or "pre-test") is a feature that concerns many
researchers about this design. Such a measure is not actually essential to true experimental
designs because respondents are randomly assigned, controlling for extraneous variable effects.
This design is of major interest, therefore, when “before” measurements are impractical or
impossible to obtain and/or when the testing and instrument effects are likely to be serious.
A common application of this design is in the testing of direct-mail advertising.
Random-sampling procedures are used to select an experimental and a control group. Directmail
pieces are sent to the experimental group and withheld from the control group. “After”
measurements of sales to each group are made and the differential is determined (O1 – O2).
Before-After with One Control Group
If “before” measurements are added to design (6), we arrive at the following
configuration:
R O1 X O2 (7)
R O3 O4
166
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This design is very similar to that of (5) but with an important difference: the experimental and
control groups are randomly selected, rather than self-selected. Most of the sources of
systematic error are controlled in this design.
This design offers three ways to evaluate the effect of the treatments: O2 – O1, O2 – O4,
and (O2 – O1) – (O4 – O3). If the results of each of these evaluations are consistent, the strength
of our inferences about the effect of the experimental treatment is substantially increased.
An example of the use of this design is in advertising tests that use a dual cable television
system with two consumer purchase panels (one from the subscribers to each cable). “Before”
measurements can be made on the test and control panels, an experimental advertising treatment
introduced on the test cable, and “after” measurements made for both panels.
There are many classical designs. In this section, we have discussed the more common
designs. A useful way to summarize what we have discussed is to look at each in terms of
sources of potential error and whether such sources are controlled for, as shown in Table 8-1.
More detailed discussions are found in Cook and Campbell (1990), Campbell and Stanley
(1966), and Shadish, Cook and Campbell (2002).
Statistical Designs
For the most part, statistical designs are “after-only” designs (6) in which there are at
least two treatment levels. In addition, such designs can examine the effects of more than one
independent variable. Two principal aspects of statistical designs are the following:
• the experimental layouts by which treatment levels are assigned to test objects
• the techniques that are used to analyze the results of the experiment
We will now briefly discuss now the major types of layouts used to obtain data. These are
discussed in more depth in Brown and Melamed (1990) and other specialized experimental
design texts such as Field and Hole (2003) and Shadish, Cook and Campbell (2002). The
analysis techniques, known generically as analysis of variance and covariance, are discussed in
later chapters.
Completely Randomized Design
The completely randomized design is the simplest type of statistical design. In this
design, the experimental treatments are assigned to test units on a random basis. Any number of
treatments can be assigned by a random process to any number of test units.
As an illustration, suppose that a marketer is interested in the effect of three versions of
an ad on customer preferences. The marketer has been able to obtain an email list of
customers and has corporate approval to run an experiment involving the three ads.
This simple design would entail randomly showing one of the ads to each survey
respondent and measuring their degree of liking of the ad they viewed. This could be described
as classification by a single factor. This design is most applicable when it is believed that
extraneous variables will have about the same effect on all test units, and when the marketer is
interested in only one independent variable. Moreover, there is no absolute guarantee that
randomization will keep extraneous influences in check. Suppose that our marketing researcher
were interested in the effect of other variables such as the introductory message or gender of the
167
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
salesperson appearing in the ads. Or, suppose that the researcher would like to generalize the
results of the experiment to other sizes of ads or products. It may be preferable to ask many
rather than few questions of nature if the researcher would like to establish the most general
conditions under which the findings are expected to hold. That is, not only may single-factor
manipulation be difficult to do in practice, but it may be inefficient as well. We now discuss
somewhat more specialized experimental designs, all of which are characterized by two or more
variables of classification.
Table 8-1 Sources of Invalidity and Selected Experimental Designs
Internal External
History
Maturation
Testing
Instrumentation
Regression
Selection
Mortality
Interactionof
SelectionandOthers
Interactionof
TestingandX
Interactionof
SelectionandX
ReactiveArrangements
One-Shot Case Study
X O
- - - - One-Group
Pretest-Posttest Design
O X O
- - - - ? + + - - - ?
Time Series
O O O X O O O
- + + ? + + + + - ? ?
Multiple Time-Series
O O O X O O O
O O O O O O
+ + + + + + + + - - ?
Static-Group Comparison
X O
O
+ ? + + + - - - Nonequivalent
Control Group Design
O X O
O O
+ + + + ? + + - - ? ?
Posttest-Only control Group Design
R X O
R O
+ + + + + + + + + ? ?
Separate-Sample Pretest-Posttest Design
R O (X)
R X O
- - + ? + + - - + + +
Pretest-Posttest Control Group Design
R O X O
R O O
+ + + + + + + + - ? ?
Solomon Four-Group Design
R O X O
R O O
R X O
R O
+ + + + + + + + + ? ?
Note: In the tables, a minus (-) indicates that the factor is not controlled, a plus (+) indicates that the factor is controlled,
a question mark (?) indicates a possible source of concern, and a blank indicates that the factor is not relevant.
SOURCE: Adapted from Campbell, D. T., &Stanley, J.C., Experimental and Quasi-Experimental Designs for Research.
Copyright ©1963 by Houghton Mifflin Company. Adapted by permission.
168
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Factorial Designs
A factorial experiment is one in which an equal number of observations is made of all
variable combinations. The variables must have at least two levels or categories each. In
essence, the factorial design is one that has combined two or more completely randomized
designs into a single experiment. This type of experiment enables the researcher to study
possible interactions among the variables of interest. Suppose we return to our Ad illustration
but now assume that the researcher is interested in studying the effects of two variables of
interest: Ad Spokesperson type (at three levels – Male, Female, Cartoon Character ) and
Message Type (at two levels – Benefits Message and Humorous Message). The design is shown
in Table 8-2. Note that each combination of MiSj occurs only once in the design. While the plan
still is to use a single sample for the experiment, the researcher intends to randomize the
presentation of each combination among the various respondents (each receiving one Ad).
In the factorial experiment we can test for all main effects (i.e., spokesperson types,
messages), and in this case, where we have replicated each combination, for the interaction of the
variables as well. If the interaction term is significant, ordinarily the calculation of main effects
is superfluous, since the experimenter will customarily be interested in the best combination of
variables. That is, in market experimentation the researcher is typically interested in the
combination of controlled variables that leads to the best payoff in terms of preference, sales,
market share, cash flow, or some other measure of effectiveness.
Table 8-2 Factorial Design—Advertising Experiment
Ad Spokesperson Gender
________________________________________________
Message Male Female Cartoon Character
Benefit Statements M1S1 M1S2 M1S3
Humorous Statements M2S1 M2S2 M2S3
Latin Square
Latin-square designs are multivariable designs that are used to reduce the number of
observations that would be required in a full factorial design. In using Latin-square designs the
researcher is usually assuming that interaction effects are negligible; in so doing, all main effects
can be estimated by this procedure.
As an illustration of a Latin-square design, suppose that the researcher is interested in
three variables (each at four levels) on store sales. For example, the researcher may be interested
in the effect of shelf placement on the sale of energy drinks and be able to manipulate the
following variables for product placement:
A: shelf height—four levels—knee level, waist level, eye level, reach level
B: shelf facings—four levels—25%, 50%, 75%, and 100% of total width of gondola
C: shelf fullness—four levels—25%, 50%, 75%, and 100% of total height of gondola section
If the researcher were to run a full factorial experiment, with one replication only, there would be
(4)3 = 64 observations required. By using a Latin-square design only 16 observations are
required (with estimation of main effects only).
169
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 8-3 Latin-Square Design—Energy Drink Shelf Placement Experiment
Variable B—Shelf Facing
Variable A—Shelf Height B1 B2 B3 B4
A1 C1 C2 C3 C4
A2 C4 C1 C2 C3
A3 C3 C4 C1 C2
A4 C2 C3 C4 C1
Table 8-3 shows one possible Latin-square design for this experiment. Notice in Table 8-3 that
each level of treatment C (shelf fullness) appears once in each row and each column. See
Edwards, 1968, pp. 175-77 for more advanced Latin-Square designs.
Randomized Block Design
Randomized-block designs, also known as treatments by blocks designs, represent a
frequently used experimental framework for dealing with multivariable classifications. These
designs are typically used when the experimenter desires to eliminate a possible source of
uncontrolled variation (a nuisance variable) from the error term in order that the effects due to
treatments will not be masked by a larger-than-necessary error term. For example, suppose that
our researcher were interested only in the effect of ad spokesperson effect on product
performance, but had designed the experiment so that more than one product was used in the
study. The effect of product type could influence preference, and the experimenter might wish to
remove this effect from the error term by “blocking” on product types. That is, each product
would be considered a test unit and each level of spokesperson would be tested in each product.
To illustrate, suppose researchers were interested in examining three levels of
spokesperson for each of four products. Respondents could be randomly assigned to the
respective treatments, or even shown multiple treatments if measurement and learning effects
(distracters of internal validity) were ignored. Results could be summarized in the form shown in
Table 8-4, where we are dealing with a two-variable classification and can separate the block
effect from the error term. Thus, if genuine treatment effects are present, this type of design will
be more likely to detect them than a single-variable classification in which the block effect
would become part of the error term.
Table 8-4 Randomized-Block Design—Spokesperson Experiment
Treatments—Spokesperson
BLOCKS—Products Level 1
Male
Level 2
Female
Level 3
Cartoon
1 X11 X12 X13
2 X21 X22 X23
3 X31 X32 X33
4 X41 X42 X43
Covariance Design
Covariance designs are appropriate in situations where some variable affects response but
is not subject to control during the experiment. For example, if test units consist of human
subjects and the response variable is the number of correct identification of trademarks that are
shown on a projection screen (where such factors as length of exposure and clarity of focus are
varied), it may be that response is affected by the general intelligence level of the viewing
170
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
subject. Suppose that it is too costly to screen subjects, and only those with approximately the
same intelligence quotient are selected. We shall assume that the researcher is able to measure
each subject’s IQ.
In this type of situation, the researcher may use covariance analysis. Roughly speaking,
the computational procedure is similar to a regression problem. The researcher, in effect,
determines the effect on response resulting from differences (in IQ) among test units and
removes this influence so that the effect of the controlled variables can be determined
independently of the effect of test differences on response.
Recapitulation
As noted earlier, the study of experimental design is basically the study of two things:
1. various experimental layouts, such as single factor, factorial, Latin-square, and
randomized block designs
2. analysis of variance and covariance techniques for testing whether the various
treatment effects are significant
Many other kinds of design layouts, including fractional factorial designs, balanced incomplete
blocks, hierarchical designs, split-plot designs, and partially balanced incomplete blocks are
available. We have only described briefly the characteristics of some of the basic statistical
designs.
We have not covered the issue of how to select an experimental design. The design
chosen must be congruent with the research question and capable of implementation with
resources that are available. Such issues as the number of independent variables, the sources and
number of extraneous variables, the nature of the dependent variable, the number of subjects
available for participation in the experiment and other methodological issues will have a bearing
on what might be an appropriate design to use. Increasingly, marketing researchers use personal
computer software to aid in this selection.
The Panel as a Natural Experimental Design
The normal course of operation of a consumer panel generates a continuing set of natural
experimental data. Customer responses to changes in any of the controllable or environmental
variables affecting purchase decisions are recorded in the normal process of conducting the
panel. Audience and deal panels provide similar response measurements.
Time-series, cross-sectional, and combination cross-sectional, time-series designs are all
inherent in panel data. To illustrate their application, suppose that we have increased the price of
our product in selected territories. We can analyze the price-increase effect, at either the
aggregated or individual household level, using the data from those territories in which price was
increased with either the after-only without control group or the before-after without control
group designs [classical designs (1) and (2)]. A cross-sectional analysis may be made by
comparing, for a given period after the increase, the purchase data for the territories in which the
price was raised with those in which no change was made [classical design (4)]. A preferable
approach here would be to use a combination cross-sectional, time-series design and compare the
change in purchases before and after the price increase in the territories in which price was not
changed (control group). Such a study could employ either classical design (3) or (5).
The limitations of each of these designs discussed earlier apply when they are used with
panel data as well. A major difficulty, or course, is in sorting out the effect of the price increase
from the extraneous producers affecting purchases over time and among territories. In this
illustration selective price increases by territory would only have been made in response to
171
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
differing conditions among the sales territories (a price increase by competitors, higher levels of
demand, etc.). History variables must therefore be analyzed carefully in using panel data.
Controlled Experimental Designs Using Panels
The controlled experimental design in conjunction with a panel is most often applied to
market tests of prospective new products, different levels of promotion, new campaign themes,
price changes, and combinations of two or more of these variables. Consider, for example, a
market test of a general price increase. The requirement of random selection of test and control
groups can be met by selecting territories at random in which to raise prices. The remaining
territories automatically constitute the control group. Depending on the kinds of information
desired, an after-only with control group, or a before-after with one control group [classical
design (7)] may be used.
The general advantages and limitations of these designs were discussed earlier. We must,
however, consider the limitations that arise from the use of the panel for measurement,
applicable to both natural and controlled experimental designs.
The Limitations of Continuous Panels
Although panels can provide highly useful marketing information that is difficult to
obtain by alternative research methods, there are some important limitations. The first of these
limitations involves selection and stems from the difficulty of obtaining cooperation from the
families or firms selected in the sample and the resulting effect on the degree of
representativeness of the panel. To be most useful for drawing inferences about the population
being studied, the sample should be drawn by a random process. The sample of families to
comprise a consumer purchase panel may be chosen randomly, but the typical panel has
experienced a high refusal rate during the period of establishment and a high attrition rate, once
in operation. Panels require a great deal of effort to recruit and maintain.
Evidence indicates that the characteristics of both those families who refuse to participate
and those who later drop from the panel are different from those who agree to participate and
remain. To reduce the bias introduced by such nonrandom attrition, replacements are typically
chosen from families with the same demographic and usage characteristics as those lost from
refusals and dropouts.
An additional source of bias is in testing effect arising from continued participation on
the panel. Since the individual is undoubtedly conditioned to some extent by the fact that data on
purchases are reported, panel members may become atypical in their purchase behavior as a
result of being a part of a panel. In short, being a panel member may influence what products
and brands are purchased as the panel member is sensitized.
Panel data may also be systematically biased through instrument effects. The majority of
panels use diaries for reporting. These are self-administered, structured questionnaires. If filled
out properly and submitted on schedule, the information is relatively accurate and inexpensive to
obtain. However, in those cases where there is incomplete data possible biases are introduced
and the total amount of data is decreased.
Despite these limitations, the use of data from panels has become widespread,
particularly in online research. If the panel is administered carefully, the resulting data are
important additions to the information required for making sound marketing decisions.
172
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
DEVELOPING EXPERIMENTAL DESIGNS IN ONLINE SURVEYS
The Qualtrics.com survey software provides advanced survey capabilities and logic,
including experimental designs and data analysis. This advanced technology now provides a
simple and efficient way to design experiments into online surveys. In addition, integrated panel
management capabilities make possible accurate tracking, profiling and monitoring of response
history, response timing and response progress for each respondent. We discuss the capacities of
Qualtrics survey software to solve research problems through management of question flow,
logic, and randomization. These features provide fundamental tools for increasing control of
field experiments. Once we have provided a view of the experimental design methods and
examples for data collection, we proceed to a discussion of the integration of database
information into experimental designs.
Structuring Advanced Questionnaires
Advanced online questionnaires have at their foundation five essential building blocks.
The researcher who understands these building blocks will produce questionnaires that are better
organized and capable of superior functionality. This functionality can include the development
of and presentation of experiments. These building blocks require a clearly identified and stated
set of research objectives for the questionnaire and the experiment:
1. Identify the populations to be sampled, along with experimental and control groups
2. Map the logic and flow of the experiment, including randomization
3. Identify the organization of blocks of questions if needed for survey flow, branching or
experimentation
4. Specify the questions (and question types) within each block
5. Define sample size quotas that are required for appropriate analyses
Note that we have not yet considered the important issues of measurement, including construct
validity, the wording of questions, or controlling survey content. These topics are discussed in
the next chapter.
Survey Realism, Interest and Respondent Involvement
Experimental design is all about presenting specific treatments to the respondent and measuring
the effects. With online experimental designs the boundaries of what is possible for treatments
has been greatly expanded through multi-media content.
Enhancing Survey Content: Graphics, Audio, Video and Interactive Experiences
Graphics increase interest and provide realism to a survey. The advantage of graphic
visualizations over text descriptions cannot be questioned. A corporate logo at the top of the
survey may provide identity and authenticity to the questionnaire. Similarly, graphics can
provide an experiential view of product and brand treatments for concept tests.
Prior to online surveys, the use of audio creatives (songs, jingles, messages with varying
content, etc.) in research was limited to telephone and personal interviews, or focus group
settings. Online technology facilitates the use of songs, commercials, legal arguments, and other
verbal, melody, or tone-based treatments in online concept tests and experiments.
Video files (MOV, MP4 and FLASH) provide rich, realistic portrayals of products and can be
added easily to online experiments for testing (Figure 8-4). The increased adoption of high-speed
Internet continues to expand the type and amount of information that can be tested.
173
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 8-4 A Nokia Flash Commercial in an Evaluation Question
While interactive technology is limited in use within mainstream market research, it is
used on many entertainment Web sites and is applicable to highly technical and youth market
segments who are accustomed to this high-involvement graphics approach. Qualtrics accepts
browser interfacing programs that may be used to provide interactive Web Services content
(programmed using Flash, Java Script, PHP or Visual Basic). These tools give researchers the
option of creating and adding to a survey, custom programs that control their own experimental
treatments and measurements.
As examples, these programs could control treatments for sophisticated experiments
using Latin-square or d-optimal Federov designs that present options in choice based conjoint
analysis. Control could be developed for unique ways of displaying, timing and rotating text or
graphics. Similarly, a researcher could for each survey respondent randomly select, say, 5 open
ended text responses from other respondents who have already completed the survey, and then
ask the current respondent to rank the 5 extracted text statements in order of preference.
Qualtrics provides through these API and Web Services tools, open access to multiple databases
in real-time and an interface for programming tools. The researcher is thereby unconstrained,
except in their own creativity.
Creating Dynamic Questions through “Piping”
Piping is one method of customizing surveys. Piping generally refers to the movement of
information from one location to another within the survey. Information may include question
text, answer text, graphics, or database information about the respondent. It may also include the
text of answer options selected, or those not selected. The point of application is that the
respondent’s questionnaire flow may be customized to reflect their individual choices or options
they did not choose, so that additional topic specific in-depth information can be collected.
Travelocity customers receive a survey a few days after they return from a trip booked online.
The invitation letter and questionnaire are not of the “Dear Traveler” variety. Instead, the
customer is addressed by name and is asked specific questions about their flight to Dallas on
Delta Airlines. If they booked a rental car, the block of questions about satisfaction with the
service of their rental car company (identified by name) is included in the survey flow. Question
blocks are also customized for hotels and cruises booked for the trip. In short, the online
customer database is interfaced with the survey database through a simple API (application
program interface) and the questionnaire is totally customized to ask about specific aspects of
their personal travel experience.
174
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Panel information linked from external databases can be used in a variety of ways. For example,
if the respondent is evaluating a given brand, factual data about a product category, information
about the category leaders, or even information about the respondent’s past purchases could be
piped from a database to allow the respondent to make more accurate relative comparisons given
their purchase history.
As a simple example of piping, consider pieces of information that are piped based on the
respondent’s answers to previous questions. The respondent first chooses the brand Hummer as
the GM car they had previously purchased. The text “Hummer,” piped to an open ended text
question about dissatisfaction the car. Here, the respondent typed, “The sticker shock at the gas
pump”, a phrase that is provide to provide context and detail in the final question that asks the
customer to make recommendations to General Motors.
Other uses of piping might include concept or logo tests where respondents select a
company logo or other concept-related graphic. Similar to the above example, the selected
graphic would follow the respondent as needed in the survey, either as part of the question text or
as a possible answer choice. At the advanced end of the spectrum, personal data for a given
respondent may be read from a CRM (customer relationship management) database, such as in
the Travelocity example cited above, and then piped into questions or answers within the
questionnaire.
Figure 8-2 Text Piping Example
The Pipe Smart Text option is accessed by clicking on the text of the question and
selecting in the Rich Text Editor (Figure 8-3).
Figure 8-3 Piping Smart Text – Rich Text Editor Screen
175
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Creating Dynamic Surveys with Skip Logic, Branching and Looping and Control
Logic to control question order based on respondent answers can be easily added to online
surveys. Survey flow logic exists in several levels of complexity, each of which will be
discussed in this section. We will begin with an introduction to simple skip logic and progress to
more advanced logic used in experimental designs.
Linear Skip Logic
Linear skip patterns move the respondent from one question to another, based on the
respondent’s answer to a single question. For example, if a respondent indicates that she is a nonconsumer
of energy drinks, she is skipped past energy drink product and brand specific
questions, again entering the linear flow of survey questions that ask about general consumption
of casual beverages. In this example the survey flow moves linearly down through the
questionnaire based on answers to an individual question. Multiple instances of skip logic could
be entered where appropriate.
Skip logic is added in Qualtrics after the question has been entered. Select “Add Skip Logic”
from the purple dropdown button to the left of the question. The logic bar will appear wherein
you can specify the answer option to skip on, the logical condition (selected, not selected,
displayed, not displayed), and the skip to destination (a specific question, end of block or end of
survey). Multiple skip logic statements are supported for each question (Figure 8-4).
Figure 8-4 Linear Skip Logic
In the above example, when the answer “$20,000” is selected, the branch is taken and the
text “$20,000” is piped from Q14 to Q33. To Pipe Smart Text from Q14 to Q33, just go to Q33
and edit the question. Select in the Rich Text Editor at the point in the question or answer
where you would like the piped text to appear. The specific piped text is specified using the
menu options. Remember that linear skip logic uses positive logic conditions setting the criteria
for the people who enter the branch, not for those who don‘t—as a default people will NOT enter
the branch.
Survey- and System-Wide Boolean Logic for Compound Branching
Branching is more sophisticated than linear skip logic and is used to skip questions based on a
complex sequence of conditions. Suppose that a respondent meets the multiple conditions
176
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
required to be defined as a member of our target demographic segment. She is “female,”
“between 18 and 34,” “married,” and has “one or more children.” Branching then is based on
multiple questions or points of information provided by the respondent, or found in the panel
data. With branching, the intersection of these identifying characteristics is the point where we
change survey flow and question logic.
The concept of system-wide Boolean logic expands the locus of control for branching. This
approach enables a survey to branch based on (a) answers from multiple questions within the
survey; (b) outside variables nonspecific to the respondent, such as a randomly assigned
experimental group defined in either the panel information or by Web Services content; (c) panel
and database information about the respondent (such as CRM systems or past surveys the
respondent has completed). Figure 8-5 contrasts these approaches.
Figure 8-5 Linear Skip Logic versus Boolean Logic and Compound
Looping with Piping
Looping with piping is a combination of features designed to reduce the length and complexity
of a survey building process when a standard set of evaluation questions are repeated over and
over. The following example shows how the use of looping and piping shortens the questionnaire
and logic building process.
As an example, suppose that a respondent is asked to select from a list of 150 educational
programs, those that were used by his or her organization. Next, the person responds to three
evaluation questions about each program selected. Rather than building a linear survey of 600
177
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
questions (do you use program x? – yes, no; if yes, then evaluate, if no, ask about next program),
the researcher would build only 4 questions (check all that apply in the list of 150 programs and
the three evaluation questions). The question “Select from the 150 programs those you have
used” would appear in the main program block and then a second block would contain the 3
evaluation questions. The pipe and loop feature would loop through only those programs
selected, identifying them by name as the three evaluation questions were repeated.
.
Note that looping based on a “Question Response” can be based on all choices presented to the
respondent in a previous question, all choices selected by the respondent, or those choices not
selected (for asking questions of the type: “why didn’t you consider this option in your
decision”), or those choices that appear or do not appear, where appearance occurs as a function
of some previously defined condition. In other words, the choices of the respondent can dictate
the flow of this type of question.
In contrast to control by “Question Response”, the experimenter can use “Static” control with a
predefined set of loops and piping. For example, a researcher has predefined a set of product
brands for evaluation. In this case, the predefined product category and brand names “fast food,
McDonalds, Arbys, Subway” could be piped and merged into the sequence of three questions
that is automatically repeated for each of the three fast food brands evaluated. In experimental
settings, this approach could be used to present treatments, or evaluate a defined set of brands,
usage occasions, employee, or similar list of interest. This predefined list is defined in the Block
Options / Loop and Merge dialogue box as shown in figure 8-6. Once the text is defined, the
Smart Text Piping option is used to specify placement of the merge text within a question.
Figure 8-6 Block Loop and Merge Static Text (Researcher Provided)
With increasing time demands on potential respondents, they often review questionnaires to
estimate the time and effort required. Streamlining questionnaires using looping and piping has
178
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
the potential of reducing the physical length of the survey and associated respondent fatigue,
thereby increasing data quality and completion rates.
Carry Answers Forward
Closely related to the streamlining effects of loop and pipe feature is the ability to carry forward
any or all respondent answer choices from one question to a later question. This is particularly
useful for experimental design where additional follow-on information is desired.
All possible answers or only the k of n items that were selected can be moved to a later
question to ask further questions about those items. For example, sports actively participated in
or the names of brands previously purchased could be moved to a next question or block of
questions for further drill down questioning. The order of item presentation can be maintained or
randomized. The ability to carry answers forward is also useful where the respondent selects an
“other (please explain)” category and provides the associated text response. This text response
can then be carried forward to gain additional information.
Quota Fulfillment
Quota fulfillment is an integral part of experimental designs. Quotas work by testing the size of
the sample groups (defined by answer categories within a given question) against target numbers
(Figure 8-7). Where experimental designs are used, quotas may be created to avoid over
sampling of some groups while under sampling other groups. Given that many studies pay
respondents for completed surveys, quotas, if used early in the questionnaire can screen and
eliminate unneeded respondents, thereby reducing total research costs. The quota specification
menu is found in the Advanced Options dropdown link in the main Edit Survey page.
Figure 8-7 Sample Size Quota Setting
This section has provided a brief overview of the embedded logic and complex information
flows that can be adapted to meet the rigors of an experimental design. Unlike paper- or
interview-based questionnaires, these “smart surveys” perform on-the-fly analysis and
incorporate recorded answers, embedded or panel data into questions. In addition, experiments
and statistically sound quota sampling are facilitated in dynamic sequences of questions that are
displayed based on a respondent’s answers and/or match to a group or quota. The next section
expands on this logical base to discuss the use of question blocks, flow, and randomization in
experimental designs.
179
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Using Survey Flow to Create Experiments
Qualtrics contains powerful, but easy to use tools for customizing question and block
presentation order in experimental designs. The complex branching previously discussed can be
used to direct the respondent through blocks of questions, advanced question types (conjoint
analysis), and even complex randomization in the survey flow of an experiment.
Question Blocks
A question block is a cohesive group of questions that receive similar treatment. Simple
questionnaires using only linear skip patterns or no skips at all will often use only a single
question block (a single default question block is created automatically in Qualtrics). More
advanced questionnaires may split questions into multiple question blocks based on topic or
logical flow.
A question block may be controlled independently of the rest of the survey and for
example, be given special instructions that apply only to questions within the block. Questions
within blocks may be arranged in a fixed order, random order, or partial random order that holds
some questions in a fixed position, while randomizing the order of other questions. The
presentation order of multiple blocks, should they be present in the survey flow, may also be
manipulated:
a) A block of questions can be displayed based on choice previously made by a respondent
(evaluate a block of brand-specific questions if the respondents indicated they were
familiar with the brand),
b) The blocks can be randomly presented or held in a fixed position while randomizing the
order of other blocks (each block is a treatment or control in an experiment)
c) Blocks can be repeated in a looping pattern with piped text that describes the purchase
situation or usage occasion under which the block of questions is to be evaluated (as in
the loop and pipe example presented above).
Any element in the survey flow can be dragged and dropped into any location in the flow. In this
way, nesting of treatments within the logical flow of an experiment is easily accomplished.
Randomizing Questions and Blocks of Questions
Randomization is used when it is not important to maintain the question order or context
of the questions. Blocks may be used to assist in maintaining context and at the same time
achieving a level of randomization. For example, context often requires that questions be
presented in a given sequential order, but blocks of questions (each block, for example,
containing questions about different brands) can be randomized. In many concept testing and
other experimental situations, randomization of questions within a block or randomization of
blocks of questions is a valuable option.
180
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 8-8 The Survey Flow Chart
Once the survey building has been completed and question blocks are created, the initial
survey flow is created. In Qualtrics, question block and element presentation can be visualized
as a color coded sequence organized the way the survey "flows" (Figure 8-8). The survey
follows the main flow line on the left, and presents questions in the order shown from top to
bottom. The secondary flows branch to the right, where each color coded "element" shows the
branching logic statement (if used) and the questions shown to the respondent. Survey flow
Elements may be added at any point in the flow.
The researcher adds survey flow “Elements” into the flow control sequence. The two
survey flow elements most often used are “Block" (question blocks), and "Branch." Other
elements include “Embedded Data,” “Randomizer,” “Web Service,” “End of Survey,” and the
“Conjoint” analysis module (Figure 8-9).
Figure 8-9 Survey Flow Elements
Adding Block and Branch Elements to the Survey Flow
Block
Blocks of questions, once created, are automatically included in the questionnaire flow.
Blocks can be added at any location in a Branch or in the Main Flow. Once a block is inserted, it
will appear as a survey flow element on its own secondary line. As shown in Figure 8-7,
respondents will first see the Default Question Block, and then conditions branching to other
blocks of questions (perhaps identifying treatments) that were developed by the researcher.
181
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Branch Logic
Branches function differently than Blocks. A Branch acts as a filter which, based on answers
to previous questions or other information, directs certain people to an area of a survey if the
specified condition is met. Those not meeting the condition proceed to the next condition in the
survey flow.
All branch logic clauses operate on responses received to a previously answered
Question, Embedded Data (generally panel data), or Quotas. The logic section can also be
triggered by choices that were Selected, Not Selected, Displayed or Not Displayed. Selected and
Non-Selected are self explanatory and can be used to identify respondents who indicate these
are, for example, purchasers or non-purchasers of a brand. Displayed and Not Displayed options
are useful when text is piped or answers are carried forward from one question to another.
Displayed is synonymous with the respondent receiving the treatment or not… and is useful in
construction of if so/not, then show/don’t show a treatment of further questions. Options are
specific to the question type. For example, a text input question includes the options: Empty,
Not Empty, Displayed, Not Displayed, etc.
Figure 8-10 Multiple Then Branch If Statements
Note: Because experimental designs often require extensive logic sequences, it is worth noting
that within the chain of logic statements, quotas have first priority, followed by the "And If"
clause, which has a higher priority than the "OR IF" clause (i.e. with a chain of clauses such as 1
AND 2 AND 3 OR 4, the system will read it as (1 AND 2 AND 3) (OR 4), meaning IF 1 AND 2
AND 3 are all fulfilled, OR if 4 is fulfilled, then the respondent will be branched.
Enhancing Surveys by Adding Embedded Data into the Survey Flow
Embedded Data is "behind the scenes" respondent data. Generally, embedded data will be
uploaded as part of panel data. Once this data is uploaded, operations such as skipping and
branching can be performed using the embedded data as a key selection variable.
For example, if we were to access the person‘s home state as an embedded data field in a
panel mailing list, we would specify the location, variable, logic and condition as:
"If: Embedded Data : Home State Equals Colorado."
Embedded data can also be static codes marking events or conditions that are referenced
by logic conditions at other points in the survey.
182
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Randomization of Answer Choices, Branches, Question Blocks and Surveys
Randomization techniques are used to control answer presentation order bias, as well as the
effects of extraneous variables not controlled by an experimental design. In new product concept
tests and other advanced measurement situations, the researcher may randomize and control the
presentation order of the following in online research and experimental designs:
• Complete randomization of all choice options
• Complete randomization of choice options and non-randomization of “other” or text
input options
• Partial randomization of choice options, where one or more specific answers (such as
the item of interest) are held in a fixed position, and all other answers are randomly
ordered
• Randomization of blocks containing treatments
• Partial randomization of blocks, where one or more of the blocks are held in fixed
presentation order, or are presented to all respondents, while other blocks are
randomly or intermittently presented (randomly present 2 of the 4 other treatment
blocks)
• Separate questionnaires containing separate treatments are presented randomly.
The randomization of choice options is the most basic type of randomization, and deserves little
special attention other than to suggest that answer choice sets without a fixed order and matrix
type questions (displaying lists of questions, each evaluated on the same scale) are best presented
in random order. Randomization at the question level is found the individual question options.
Figure 8-9 Block Randomizer Screen
Randomization of treatment blocks works by inserting the element "Block Randomizer" into the
survey flow. Do this by clicking on the "Add New Element" link, choosing "Block Randomizer,"
and then choosing the Blocks you wish to place it in. The randomized blocks will appear in the
survey in the place where you insert the block randomizer.
For experimental research treatments can be randomized with all, or random subsets of the
treatments presented. To select a subset of blocks, make sure all blocks to be randomized have
183
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
been added into the Block Randomizer. The number in the dropdown reflects the number of
blocks you want the respondents to see. You may set it to show all of the blocks in the
randomizer, or a subset of those blocks (i.e., If 4 blocks are added to the randomizer and each
respondent is to randomly respond to two of those blocks, put a "2" in the box.
Randomization using Multiple Questionnaires
Occasionally, experimental treatments are so extensive that a completely different questionnaire
is required for each treatment. In such cases, the questionnaires are prepared and randomly
assigned to the respondents. Such a design may be achieved by either using a “portal survey”
where the links to the other surveys are randomly presented via questions or block
randomization, or by inserting a separate java script into a question. The java script would be
similar to the one that follows to randomly assign respondents to a survey.
Sample Experimental Designs and Associated Survey Flows
Script to Open Survey in the Same Window, or in a Separate Window:
<SCRIPT LANGUAGE="JAVASCRIPT" TYPE="TEXT/JAVASCRIPT">
Event.observe(window, 'load', function() {
<!-- Javascript to redirect user to a randomly selected survey //-->
// Array structure - as many links as you need
var seedcount = 7
var index = 3;
var linkArray = new Array(index);
// Load array with available survey links here:
linkArray[0] = "http:// new.qualtrics.com/SE?SID=SV_6LR6QjjYiHkSe21&SVID=Prod";
linkArray[1] = "http:// new.qualtrics.com/SE?SID=SV_6LR6QjjYiHkSe22&SVID=Prod";
linkArray[2] = "http:// new.qualtrics.com/SE?SID=SV_6LR6QjjYiHkSe23&SVID=Prod";
// Generate random number that will correspond to an index in the above array.
// The integer in the equation is the total number of links you have minus 1
now = new Date();
var seed = now.getSeconds();
randNum = Math.round(1000000*Math.random(seed))
for (loopcount = 0; loopcount < seedcount; loopcount++) {
randNum += Math.round(1000000*Math.random(randNum * seed));
}
randNum = Math.round(1000000*Math.random(randNum)) % index;
// Use one of the two statements below by changing the comment
// Redirect user to appropriate random survey page in the current window
window.location.replace(linkArray[randNum]);
// Or open the appropriate random survey page in a new window
// window.open(linkArray[randNum]);
});
</script>
184
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Field Experimentation in Marketing
Unfortunately, experimental methods are not always the methods of choice in the study
of marketing phenomena because depending on the mode of field experimentation, it may be
quite expensive, subject to large amounts of uncontrolled variation, or may produce results that
are difficult to generalize to other products, market areas, or time periods
There is little question that field experimentation in marketing can be costly undertaking.
Consider the case of a sales manager who wishes to determine the effects of varying amounts of
sales effort on product sales of a nationally distributed brand. Sales in a given time period could
be affected by point-of-purchase advertising, personal sales effort, broadcast promotion,
competitors’ selling efforts, relative prices, seasonal effects, past promotional expenditures, and
so on. However the capabilities of online surveys software like Qualtrics.com has provided one
possible approach to reduce costs while maintaining the precision associated with experimental
research.
Experimentation represents another way of information gathering and should be
approached in terms of its cost versus value (compared with other techniques) for decisionmaking
purposes, with regard to both present and anticipated problems. Unfortunately, it is
much easier to state this objective than to design suitable techniques for implementing it.
SUMMARY
In this chapter our primary objectives were (1) to introduce the conceptual bases
underlying marketing experimentation, and (2) to discuss basic models of experimental design.
The statistical techniques for analyzing experimental data will be discussed in later chapters.
The first section of the chapter covered the nature of experimentation. The critical
aspects of this discussion were the various ingredients of an experiment. We next turned to the
potential sources of invalidity, internal and external, associated with an experiment. These are
the experimental errors that relate to extraneous factors that can affect an experimental outcome.
We next described some of the major classical and statistical experimental designs.
Some of these designs—single-factor, factorial, randomized blocks, Latin squares—were
illustrated.
The Qualtrics research suite was then presented as a powerful and integrative tool for
researchers who need to conduct advanced field studies, including experimentation. We have
introduced an array of advanced online survey functionalities and tools that allow the researcher
to create more visually interesting, involving, and realistic approaches to conducting research.
We have also discussed advanced logic and flow capabilities that support the creation of
experiments.
We concluded the chapter with a discussion of some of the problems involved in
marketing experimentation and the relationship of this means of data collection to the cost and
value of information.
185
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
Banks, S. (1965). Experimentation in Marketing. New York: McGraw-Hill
Bay, W. (1998). NewsStand: CNN & Entertainment Weekly Report, September 28.
Brown, S.R. and Melamed, L.E. (1990). Experimental Design and Analysis. Thousand Oaks,
CA: Sage Publications,Inc.
Campbell, D.T. and Stanley, J.C. (1966). Experimental and Quasi-Experimental Designs for
Research. Chicago: Rand McNally
Christensen, L.B. (2004). Experimental Methodology. Boston: Allyn & Bacon, Ninth Edition
Cook, T.D. and Campbell, D.T. (1990). Quasi-Experimentation: Design and Analysis Issues for
Field Settings. Boston: Houghton Mifflin Co.
Edwards, A.E. (1968). Experimental Design in Psychological Research. New York: Holt,
Rinehart & Winston, Third Edition
Field, A. and Hole, G. (2003). How to Design and Report Experiments. Thousand Oaks, CA:
Sage Publications
Perdue, B.C. and Summers, J.O. (1986). “Checking the Success of Manipulations in Marketing
Experiments,” Journal of Marketing Research, 23 (November), 317-26
Rosenthal, R. and Rosnow, R.L. (1975). The Volunteer Subject. New York: John Wiley & Sons
Rosenthal, R. and Rosnow, R.L. (1991). Essentials of Behavioral Research: Methods and Data
Analysis. New York; McGraw-Hill, Second Edition
Sawyer, A. G. (1975). “Demand Artifacts in Laboratory Experiments in Consumer Research,”
Journal of Consumer Research, 1 (March)
Shadish, W., Cook, T.D., and Campbell, D.T. (2002). Experimental and Quasi-Experimental
Designs for Causal Inference. Boston: Houghton Mifflin Co.
186
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Appendix 8.1
Sample Experimental Designs
Case 1: Trademark Confusion Testing 1
“Schema theory suggests that when consumers are offered a junior mark in a product category
that is different from the senior mark’s category (e.g., a CADILLAC brand notebook computer), the
likelihood that consumers will categorize the product as a subtype of the senior mark (i.e., the likelihood
that consumers will suffer confusion) is related to the product’s incongruity with the senior mark’s
schema.
With the goal of creating stimuli to test the hypotheses, a well known automotive brand
(CADILLAC) was chosen as the senior mark. Notebook computers were chosen as a product category for
the junior mark because notebook computers were anticipated to be relevant to online participants. A
fictitious junior mark (CADILLAC brand notebook computers) as then created, along with an
advertisement that described an upscale, luxurious notebook computer that could be offered by a firm
attempting to infringe on the CADILLAC trademark. The following advertising text was introduced with
the claim that it would “appear in the next edition of a popular computer-sales catalog.”
Cadillac brand notebook computer! This luxury notebook has an upgraded keyboard for better
feel and absolutely silent typing. A dark mahogany inlay surrounds the screen. Matching dark
mahogany accents are found on the wireless mouse and the leather carrying case. Intel Core 2
Quad Q6700, 4GB RAM, 200GB Hard Drive. Price available upon request.
H1 suggests that viewers of this advertisement will be more likely to suffer confusion by
believing CADILLAC MOTORS to be the source of the notebook computer if another maker of luxury
automobiles has already extended its brand into the luxury notebook computer product category.
To test this hypothesis, three stimuli were created. The first stimulus (see Figure 1) was an
experimental control designed to represent a typical advertisement for a notebook computer at the time of
the study. The advertisement emphasized computing power and did not attempt to create an upscale
positioning with luxurious product attributes.
The second stimulus informed participants of the launch of an upscale notebook computer (see Figure 2).
Luxurious product attributes such as a titanium case were described in support of an upscale positioning.
This second stimulus was another experimental control because the upscale notebook computer was
described as being offered by a well-known computer company (Dell) rather than by an automotive
company.
The third stimulus (see Figure 3) informed participants of the launch of an upscale notebook
computer as a brand extension of a well-known automotive brand (MERCEDES-BENZ). To encourage
participants to subtype the notebook computer in the MERCEDES-BENZ schema, the ad included a
photo of a MERCEDES-BENZ car and a close-up of a MERCEDES-BENZ hood ornament. By
comparing viewers of this stimulus with viewers of the two control stimuli, H1 can be tested.
Participants were randomly assigned to view either the typical DELL computer advertisement
(condition 1, n = 167), the luxury DELL computer advertisement (condition 2, n = 165), or the luxury
MERCEDES-BENZ computer advertisement (condition 3, n = 163). The experimental method was used
to provide a rigorous test of the causality claimed in H1. Depending on the assigned experimental
condition, one of the three stimuli was shown as part of an online questionnaire instrument. The stimulus
was shown in full-color, and the time of exposure was self-paced. Participants then completed a filler task
to mask the purposes of the study. They were then shown the advertisement for the CADILLAC notebook
computer and asked to complete the source-confusion battery. The study’s independent variables were
then measured.”
1
Source: Lee, T. R., Derosia, E. D. and Christensen, G. L., Sophistication, Bridging The Gap, And The
Likelihood Of Confusion: An Empirical And Theoretical Analysis, The Trademark Reporter, Vol. 98, No.
4, July-August 2008. Used by permission of Professor Glen Christensen, Brigham Young University.
187
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 1 (Condition 1: Control) Figure 2 (Condition 2 (Control) Figure 3 Condition 3 (Treatment)
The following is the flow chart shows the above three group experimental design with control
188
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Case 2: Determining Price Sensitivity
A major investment firm is interested in determining the price sensitivity of fund managers to
specially prepared research studies about selected new equity offerings. The research design is to include
a conjoint analysis. This analysis is followed by an experimental design (unrelated to the conjoint
analysis) that measures the fund manager’s sensitivity to the report price given their current payment
expectations for a single research study that is to be prepared exclusively for their firm. In the design,
report price must vary at fixed levels to reflect volume discounts (more reports per year) and report
exclusivity (proprietary verses syndicated).
The Experimental design in this example would split respondents based on the price they
currently expect to pay for a single exclusive research report ($20k, $30k, $40k, $50k). From this base
point, price acceptability is measured using a based on two other factors: (1) The number of studies
purchased per year (4, 12, 24), and (2) The exclusivity of the research reports ( The report will be
distributed to 1, 4, 10, 25 or 100 firms). The prices tested will vary as shown in Table 1.
Table 1: Design and Price Matrix
Expected Price 20K 30K 40K 50K
Studies to
Purchase
4 Studies
(1 per quarter)
4 Studies
(1 per quarter)
4 Studies
(1 per quarter)
4 Studies
(1 per quarter)
Exclusivity 1 @$204k
4@$163.2K
10@$81.6K
25@$40.8K
100@$20.4K
1 @$108k
4@$86.4K
10@$43.2K
25@$21.6K
100@$10.8K
1 @$144k
4@$115.2K
10@$57.6K
25@$28.8K
100@$14.4K
1 @$180k
4@$144K
10@$72K
25@$36K
100@$18K
Studies to
Purchase
12 Studies
(about 1/month)
12 Studies
(about 1/month)
12 Studies
(about 1/month)
12 Studies
(about 1/month)
Exclusivity 1 @$360k
4@$288K
10@$144K
25@$72K
100@$36K
1 @$306k
4@$244.8K
10@$122.4K
25@$61.2K
100@$30.6K
1 @$408k
4@$326.4K
10@$163.2K
25@$81.6K
100@$40.8K
1 @$510k
4@$408K
10@$204K
25@$102K
100@$51K
Studies to
Purchase
24 Studies
(about 2/month)
24 Studies
(about 2/month)
24 Studies
(about 2/month)
24 Studies
(about 2/month)
Exclusivity 1 @$72k
4@$57.6K
10@$28.8K
25@$14.4K
100@$7.2K
1 @$540k
4@$432K
10@$216K
25@$108K
100@$54
1 @$720k
4@$576K
10@$288K
25@$144K
100@$72K
1 @$900k
4@$720K
10@$360K
25@$180K
100@$90K
The survey flow next presents a default question block of classification and qualification
questions, and is followed by the conjoint analysis module. The flow next branches based on the
respondent’s stated acceptable price, to a matrix evaluating exclusivity x volume discount. This matrix
uses piped smart text to reference the appropriate acceptable price level specified by the respondent
(Figure 1). Finally, Branch elements direct the respondent to the appropriate pricing sensitivity section.
189
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 1 Piped Text to Branched Matrix Question
Figure 2 The Survey Flow
190
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 3 Number of Studies x Exclusivity for Branch (Cell 1-1 of the Design – Price Matrix)
191
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 9
MEASURING RESPONDENT INFORMATION:
ATTITUDES, SATISFACTION, LOYALTY AND BEHAVIOR
INFORMATION FROM RESPONDENTS
All marketing decisions involve recognizing alternatives to, and making
predictions of, the behavior of market participants. Marketing decisions ultimately hinge,
in whole or in part, on a prediction of the behavior of consumers, industrial users,
marketing intermediaries, competitors, and, at times, the government. Whether the
decision is to introduce a particular new product, raise the price of an existing product,
change distribution channels, or determine an advertising budget, the solution involves
forecasting the behavior of one or more groups of market participants.
We now consider the most frequently studied variables used to predict behavior:
demographics, attitudes, intentions, satisfaction and loyalty. The information used to
predict behavior is grouped into behavioral correlates (past or intended behavior) and
nonbehavioral correlates (demographics, attitudes, lifestyles, values).
Behavioral Correlates
Behavioral correlates address the question of what past behaviors or intended
behaviors predict future behavior.
Past Behavior
Past behavior is widely used as a predictor of future behavior. Each of us relies
heavily upon our knowledge of others’ past behavior in our everyday relationships with
our family, friends, and associates. When we state that we “know” someone well, we are
implicitly saying that we believe we are able to predict that person’s behavior in a wide
range of social and interpersonal situations. In economics applications, we examine past
trends, seasonal averages, and cyclical patterns for forecasting.
Regardless of the nature of the variable or variables to be forecasted, a basic
premise of using past behavior to predict future behavior is that there is a relationship
between the two that, to some extent, is stable. Recognizing that the degree of stability is
sometimes difficult to determine, and that we do not always understand the underlying
causal relationships, we nonetheless must believe that there is some continuity and
stability in the behavior patterns of people.
A typical consumer brand purchase study would concern itself in part with
determining such facts as what brands have been used, the last brand bought, where and
with what frequency purchases are made, what the exposure to company advertising has
been, and similar aspects of past behavior.
Information about past behavior toward products is often classified into three
categories: acquisition, use, and possession. Within each of these behavioral areas, we
collect detailed information about who, what, when, where, how much, and in what
situation the behavior occurs. This data produces detail useful in understanding product
consumption patterns. The particular study’s requirements will dictate which of these
types of information will be most useful. Table 9.1 shows the requirements for a study on
192
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
tomato juice to determine, among other things, whether a new type of container should be
developed. Often, such information comes from secondary sources, as previously
discussed.
Intended Behavior
Intentions may be defined as presently planned actions to be taken in a specified
future period of time. What more logical method of predicting the future behavior of
respondents could be used than determining their intentions? After all, intentions are selfpredictions
of behavior, and thus, if obtained from people whose behavior we want to
predict, would seemingly be the most direct and reliable method of prediction.
Intentions are relevant and commonly sought as predictors of behavior. However,
consideration of our own experiences in terms of what we have planned to do vis-à-vis
what we have actually done later should raise some questions concerning the reliability of
intentions as a predictive tool. The question “What will you do?” must always be
answered conditionally.
TABLE 9.1 Past Behaviors: A Study of Tomato Juice Usage Patterns
Acquisition Use Possession
Who Who in your family
usually does the
shopping
Who in your family
drinks tomato juice
What What brand of
tomato juice did you
buy last time?
What dishes do you
cook or prepare with
tomato juice?
What brands of tomato juice
do you now have on hand?
When How long has it been
since you last bought
tomato juice?
Where Do you usually do
your shopping at a
particular store or
supermarket?
(if Yes, where?)
sssHow Much What size can of
tomato juice do you
usually buy?
About how often do
you buy tomato juice?
About how many cans
do you buy at a time?
About how much
tomato juice does
your family drink in a
week?
For which purpose
does your family use
more juice?
Do you now have any
unopened cans of tomato
juice on hand?
(If Yes, about how many
cans?)
Usage Situation How does your family
use tomato juice?
Beverage □
Cooking □
Both □
Beverage □
with friends
How do you store tomato
juice after it is opened?
Can □
Bottle □
Plastic □
Other □
The degree of assurance that planned actions will be translated into actual actions
varies widely depending on circumstances and future happenings, many of which are
outside the respondent’s control.
193
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The results of a study of expected and purchase rates of a few products and
services are shown in Table 9.2. Researchers collected intentions data from a consumer
panel sample using a 0 to 10 scale to measure purchase probabilities. Verbal definitions
were assigned to each point on the scale. A 10 was defined as “absolutely certain of
buying” and a 0 as “absolutely no chance of buying.” The definition of a 5 was given as
“five chances out of ten of buying,” and the other points between 1 and 9 inclusively
were similarly defined.
TABLE 9.2 Intentions Fulfilled and Not Fulfilled During a 60-Day Period
Product/Service
Intentions-Based
Expected Purchase
Rate % Purchase % Difference
Ride local public transportation
Purchase tax-sheltered investment
Purchase stereo system
Take a trip on cruise ship
Purchase new automobile
22.5
11.4
17.6
4.2
14.3
21.7
7.2
15.6
3.7
14.1
-0.8
-4.2
-2.0
-0.5
-0.2
Expected purchase rates were calculated as the average purchase probability for
each item. The actual rate was determined by reinterviewing the panel members 60 days
later to find out what purchases they had actually made.
Intentions to buy are often conditioned by judgments and expectations of future
events or future situations, as well as by our past experiences. Such variables as expected
change in financial status, price expectations, general business forecasts, and predictions
of need all contribute to the final intention decision. Since each of these is subject to
variation, it seems plausible to suppose that the intender views them as such and that his
or her stated intention is based on a subjective probability of purchase. This supposition
is supported by the fact that intentions data with assigned probabilities have generally
proven to be more accurate than those expressed in “either/or” form.
Past experiences can often be inferred from a customer’s level of satisfaction.
Increasingly, customer satisfaction surveys fill a critical role in many firms’ customer
relationship management (CRM) systems (James, 2002). Customer satisfaction relates to
intentions as it can significantly affect the repurchase decision of consumers.
Nonbehavioral Correlates
So far we have discussed how people’s past behaviors and their intentions are
correlates of what they will do. We now need to examine the nonbehavioral correlates
that are useful for predicting their future behavior.
Socioeconomic Characteristics
How is information on the social and economic characteristics of respondents
useful for forecasting what people will do? The answer can be readily suggested by an
illustration. The Radio Corporation of America (RCA), when introducing color television
in the 1950s, was very much interested in the age, income, educational, and occupational
composition of the market. They judged that the initial market willing to pay for a
premium priced color television set would be families proportionally higher in income
and educational levels and older, than either the black-and-white set owners or the
population as a whole. These judgments were subsequently confirmed by a study of early
purchasers of color sets. This information was useful for both pricing and promotional
194
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 9.1 Measuring Intentions
195
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
decisions, since an association was found to exist between families with these
characteristics and the purchase of color televisions sets.
In studies of consumers where there is a basis for believing that such associations
might exist, researchers obtain information on one or more socioeconomic
characteristics; those most frequently obtained are income, occupation, level of
education, age, sex, marital status, and size of family. While socioeconomic
characteristics are by far the most widely used bases for classification of consumers,
other bases exist. Among these are attitudes, preferences, personality traits, perceived
risk, and such measures of actual buying behavior as amount purchased and brand
loyalty. It may be interesting to know, for example, that owners of SUVs show different
personality traits than owners of other vehicles; such knowledge will be useful in
marketing automobiles, however, only if it can be used to develop and evaluate appeals
for each type of buyer. Doing so can enhance segmentation, positioning, and market
targeting.
In general, the identification of consumer segments is useful in marketing so long
as the following four statements apply:
1. Substantial: The value in terms of potentially increased sales makes it
worthwhile to do so.
2. Differentiable: There are practical means of differentiating purchase behavior
among market segments. There is homogeneity within and heterogeneity
between segments.
3. Operational: There is a cost effective means of reaching the targeted market
segment
4. Responsive: The differentiated market segments respond differentially to
marketing offerings tailored to meet their needs.
Two commonly used and widely accepted classifications of consumers are by
stage of the life cycle and by lifestyle. One classification identifies the household lifecycle
groups as the following:
1. Young unmarrieds
2. Young marrieds, no children
3. Young marrieds, with children, youngest child under six
4. Older marrieds, with children, youngest child six or older
5. Older marrieds, with children maintaining separate households
6. Solitary survivors or older single people
Some writers have expanded the number of stages by distinguishing in the last
two stages whether a person is in the labor force or retired. See Wells and Gubar (1966)
and Wagner and Hanna (1983) for more detailed explanations of the life-cycle concept
and marketing research.
The life-cycle stage has obvious implications with respect to purchases associated
with family formation (furniture, appliances, household effects, and housing) and
addition of children (food, clothing, toys, expanded housing). Other, less obvious
relationships exist as well. New-car buying reaches its peak among the older married
couples whose children have passed the age of six. A second stage of furniture buying
196
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
takes place when children begin to date and have parties at home. Dental work, travel,
and purchases of insurance are examples of service purchases associated with the life
cycle.
Lifestyle has a close association with membership in a social class. It is a basis for
segmenting customers by values, activities, interests and opinions, as well as by income.
These differences tend to be expressed through the products bought and stores
patronized, as well as the area in which one lives, club membership, religious affiliation,
and other means. The media used for expression are often either consciously or
subconsciously, symbolic representations of the class to which the person perceives he or
she belongs (or would like to belong). When used with personality traits, lifestyle
variables form the basis of psychographic classification, as illustrated in Exhibit 9.2. An
illustration of psychographic questions is shown in Table 9.3.
As an example, let us consider the life styles of the Harley Owners Group (HOG).
By examining a group of questions used in a segmentation study of values and
motorcycle use, we find a divergent group of lifestyles that have embraced the mystique
of owning a Harley.
EXHIBIT 9.2 Harley Owners Group (HOG) Classification by Psychographics
Psychographic research has suggested many different segmentation schemes. Such schemas
represent interesting demographic and product markets, and provide a much more colorful
description of the group as a whole as well as the diversity within.
Research by William Swinyard (1994a, 1994b) suggests that Harley-Davidson owners are a
diverse group consisting of six distinct segments with very different motorcycling lifestyles:
• Tour Glides find the appeal of motorcycling in long distance touring. They like riding
long distances, use their bike both for touring and everyday transportation, are more
interested in the comfort of their motorcycle than its speed, prefer riding with a
passenger, and wear a helmet.
More than the average Harley rider, Tour Glides are religiously traditional, have
somewhat old-fashioned tastes and habits, are disciplinarians with their children,
like reading, and feel they live a full and interesting life. They are less ambitious than
others, and are distinctively unattracted by social gatherings and danger.
• Vanilla Dream Riders. The Vanilla Dream Riders are more interested in the dream of
motorcycling than in motorcycling itself, and are otherwise just plain vanilla—a
relatively undistinguished group.
This is the largest, oldest, wealthiest, and among the best educated segment of
Harley owners, who have the newest motorcycles yet ride them least, and spend
little in accessorizing them. You see the Dream Riders taking riding on short trips
around town (often by themselves), wearing a helmet, and riding a stock bike. They
are distinctively unaffiliated with the “live to ride” ethic, and receive relatively little
psychic satisfaction from riding. Their motorcycle is merely a possession, having no
real place as a “family member.” They are conservative in their moral values, marital
roles, and daily behavior.
• The Hard Core segment is on the fringe of society, and identifies with the stereotypical
biker subculture.
They are the youngest, next-to-least well-educated, and certainly the poorest, yet
spend nearly 50 percent more than any other segment on accessorizing their
motorcycles. Virtually all are blue-collar workers. In relative terms, Hard Core
members are much more likely than others to feel like an outlaw, and believe
197
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
people would call them and their friends “dirty bikers.” Note, however, that they
still only “slightly agree” that these lifestyles describe them well. More than others,
the Hard Core likes to be outrageous, enjoys danger, favors legalizing marijuana,
and embraces the ethic of “eat, drink, and be merry, for tomorrow we die.”
• Hog Heaven finds great psychic and spiritual satisfaction in owning and riding a Harley.
More than others, these riders feel like an “old wild west cowboy” and closer to
nature when they ride. They have many motorcycle friends, and when group riding,
they feel the group becomes “one.” They do not like helmets, and feel cars are like a
“cage.”
This segment is distinctively mechanically inclined, likes to work on their
motorcycles, but spends little on accessories. They have old-fashioned tastes and
habits, and read relatively little. They are less likely than others to believe in a life
after death, but often think about how short life really is.
• Zen Riders too find solace and spiritual satisfaction, but find it in motorcycling itself, and
escape life’s stresses in doing so.
They include the highest percentage of married riders, but otherwise are typical of
Harley owners in most demographic characteristics. More than others, Zen Riders
find motorcycling fulfilling in many dimensions: their motorcycle seems alive and
they like all motorcycles, from dirt bikes to four-cylinder Japanese motorcycles.
Zen Riders are more impulsive and are willing to take chances and to run risks. They
believe they are more ambitious than other segments. They like to party, and have
trouble relaxing in everyday life, but they are “modern” husbands and are opposed
to legalizing marijuana.
• Live to Ride rides more than any other segment, and motorcycles represent a total
lifestyle to them. They “ride to live and live to ride.”
Members of this, the smallest segment, are most likely to have bought their
motorcycle new, and ride it the most by a wide margin. They simply love riding.
More than other Harley owners, they use their bike for everyday transportation,
enjoy riding long distances, and use their bike for touring. They find riding to be a
magical experience, and motorcycling is a total lifestyle to them.
If they did not have family, members of this segment would quit their jobs and take
off. They agree with an “eat, drink, and be merry” premise, like to create a stir, like
danger, and get lots of satisfaction from their hobbies. They care little about their
appearance and tend not to believe in a life after death.
Life style segmentation offers significant implications for marketing and advertising to
these segments in terms of not only better communicating with each group, but in
creating the product configuration and accessories that appeal to each group.
A common designation of social classes is the one originally used by Warner and
others (1960) that designates the familiar upper, middle, and lower class designations,
each divided into upper and lower segments. Thus, the Warnerian classification results in
six classes, ranging from the UU (upper upper) down through the LL (lower lower).
Somewhat newer is the value and lifestyles (VALS) schema, which classifies the
American public into: survivors, sustainers, belongers, emulators, achievers, I-am-me’s,
experientials, societally conscious, and integrateds (Mitchell, 1983).
198
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 9.3 Psychographic Measurement Questions for Three Motorcycle Owner
Segments: Most and Least Frequent Motorcycle Lifestyle Descriptors
Although less direct and more subtle than life-cycle stage in its effect on overt
buying behavior, there can be little question that an upper-middle-class household will
show more similarity in purchasing and consumption patterns of food, clothing, furniture,
and housing to another upper-middle-class household than it will to a blue-collar, upper-
199
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
lower-class household. The media to which the managerial-professional, upper-middleclass
family is exposed, and the appeals to which it responds, are also likely to be closer
to those of other managerial-professional families than to those of the blue-collar family.
Similarly, on the basis of VALS, a processor and packager of tofu may find that because
“experientials” have a greater appreciation for natural things, they are heavier users of
tofu. The marketer can then direct the product at this lifestyle group.
Another approach to values and lifestyles is the List of Values (LOV) developed
by Kahle (1983). These values fit well into the qualitative research discussed in Chapter
6, (means-end laddering to produce hierarchical value maps and ZMET analysis):
• Self-respect
• Security
• Warm relationships with others
• Sense of accomplishment
• Self-fulfillment
• Being well-respected
• Sense of belonging
• Fun and enjoyment in life
Classification of consumers is vital if we are to learn more about consumer
behavior and utilize this information to develop more efficient marketing techniques. But
caution is needed to ensure that managers do not segment too finely and use
categorizations when they should not. Given the new software available that simplifies
using such advanced techniques as neural networks, latent-class models, fuzzy or
overlapping clustering, and even occasion based segmentation, there will be a tendency to
use a backhoe when a shovel is more appropriate. The technology in this area is far ahead
of other aspects of the marketing process, and ahead of most managers’ needs (Freeman,
2001). Advanced statistical techniques for consumer classification purposes are discussed
in later chapters.
Although the discussion thus far has focused on consumers, similar classification
requirements exist and are used in studies of industrial users and marketing
intermediaries. Comparable characteristics of these firms include sales volume, number
of employees, and the type of products manufactured or handled.
Extent of Knowledge
Prediction of what actions respondents will take is often aided by knowing “how
much they know.” This is especially so when making advertising budget and media
allocation decisions, where consumers’ choices are strongly affected by their levels of
awareness and extent of knowledge of the product and its attributes. To illustrate
questions measuring consumer awareness, a study of homeowners examined their
knowledge of conventional and variable-rate mortgages. One question sought knowledge
of interest rate differences:
200
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Attitudes and Opinions
Investigators in the behavioral science fields of psychology, sociology, and
political science have made extensive studies of attitudes and opinions over a wide range
of subject areas. The study of people’s behavior in business and economic contexts is
also a behavioral science. As such, it has been a natural consequence that marketing
research has adopted, adapted and applied many techniques, including attitude-opinion
studies to obtain information applicable to the solution of marketing problems.
The terms attitude and opinion have frequently been differentiated in
psychological and sociological investigations. A commonly drawn distinction has been to
view an attitude as a predisposition to act in a certain way, and an opinion as a
verbalization of an attitude. Thus, a statement by a respondent that he or she prefers
viewing Blue-Ray DVD’s to HDTV color television programs would be an opinion
expressing (one aspect of) the respondent’s attitude toward high-definition television.
When used to predict actions that the respondent will take, this distinction
between attitude and opinion rapidly becomes blurred. Since the major purpose of
attitude-opinion research in marketing is to predict behavior, this differentiation is, at
best, of limited usefulness. We shall therefore use the terms interchangeably. Attitude
research in marketing has been conducted with the use of both qualitative and
quantitative techniques. In either form, researchers encounter problems that are more
severe than those involved in obtaining any other type of descriptive information
discussed. Despite these problems, which we will discuss in later chapters in some detail,
attitude-opinion research has been widely used to provide information for choosing
among alternatives. Its greatest use has been in the areas of product design (including
packaging and branding) and advertising. Other uses have been in selecting store
locations, developing service policies, and choosing company and trade names. In fact,
attitudes and opinions are central in customer satisfaction studies.
Measuring Attitudes
Expectancy Value Measures of Behavioral Intention (BI) and Attitudes (A)
Expectancy value models were first developed in the 1960’s as a method of
predicting behavioral intentions (a precursor of actual behavior). Expectancy value
models use attitudes to predict behavioral intention (intention to try, purchase,
recommend, or re-purchase a product or service). This methodology has become a
mainstay of marketing research and is found to perform well in predicting both consumer
behavior and consumer satisfaction/dissatisfaction.
The Expectancy value model uses attitudes and beliefs in a mathematical formulation that
is read as follows:
201
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Behavior (purchase of brand X) is approximated by Behavioral Intention (intention to
purchase brand X), which in turn is approximated by the Overall Attitude toward brand X.
• The Overall Attitude toward brand X equals the sum of all salient attitudes about brand X,
weighted by how important each attitude is in the purchase decision process.
• Normative beliefs are added to this compensatory evaluative process by considering the
norms surrounding an attitude ( My friends insist I should buy an environmentally friendly
car that gets 50 mpg), and applying weights reflecting the buyer’s motivation to comply with
the norm (Will they praise me for buying a 50 mpg car? – or Do I really care about getting
50 mpg?).








+=≈≈ ∑∑ ==
i
k
i
i
k
i
iio mcnbwbawABIB **
1
2
1
1
where:
B = Behavior
BI = Behavioral Intention, which is measured using a question such as “how likely are you
to buy a Toyota Prius sometime during the next year?" with a five or seven-point Likert or
semantic differential scale labeled "definitely will purchase" and "definitely will not
purchase" at the endpoints.
Ao = Overall Attitude toward the object (brand, product, service or company)
w1, w2 = Weights that indicate the relative influence of the overall attitude toward the object
and the normative influence to purchase the product
∑ai * bi = the sum of individual attitudes toward the object, weighted by importance. The
overall attitude is formed by the multiplicative product of ai (the person’s affective evaluation
- liking of the attribute i), and bi (here defined as the importance of attribute i in the purchase
decision). The sum is taken over the k attributes that are defined as salient in the purchase
decision.
∑nbi * mci = The normative component of the decision process that asks “should the norms
surrounding this attribute be important to me?” This is computed as the multiplicative
product of nbi (the norms governing attitude i), and mci (the motivation of the respondent to
comply with those norms).
Attitude (ai*bi)
ai is the affective (liking) component of the evaluation of attribute i. A representative
measure of ai would be using the context "In terms of buying a Toyota Prius”, to evaluate “Gets
50 miles/gallon”. The evaluation would use a five or seven point scale with endpoints ranging
from “Poor” to “Excellent”, or “Not at all Desirable” to “Very Desirable”. This is found on the
right side of Figure 9.1.
bi - the importance of attribute I in the context of behavior B. This is sometimes
measured as the probability of attribute i being associated with brand X. At other times it is
measured as the importance of attribute i in achieving behavior B. Using the context of
purchasing a Toyota Prius, the attribute would be “Gets 50 miles per gallon” could be rated on a
seven point scale with endpoints labeled "Very Unlikely" and "Very Likely", or as in the
example below, is measured as “Not at all Important” to “Very Important”. This is found on the
left side of Figure 9.1.
202
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 9.1 Expectancy Value Importance-Performance Rating Scales
Σ nbi * mci = The overall normative component of the decision process.
This is computed as the multiplicative product of nbi (the norms governing belief i), and
mci (the motivation of the respondent to comply with those norms) see Figure 9.2.
Figure 9.2 Normative Rating Scales
The normative component is often considered to be absorbed in the overall attitude
component and depending on the application, is often ignored.
34
102024
)1)(10()2)(10()3)(8(
3
1
Pr
=
−+=
−++=
= ∑=
e
e
e
i
iiius
A
A
A
ebAttitude
40
82012
)1)(8()2)(10()3)(4(
3
1
e
e
e
i
iiFusion
A
A
A
ebAttitude
++=
=+=
= ∑=
b2=10
Foreign
Warranty
40 MPG
Domestic
Warranty
Ford
Fusion
b2=10
b2=10
b3=8
b1=4
A0=34
e1=+3
A2=40
e3=-1
e2=+2
e1=+3
e3=+1
e2=+2
Toyota
Prius
50 MPG
b1=8
203
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The expectancy value model is an aggregation of attitudes and beliefs about the most
important attributes that predict behavior. This compensatory type of decision model results in a
powerful approach to measuring customer attitudes and predicting customer behavior, and is
well used in research. It is not surprising that many popular methodologies, including conjoint
analysis, are theoretically based on this measurement model.
Measuring Satisfaction
Customer satisfaction is the most common of all marketing surveys and is part of
the “big three” research studies in marketing that include market segmentation and
concept testing. Customer satisfaction has been defined as the state of mind that
customers have about a company and its products/services when their expectations have
been met or exceeded over the lifetime of product or service use (Cacioppo, 2000, p. 51).
The positive attitudes that lead to customer satisfaction usually, in turn, lead to customer
loyalty and product repurchase. But measuring satisfaction is not measuring loyalty. The
following are typical measures of overall satisfaction:
1. Overall, how satisfied are you with (brand name)?
2. Would you recommend (brand name)?
3. Do you intend to repurchase (brand name)?
According to William Neal (2000), these questions are usually measuring the
same thing—satisfaction with the product or service. Satisfaction is a condition that is
necessary, but not sufficient to predict loyalty. Customer satisfaction can be measured by
traditional surveys, by using comment cards, and for business-to-business situations by
field or call reports, to name just some of the methods (Cochran, 2001). Many companies
routinely measure satisfaction in consumer tracking studies (see Exhibit 9.4).
The attitudes and opinions of prospective buyers will clearly affect expectations
and the resulting satisfaction in a purchase decision. Consequently, the marketing
manager should be as well informed as possible about both the nature of their relevant
attitudes and opinions and the intensity with which they are held. Subaru of America, for
example, has a program that includes a Purchase Experience Survey and a Service
Experience Survey that goes out to all customers who have purchased a Subaru or had it
serviced. Findings from these surveys are reported back to the dealer, who then acts on
them. In addition, Subaru sends out a Product Survey to a sample of new Subaru owners
every year. This survey examines the quality of the product and whether new owners are
satisfied with the performance, fit, and finish of their new vehicle.
Measuring satisfaction and building a satisfaction survey requires at least a basic
knowledge of the satisfaction measurement literature, combined with knowledge and
experiences with the company’s customers. We will first introduce the theoretical and
methodological underpinnings of satisfaction research by first defining the concept of
customer satisfaction and how satisfaction is used in business. Next, different
satisfaction survey measures are discussed and presented. Finally, the components of a
satisfaction survey are presented, along with sample satisfaction survey questions. The
Qualtrics.com survey library contains sample satisfaction questionnaires, questions and
scales. This discussion of satisfaction research provides the basis for understanding how
to measure satisfaction and why the suggested should be used.
204
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 9.4 Common Ingredients of a Customer Satisfaction Survey
Product Use
Frequency of product use
Primary use location
Primary precipitating events or situations for product use or need
Usage rates and trends
Product Familiarity
Degree of actual product use familiarity
Knowledge (read product information, read product label, etc.)
Knowledge and Involvement with product and the purchase process
Awareness of other brands
Reasons for original product purchase (selection reasons)
Primary benefits sought from the product
Product Evaluation
Attribute evaluation matrix: (quality, price, trust, importance, performance, value)
Perceived benefit associations matrix
Importance, performance
Identification of primary benefits sought
Comparison to other brands (better, worse)
What is the best thing about the brand, what could be done better
Message and Package Evaluation
Packaging size, design
Advertising Promise, message fulfillment evaluation
Value Analysis
Expectation of price
Expectation of relative price (full price, on sale)
Current price paid
Satisfaction Measurements
Overall Satisfaction
Reasons for Satisfaction Evaluation
Satisfaction with attributes, features, benefits
Satisfaction with use
Expected and Ideal Satisfaction-Performance Measures
Likelihood of recommending
Likelihood of repurchasing
What Is Customer Satisfaction?
Customer satisfaction measures indicate how well a company’s products or services meet
or exceed customer expectations. These expectations will reflect many aspects of the company’s
business activities including the actual product, service and company. Customer satisfaction
measures will tap the customer’s lifetime of product and service experience.
205
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Why is Customer Satisfaction So Important?
Effective marketing focuses on two activities: retaining existing customers and adding
new customers. Customer satisfaction measures are critical to any company, be it consumer or
B2B, because customer satisfaction is a strong predictor of customer retention, customer loyalty
and product repurchase.
When to Conduct Customer Satisfaction Surveys
When to measure customer satisfaction depends on the kind of product or service
provided, the kinds of customers served, the number of customers served, the longevity
and frequency of customer/supplier interactions, and the intended use of the results.
Three very different approaches may each produce meaningful and useful results
that are appropriate for specific situations, uses and needs:
• Post Purchase Evaluations reflect the satisfaction of the individual customer at
the time of product or service delivery (or shortly afterwards). This type of
satisfaction survey is typically used as part of a CRM (Customer Relationship
Management System) and focuses on securing a long term relationship with the
individual customer
• Periodic Satisfaction Surveys provide an occasional snapshot of customer
experiences and expectations, and are conducted for specific groups of consumers
on a periodic basis.
• Continuous Satisfaction Tracking is often part of a management initiative to
assure quality is at high levels over time, and can involve post purchase
evaluations or a succession of periodic satisfaction surveys conducted on a regular
basis (daily, quarterly or monthly basis). Satisfaction feedback is obtained from
the individual customer at the time of product or service delivery (or shortly
afterwards). Satisfaction tracking surveys are.
Satisfaction surveys are developed to provide an understanding of customers'
expectations and satisfaction. Satisfaction surveys typically require multiple questions
that address satisfaction with different dimensions of the product or service concept. That
is, satisfaction measurement includes measures of overall satisfaction, satisfaction with
individual product and service attributes, and satisfaction with the benefits recorded as a
result of purchase. Satisfaction measurement is like peeling away layers of an onion-each
layer reveals yet another deeper layer, closer to the core.
Each of the three methods of conducting satisfaction surveys are helpful in
obtaining customer feedback for assessing overall accomplishments, degree of success,
and areas for improvement. Ulwick (2005) advocates the combination of satisfaction and
importance to identify unfulfilled opportunities in the marketplace. In this case,
performance gaps not being adequately served by the products in the market are
identified, evaluated for feasibility, and targeted for development.
206
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Customer Satisfaction Survey Measures
Customer satisfaction surveys will often include several different multiple measures of
satisfaction, including:
• Overall measures of customer satisfaction
• Affective measures of customer satisfaction
• Cognitive measures of customer satisfaction
• Behavioral measures of customer satisfaction
• Expectancy value measures of customer satisfaction
We will now explain each in detail. Because general measures of customer satisfaction
usually involve product fulfillment, we will discuss product use scenarios focusing on where and
how the product is used.
Satisfaction Measurement: Overall Measures of Satisfaction
Like when we have a great food experience at a favorite local restaurant, elevated levels
of customer satisfaction usually leads to customer loyalty and product repurchase. But measures
of satisfaction are different than measures of loyalty. Satisfaction measurement questions
typically include items like those found in Exhibit 9.5 and like the following:
1. An overall satisfaction measure (emotional):
Overall, how satisfied are you with “Yoni fresh yogurt”?
This question reflects the overall opinion of a consumer’s satisfaction experience. It is
noteworthy that we can meaningfully measure attitudes towards a product that a consumer
has never used, but we cannot measure satisfaction for a product or brand that has never been
used. There is no experience base for such a measure of satisfaction.
2. A loyalty measure (affective, behavioral):
Would you recommend “Yoni” to your family and friends?
3. A series of attribute satisfaction measures (affective and cognitive):
How satisfied are you with the “taste” of Yoni fresh yogurt?
How important is “taste” in your decision to select Yoni fresh yogurt?
Satisfaction and attitude are closely related concepts. The psychological concepts of attitude
and satisfaction may both be defined as the evaluation of an object and the individual’s
relationship to it. The distinction is that here, satisfaction is a "post experience" state
representing the emotional affect produced by the product’s quality or value.
4. Intentions to repurchase (behavioral measures):
Do you intend to repurchase Yoni fresh yogurt?
Satisfaction can influence post-purchase/post-experience actions other than usage (these
other actions might include word of mouth communications and repeat purchase behavior).
Additional post-experience actions might reflect a heightened level of product involvement
207
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
that in turn results in increased product or information search activity, reduced trial of
alternative products, and even changes in preferences for shopping locations and choice
behavior.
As shown in Figure 9.3, customer satisfaction is diagrammatically shown to be
influenced by perceived quality of product and service attributes, associated product features and
benefits, and is moderated by the customer’s expectations regarding the product or service. The
researcher may want to define and develop measures for each of these constructs that influence
customer satisfaction.
Affective Measures of Customer Satisfaction
Attitudes toward a product can be developed as a result of product information or any
experience with the product, whether perceived or real. Affect (liking/disliking) is best
measured in the context of product attributes or benefits. Again, it may be meaningful to measure
attitudes towards a product or service that a consumer has never used, but it is not meaningful to
measure satisfaction when a product or service has not been used.
Figure 9.3 Building a Customer Satisfaction Survey
208
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 9.5 Sample Satisfaction Measures from the Qualtrics Question Library
209
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
210
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
211
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
212
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Cognitive Measures of Customer Satisfaction
A cognitive element is defined as an appraisal or conclusion that the product was useful
(or not useful), fit the situation (or did not fit), exceeded the requirements of the
problem/situation (or did not exceed), or was an important part of the product experience (or was
unimportant). Cognitive responses are often specific to the situation for which the product was
purchased and specific to the consumer’s intended use of the product, regardless if that use is
correct or incorrect.
Behavioral Measures of Customer Satisfaction
It is sometimes believed that dissatisfaction is synonymous with regret or disappointment
while satisfaction is linked to ideas such as, "it was a good choice" or "I am glad that I bought
it." When phrased in behavioral response terms for a future or hypothetical behavior, consumers
indicate that “purchasing this product would be a good choice” or “I would be glad to purchase
this product.” Often, behavioral measures reflect the consumer’s past experience with
individuals associated with the product (i.e. customer service representatives) and the intention to
repeat that experience.
Expectations Measures of Customer Satisfaction
As might be expected, many different approaches to measuring satisfaction exist in the
consumer behavior literature. Parasuraman, Ziethaml and Berry (1998) refined their earlier work
to identify five generic dimensions of a service satisfaction scale called SERVQUAL: tangibles,
reliability, responsiveness, assurance and empathy. SERVQUAL has become the standard for
service quality measurement, but it should be recognized that satisfaction dimensions seem to
vary depending on the application (high tech vs. health care vs. leisure services, etc.). None the
less, this important work is often used for the measurement of customer satisfaction in a service
environment.
Figure 9.4 Product and Service Questions
213
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
A slightly different diagnostic approach to satisfaction measurement is one that examines
the gap between the customer's expectations of performance and the consumer’s actual
experience. This “satisfaction gap" involves measuring both perception of performance and
expectation of performance along specific product or service attributes dimensions. This
approach can be applied to an individual product or service experience, a brand, or even a
complete product category, as proposed by Ulwick (2005).
Again we see that customer satisfaction is largely a reflection of both expectations and
experiences that the customer has with a product or service. Expectations may reflect past
product experiences, but will also reflect the purchase evaluation process that occurs when
shopping for a product or service. For example, when we make major purchases, we research
the product or service, collecting information from the advertising, salespersons, and word-ofmouth
from friends and associates. This information influences our expectations and ability to
evaluate quality, value, and the ability of the product or service to meet our needs.
Customer Expectations that Influence Satisfaction
Customers hold both explicit and implicit performance expectations for attributes,
features and benefits of products and services. The nature of these expectations will dictate the
form and even the wording of a satisfaction questionnaire.
Explicit expectations are mental targets for product performance, such as well identified
performance standards. For example, if expectations for a color printer were for 17 pages per
minute and high quality color printing, but the product actually delivered 3 pages per minute and
good quality color printing, then the cognitive evaluation comparing product performance and
expectations would be 17 PPM – 3 PPM + High – Good, with each item weighted by their
associated importance.
Implicit expectations represent the norms of performance that reflect accepted standards
established by business in general, other companies, industries, and even cultures. An implicit
reference might include wording such as “Compared with other companies…” , or “Compared to
the leading brand…”
Static performance expectations address how performance and quality for a specific
application are defined. Performance measures for each application are unique, though general
expectations relate to quality of outcome and may include those researched by Parasuraman or
others such as: accessibility, customization, dependability, timeliness, and accuracy, tangible
cues which augment the application, options, cutting edge technology, flexibility, and user
friendly interfaces. Static performance expectations are the visible part of the iceberg; they are
the performance we see and -- often erroneously – are assumed to be the only dimensions of
performance that exist.
Dynamic performance expectations are about how the product or service evolves over
time and includes the changes in support and product or service enhancement needed to meet
future business or use environments. Dynamic performance expectations may help to produce
“static” performance expectations as new uses, integrations, or system requirements develop and
become more stable.
214
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Technological expectations focus on the evolving state of the product category. For
example, mobile phones are continually evolving. Mobile service providers, in an effort to deal
with the desire to switch to new technology phones, market rate plans with high cancellation
penalties for switching providers, but with liberal upgrade plans for the phones they offer. The
availability of low profile phones with email, camera, MP3, email, blue tooth technology, and
increased storage will change technology expectations as well as the static and dynamic
performance expectations of the product. These highly involving products are not just feature
based, but have expectations that enhance perceptions of status, ego, self-image, and can even
invoke emotions of isolation and fear when the product is not available.
Interpersonal expectations reflect the relationship between the customer and the product
or service provider. Person to person relationships are increasingly important, especially where
products require support for proper use and functioning. Expectations for interpersonal support
include technical knowledge and ability to solve the problem, ability to communicate, time to
problem resolution, courtesy, patience, enthusiasm, helpfulness, assurance that they understood
my problem and my situation, communication skills, and customer perceptions regarding
professionalism of conduct, often including image, appearance.
Situational Expectations
In building a customer satisfaction survey, it is also helpful to consider reasons why prepurchase
expectations or post-purchase satisfaction may or may not be fulfilled or even
measurable. In some situations the following conditions may occur:
1) Expectations may not reflect unanticipated service attributes;
2) Expectations may be quite vague, creating wide latitude of acceptable performance
and expected satisfaction;
3) Expectation and product performance evaluations may not be cognitive, but instead
sensory, as in expectations of taste, style or image. Such expectations are not only
difficult to evaluate and understand, but may change over time.
4) The product use may attract so little attention as to produce no conscious affect or
cognition (evaluation), and result in measures that are meaningless satisfaction or
dissatisfaction measures;
5) There may have been unanticipated benefits or consequences of purchasing or using
the product (such as a use, usage situation, or feature not anticipated with purchase);
6) The original expectations may have been unrealistically high or low;
7) The product purchaser, influencer and user may have each been a different individual,
each having different expectations.
When fulfilled, expectations result in customer satisfaction (or when expectations are not
fulfilled, result in dissatisfaction and complaining behavior). Each of the above discussed types
of expectation should be considered in the context of the unique research project to determine if
special consideration is warranted during questionnaire development. The research study may
also benefit from consideration of expectations related to perceived quality and value.
215
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Satisfaction Measurement: Perceived Quality Measures
Perceived quality is often measured through three measures: overall quality, perceived
reliability, and the extent to which a product or service is able to fulfill the customer’s needs.
Customer experiences that result in attributions of quality are the single greatest predictor of
customer satisfaction.
Satisfaction Measurement: Perceived Value Measures
Perceived value may conceptually be defined as the overall price divided by quality or
the overall quality divided by price. Perceived value is measured in many ways including overall
evaluation of value, expectations of price that would be paid, and more rigorous methodologies
including the Van Westendorp pricing analysis and conjoint analysis. Note that these are
Qualtrics advanced option question types, and that the Qualtrics online survey university
contains white papers and tutorials about these topics.
The consumer behavior literature has long shown that price is a primary indicator of
quality when other attributes and benefits are relatively unknown. However when repeat
purchases are made in some product categories, price may be reduced in importance. This may
reflect a decision simplification strategy that makes shopping or purchase decisions easier, or
may reflect increased customer loyalty.
Satisfaction Measurement: Customer Loyalty Measures
Customer loyalty reflects the likelihood of repurchasing products or services. Customer
satisfaction is a major predictor of repurchase, but is strongly influenced by explicit performance
evaluations of product performance, quality, and value.
Loyalty is often measured as a combination of measures including overall satisfaction,
likelihood of repurchase, and likelihood of recommending the brand to a friend. For a common
measure of loyalty might be the sum of the following three questions measured on an agreement
scale (Strongly agree – Strongly disagree):
1. Overall, how satisfied are you with [brand]?
2. How likely are you to continue to choose/repurchase [brand]?
3. How likely are you to recommend [brand] to a friend or family member?
Measures of loyalty are generally not perfect predictors of the brand switching that often
occurs in an actual consumer purchase situation. Switching may occur because of out of stock
situations, inconvenience of going to a store that has the preferred brand, or the lack of
commitment to the brand. This latter condition has been addressed by Gallup in the
identification of four emotional states that are indicative of the degree of commitment and brand
loyalty (Table 9.3).
Table 9.3 Brand Loyalty Measures
Construct Measure
Confidence in the brand [Brand] is a name I can always trust.
[Brand] always delivers on what they promise.
Integrity of the brand [Brand] always treats me fairly.
If a problem arises, I can always count on [brand]
to reach a fair and satisfactory resolution
Pride in the brand I feel The to be a [brand] customer.
[Brand] always treats me with respect.
Passion for the brand [Brand] is the perfect company for people like me.
I can't imagine a world without [brand].
216
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 9.5 Gallup Customer Loyalty Example
The application of these loyalty scales enables the differentiation of brands, companies
and even industries according to how well they are “engaging” their customers (Figure 9.5).hat
This approach to measuring customer loyalty opens the door for the creation of
performance benchmarks and for management decision making. An improved customer
experience can be tracked and improved as it instills confidence in the brand through an image of
trust, confidence, and fairness. Customer loyalty is created by front-line employees who interact
with the customer. It must always be remembered that all companies face the challenge of
meeting increasing customer expectations. Dealing with expectations in a timely and effective
manner increases customer loyalty and retention.
Measuring Customer Expectations and Satisfaction
Expectations are beliefs that are measured as the (likelihood or probability that a product
or service (with certain attributes, features or characteristics) will produce certain outcomes
(benefits-values). As we have discussed, these expectations are based on the consumer’s
reservoir of affective, cognitive and behavioral experiences. Expectations are seen as related to
satisfaction and can be measured in the following ways:
1) As an Importance-Value relationship that fulfills the expectations for the product/service.
This can include overall expectations or expectations for the individual product attributes
most important in the purchase decision or experience.
2) Affect-Satisfaction Expectations: The (liking/disliking) of the product/service or of the
attributes that comprise the product/service.
3) Fulfillment of Expectations: the expected level of performance vs. the desired expectations.
This is “Predictive Fulfillment” and is a respondent specific index of the performance level
necessary to satisfy.
4) Expected Value from Use: Satisfaction is often determined by the frequency of use. If a
product/service is not used as often as expected, the result may not be as satisfying as
anticipated. For example a Harley Davidson motorcycle that is purchased with high
expectations for a fun and adventurous lifestyle, but sits in the garage, an unused year
subscription to the local fitness center/gym or a little used season pass to the local ski resort
or amusement park may produce more dissatisfaction with the decision to purchase than with
the actual product/service.
217
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
SUMMARY
Customer satisfaction is the most important element of business success and is a key
measure of fulfillment of business strategies, including those involving segmentation and
concept development.
In this chapter our discussion has provided an introduction to satisfaction measurement
that illustrates the theoretical and methodological underpinnings of satisfaction research for
business. The different satisfaction survey measures discussed were presented as components of
a satisfaction survey, along with sample satisfaction survey questions. The Qualtrics survey
library contains sample satisfaction questionnaires, and the Qualtrics question library contains
sample individual questions. These templates will assist you in further understanding what
measures should be included in your satisfaction research and why those measures are necessary
and of value.
REFERENCES
Cochran, C. (2001, November). Customer satisfaction: The elusive quality. Quality
Digest, 45–50.
Freeman, 2001 Freeman, L. (2001, September 24). Small, smaller, smallest: New
analytical tools can slice markets too thin. Marketing News, 35, 1ff.
James, D. (2002, May 13). Better together. Marketing News, 36, 15–16.Kahle (1983
Mitchell, A. (1983). The nine American lifestyles. New York: Warner Books
Neal, W. D. (2000, June 5). When measuring loyalty satisfactorily, don’t measure CS.
Marketing News, 34, 19.
Parasuraman, A, V. Zeithaml, and L. Berry."SERVQUAL: A Multiple-Item Scale for
Measuring Customer Perceptions of Service Quality," Journal of Retailing, Spring 1988,
pp. 12-40.
Ulwick, A. (2005) What Customers Want: Using Outcome-Driven Innovation to Create
Breakthrough Products and Services, New York: McGraw Hill
Wagner, J., & Hanna, S. (1983, December). The effectiveness of family life cycle
variables in consumer expenditures research. Journal of Consumer Research, 10, 281–
291.
Wells, W. D., & Gubar, G. (1966, November). Life cycle concept in marketing research.
Journal of Marketing Research, 3, 355–363.
218
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This page is intentionally blank
219
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 10
GENERAL CONCEPTS OF MEASUREMENT
AND SCALING
Questions focus on the problem we are trying to solve, while answers are more closely associated with the
measurement scale we are using to achieve our analysis.
Survey research, as a source of marketing information addresses many topics of
practical interest, including concept testing for new products, corporate image measurement, ad
copy evaluation, purchase intentions, customer satisfaction, and so forth. Regardless of the
research topic, useful data is obtained only when the researcher exercises care in making
procedural decisions such as:
1. Defining what is to be measured
2. Deciding how to make the measurements
3. Deciding how to conduct the measuring operations
4. Deciding how to analyze the resulting data
Definitions and decisions play a significant role in scientific inquiry, especially in
marketing research and the behavioral sciences.
In the first section of this chapter, we focus on conceptual and operational definitions and
their use in research. Increasingly, behavioral scientists are paying greater attention to defining
the concepts measured in their specific disciplines, and refining operational definitions that
specify how to measure and quantify the variables defining the concepts.
In the next section we discuss measurement scales and their relationship to the
interpretation of statistical techniques. This section serves as useful background for the
discussion of statistical techniques covered in later chapters. We then discuss the pragmatics of
writing good questions.
The overall quality of a research project depends not only on the appropriateness and
adequacy of its research design and sampling techniques, but also on the measurement
procedures used. The third section of this chapter looks at measurement error and how we may
control the reliability and validity in these measurements.
Definitions in Marketing Measurement
Marketers measure marketing program success as increased brand awareness, ad
awareness, ratings of brand likeability and uniqueness, new product concept ratings and purchase
intent, and customer satisfaction (Morgan, 2003). Researchers will often model these constructs.
Models are representations of reality and therefore raise the fundamental question of how
well each model confidently represents reality on all significant issues. The quality of a model is
judged against the criteria of validity and utility. Validity refers to a model’s accuracy in
describing and predicting reality, whereas utility refers to the value it adds to the making of
decisions. A sales forecasting model that does not forecast sales with reasonable accuracy is
probably worse than no sales forecasting model at all.
Model quality also depends on completeness and validity, two drivers of model accuracy.
Managers should not expect a model to make decisions for them, but instead models should be
220
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
viewed as one additional piece of information to help make decisions.
Clearly, managers will probably benefit from models that are simple enough to
understand and deal with. But models used to help make multi-million-dollar decisions should be
more complete than those used to make hundred-dollar decisions. The required sophistication of
a model to be used depends on the model’s purpose. We measure the value of a model based on
its efficiency in helping us arrive at a decision. Models should be used only if they can help us
arrive at results faster, with less expense, or with more validity.
Building Blocks for Measurement and Models
We cannot measure an attitude, a market share, or even sales, without first specifying
how it is defined, formed, and related to other marketing variables. To better understand this, we
must briefly study the building blocks of measurement theory: concepts, constructs, variables,
operational definitions, and propositions.
Concepts and Constructs
A concept is a theoretical abstraction formed by a generalization about particulars.
“Mass”, “strength”, and “love” are all concepts, as are “advertising effectiveness”, “consumer
attitude”, and “price elasticity”. Constructs are also concepts, but they are observable,
measurable, and are defined in terms of other constructs. For example, the construct “attitude”
may be defined as “a learned tendency to respond in a consistent manner with respect to a given
object of orientation.”
Variables
Researchers loosely call the constructs that they study variables. Variables are constructs
in measured and quantified form. A variable can take on different values (i.e., it can vary).
Operational Definitions
We can talk about “consumer attitudes” as if we know what it means, but the term makes
little sense until we define it in a specific, measurable way. An operational definition assigns
meaning to a variable by specifying what is to be measured and how it is to be measured. It is a
set of instructions defining how we are going to treat a variable. For example, the variable
“height” could be operationally defined in a number of different ways, including measures in
inches with a precision ruler with the person (1) wearing shoes, or (2) not wearing shoes, (3) by
an altimeter or barometer, or (4) for a horse, by the number of “hands”
As another example, measuring “purchase intentions” for Brand X window cleaner might
be operationally defined as the answer to the following question:
The researcher could have just as appropriately defined a measurement of “purchase
intention” in other ways. For example, the concepts of “attitude” and “importance”, which have
been summed as the multiplicative product of a series of attributes like cleaning ability, fresh
smell, etc. This would appear as:
221
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
P ≈ BI = Σ Ai × Bi
Purchase Purchase Attitudes Importance of
Behavior Intention about Brand X Window Attributes of
Toward For Brand X Cleaner Attributes Brand X Window Cleaner
Propositions
A proposition defines the relationships between variables, and specifies both the
variables influencing the relationship and the form of the relationship. It is not enough to simply
state that the concept “sales” is a function of the concept “advertising”, such that S = f (Adv).
Intervening variables must be specified, along with the relevant ranges for the effect, including
where we would observe saturation effects, threshold effects, and the symbolic form of the
relationship.
Integration into a Systematic Model
A proposition is quite similar to a model. A model is produced by linking propositions
together to provide a meaningful explanation for a system or a process. When concepts,
constructs, variables and propositions are integrated into a model for a research plan, we should
conceptually ask the following questions:
• Are concepts and propositions specified?
• Are the concepts relevant to solving the problem at hand?
• Are the principal parts of the concept clearly defined?
• Is there consensus as to which concepts are relevant in explaining the problem?
• Are the concepts properly defined and labeled?
• Is the concept specific enough to be operationally reliable and valid?
• Do clear assumptions made in the model link the concepts?
• Are the limitations of the model stated?
• Can the model explain and predict?
• Can the model provide results for managerial decision making?
• Can the model be readily quantified?
• Are the outcomes of the model supported by common sense?
If the model does not meet the relevant criteria, it probably should be revised. Concept
definitions may be made more precise; variables may be redefined, added, or deleted; operational
definitions and measurements may be tested for validity; and/or mathematical forms revised.
INACCURACIES IN MEASUREMENT
Before delving into measurement scales and question types, it is helpful to remember that
measurements in marketing research are rarely “exact.” Inaccuracies in measurement arise from
a variety of sources or factors. A portion of this variation among individual scores may represent
true differences in what is being measured, while other variation may be error in measurement.
For any given research project, not all will necessarily be operative, but the many possible
sources causing variations in respondent scores can be categorized as follows:
• True differences in the characteristic or property
• Other relatively stable characteristics of individuals which affect scores (intelligence, extent of education,
information processed)
• Transient personal factors (health, fatigue, motivation, emotional strain)
• Situational factors (rapport established, distractions that arise)
222
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Variations in administration of measuring instrument, such as interviewers
• Sampling of items included in the instrument
• Lack of clarity (ambiguity, complexity, interpretation of words and context)
• Mechanical factors (lack of space to record response, appearance of instrument)
• Factors in the analysis (scoring, tabulation, statistical compilation)
• Variations not otherwise accounted for (chance), such as guessing an answer
In the ideal situation, variation within a set of measurements would represent only true
differences in the characteristic being measured. For instance, a company wanting to measure
attitudes toward a possible new brand name and trademark would like to feel confident that
measurement differences concerning the proposed names represent only the individuals’
differences in this attitude. Obviously the ideal situation for conducting research seldom, if ever
exists. Measurements are often affected by characteristics of individual respondents such as
intelligence, education level, and personality attributes. Therefore, the results of a study will
reflect not only differences among individuals in the characteristic of interest but also differences
in other characteristics of the individuals. Unfortunately, this type of situation cannot be easily
controlled unless the investigator knows all relevant characteristics of the population members
such that control can be introduced through the sampling process.
There are many influences in a measurement other than the true characteristic of
concern—that is, there are many sources of potential error in measurement. Measurement error
has a constant (systematic) dimension and a random (variable) dimension. If the error is truly
random, (it is just as likely to be greater than the true values as less) then the expected value of
the sum of all errors for any single variable will be zero, and therefore less worrisome than
nonrandom measurement error (Davis, 1997). Systematic error is present because of a flaw in the
measurement instrument or the research or sampling design. Unless the flaw is corrected, there is
nothing the researcher can do to get valid results after the data are collected. These two subtypes
of measurement error affect the validity and reliability of measurement, topics that are discussed
in the later part of this chapter. But now that we are aware of the conceptual building blocks and
errors in measurement that should be considered in developing measurement scales, we will
consider the types of measurement and associated questions that are commonly used in
marketing research today.
MEASUREMENT CONCEPTS
Measurement can be defined as a way of assigning symbols to represent the properties of
persons, objects, events, or states. These symbols should have the same relevant relationship to
each other as do the things they represent. Another way of looking at this is that measurement is
“the assignment of numbers to objects to represent amounts or degrees of a property possessed
by all of the objects” (Torgerson, 1958, p. 19). If a characteristic, property, or behavior is to be
represented by numbers, a one-to-one correspondence between the number system used and the
various quantities (degrees) of that being measured must exist. There are three important
characteristics or features of the real number series:
1. Order. Numbers are ordered.
2. Distance. Differences exist between the ordered numbers.
3. Origin. The series has a unique origin indicated by the number zero.
A scale of measurement allows the investigator to make comparisons of amounts and changes in
the variable being measured. It is important to remember that it is the attributes or characteristics
of objects we measure, not the objects themselves.
223
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Primary Types of Scales
To many people, the term scale suggests such devices as a bathroom scale, pan balances,
yard sticks, gasoline gauges, measuring cups, and similar instruments for finding length, weight,
volume, and the like. We ordinarily tend to think about measurement in the sense of well-defined
scales possessing a natural zero and constant unit of measurement. In the behavioral sciences
(including marketing research), however, we must frequently settle for less-precise data. Scales
can be classified into four major categories, designated as Nominal, Ordinal, Interval, and
Ratio scales.
Each scale possesses its own set of underlying assumptions about order, distance and
origin, and how well the numbers correspond with real-world entities. As our rigor in
conceptualizing concepts increases, we can upgrade our measurement scale. One example is the
measurement of color. We may simply categorize colors (nominal scale), or measure the
frequency of light waves (ratio scale).
The specification of scale is extremely important in all research, because the type of measurement
scale dictates the specific analytical (statistical) techniques that are most appropriate to use in
analyzing the obtained data.
Nominal Scales
Nominal scales are the least restrictive and, thus, the simplest of scales. They support
only the most basic analyses. The nominal scale serves only as labels or tags to identify objects,
properties or events. The nominal scale does not possess order, distance, or origin. For example,
we can assign numbers to baseball players. We have a one-to-one correspondence between
number and player and are careful to make sure that no players receive the same number (or that
a single player is assigned two or more numbers). The classification of supermarkets into
categories that “carry our brand” versus those that “do not carry our brand” is further illustration
of the nominal scale.
It should be clear that nominal scales permit only rudimentary mathematical operations.
We can count the stores that carry each brand in a product class and find the modal (highest
number of mentions) brand carried. The usual statistical operations involving the calculations of
means, standard deviations, etc. are not appropriate or meaningful for nominal scales.
Ordinal Scales
Ordinal scales are ranking scales and possess only the characteristic of order. These
scales require the ability to distinguish between objects according to a single attribute and
direction. For example, a respondent may be asked to rank a group of floor polish brands
according to “cleaning ability”. An ordinal scale results when we assign the number 1 to the
highest-ranking polish, 2 to the second-highest ranking polish, and so on. Note, however, that the
mere ranking of brands does not quantify the differences separating brands with regard to
cleaning ability. We do not know if the difference in cleaning ability between the brands ranked
1 and 2 is larger, less than, or equal to the difference between the brands ranked 2 3. In dealing
with ordinal scales, statistical description can employ positional measures such as the median,
quartile, and percentile, or other summary statistics that deal with order among.
An ordinal scale possesses all the information of a nominal scale in the sense that
equivalent entities receive the same rank. Also, like the nominal scale, arithmetic averaging is
not meaningful for ranked data.
224
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Interval Scales
Interval scales possess a constant unit of measurement and permit one to make
meaningful statements about differences separating two objects. This type of scale possesses the
properties of order and distance, but the zero point of the scale is arbitrary. Among the most
common examples of interval scaling are the Fahrenheit and Centigrade scales used to measure
temperature, and various types of indexes like the Consumer Price Index. While an arbitrary zero
is assigned to each temperature scale, equal temperature differences are found by scaling equal
volumes of expansion in the liquid used in the thermometer. Interval scales permit inferences to
be made about the differences between the entities to be measured (say, warmth), but we cannot
meaningfully state that any value on a specific interval scale is a multiple of another.
An example should make this point clearer. It is not empirically correct to say that 50°F
is twice as hot as 25°F. Converting from Fahrenheit to Centigrade, we find that the
corresponding temperatures on the centigrade scale are 10°C and –3.9°C, which are not in the
ratio 2:1. We can say, however, that differences between values on different temperature scales
are multiples of each other. That is, the difference of 50°F–0°F is twice the difference of 25°–
0°F. The corresponding differences on the Centigrade scale are 10°C – (–17.7°C) = 27.7°C and –
3.9°C – (–17.7°C) = 13.8°C are in the same 2:1 ratio.
Interval scales are unique up to a transformation of the form y = a + bx; b > 0. This means
that interval scales can be transformed from one to another by adding or multiplying a constant.
For example, we can convert from a Fahrenheit to Celsius using the formula:
TC = 5/9 (TF – 32)
Most ordinary statistical measures (such as arithmetic mean, standard deviation, and
correlation coefficient) require only interval scales for their computation.
Ratio Scales
Ratio scales represent the elite of scales and contain all the information of lower-order
scales and more besides. These are scales like length and weight that possess a unique zero point,
in addition to equal intervals. All types of statistical operations can be performed on ratio scales.
An example of ratio-scale properties is that 3 yards is three times 1 yard. If transformed
to feet, then 9 feet and 3 feet are in the same 3:1 ratio. It is easy to move from one scale to
another merely by applying an appropriate positive multiplicative constant; this is the practice
followed when changing from grams to pounds or from feet to inches.
Relationships Among Scales
To provide some idea of the relationships among nominal, ordinal, interval, and ratio
scales, the marketing researcher who uses descriptive statistics (arithmetic mean, standard
deviation) and tests of significance (t-test, F-test) should require that the data are (at least)
interval-scaled .
From a purely mathematical point of view, you can obviously do arithmetic with any set
of numbers—and any scale. What is at issue here is the interpretation and meaningfulness of the
results. As we select more powerful measurement scales, our abilities to predict, explain, and
otherwise understand respondent ratings also increase.
225
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 7.1 Scales of Measurement
Scale
Mathematical
Group Structure Permissible Statistics Typical Elements
Nominal Permutation group
y = f(x), where f(x) means any
one-to-one correspondence
Mode
Contingency Coefficient
Numbering of football players
Assignment of type or model
numbers to classes
Ordinal Isotonic group
y = f(x), where f(x) means any
strictly increasing function
Median
Percentile
Order correlation
Sign test; run test
Hardness of minerals
Quality of leather, lumber,
wool, etc.
Pleasantness of odors
Interval General linear group
y = a+bx
b > 0
Mean
Average deviation
Standard deviation
Product-moment correlation
t-test, F-test
Temperature (Fahrenheit and
centigrade)
Energy
Calendar dates
Ratio Similarity group
y = cx
c > 0
Geometric mean
Harmonic mean
Coefficient of variation
Length, width, density,
resistance
Pitch scale, loudness scale
( Stevens 1946, p. 678)
Basic Question and Answer Formats
Underlying every question is a basic reason for asking it. This reason reflects the
construct to be measured, the problem to be solved or hypothesis to be tested. Constructing a
question that reflects this reason will result in a higher probability that the desired response will
be obtained. Table 7.2 shows nine different types of questions (based on the nature of content),
the broad reason underlying asking each type of question, and some examples of each type.
Table 7.2 Basic Question Types
Type of Question Goal of Question Positioning of Question
Factual or
behavioral
To get information. Questions beginning with what, where, when,
why, who and how.
Explanatory To get additional information or to
broaden discussion.
How would that help? How would you go
about doing that? What other things should be
considered?
Attitudinal To get perceptions, motivations, feelings,
etc., about an object or topic.
What do you believe to be the best?
How strongly do you feel about XYZ?
Justifying To get proof to challenge old ideas and to
get new ones.
How do you know?
What makes you say that?
Leading To introduce a thought of your own. Would this be a possible solution?
What do you think of this plan?
Hypothetical To use assumptions or suppositions. What would happen if we did it this way? If it
came in blue would you buy it today?
Alternative To get a decision or agreement. Which of these plans do you think is best? Is
one or two o’clock best for you?
Coordinative To develop common agreement.
To take action.
Do we all agree that this is our next step?
Comparative To compare alternatives or to get a
judgment anchored by another item.
Is baseball more or less exciting to watch on
TV than soccer?
226
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Based on this structure, and the information in Table 7.3, which deals with standard
answer formats, we are able to distinguish four basic question/answer types:
1. Free-answer (open-ended text)
2. Choice answers: dichotomous, single choice and multiple choice (select k of n)
3. Rank order answers
4. Constant sum answers
Table 7.3 Standard Answer Formats Based on Task
Measurement
Scale* Format Type Description
N,O,I Select 1/n—pick-1: The respondent is given a list of n options and is required to choose one option only.
N,O Select k/n—pick-k: The respondent gets a set of n options to select from but this time chooses up to k options (k =
n).
N,O Select k1/k2/n—pick-
and-pick:
The respondent is asked to select k1 options in Category 1 and k2 options in Category 2. Each
option can be selected in only one of the two categories.
N, I Sort and rank: The respondent picks k items and allocates them into L buckets, then items allocated to each
bucket are assigned ranks.
N,O Rank k/n—rank: In this question the respondent gets n options and is asked to rank the top k (k = n).
N,O Select k1/n and Rank
k2/k1—pick and rank:
This question type is similar to pick-k, but in addition to selecting k1 options from a list of n
options, the respondent is then asked to rank some fraction, k2/k1 of those selected.
O,I Integer Rating: The respondent is asked to rate on a linear scale of 1 to n the description on the screen or
accompanying prop card (for example, 1 for completely disagree to 5 for completely agree).
Only integer responses are accepted.
O,I Continuous Rating: This is similar to integer rating, except that the response can be any number (not necessarily
an integer number) within the range (for example, 5.2 on a scale of 0 to 10).
R Constant Sum: The respondent is provided with a set of attributes (5, 10, etc.) and is asked to distribute a total
of p points across those attributes.
N,O Yes/No: This question entails a yes/no answer and is of course, a Select 1/2---pick-1 question type.
I Integer—integer-#: The respondent is asked for a fact that can be expressed in integer number form. A valid range
can be provided for error checking. Example: Age.
I,R Real—real-#: Similar to integer-# except that the answer expected is in the form of a real (not necessarily an
integer) number. Example: Income. A valid range can be provided for error checking.
C Character: The respondent types in a string of characters as a response. Example: Name.
I Multiple Integer
Ratings:
This question type is identical to integer-scale except that multiple questions (classified as
“options”) can appear on a single screen. Each question is answered and recorded separately.
I,R Multiple Real
Number Ratings:
This question type is identical to real-scale except that multiple questions (classified as
“options”) can appear on a single screen. Each question is answered and recorded separately.
*Legend: N = Nominal, O = Ordinal, I = Interval, R = Ratio, C = Alpha-Numeric Text Characters
Free Answer or Open-Ended Text Questions/Answers
The free answer (or open-ended text question) has no fixed alternatives to which the
answer must conform. The respondent answers in his or her own words and at the length he or
she chooses, subject of course to any limitations imposed by the questionnaire itself.
Interviewers are usually instructed to make a verbatim record of the answer.
While free-answer questions are usually shorter and less complex than multiple-choice
and dichotomous questions, they place greater demands on the ability of the respondents to
express themselves. As such, this form of question provides the opportunity for greater
ambiguity in interpreting answers. To illustrate, consider the following verbatim transcript of one
female respondent’s reply to the question:
227
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
What suggestions could you make for improving tomato juice?
“I really don’t know. I never thought much about it. I suppose that it would be nice if you could buy
it in bottles because the can turns black where you pour the juice out after it has been opened a day
or two. Bottles break, though.”
Did she have “no suggestion”, “suggest packaging in a glass container”, or “suggest that some
way be found to prevent the can from turning black around the opening”?
One way to overcome some of these problems, at least in personal and telephone surveys
is to have interviewers probe respondents for clarity (rather than additional information). One
practitioner has gone so far as to suggest that questionnaires should clearly instruct interviewers
to probe only once for additional information, and to continue to probe for clarity until the
interviewer understands a respondent’s reply.
Compared with other question forms (see Exhibit 7.1), we may tentatively conclude that
the free-answer question provides the lowest probability of the questions being ambiguous, but
the highest probability of the answers being ambiguous,.
Exhibit 7.1 Open-Ended Questions and Answers
The advantages of the open-ended format are considerable, but so are its disadvantages
(Sudman and Bradburn, 1982). In the hands of a good interviewer, the open format allows and
encourages respondents to give their opinions fully and with as much nuance as they are capable of.
It also allows respondents to make distinctions that are not usually possible with the fixed alternative
formats, and to express themselves in language that is comfortable for them and congenial to their
views. In many instances it produces vignettes of considerable richness and quotable material that
will enliven research reports.
The richness of the material can be a disadvantage if there is need to summarize the data into
simple response categories. Coding of free-response material is known as content analysis and is not
only time consuming and costly, but also introduces some amount of coding error.
Open-ended questions also take somewhat more time and psychological work to answer than
closed questions. They also require greater interviewer skill to recognize ambiguities of response and
to probe and draw respondents out, particularly those who are reticent and not highly verbal, to
make sure that they give answers that can be coded. Open-ended response formats may work better
with telephone interviews, where a close supervision of interview quality can be maintained, although
there is a tendency for shorter answers to be given on the telephone. No matter how well controlled
the interviewers may be, however, factors such as carelessness and verbal facility will generate
greater individual variance among respondents than would be the case with fixed alternative
response formats.
Dichotomous and Multiple-Choice Answers
The select k of n format is the workhorse of survey building, and provides the general
form for both dichotomous and multiple-choice answer types. Three general forms of questions
are frequently used:
Select Exactly 1 of n Answers:
When selecting k = 1/n, the type of answer scale is dependent on n, the number of
228
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
answers. A dichotomous question has two fixed answer alternatives of the type “Yes/No”, “In
favor/Not in favor”, “Use/Do not use”, and so on. The question quoted earlier, “Do you like the
taste of tomato juice?”is an example of a dichotomous question. Multiple-choice questions are
simply an extension of the dichotomous question that have more answer points and often take the
form of an ordered or interval measurement scale.
Traditional multiple–choice answers also are of the select 1 of n answer form, but have
more than two available answers. For example, an agreement scale could have three, five, or
seven available answers.
Three answers: Agree/Neutral/Disagree
Five answers: Strongly Agree/Agree/Neither/Disagree/Strongly Disagree
Seven answers: Very Strongly Agree/Strongly Agree/Agree/Neither Agree nor Disagree/ Disagree/
Strongly Disagree/Very Strongly Disagree
As with all select 1 of n answers, the specific text associated with the answer options is
variable and could measure many different constructs such as affect (liking), satisfaction, loyalty,
purchase likelihood, and so forth.
Select Exactly k of n Answers
When questions are developed that accept or require multiple responses within a set of
answers, the form “exactly k of n” or “as many as k of n” can be used. This general form asks the
respondent to indicate that several answers meet the requirements of the question. In this case,
the data collected would be of type categorical or even loosely ordered if presence or absence of
a characteristic is being measured (data is coded as 0 if not selected, and 1 if selected). This type
of question might be
Select as Many as k of n Answers
A variable number of answers may also be appropriate, particularly where long lists of
attributes or features are given. In these cases, the respondent is asked to select as many as k of
the n possible answers, where k can be any number from 2 to n. For example, in the previous
question, the respondent could select as many as three (one, two, or three) of 10 possible
answers. The question might be reworded to read something like . . .
Please identify which service activities are most likely to be outsourced in the next 12 months
(check all that apply).
Whitlark and Smith (2004) show an application of pick k of n data that asks respondents
to pick a small number of attributes that they feel best describe a brand from a list of 10, 20, or
even 30 attributes. Collecting the pick data is much faster than asking a respondent to rate brands
229
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
with respect to a long list of attributes. In an online survey environment, respondents can quickly
scan down columns or across screens and quickly complete the pick data task for a familiar
brand, thereby saving time and reducing respondent fatigue and dropout rates.
Having people describe a brand by picking attributes from a list is a quick and simple
way to assess brand performance and positioning. Whitlark and Smith (2004) show that when
respondents are asked to pick from one third to one half of the viewed items, the pick k of n data
can be superior to scaled data in terms of reliability and power to discriminate between attributes.
Rank-Order Questions/Answers
The next level of measurement rank-orders the answers and thereby increases the power
of the measurement scale over categorical measurement by including the characteristic of order
in the data. Whereas the categorical data associated with many dichotomous or multiple-choice
items does not permit us to say that one item is greater than another, rank-order data allows for
the analysis of differences. Rank-order questions use an answer format that requires the
respondent to assign a rank position to all items, or a subset of items in the answer list. The first,
second, and so forth up to the nth item would be ordered. Procedures for assigning position
numbers can be very versatile, resulting in different types of questions that can be asked. Typical
questions might include identifying preference rankings, or attribute associations from first to
last, most recent to least recent or relative position (most, next most, and so forth, until either a
set number of items is ordered or all items may be ordered).
When this type of question is administered online or using a CATI (Computer Aided
Telephone Interviewing) system, additional options for administration may include
randomization and acceptance/validation of ties in the ranking. Randomization of the answer list
order helps to control for presentation order bias. It is well established that in elections, being the
first in a ballot candidate list increases chances of receiving the voter’s election.
Tied rankings are another issue to be considered for rank-order questions. When ties are
permitted, several items may be evaluated as having the same rank. In general, this is not a good
idea because it weakens the data. However, if ties truly exist, then the ranking should reflect this.
Rank-order questions are generally a difficult type of question for respondents to answer,
especially if the number of items to be ranked goes beyond five or seven.
Constant Sum Questions/Answers
A constant sum question is a powerful question type that permits collection of
ratio data, meaning that the data is able to express the relative value or importance of the options
(option A is twice as important as option B). This type of question is used when you are
relatively sure of the answer set (i.e., reasons for purchase, or you want to evaluate a limited
number of reasons that you believe are important). The following example of a constant sum
question from Qualtrics, uses sliding scales to select a sum of 100 points:
230
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
ADVANCED MEASUREMENT AND SCALING CONCEPTS
Continuing our discussion of scales, we now focus on some of the more common scaling
techniques and models. We focus on broad concepts of attitude scaling—the study of scaling for
the measurement of managerial and consumer or buyer perception, preference, and motivation.
All attitude (and other psychological) measurement procedures are concerned with having
people—consumers, purchasing agents, marketing managers, or whomever—respond about
certain stimuli according to specified sets of instructions. The stimuli may be alternative products
or services, advertising copy themes, package designs, brand names, sales presentations, and so
on. The response may involve judging which copy theme is more pleasing than another, which
package design is more appealing than another, what mental images do new brand names evoke,
which adjectives best describe each salesperson, and so on.
Scaling procedures can be classified in terms of the measurement properties of the final
scale (nominal, ordinal, interval, or ratio), the task that the subject is asked to perform, or in still
other ways, such as whether the scale measures the subject, the stimuli, or both (Torgerson,
1958).
We begin with a discussion of various methods for collecting ordinal-scaled data (paired
comparisons, rankings, ratings, etc.) in terms of their mechanics and assumptions regarding their
scale properties. Then specific procedures for developing these actual scales are discussed.
Techniques such as Thurstone Case V scaling, semantic differential, the Likert summated scale,
and the Thurstone differential scale are illustrated. The chapter concludes with some issues and
limitations of scaling.
231
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Advanced Ordinal Measurement Methods
The variety of ordinal measurement methods includes a number of techniques:
• Paired comparisons
• Ranking procedures
• Ordered-category sorting
• Rating techniques
We discuss each of these data collection procedures in turn.
Paired Comparisons
As the name suggests, paired comparisons require the respondent to choose one of a pair
of stimuli that “has more of”, “dominates”, “precedes”, “wins over”, or “exceeds” the other with
respect to some designated property of interest. If, for example, six laundry detergent brands are
to be compared for “sudsiness”, a full set of paired comparisons would involve (n x n – 1)/2 = (6
× 5) / 2, or 15, paired comparisons (if order of presentation is not considered). Respondents are
asked which one of each pair has the most sudsiness.
A sample question format for paired comparisons is shown in Table 7.4. The order of
presentation of the pairs and which item of a pair is shown first are typically determined and/or
presented randomly. Consider the following hypothetical brand names (and numerical
categories): Arrow (1), Zip (2), Dept (3), Advance (4), Crown (5), and Mountain (6).
Table 7.4 Example of the Paired Comparisons Question
Original Data
Brand
Brand
Arrow Zip Advance Dept Crown Mountain
Arrow X 0 1 1 1 1
Zip 1 X 1 1 1 1
Advance 0 0 X 0 0 0
Dept 0 0 1 X 0 0
Crown 0 0 1 1 X 1
Mountain 0 0 1 1 0 X
232
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Permutated Rows and Columns
A cell value of 1 implies that a row brand exceeds the column brand, “0” otherwise.
Brand
Brand
Zip Arrow Crown Mountain Dept Advance Total
Zip X 1 1 1 1 1 5
Arrow 0 X 1 1 1 1 4
Crown 0 0 X 1 1 1 3
Mountain 0 0 0 X 1 1 2
Dept 0 0 0 0 X 1 1
Advance 0 0 0 0 0 X 0
Ranking Procedures
Ranking procedures require the respondent to order stimuli with respect to some
designated property of interest. For example, instead of using the paired-comparison technique,
respondent might have been asked to directly rank the detergents with respect to sudsiness.
Similarly, ranking can be used to determine key attributes for services.
In a survey conducted by Subaru of America, new Subaru car purchasers were asked
questions regarding the purchase and delivery processes. One question required ranking:
A variety of ordering methods may be used to order k items from a full set of n items.
These procedures, denoted by Coombs (1964) as “order k/n” (k out of n), expand the repertory of
ordering methods quite markedly. The various ordering methods may pre-specify the value of k
(“order the top three out of six brands with respect to sudsiness”) as illustrated by the Subaru
study, or allow k to be chosen by the respondent (“select those of the six brands that seem to
exhibit the most sudsiness, and rank them”).
When the groups can be ordered by some category, a procedure known as category
sorting can be used. For example, if it is desired that a respondent rank all items in a longer list
of items, the pick-group-rank procedure may be used. The respondent first sorts the items into a
number of ordered categories or piles (each of which has a relatively small, equal number of
items). Then, the request is made to rank the items within each pile. This task can be completed
in personal interviews or in online surveys using something like the Qualtrics pick-group-rank
question that facilitates the ordered ranking within the ordered-category sorting tasks.
233
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Ordered-Category Sorting
Pick-Group-Rank is one of a variety of data collection procedures that have as their
purpose the assignment of a set of stimuli to a set of ordered categories. For example, if 15
varieties of laundry detergents represented the stimulus set, the respondent might be asked to
complete the following task:
The pick-group-rank could be used with ordered categories to sort all of a large list of
items, where there is:
1. free versus forced assignment of names to grouping categories
2. free verses forced assignment of stimuli to grouping categories
3. the assumption of equal intervals between category boundaries versus the
weaker assumption of category boundaries that are merely ordered with
regard to the attribute of interest
In ordinal measurement methods one assumes only an ordering of category boundaries.
The assumption of equal intervals separating boundaries is part of the interval/ratio measurement
set of methods. Ordered-category sorting appears especially useful when the researcher is
dealing with a relatively large number of stimuli (over 15 or so) and it is believed that a subject’s
discrimination abilities do not justify a strict (no ties allowed) ranking of the stimulus objects.
Rating Techniques
Rating scales are ambiguous as to whether or not they meet the criterion of equal
intervals separating the category boundaries. In some cases, the scaled responses are considered
by the researcher to be only ordinal, while in other cases, the researcher treats them as interval-
234
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
or ratio-scaled. The flexibility of rating procedures makes them appropriate for either the ordinal
or interval/ratio measurement data collection methods (depending on the nature of the scale
values).
The rating task typically involves having a respondent place that which is being rated (a
person, object, or concept) along a continuum or in one of an ordered set of categories. Ratings
allow the respondent to designate a degree or an amount of a characteristic or attribute as a point
on a scale. The task of rating is one of the most popular and easily applied data collection
methods, and is used in a variety of scaling approaches, such as the semantic differential and the
Likert summated scale.
Rating scales can be either monadic or comparative. In monadic scaling, each object is
measured (rated) by itself, independently of any other objects being rated. In contrast,
comparative scaling objects are evaluated in comparison with other objects. For example, a
recent in-flight survey conducted by United Airlines asked the following questions:
The rating is monadic. United then asked respondents another question:
Ratings are used very widely because they are easier and faster to administer and yield
data that are amenable to being analyzed as if they are interval-scaled. But there is a risk of lack
of differentiation among the scores when the particular attributes are worded positively or are
positive constructs, such as values, and the respondents end-pile their ratings toward the positive
end of the scale. Such lack of differentiation may potentially reduce the variance of the items
being rated and reduce the ability to detect relationships with other variables.
McCarty and Shrum (2000) offer an alternative to simple rating. Respondents first picked
their most and least important values (or attributes or factors), and then rated them. The
remaining values were then rated. Their results indicate that, compared with a simple rating of
values, the most-least procedure reduces the level of end-piling and increases the differentiation
of values ratings, both in terms of dispersion and the number of different rating points used.
235
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Respondents may have trouble choosing a rating scale value on the high end, but not so at
the low end. One person’s rating of 9 or 10 may be equal in meaning to another’s 7 or 8. Semon
(1999) suggests that one way to find the real difference in perception or attitude is to ask each
respondent three questions at the start of an interview:
1. On this scale, how do you rate the brand you now use or that you know best?
2. How do you rate the best brand you know about?
3. What rating represents the minimum acceptable level?
Questions such as these are often asked in product and brand studies, to interpret ratings
and provide anchor points for a respondent’s ratings. A respondent’s actual ratings can be
translated into responses relative to one or more of these anchors to produce real-meaning
relative ratings that can be reliably aggregated and analyzed without depending upon
assumptions that may be questionable.
Rating methods can take several forms, numerical, graphic and verbal. Often two or more
of these formats appear together, as illustrated in Figures 7.1 and 7.2. Many other types of rating
methods are in use (Haley & Case, 1979).
Figure 7.1 Examples of Rating Scales Used in Marketing Research
a.
b.
c.
d.
e.
In many instances where rating scales are used, the researcher assumes not only that the
items are capable of being ranked, but also that the descriptive levels of progress are in equalinterval
steps psychologically. That is, the numerical correspondences shown in Panels (c) and
(e) of Figure 7.1 may be treated—sometimes erroneously—as interval- or ratio-scaled data. Even
in cases represented by Panels (a), (b), and (d), it is not unusual to find that the researcher assigns
successive integer values to the various category descriptions and subsequently works with the
data as though the responses were interval-scaled.
Treating rating scales as interval or ratio measurements is a practice that is well documented and
widespread. Research shows that there is little error in treating the data as being of a higher level
of measurement than it is. Research evidence supports this practice, in that often when ordinal
data are treated as interval and parametric analysis are used, the conclusions reached are the
same as when the data are treated as ordinal and tested using non-parametric analyses.
236
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 7.2 A Rating Thermometer
Constructing a Behaviorally Anchored Rating Scale
One type of itemized rating scale that has merit in cases where leniency error (lack of
discrimination) may be troublesome is the behaviorally-anchored rating scale, or BARS (see
Figure 7.3). This scale uses behavioral incidents to define each position on the rating scale rather
than verbal, graphic, or numeric labels. Thus, providing specific behavioral anchors can reduce
leniency errors and increase discrimination. Developing scales such as these requires a great
amount of testing and refinement to find the right anchors for the situation under examination.
Figure 7.3 Behaviorally Anchored Rating Scale (BARS)
The basic process of developing a behaviorally anchored rating scale consists of four steps:
1. Construct definition—the construct being measured must be explicitly defined and the key
dimensions identified
237
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
2. Item generation—statements must be generated describing actual behaviors that would
illustrate specific levels of the construct for each dimension identified
3. Item testing—to unambiguously fit behavioral statements to dimensions
4. Scale construction—lay out the scale with behavioral statements as anchors
In following this process, sets of judges are used. It should be clear that developing
BARS is a time-consuming and costly task. Thus, they should be reserved for those applied
settings where they can minimize the errors they are designed to curtail, especially leniency
error. As an example, families with elderly members were surveyed to determine their need for
in home health-care services. A BARS was used for one critical measure of how well elderly
members of the household were able to perform everyday living activities:
Table 7.5 identifies nine questions that must be answered when a scale is constructed.
Table 7.5 Issues in Constructing a Rating Scale
1.Should negative numbers be used?
2.How many categories should be included?
3.Related to the number of categories is: Should there be an odd number or an even number?
That is, should a neutral alternative be provided?
4.Should the scale be balanced or unbalanced?
5.Is it desirable to not force a substantive response by giving an opportunity to indicate “don’t
know,” “no opinion,” or something similar?
6.What does one do about halo effects—that is, the tendency of raters to ascribe favorable
property levels to all attributes of a stimulus object if they happen to like a particular object in
general?
7.How does one examine raters’ biases—for example, the tendency to use extreme values or,
perhaps, only the middle range of the response scale, or to overestimate the desirable
features of the things they like (i.e., the generosity error)?
8.How should descriptive adjectives for rating categories be selected?
9.How anchoring phrases for the scale’s origin should be chosen?
Some research on these questions has been conducted showing that errors are not made
when a neutral option is provided. Our suggestion is that it always be included, unless the
researcher has a compelling reason to not do so (e.g., the problem situation/sample mix is such
that each sample member can be expected to have a non-neutral attitude). Expected voting in a
survey of voters is an example.
238
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Question 4 deals with the interesting issue of Balance, referring to having an equal
number of negative response alternatives as positive ones. When using importance scales for
attributes, the alternatives provided may be “very important”, “important”, “neither important
nor unimportant”, “unimportant”, and “very unimportant”, or additional categories may be
included. Thomas Semon (2001) has questioned the use of balance (or symmetry, as he calls it)
in importance scales. He argues that importance is not a bipolar concept. Importance ranges from
some positive amount to none, not a negative amount. Although this appears to have conceptual
appeal, researchers continue to use successfully use importance scales from some mid-point—
specified or implied. There would seem to be three keys to successful importance scale use:
1. Isolating any findings of unimportance
2. Recognizing that importance is ordinally scaled
3. Accurately interpreting the relative nature of importance findings
Answers to questions such as these will vary by the researcher’s approach, and by the
problem being studied. The effects of research design on reliability and validity of rating scales
are discussed in two excellent review papers (Churchill and Peter, 1984; Peter and Churchill,
1986).
Exhibit 7.4 Measuring Preferences of Young Children Calls for Creativity
The children’s market is a multi-billion dollar market in direct purchasing power and an even greater
market in purchasing influence. Among the areas of most concern are better scaling techniques for
measuring children’s product preferences. Widely used approaches for assessing children’s
preferences are itemized rating scales using a series of stars (a scale from 1 to 5 stars) or a series of
facial expressions (a scale anchored at one end with a happy face and at the other end with a sad
face), as illustrated below:
(a) Smiling Faces Scale
(b) Star Scale
    
Children are asked to indicate how much they like a product, or how much they like a particular feature
of a product, by pointing to one of the visual anchors on the scale.
Although these scales have done well in varied research applications, there is often a problem with
leniency that emerges, particularly when used with young children under the age of eight. This error
emerges when young children consistently use the extreme positions (usually on the positive side) with
relatively little use of intermediate scale positions. If this is done for all products tested, the overall
sensitivity of existing (traditional) rating scales is lowered, resulting in inconclusive findings about
children’s preferences.
In summary, rating methods—depending on the assumptions of the researcher—can be
considered to lead to ordinal-, interval-, or even ratio-scaled responses. The latter two scales are
taken up next. We shall see that rating methods figure prominently in the development of
quantitative-judgment scales.
239
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Interval/Ratio Procedures
Direct-judgment estimates, fractionation, constant sum and rating methods assume more
than ordinal properties about respondents’ judgments) are all variants of interval/ratio procedures
or metric measurement methods.
Direct-Judgment Methods
In direct-judgment methods, the respondent is asked to give a numerical rating to each
stimulus with respect to some designated attribute. In the case of continuous rating scales, the
respondent is free to choose his or her own number along some line that represents his or her
judgment about the magnitude of the stimulus relative to some reference points (Figure 7.4).
These continuous scales work effectively in a semantic differential context (to be
discussed later in this chapter) (Albaum, Best, & Hawkins, 1981), and appear to be insensitive to
fluctuations in the length of the line used (Hubbard, Little, & Allen, 1989).
The limited-response category sub case, illustrated by Panel (b) in Figure 7.4, is nothing
more than a straight rating procedure, with the important addition that the ratings are now treated
as either interval- or ratio-scaled data (depending on the application).
Figure 7.2 Sample Interval-Ratio Scales
(a) Unlimited-Response Categories
Don’t
Like
At All
Like
Very
Much
(b) Limited-Response Categories
(c) Fractionation
Result:
The Relative degree of sudsiness compared to Brand A
Brand Response
B 0.75
C 0.80
D 2.40
E 0.50
(d) Constant Sum
Fractionation
Fractionation is a procedure in which the respondent is given two stimuli at a time (e.g., a
standard laundry detergent and a test brand) and asked to give some numerical estimate of the
ratio between them, with respect to some attribute, such as sudsiness. The respondent may
answer that the test brand, in his or her judgment, is three-fourths as sudsy as the standard. After
this is done, a new test brand is compared with the same standard, and so on, until all test items
are judged. Panel (c) in Figure 7.2 illustrates this procedure.
240
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In other cases, the test item can be more or less continuously varied by the respondent.
For example, in an actual test of the attribute is sweetness of lemonade, the respondent may be
asked to add more sweetener until the test item is “twice as sweet” as the standard.
Constant Sum
Constant-sum methods have become quite popular in marketing research, primarily
because of their simplicity and ease of instructions. In constant-sum methods the respondent is
given some number of points—typically 10 or 100—and asked to distribute them over some set
of stimuli or attribute alternatives in a way that reflects their relative importance or magnitude
(Figure 7.2d). Constant sum forces the respondent to make comparative evaluations across the
stimuli, and effectively standardizes each scale across persons, since all scores must add to the
same constant. Generally, it is assumed that a subjective ratio scale is obtained by this method.
In summary, unlike ordinal measurement methods, the major assumption underlying
ratio/interval measurement methods) is that a unit of measurement can be constructed directly
from respondents’ estimates about scale values associated with a set of stimuli. The respondent’s
report is taken at face value and any variation in repeated estimates (over test occasions within
respondent or over respondents) is treated as error; repeated estimates are usually averaged over
persons and/or occasions. The problems associated with Interval-Ratio Scaling methods include
the following:
1. Respondents’ subjective scale units may differ across each other, across testing
occasions, or both.
2. Respondents’ subjective origins (zero points) may differ across each other, across
occasions, or both.
3. Unit and origin may shift over stimulus items within a single occasion.
4. Subjective distance between stimuli may not equal one’s perception of the distance on
the scale.
These problems should not be treated lightly, but considered in the design of the question
and scale points.
Most ratings measurement methods have the virtue of being easy to apply. Moreover,
little additional work beyond averaging is required to obtain the unit of measurement directly.
Indeed, if a unique origin can be established (e.g., a zero level of the property), then the
researcher obtains both an absolute origin and a measurement unit. As such, a subjective ratio
scale is obtained.
241
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Techniques for Scaling Stimuli
Mission Impossible?
You are making good progress in your first internship and have just been asked by the
head of research to present a summary of how your brand is perceived relative to the four
major competitor brands… for tomorrow’s meeting.
You find the results of a survey conducted just before you came on board. The survey
used a paired comparison task to rate preference (10 pairs: 1 vs. 2, 1 vs. 3, 1 vs. 4, etc.), and
another simple rank order question for the 5 brands. But your manager specifically asked
that preference be displayed on a single continuous (interval or ratio scale).
You’re not yet in full blown panic… you average the preference rankings and put them
into a symmetric square matrix (Table 7-7)…. but how do you convert a data matrix into a
single dimension interval preference scale?
The answer to this problem is that ranking methods may undergo a further transformation (via an
intervening scaling model) to produce set of scale values that are interval-scaled. One such
transformation, Thurstone’s Case V method, is capable of transforming ordinal data obtained
from ranking methods. It should be noted that technically speaking, the raw data obtained from
ratings methods also requires an intervening model in order to prepare an interval scaled
summary measure. However, in this case the model may be no more elaborate than averaging the
raw data across respondents and/or response occasions.
Osgood’s semantic differential is an illustration of a procedure for dealing with raw data
obtained from interval-ratio scale ratings methods. We consider each of these techniques in turn.
Case V Scaling
Thurstone’s Case V Scaling model, based on his Law of Comparative Judgment, permits
the construction of a unidimensional interval scale using responses from ordinal measurement
methods, such as paired comparisons (Thurstone, 1959). This model can also be used to scale
ranked data or ordered-category sorts. Several sub cases of Thurstone’s model have been
developed. We shall first describe the general case and then concentrate on Case V, a special
version particularly amenable to application in marketing situations.
Essentially, Thurstone’s procedure involves deriving an interval scale from comparative
judgments of the type “A is fancier than B”, “A is more prestigious than B”, “A is preferred to
B”, and so on. Scale values may be estimated from data in which one individual makes many
repeated judgments on each pair of a set of stimuli or from data obtained from a group of
individuals with few or no replications per person.
The example should make the Case V procedure easier to follow. Assume that the survey
you found had asked 100 homemakers to compare five brands of “fortified juice” with respect to
“overall preference of flavor”. The homemakers sipped a sample of each brand paired with a
sample of every other brand (a total of ten pairs) from paper cups that were marked merely with
identifying numbers. Table 7.7 shows the empirically observed proportion for each comparison.
From this table we see that 69 percent of the respondents preferred Juice C to Juice A and
the remainder, 31 percent preferred Juice A to Juice C (if we arbitrarily let column dominate
row). It is customary to set self-comparisons (the main-diagonal entries of Table 10.4) to 0.5; this
has no effect on the resulting scale values (Edwards, 1957). From the data of this table we next
242
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
prepare Table 7.8, which summarizes the Z-values appropriate for each proportion. These Zvalues
were obtained from Table A.1 in Appendix A at the end of this book. If the proportion is
less than 0.5, the Z-value carries a negative sign; if the proportion is greater than 0.5, the Z-value
carries a positive sign. The Z-values are standard unit variates associated with a given proportion
of total area under the normal curve. The Thurstonian model assumes normally distributed scale
differences in mean = 0 and standard deviation = 1.0.
For example, from Table 7.7 we note that the proportion of respondents preferring Juice
B over Juice A is 0.82. We wish to know the Z-value appropriate thereto. This value (labeled Z
in the standard unit normal table of Table Appendix A.1) is 0.92. That is, 82 percent of the total
area under the normal curve is between Z = – ∞ and Z =0.92. All remaining entries in Table 10.5
are obtained in a similar manner, a minus sign being prefixed to the Z-value when the proportion
is less than 0.5.
Column totals are next found for the entries in Table 7.8. Scale values are obtained from
the column sums by taking a simple average of each column’s Z-values. For example, from
Table 10.5, we note that the sum of the Zs for the first column (Juice A) is –0.36. The average Z
for column A is simply:
072.0
5
36.0
−=−=Ζ
This scale value expresses Juice A as a deviation from the mean of all five scale values.
The mean of the five values, as computed from the full row of Zs, will always be zero under this
procedure. Similarly, we find the average Z-value for each of the remaining four columns of
Table 7.8.
Table 7.7 Observed Proportions Preferring Brand X to Brand Y
Proportion
Preferring Row
Brand to
Column Brand
Proportion Preferring Column Brand to Row Brand
A B C D E
A 0.50 0.82 0.69 0.25 0.35
B 0.18 0.50 0.27 0.07 0.15
C 0.31 0.73 0.50 0.16 0.25
D 0.75 0.93 0.84 0.50 0.59
E 0.65 0.85 0.75 0.41 0.50
Table 7.8 Z-Values Related to Preference Proportions in Table 7.7
Brand
Brand A B C D E
A 0 0.92 0.50 -0.67 -0.39
B -0.92 0 -0.61 -1.48 -1.04
C -0.50 0.61 0 -0.99 -0.67
D 0.67 1.48 0.99 0 0.23
E 0.39 1.04 0.67 -0.23 0
Total -0.36 4.05 1.55 -3.37 -1.87
Mean (Z) -0.072 0.81 0.31 -0.674 -0.374
R 0.602 1.484 0.984 0 0.3
Next, since the zero point of an interval scale is arbitrary, we can transform the minimum
scale so that it becomes zero. Since Juice D has the lowest value (RD = ZD = –0.674), we force it
to become the reference point (or origin) of zero by adding .674. We then simply add 0.674 to
each of the other Z-values to obtain the Case V scale values of the other four brands. These are
denoted by R* and appear in the last row of Table 7.8.
The scale values of Juices A through E indicate the preference ordering B > C > A > E >
D. Moreover, assuming that an interval scale exists, we can say, for example, that the difference
in “goodness of flavor” between Juices B and A is 2.3 times the difference in “goodness of
243
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
flavor” between Juices C and A, since
B – A = 2.3 (C – A)
1.484 – 0.602 = 2.3(0.984 – 0.602)
0.882 = 2.3(0.382) (within rounding error).
The test of this model is how well scale values can be used to work backward—that is, to
predict the original proportions. The Case V model appears to fit the data in the example quite
well. For any specific brand, the highest mean absolute proportion discrepancy is 0.025 (Juice
A). Moreover, the overall mean absolute discrepancy is only .02 (rounded). Even the simplest
version (Case V) of the Thurstonian model leads to fairly accurate predictions. The R* scale
values of the Case V model preserve the original rank ordering of the original proportions data.
The Semantic Differential
The semantic differential (Osgood, Suci, & Tannenbaum, 1957) is a ratings procedure
that results in (assumed interval) scales that are often further analyzed by such techniques as
factor analysis (see Chapter 12). Unlike the Case V model, the semantic differential provides no
way to test the adequacy of the scaling model itself. It is simply assumed that the raw data are
interval-scaled; the intent of the semantic differential is to obtain these raw data for later
processing by various multivariate models.
The semantic differential procedure permits the researcher to measure both the direction
and the intensity of respondents’ attitudes (i.e., measure psychological meaning) toward such
concepts as corporate image, advertising image, brand or service image, and country image.
As shown in Figure 7.3, the respondent may be given a set of pairs of antonyms, the
extremes of each pair being separated by seven intervals that are assumed to be equal. For each
pair of bi-polar adjectives (e.g., powerful/weak), the respondent is asked to judge the concept
along the seven-point scale with implicit descriptive phrases.
Figure 7.3 Corporate Profile Obtained by the Semantic Differential
In practice, however, profiles would be built up for a large sample of respondents, with
many more bipolar adjectives being used than given here.
By assigning a set of integer values, such as +3, +2, +1, 0, –1, –2, –3, to the seven
gradations of each bipolar scale in Figure 7.4, the responses can be quantified under the
assumption of equal-appearing intervals. These scale values, in turn, can be averaged across
respondents to develop semantic differential profiles. For example, Figure 10.5 shows a chart
comparing evaluations of Companies X and Y. The average score for the respondents show that
the Company X is perceived as very weak, unreliable, old-fashioned, and careless, but rather
warm. Company Y is perceived as powerful, reliable, and careful, but rather cold as well; it is
almost neutral with respect to the modern/old-fashioned scale.
244
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 7.4 Average-Respondent Profile Comparisons of Companies X and Y
Company X = Company Y = ----------
In marketing research applications, the semantic differential often uses bipolar
descriptive phrases rather than simple adjectives, or a combination of both types. These scales
are developed for particular context areas, so the scales have more meaning to respondents, thus
leading usually to a high degree of reliability.
The same issues of scale construction presented in Table 7.5 apply to the semantic
differential. In addition, the researcher must select an overall format for presentation of the
scales. As Figure 7.5 illustrates, there are many formatting variations to semantic differential
scaling, such as scales that include the numbers 1-7 at the scale points, or numerical comparative
scale where respondents make their judgments for KMart, Wal-Mart, and Sears on one attribute
before moving to the next one (Golden, Brockett, Albaum, & Zatarain, 1992).
Figure 7.5 Alternate Formats for the Semantic Differential
245
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The number and type of stimuli to evaluate and the method of administration (personal
interview, mail, telephone and email) should determine at least which format the researcher
should use. Comparative studies of data quality (including reliability) seem to indicate that the
choice of a format may be appropriately made on the basis of ease of subject understanding, ease
of coding and interpretation for the researcher, ease of production and display, and cost. If a
large number of stimuli are to be evaluated, this would tend to favor use of the graphic
positioning or numerical comparative scales.
A Concluding Remark
The semantic differential technique is appropriate for use in a variety of applications:
• Comparing corporate images, both among suppliers of particular products and against an ideal
image of what respondents think a company should be
• Comparing brands and services of competing suppliers
• Determining the attitudinal characteristics of purchasers of particular product classes or brands
within a product class, including perceptions of the country of origin for imported products
• Analyzing the effectiveness of advertising and other promotional stimuli toward changing
attitudes
The comparatively widespread use of the semantic differential by marketing researchers
suggests that this method provides a convenient and reasonably reliable way for scaling stimuli
(scaling images of brands, corporations, services, etc.), and developing profiles of
consumer/buyer attitudes on a wide variety of topics.
Techniques for Scaling Respondents
In contrast to the approaches for scaling stimuli just discussed, researchers also have available
techniques whose primary purpose is to scale respondents along some attitude continuum of
interest. Two better-known procedures for doing this are the summated scale, and the Q-sort
technique. Each of these is described in turn.
The Summated Scale
The summated scale was originally proposed by Rensis Likert (pronounced “lick-ert”), a
psychologist (Likert, 1967; Kerlinger, 1973). To illustrate, assume that the researcher wishes to
scale some characteristic, such as the public’s attitude toward travel and vacations.
To illustrate the Likert scale, a set of seven statements regarding travel and vacations
used in a study by a travel company are shown in Figure 7.6. Each of the seven test items has
246
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
been classified as “favorable” (items 1, 3, and 7) or “unfavorable” (items 2, 4, 5, and 6). Each
subject would be asked to indicate their agreement with the statement. The responses are scored
+2 for “strongly agree”, +1 for “agree”, 0 for “neither”, –1 for “disagree”, and –2 for “strongly
disagree”. Since, we reverse scaled items 2, 4, 5, and 6 on “unfavorable” statements, we would
reverse the order of the scale values so as to maintain a consistent direction ( +2 would stand for
“strongly disagree,” and so on).
Suppose that a subject evaluated the seven items as follows such that the respondent
would receive a total score of:
+ 2 + 1 + 1 + 2 + 1 + 2 + 2 = 11
Suppose that another respondent responded to the seven items by marking (1) strongly
disagree, (2) neither, (3) disagree, (4) strongly agree, (5) strongly disagree, (6) strongly agree,
and (7) neither. This person’s score would be:
– 2 + 0 – 1 – 2 – 2 – 2 + 0 = –9
This listing indicates that the second respondent would be ranked “lower” than the first—
that is, as having a less-favorable attitude regarding travel and vacations. However, as indicated
earlier, a given total score may have different meanings.
Figure 7.6 A Direction-Intensity Scale for Measuring Attitudes Toward Travel and Vacations
In applying the Likert summated-scale technique, the steps shown in Table 7.9 are
typically carried out.
Table 7.9 Steps in Constructing a Likert Summated Scale
1. The researcher assembles a large number (e.g., 75 to 100) of statements concerning the public’s
sentiments toward travel and vacations.
2. Each of the test items is classified by the researcher as generally “favorable” or “unfavorable” to
the attitude under study. No attempt is made to scale the items; however, a pretest is
conducted that involves the full set of statements and a limited sample of respondents. Ideally,
the initial classification should be checked across several judges.
3. In the pretest the respondent indicates approval (or not) with every item, checking one of the
following direction-intensity descriptors:
247
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
4. Each response is given a numerical weight (e.g., +2, +1, 0, −1, −2, or +1 to +5).
5. The individual’s total-attitude score is represented by the algebraic summation of weights
associated with the items checked. In the scoring process, weights are assigned such that the
direction of attitude— favorable to unfavorable—is consistent over items. For example, if a + 2
were assigned to “strongly approve/agree” for favorable items, a + 2 should be assigned to
“strongly disapprove/disagree” for unfavorable items.
6. On the basis of the results of the pretest, the analyst selects only those items that appear to
discriminate well between high and low total scorers. This may be done by first finding the
highest and lowest quartiles of subjects on the basis of total score. Then, the mean differences
on each specific item are compared between these high and low groups (excluding the middle
50 percent of subjects).
7. The 20 to 25 items finally selected are those that have discriminated “best” (i.e., exhibited the
greatest differences in mean values) between high versus low total scorers in the pretest.
8. Steps 3 through 5 are then repeated in the main study.
When analysis is completed, many researchers assume only ordinal properties regarding
the placement of respondents along the continuum. Nonetheless, two respondents could have the
same total score even though their response patterns to individual items were quite different.
That is, the single (summated) score ignores the details of just which items were agreed with and
which ones were not. Moreover, the total score is sensitive to how the respondent reacts to the
descriptive intensity scale.
Often, a researcher will reverse the polarity of some items in the set (i.e., word items
negatively) as a way to overcome the possibility of acquiescence bias (being overly agreeable).
Having positively and negatively worded statements hopefully forces respondents with strong
positive or negative attitudes to read carefully and use both ends of a scale. A researcher should
reverse the polarity of some items, but may need to adjust the scoring, as appropriate. That is, a
“strongly agree” response to a positive statement and a “strongly disagree” to a negative
statement should be scored the same, and so forth.
Another approach to wording the summated scale adapts the statements into a set of nondirectional
questions, thereby alleviating the problems associated with mixed-wording scales
(Wong, Rindfleisch, & Burroughs, 2003). As an illustration, a non-directional format for one
item would be:
“How much pleasure do you get from traveling? [Very little…A great deal]”
In contrast, the normal Likert format for this item is:
“Traveling gives me a lot of pleasure [strongly agree, agree, neither agree nor
disagree, disagree, strongly disagree]”
Some final comments are in order. When using this format, Likert (1967) stated that a
key criterion for statement preparation and selection should be that all statements be expressions
of desired behavior and not statements of fact. Because two persons with decidedly different
attitudes may agree on fact, it is recognized that direction is the only meaningful measure
248
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
obtained when using statements of facts.
The second concern is that the traditional presentation of a Likert scale is one-stage, with
both intensity and direction combined. As stated earlier, this may lead to reluctance on the part
of respondents to either give extreme scores or use the extreme position on an individual scale
item (central tendency error). To compensate for this situation the longer two-stage format,
whereby direction and intensity are separate evaluations, can be used.
The Q-Sort Technique
The Q-sort technique has aspects in common with the summated scale. Very simply, the
task required of a respondent is to sort a number of statements (on individual cards or a pick and
group Qualtrics question) into a predetermined number of categories (usually 11) with a
specified number having to be placed in each category.
In illustrating the Q-sort technique, assume that four respondents evaluate the test items
dealing with travel and vacations. For purposes of illustration, only three groups will be used.
The respondents are asked to sort items into:
Group 1
LEAST AGREED WITH
Group 2
NEUTRAL ABOUT
Group 3
MOST AGREED WITH
-1 0 +1
Suppose that the responses toward seven items by the four respondents, A, B, C, and D,
result in the following scale values:
Respondent
Item A B C D
1 +1 +1 –1 –1
2 0 0 0 0
3 +1 0 0 –1
4 -1 –1 +1 +1
5 0 0 0 0
6 –1 –1 +1 +1
7 0 +1 –1 –1
As can be noted, the respondent pairs A & B and C & D seem to be the “most alike” of
the six distinct pairs that could be considered. We could, of course, actually correlate each
respondent’s scores with every other respondent and, similar to semantic differential
applications, and then conduct factor or cluster analyses (see Chapter 14) to group the
respondents or items. Typically, these additional steps are undertaken in Q-sort studies.
Multi-Item Scales
Each of the types of scales discussed in this chapter can be used either alone or part of a
multi-item scale used to measure some construct. A multi-item scale consists of a number of
closely related individual rating scales whose responses are combined into a single index,
composite score, or value (Peterson, 2000). Often the scores are summated to arrive at a total
score. Multi-item scales are used when measuring complex psychological constructs that are not
easily defined by just one rating scale or captured by just one question.
The major steps in constructing a multi-item scale. The first, and perhaps most critical,
step is to clearly and precisely define the construct of interest. A scale cannot be developed until
249
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
it is clear just what the scale is intended to measure. This is followed by design and evaluation of
the scale. A pool of items is developed and then subject to analysis to arrive at the initial scale.
Along the way, a pilot study is conducted to further refine the scale and move toward the final
version. Validation studies are conducted to arrive at the final scale. Of concern is construct
validation, in which an assessment is made that the scale measures what it is supposed to
measure. At the same time that validity data are collected, normative data can also be collected.
Norms describe the distributional characteristics of a given population on the scale. Individual
scores on the scale then can be interpreted in relation to the distribution of scores in the
population (Spector, 1992, p. 9).
A good multi-item scale is both reliable and valid. Reliability is assessed by the scale’s
stability (test-retest reliability) and internal consistency reliability (coefficient alpha). According
to Spector (1992), there are several other characteristics of a good multi-item scale:
• The items should be clear, well-written, and contain a single idea.
• The scale must be appropriate to the population of people who use it, such as having an
appropriate reading level.
• The items should be kept short and the language simple.
• Consider possible biasing factors and sensitive items.
Table 7.11 gives an example of a multi-item scale developed to measure consumer
ethnocentrism within a nation, the CETSCALE (Shimp & Sharma, 1987). This scale is formatted
as a 7 point agreement question in the Likert scale format. A Cetscale score for an individual
respondent is obtained as a sum of item ratings, and ranges from 17 to 119 with higher numbers
indicating greater consumer ethnocentrism. A compilation of multi-item scales frequently used in
consumer behavior and marketing research is provided by Bearden and Netemeyer (1999).
Table 7.11 Example of a Multi-Item Scale: Consumer Ethnocentrism (CETSCALE)
1. American people should always buy American-made products instead of imports.
2. Only those products that are unavailable in the United States should be imported.
3. Buy American-made products. Keep America working.
4. American products first, last and foremost.
5. Purchasing foreign-made products is un-American.
6. It is not right to purchase foreign products.
7. A real American should always buy American-made products.
8. We should purchase products in America instead of letting other countries get rich off us.
9. It is always best to purchase American products.
10. There should be very little trading or purchasing of goods from other countries unless out of
necessity.
11. Americans should not buy foreign products, because this hurts American business and
causes unemployment.
12. Curbs should be put on all imports.
13. It may cost me in the long run, but I prefer to support American products.
14. Foreigners should not be allowed to put their products on our markets.
15. Foreign products should be taxed heavily to reduce their entry into the United States.
16. We should buy from foreign countries only those products that we cannot obtain within our
own country.
17. American consumers who purchase products made in other countries are responsible for
putting their fellow Americans out of work.
NOTE: Items composing the 10-item reduced version are items 2, 4 through 8, 11, 13, 16, and 17.
250
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Predictions from attitude scales, preference ratings, and the like still need to be
transformed into measures (sales, market share) of more direct value to the marketer. We still do
not know, in many cases, how to effectively translate verbalized product ratings, attitudes about
corporations, and so on into the behavioral and financial measures required to evaluate the
effectiveness of alternative marketing actions.
The Art of Writing Good Questions
Strength of Question Wording
The wording of questions is a critical consideration when obtaining information from
respondents. Consider that the following question differing only in the use of the words
“should”, “could” and “might” was shown to three matched samples of respondents (Payne,
1951, pp. 8–9).
Do you think anything should be done to make it easier for people to pay doctor or
hospital bills? (82 percent replied “Yes”.)
For the sample shown the sentence with the word “could”, 77 percent replied “Yes”, and
with “might”, 63 percent replied “Yes”. These three words are sometimes used as synonyms, and
yet at the extreme, responses are 19 percentage points apart.
As another example, Rasinski (1989) posed a question where labels for the topic issue
was changed:
Are we spending too much, too little, or about the right amount on welfare?
In this case, 23.1 percent of respondents replied “too little”, but when the label was
changed to assistance to the poor, 62.8 percent replied “too little”. Questions portraying a more
descriptive and positive position may show a large difference in the evaluation score.
Reducing Question Ambiguity
Reducing ambiguity and bias is critical in both the respondent’s understanding and proper
consideration of the question and in the researcher’s understanding of the answer’s meaning. In
this section we discuss issues of question structure and form that can greatly influence and
improve the quality of your questionnaire.
The Qualtrics “Survey University” and survey and question libraries provide many
suggestions and helpful examples for writing unambiguous questions. Writing questions is an
art, which like all arts requires a great amount of work, practice, and help from others. In Exhibit
7.4 we provide an overview of the common pitfalls we often see in "bad questionnaires" that lead
to various forms of ambiguity.
251
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 7.4 Key Considerations for Reducing Question Ambiguity
1. Strength of Question Wording The wording of questions is a critical consideration when
obtaining information from respondents. One study “should”, “could” and “might” was shown
to three matched samples of respondents (Payne, 1951, pp. 8–9). Do you think anything should
be done to make it easier for people to pay doctor or hospital bills? (82 percent replied “Yes”.)
For the sample shown the sentence with the word “could”, 77 percent replied “Yes”, and with
“might”, 63 percent replied “Yes”. These three words are sometimes used as synonyms, and yet
at the extreme, responses are 19 percentage points apart. Questions portraying a more
descriptive and positive position may show a large difference in the evaluation score.
2. Avoid loaded or leading words or questions. Slight wording changes can produce great
differences in results. Could, Should, Might all sound almost the same, but may produce a 20%
difference in agreement to a question (The supreme court could.. should.. might.. have forced
the breakup of Microsoft Corporation). Strong words that represent control or action, such as
prohibit produces similar results (Do you believe that congress should prohibit insurance
companies from raising rates?) Sometimes wording is just biased: You wouldn't want to go to
Rudolpho's Restaurant for the company's annual party would you?
3. Framing effects. Information framing effects reflect the difference in response to objectively
equivalent information depending upon the manner in which the information is labeled or
framed. Levin, Schneider, and Gaeth (1998) and Levin et al. (2001) identify three distinct types
of framing effects:
• Attribute framing effects occur when evaluations of an object or product are more favorable
when a key attribute is framed in positive rather than negative terms.
• Goal framing effects occur when a persuasive message has different appeal depending on
whether it stresses the positive consequences of performing an act to achieve a particular
goal or the negative consequences of not performing the act.
• Risky choice framing effects occur when willingness to take a risk depends upon whether
potential outcomes are positively framed (in terms of success rate) or negatively framed (in
terms of failure rate).
Which type of potential framing effects should be of concern to the research designer depends
upon the nature of the information being sought in a questionnaire. At the simplest level, if
intended purchase behavior of ground beef was being sought, the question could be framed as
“80 percent lean” or “20 percent fat.” This is an example of attribute framing. It should be
obvious that this is potentially a pervasive effect in question design, and is something that needs
to be addressed whenever it arises. More detailed discussion of these effects is given by
Hogarth (1982).
4. Misplaced questions. Questions placed out of order or out of context should be avoided. In
general, a funnel approach is advised. Broad and general questions at the beginning of the
questionnaire as a warm-up (What kind of restaurants do you most often go to?). Then more
specific questions, followed by more general easy to answer questions (like demographics) at
the end of the questionnaire.
5. Mutually non-exclusive response categories. Multiple choice response categories should be
mutually exclusive so that clear choices can be made. Non-exclusive answers frustrate the
respondent and make interpretation difficult at best.
6. Nonspecific questions. Do you like orange juice? This is very unclear...do I like what? Taste,
texture, nutritional content, Vitamin C, the current price, concentrate, fresh squeezed? Be
specific in what you want to know about. Do you watch TV regularly? (what is regularly?).
7. Confusing or unfamiliar words. Asking about caloric content, acrylamide, phytosterols, and
other industry specific jargon and acronyms are confusing. Make sure your audience
understands your language level, terminology and above all, what you are asking.
252
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
8. Non-directed questions give respondents excessive latitude.
What suggestions do you have for improving tomato juice? The question is about taste, but the
respondent may offer suggestions about texture, the type of can or bottle, mixing juices, or
something related to use as a mixer or in recipes.
9. Forcing answers. Respondents may not want, or may not be able to provide the information
requested. Privacy is an important issue to most people. Questions about income, occupation,
finances, family life, personal hygiene and beliefs (personal, political, religious) can be too
intrusive and rejected by the respondent.
10. Non-exhaustive listings. Do you have all of the options covered? If you are unsure, conduct a
pretest using the "Other (please specify) __________" option. Then revise the question making
sure that you cover at least 90% of the respondent answers.
11. Unbalanced listings. Unbalanced scales may be appropriate for some situations and biased in
others. When measuring alcohol consumption patterns, one study used a quantity scale that
made the heavy drinker appear in the middle of the scale with the polar ends reflecting no
consumption and an impossible amount to consume. However, we expect all hospitals to offer
good care and may use a scale of excellent, very good, good, fair. We do not expect poor care.
12. Double barreled questions. What is the fastest and most convenient Internet service for you?
The fastest is certainly not the most economical. The double barreled question should be split
into two questions.
13. Independent answers. Make sure answers are independent. For example the question "Do you
think basketball players as being independent agents or as employees of their team?" Some
believe that yes, they are both.
14. Long questions. Multiple choice questions are the longest and most complex. Free text answers
are the shortest and easiest to answer. When you Increase the length of questions and surveys,
you decrease the chance of receiving a completed response.
15. Questions on future intentions. Yogi Berra (Famous New York Yankees Baseball Player) once
said that making predictions is difficult, especially when they are about the future. Predictions
are rarely accurate more than a few weeks or in some case months ahead.
VALIDITY AND RELIABILITY OF MEASUREMENT
The content of a measurement instrument includes a subject, theme, and topics that relate
to the characteristics being measured. However the measuring instrument does not include all of
the possible items that could have been included. When measuring complex psychological
constructs such as perceptions, preferences, and motivations, hard questions must be asked to
identify the items most relevant in solving the research problem:
1. Do the scales really measure what we are trying to measure?
2. Do subjects’ responses remain stable over time?
3. If we have a variety of scaling procedures, are respondents consistent in their scoring
over those scales that purport to be measuring the same thing?
By solving these problems, we establish the validity and reliability of scaling techniques.
Note that our focus is only on the general concepts and measures of validity and
reliability that are used in cross-sectional studies. There is little documented research on issues of
measure reliability and validity for time-series analysis.
253
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Validity
Validity simply means that we are measuring what we believe we are measuring. The
data must be unbiased and relevant to the characteristic being measured. The validity of a
measuring instrument reflects the absence of systematic error. Systematic error may arise from
the instrument itself, the user of the instrument, the subject, or the environment in which the
scaling procedure is being administered. Since in practice we rarely know true scores, we usually
have to judge a scaling procedure’s validity by its relationship to other relevant standards.
The validity of a measuring instrument hinges on the availability of an external criterion
that is thought to be correct. Unfortunately the availability of such outside criteria is often low.
What makes the problem even more difficult is that the researcher often is not interested in the
scales themselves, but the underlying theoretical construct that the scale purports to measure. It is
one thing to define IQ as a score on a set of tests; it is quite another to infer from test results that
a certain construct, such as intelligence, or a dimension of intelligence is being measured.
In testing the validity of a scale, the researcher must be aware that many forms of validity
exist, including (1) Content validity, (2) Criterion validity, and (3) Construct validity.
Content Validation
Content Validity concerns how the scale or instrument represents the universe of the
property or characteristic being measured. It is essentially judgmental and is ordinarily measured
by the personal judgments of experts in the field. That is, several content experts may be asked to
judge whether the items being used in the instrument are representative of the field being
investigated. Closely related to this approach for assessing content validation is a method
involving known groups. For instance, a scale purported to measure attitudes toward a brand
could be tested by administering it to a group of regular buyers of the product (which
presupposes a favorable attitude) and compared with those from a group of former buyers or
other non-buyers (who presumably have a negative attitude). If the scale does not discriminate
between the two groups, then its validity with respect to measuring brand attitude is highly
questionable. Caution must be exercised in using this method in that other group differences
besides their known behavior might exist, and account for the differences in measurement.
Face Validity is a preliminary or exploratory form of content validity. It is based on a
cursory review of items by non-experts such as one’s wife, mother, tennis partner, or similarly
convenient to access persons. A simple approach is to show the measurement instrument to a
convenient group of untrained people and ask whether or not the items look okay (Litwin, 2003).
Logical Validation refers simply to an intuitive, or common-sense, evaluation. This type
of validation is derived from the careful definition of the continuum of a scale and the selection
of items to be scaled. Thus, in an extreme case, the investigator reasons that everything that is
included is done so because it is obvious that it should be that way. Because things often do not
turn out to be as obvious as believed, it is wise for the marketing researcher not to rely on logical
validation alone.
Example: of research lacking content validity is the Coca-Cola Company’s introduction many
years ago of New Coke. Since the product represented a major change in taste, thousands of
consumers were asked to taste New Coke. Overwhelmingly, people said they liked the new
flavor. With such a favorable reaction, why did the decision to introduce the product turn out to
be a mistake? Executives of the company acknowledge that the consumer survey conducted
omitted a crucial question. People were asked if they like the new flavor, but they were not asked
if they were willing to give up the old Coke. In short, they were not asked if they would buy the
new product in place of the old one.
254
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Criterion Validation
Criterion Validity, also known as pragmatic validity, has two basic dimensions known as
predictive validity and concurrent validity. They question if the instrument works and are better
decisions can be made with it than without it?
The New Coke example also illustrates a case of poor predictive validity. The measures
of liking, and so on, were not very good predictors of purchase, which was the real measure of
managerial interest.
In concurrent validity, a secondary criterion, such as another scale, is used to compare
results. Concurrent validity can be assessed by correlating the set of scaling results with some
other set, developed from another instrument administered at the same time. Often product
researchers will ask a question like “Overall, how much do you prefer Brand A soft drink?”, and
then follow with another question such as, “Given the following four brands, indicate the
percentage of your total soft drink purchases that you would make for each brand.”
Alternatively, the correlation may be carried out with the results of the same question
asked again later in the survey or on another testing occasion.
Construct Validation
In construct validation the researcher is interested both in the question, “Does it work?”
(i.e., predict), and in developing criteria that permit answering theoretical questions of why it
works and what deductions can be made concerning the theory underlying the instrument.
Construct validity involves three subcases: convergent, discriminant, and nomological validity.
Convergent Validity: The correspondence in results between attempts to measure the
same construct by two or more independent methods. These methods need not all be scaling
techniques.
Discriminant Validity: Refers to properties of scaling procedures that do differ when they
are supposed to—that is, in cases where they measure different characteristics of stimuli and/or
subjects. That is, more than one instrument and more than one subject characteristic should be
used in establishing convergent-discriminant validity. Discriminant validity concerns the extent
to which a measure is unique (and not simply a reflection of other variables), and as such it
provides the primary test for the presence of method variance.
Nomological Validity: “Understanding” a concept (or construct). In nomological validity
the researcher attempts to relate measurements to a theoretical model that leads to further
deductions, interpretations, and tests, gradually building toward a nomological net, in which
several constructs are systematically interrelated.
Ideally, the marketing researcher would like to attain construct validity, thus achieving
not only the ability to make predictive statements but understanding as well. Specifically, more
emphasis should be placed on the theories, the processes used to develop the measures, and the
judgments of content validity.
Reliability
Reliability is concerned with the consistency of test results over groups of individuals or
over the same individual at different times. A scale may be reliable but not valid. Reliability,
however, establishes an upper bound on validity. An unreliable scale cannot be a valid one.
Reuman (1982, p. 1099) states that “according to classical test theory, highly reliable measures
are necessary, but not sufficient, for demonstrating high construct validity or high criterion
validity.”
255
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The achievement of scale reliability is, of course, dependent on how consistent the
characteristic being measured is from individual to individual (homogeneity over individuals)
and how stable the characteristic remains over time. Just how reliable a scaling procedure turns
out to be will depend on the dispersion of the characteristic in the population, the length of the
testing procedure, and its internal consistency. Churchill and Peter (1984) concluded that rating
scale estimates were largely determined by measuring characteristics such as number of items in
a scale, type of scale, and number of scale points. They further concluded that sampling
characteristics and measurement development processes had little impact.
In general, a measurement of the reliability of a scale (or measurement instrument) may
be measured by one of three methods: test-retest, alternative forms, or internal consistency. The
basics of reliability in a marketing context are reviewed by Peter (1979).
Test-Retest
The test-retest method examines the stability of response over repeated applications of
the instrument. Do we achieve consistent results, assuming that the relevant characteristics of
the subjects are stable over trials? One potential problem, of course, is that the first measurement
may have an effect on the second one. Such effects can be reduced when there is a sufficient
time interval between measurements. If at all possible, the researcher should allow a minimum of
two weeks to elapse between measurements. Reliability may be estimated by any appropriate
statistical technique for examining differences between measures.
Alternative Forms
The alternative forms method attempts to overcome the shortcomings of the test-retest
method by successively administering equivalent forms of the measure to the same sample.
Equivalent forms can be thought of as instruments built such that the same types and structures
of questions are included on each form, but where the specific questions differ. The forms of the
measurement device may be given one after the other or after a specified time interval,
depending upon the investigator’s interest in stability over time. Reliability is estimated by
correlating the results of the two equivalent forms.
Internal Consistency
Internal consistency refers to estimates of reliability within single testing occasions. In a
sense it is a modification of the alternative form approach, but differs in that alternatives are
formed by grouping variables. The basic form of this method is split-half reliability, in which
items are divided into equivalent groups (say, odd- versus even-numbered questions, or even a
random split) and the item responses are correlated. In practice, any split can be made.
A potential problem arises for split-half in that results may vary depending on how the
items are split in half. A way of overcoming this is to use coefficient alpha, known also as
Cronbach’s alpha, which is a type of mean reliability coefficient for all possible ways of splitting
an item in half (Cronbach, 1951). Whenever possible, alpha should be used as a measure of the
internal consistency of multi-item scales. Alpha is perhaps the most widely used measure of
internal consistency for multiple-item measures within marketing research. One caution
however, is that there should be a sufficient number of items in the measure so that alpha
becomes meaningful. Alpha has been used for as few as two items, and this essentially amounts
to a simple correlation between the two. Although there is no generally acceptable heuristic
covering the number of items, common sense would indicate that the minimum number of items
should be four or perhaps even six. What is clear, however, is that alpha is a function of the
256
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
number of items in a scale (i.e., the more items, the greater alpha will tend to be), and also a
function of the intercorrelation of the items themselves (Cortina, 1993; Voss, Stem, &
Fotopoulos, 2000). Consequently, when interpreting an obtained alpha, the number of items must
always be kept in mind.
The usual application of coefficient alpha is to calculate it using a statistical analysis
packages, report it, and assess whether the value obtained exceeds some rule-of-thumb minimum
value, typically 0.70. There now exist methods to make inferential tests about the size of alpha,
and to attach confidence intervals to the measure (Iacobucci & Duhachek, 2003).
In many projects, measurements or evaluations are made by more than a single evaluator.
Sometimes this is done when coding answers to open-ended questions. In these situations, the
researcher is interested in the reliability of these evaluations. This is known as interrater or
interobserver reliability. The most common measure used is a correlation (Litwin, 2003).
A Concluding Comment
Although it is not our objective to pursue in detail the methods by which reliability or
validity can be tested, we hope to provide an appreciation of the difficulties encountered in
designing and analyzing psychological measure. One question that has not been answered is,
“What is a satisfactory level of reliability, or what minimum level is acceptable?” There is no
simple definitive answer to this question. Much depends on the investigator’s or decision
maker’s primary purpose in measurement and on the approach used to estimate reliability. In
trying to arrive at what constitutes satisfactory reliability, the investigator must at all times
remember that reliability can affect certain qualities of a study including (1) Validity, (2) The
ability to show relationships between variables, and (3) The making of precise distinctions
among individuals and groups.
SUMMARY
This chapter focused on general concepts of measurement. We discussed the role of
definitions and defined concepts, constructs, variables, operational definitions, and propositions.
We then turned to measurement and examined what it is and how measurement relates to
development of scales. Also discussed, but rather briefly, were alternative sources that cause
variations within a set of measurements derived from a single instrument. This was followed by a
description of different types of scales that are commonly used in marketing research. Advanced
scaling techniques for scaling stimuli and respondents were also discussed. We concluded with a
brief overview of measurement validity and reliability, and the various types of each that are of
concern to an investigator.
257
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
Albaum, G., Best, R., & Hawkins, D. (1981). Continuous vs. discrete semantic differential scales.
Psychological Reports, 49, 83–86.
Bearden, W. O., & Netemeyer, R. G. (1999). Handbook of Marketing Scales (2nd ed.). Thousand Oaks,
CA: Sage.
Churchill, G. A., Jr., & Peter, J. P. (1984, November). Research design effects on the reliability of
rating scales: A meta analysis. Journal of Marketing Research, 21, 360–375.
Coombs, C. H. (1964). A Theory of Data. New York: Wiley.
Cortina, J. (1993). What is coefficient alpha? An examination of theory and applications. Journal of
Applied Psychology, 78(1), 98–104.
Cronbach, L. J. (1951, September). Coefficient alpha and the internal structure of tests. Psychometrika,
16, 297–334.
Davis, D. W. (1997). Nonrandom measurement error and race of interviewer effects among AfricanAmericans.
Public Opinion Quarterly, 61, 183–207.
Golden, L., Brockett, P., Albaum, G., & Zatarain, J. (1992). The golden numerical comparative scale
format for economical multi-object/multiattribute comparison questionnaires. Journal of Official
Statistics, 8 (1), 77–86.
Guttman, L. (1985). Measuring the true-state of opinion. In R. Ferber & H. Wales (Eds.), Motivation and
market behavior. Homewood, IL: Richard D. Irwin, 393–415.
Haley, R., & Case, P. (1979, Fall). Testing thirteen attitude scales for agreement and brand determination.
Journal of Marketing, 43, 20–32.
Hogarth, R. M. (Ed.) (1982). Question framing and response consistency. San Francisco: Jossey-Bass.
Hubbard, R., Little, E. L., & Allen, S. J. (1989). Are responses measured with graphic rating scales
subject to perceptual distortion? Psychological Reports, 69, 1203–1207.
Iacobucci, D., & Duhachek, A. (2003). Applying confidence intervals to coefficient alpha. Unpublished
working paper, Kellogg School of Management, Northwestern University, Evanston, IL.
Kerlinger, F. (1973). Foundation of behavioral research (2nd ed.). New York: Holt, Rinehart & Winston.
Levin, I. P., Gaeth, G. J, Evangelista, F., Albaum, G., & Schreiber, J. (2001). How positive and negative
frames influence the decisions of persons in the United States and Australia. Asia Pacific Journal of
Marketing and Logistics, 13(2), 64–71.
Levin, I. P., Schneider, S. L., & Gaeth, G. J. (1998, November). All frames are not created equal: A
typology of framing effects. Organizational Behavior and Human Decision Processes, 76(2), 149–188.
Likert, R. (1967). The method of constructing an attitude scale. In M. Fishbein (Ed.), Readings in attitude
theory and measurement. New York: Wiley, pp. 90–95.
258
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Litwin, M. S. (2003). How to assess and interpret survey psychometric (2nd ed.). Thousand Oaks, CA:
Sage.
McCarty, J. A., & Shrum, L. J. (2000). The measurement of personal values in survey research: A test of
alternative rating procedures. Public Opinion Quarterly, 64, 271–298.
Morgan, M. (2003, March 31). Be careful with survey data. Marketing News, 37, 26.
Osgood, C. E., Suci, G. J., & Tannenbaum, P. H. (1957). The measurement of meaning. Urbana:
University of Illinois Press.
Peter, J. P., & Churchill, G. A., Jr. (1986, February). Relationships among research design choices
and psychometric properties of rating scales: A meta-analysis. Journal of Marketing Research, 23, 1–10.
Peterson, R. A. (2000). Creating effective questionnaires. Thousand Oaks, CA: Sage.
Payne, S. L. (1951). The art of asking questions. Princeton, NJ: Princeton University Press.
Rasinski, K. A. (1989). The effect of question wording on public support for government spending.
Public Opinion Quarterly, 53, 388–394.
Reuman, D. A. (1982). Ipsative behavioral variability and the quality of thematic apperceptive
measurement of the achievement motive. Journal of Personality and Social Psychology, 43(5), 1098–
1110.
Semon, T. T. (1999, August 30). Scale ratings always betrayed by arithmetic. Marketing News, 33, 7.
Semon, T. T. (2001, October 8). Symmetry shouldn’t be goal for scales. Marketing News, 35, 9.
Shimp, T. A., & Sharma, S. (1987, August). Consumer ethnocentrism: Construction and validation of the
CETSCALE. Journal of Marketing Research, 24, 280–289.
Spector, P. E. (1992). Summated rating scale construction: An introduction. Newbury Park, CA: Sage.
Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103, 677–680.
Sudman, S., & Bradburn, N. (1983). Asking questions. San Francisco: Jossey-Bass.
Thurstone, L. L (1959). The measurement of values. Chicago: University of Chicago Press.
Torgerson, W. S. (1958). Theory and methods of scaling. New York: Wiley.
Voss, K. E., Stem, D. E, Jr., and Fotopoulos, S. (2000). A comment on the relationship between
coefficient alpha and scale characteristics. Marketing Letters, 11(2), 177—191.
Whitlark, D., & Smith, S. (2004). "Pick and Choose “ Marketing Research Volume 16, Issue 4, Pages 8-
14, American Marketing Association, Chicago, December, 2004
Wong, N., Rindfleisch, A., & Burroughs, J. E. (2003, June). Do reverse-worded items confound measures
in cross-cultural research? The case of the material values scale. Journal of Consumer Research, 30, 72–
91.
259
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 11
HYPOTHESIS TESTING AND UNIVARIATE
ANALYSIS
Scientific research is directed at the inquiry and testing of alternative explanations of
what appears to be fact. For behavioral researchers, this scientific inquiry translates into a desire
to ask questions about the nature of relationships that affect behavior within markets. It is the
willingness to formulate hypotheses capable of being tested to determine (1) what relationships
exist, and (2) when and where these relationships hold.
The first stage in the analysis process is identified to include editing, coding, and making
initial counts of responses (tabulation and cross tabulation). In the current chapter, we then
extend this first stage to include the testing of relationships, the formulation of hypotheses, and
the making of inferences.
In formulating hypotheses the researcher uses “interesting” variables, and considers their
relationships to each other, to find suggestions for working hypotheses that may or may not have
been originally considered. In making inferences, conclusions are reached about the variables
that are important, their parameters, their differences, and the relationships among them. A
parameter is a summarizing property of a collectivity—such as a population—when that
collectivity is not considered to be a sample (Mohr, 1990, p.12).
Although the sequence of procedures, (a) formulating hypotheses, (b) making inferences
and (c) estimating parameters is logical, in practice these steps tend to merge and do not always
follow in order. For example, the initial results of the data analysis may suggest additional
hypotheses that in turn require more and different sorting and analysis of the data. Similarly, not
all of the steps are always required in a particular project; the study may be exploratory in nature,
which means that it is designed more to formulate the hypotheses to be examined in a more
extensive project, than to make inferences or estimate parameters.
AN OVERVIEW OF THE ANALYSIS PROCESS
The overall process of analyzing and making inferences from sample data can be viewed
as a process of refinement that involves a number of separate and sequential steps that may be
identified as part of three broad stages:
1. Tabulation: identifying appropriate categories for the information desired, sorting the data by
categories, making the initial counts of responses, and using summarizing measures to provide
economy of description and thereby facilitate understanding.
2. Formulating additional hypotheses: using the inductions derived from the data concerning the
relevant variables, their parameters, their differences, and their relationships to suggest working
hypotheses not originally considered.
3. Making inferences: reaching conclusions about the variables that are important, their parameters,
their differences, and the relationships among them.
260
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The Data Tabulation Process
Seven steps are involved in the process of data tabulation:
1. Categorize. Define appropriate categories for coding the information collected.
2. Editing and Coding Data. Assign codes to the respondent’s answers.
3. Create the Data File. Enter the data into the computer and create a data file.
4. Error Checking and Handling Missing Data. Check the data file for errors by performing a
simple tabulation analysis to identify errors in coding or data entry. Once errors are identified,
data may be edited or recoded to collapse, combine or delete responses or categories.
5. Generate New Variables. New variables may be computed by data manipulations that multiply,
sum, or otherwise transform variables.
6. Weight Data Subclasses. Weights are often used to adjust the proportionate representation of
sample subgroups so that they match the proportions found in the population.
7. Tabulate. Summarize the responses to each variable included in the analysis.
As simple as these steps are from a technical standpoint, data management is most
important in assuring a quality analysis and thereby merit an introductory discussion. A more in
depth discussion of survey-based data management is provided by Fink (2003, Chap.1).
Defining Categories
The raw input to most data analyses consists of the basic data matrix, as shown in Table
11.1. In most data matrices, each row contains a respondent’s data and the columns identify the
variables or data fields collected for the respondent. The analyses of a column of data might
include a tabulation of data counts in each of the categories or the computation of the mean and
standard deviation. This analysis is often done simply because we want to summarize the
meaning of the entire column of values. In so doing we often (willingly) forgo the full
information provided by the data in order to understand some of its basic characteristics, such as
central tendency, dispersion, or categories of responses. Because we summarize the data and
make inferences from it, it is doubly important that the data be accurate.
Tabulation of any sizable array of data often requires that responses be grouped into
categories or classes. The identification of response categories early in the study has several
advantages. Ideally, it forces the analyst to consider all possible interpretations and responses to
the questionnaire. It often leads to improvements in the questionnaire or observation forms. It
permits more detailed instruction of interviewers and results in higher consistency in interpreting
responses. Editing problems are also reduced.
The definition of categories allows for identification of the database columns and values
assigned to each question or variable and to indicate the values assigned to each response
alternative. Depending on the data collection method, data code sheets can be prepared and precoded.
Data files are often formatted as comma separated variable (CSV) files, meaning that
each variable appears in the same relative position for each respondent with a comma separating
each of the variables. The major data analysis software programs read data files and then display
them in a spreadsheet-like database (see Table 11.1). Often the data are entered directly into a
Microsoft Excel spreadsheet for import into the statistical program to be used for analysis.
Where data is not collected and formatted electronically, pre-coding of printed questionnaires
will eliminate transcription and thereby decreasing both processing errors and costs. Most of
today’s computer-based software for telephone (CATI) or internet surveys (Qualtrics.com)
automate this entire process. They not only define the question categories in the database but also
automatically build the database and record the completed responses as they are submitted. The
261
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
data may then be analyzed online, exported to Microsoft Excel™ or imported into a dedicated
statistical analysis program such as PASW (formerly known as SPSS). Response categories are
coded from 1 for the first category to the highest value for the last category. Category values can
be recoded to assign different numbers as desired by the researcher.
As desirable as the early definition of categories is, it can sometimes only be done after
the data have been collected. This is usually the case when open-end text questions, unstructured
interviews, and projective techniques are used.
The selection of categories is controlled by both the purposes of the study and the nature
of the responses. Useful classifications meet the following conditions:
1. Similarity of response within the category. Each category should contain responses that, for
purposes of the study, are sufficiently similar that they can be considered homogenous.
2. Differences of responses between categories. Differences in category descriptions should be
great enough to disclose any important distinctions in the characteristic being examined.
3. Mutually exclusive categories. There should be an unambiguous description of categories,
defined so that any response can be placed in only one category.
4. Categories should be exhaustive. The classification schema should provide categories for all
responses.
The use of extensive open-end questions often provides rich contextual and anecdotal
information, but is a practice often associated with fledgling researchers. Open-end questions, of
course, have their place in marketing research. However, the researcher should be aware of the
inherent difficulties in questionnaire coding and tabulation, not to mention their tendency to be
more burdensome to the respondent. All of this is by way of saying that any open-end question
should be carefully checked to see if a closed-end question (i.e., check the appropriate box) can
be substituted without doing violence to the intent of the question. Obviously, sometimes this
substitution should not be made.
Editing and Coding
Editing is the process of reviewing the data to ensure maximum accuracy and clarity. This
applies to the editing of the collection forms used for pretesting as well as those for the full-scale
project. Careful editing during the pre-test process will often catch misunderstandings of
instructions, errors in recording, and other problems so as to eliminate them for the later stages of
the study. Early editing has the additional advantage of permitting the questioning of
interviewers while the material is still relatively fresh in their minds. Obviously, this has limited
application for printed questionnaires, though online or CATI surveys can be edited even when
data is being collected.
Editing is normally centralized so as to ensure consistency and uniformity in treatment of
the data. If the sample is not large, a single editor usually edits all the data to reduce variation in
interpretation. In those cases where the size of the project makes the use of more than one editor
mandatory, it is usually best to assign each editor a different portion of the data collection form
to edit. In this way the same editor edits the same items on all forms, an arrangement that tends
to improve both consistency and productivity.
Typically, interviewer and respondent data are monitored to ensure that data requirements
are fulfilled. Each collection form should be edited to ensure that data quality requirements are
fulfilled. Regarding data obtained by an interviewer (and to an extent self-report) the following
should be specifically evaluated:
262
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
1. Legibility of entries. Obviously the data must be legible in order to be used. Where not legible,
although it may be possible to infer the response from other data collected, where any real doubt
exists about the meaning of data it should not be used.
2. Completeness of entries. On a fully structured collection form, the absence of an entry is
ambiguous. It may mean either that the respondent could not or would not provide the answer,
that the interviewer failed to ask the question, or that there was a failure to record collected data.
3. Consistency of entries. Inconsistencies raise the question of which response is correct. (If a
respondent family is indicated as being a non-watcher of game shows, for example, and a later
entry indicates that they watched Wheel of Fortune twice during the past week, an obvious
question arises as to which is correct.) Discrepancies may be cleared up by questioning the
interviewer or by making callbacks to the respondent. When discrepancies cannot be resolved,
discarding both entries is usually the wisest course of action.
4. Accuracy of entries. An editor should keep an eye out for any indication of inaccuracy in the
data. Of particular importance is the detection of any repetitive response patterns in the reports of
individual interviews. Such patterns may well be indicative of systematic interviewer or
respondent bias or dishonesty.
Coding is the process of assigning respondent answers to data categories and numbers are
assigned to identify them with the categories. Pre-coding refers to the practice of assigning codes
to categories. Sometimes these codes are printed on structured questionnaires and observation
forms before the data are collected. Using these predefined codes, the interviewer is able to code
the responses when interpreting the response and marking the category into which it should be
placed.
Post-coding is the assignment of codes to responses after the data are collected, and is
most often required when responses are reported in an unstructured format (open-ended text or
numeric input). Careful interpretation and good judgment are required to ensure that the meaning
of the response and the meaning of the category are consistently and uniformly matched.
When not using CATI or online data collection technologies, a formal coding manual or
codebook is often created and made available to those who will be entering or analyzing the data.
The codebook used for a study of supermarkets in the United States is shown in Figure 11.1 as
an illustration.
Like good questionnaire construction, good coding requires training and supervision. The
editor-coder should be provided with written instructions, including examples. He or she should
be exposed to the interviewing of respondents and become acquainted with the process and
problems of collecting the data, thus providing aid in its interpretation. The coder also should be
aware of the computer routines that are expected to be applied, insofar as they may require
certain kinds of data formats.
Whenever possible (and when cost allows) more than one person should do the coding,
specifically the post-coding. By comparing the results of the various coders, a process known as
determining inter-coder reliability, any inconsistencies can be brought out. In addition to the
obvious objective of eliminating data coding inconsistencies, the need for recoding sometimes
points to the need for additional categories for data classification and may sometimes mean that
there is a need to combine some of the categories. Coding is an activity that should not be taken
263
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
lightly. Improper coding leads to poor analyses and may even constrain the types of analysis that
can be completed.
Qualtrics has an interesting feature that uses a “wizard” to take a survey by selecting
random choices and following the various logic paths available. The resulting test data conforms
to the “sample size” specified by the researcher and the pre-specified logic and coding can be
checked for errors that the researcher has made.
Figure 11.1 Codebook for Comparative Supermarket Study
Question
Number Variable
01 Sample group
1=Group A 2=Group B
02 Respondent ID number xxx=actual number
1 03 Residential district 1=SE 2=SW 3=NW 4=NE
2 04 How often shop at Albertson’s xx=actual number
2 05 How often shop at Raley’s xx=actual number
2 06 How often shop at Wal-Mart xx=actual number
2 07 How often shop at Smith’s xx=actual number
3 08 Primary shopper 1=self 2=spouse 3=parent(s) 4=housekeeper 5=other
4A 09 Store most likely to shop at 1=Albertson’s 2=Raley’s 3=Wal-Mart 4=Smith’s 5=Other
4B 10 Time to get to store xxx=actual number
4C 11 How to get there 1=car/taxi 2=bus 3=walk
5 12 Amount spent at Albertson’s xxxxx=actual number
5 13 Amount spent at Raley’s xxxxx=actual amount
5 14 Amount spent at Wal-Mart xxxxx=actual number
5 15 Amount spent at Smith’s xxxxx=actual number
16 BLANK
6 17 Supermarket evaluated 1=Albertson’s 2=Smith’s
6 18-35 Semantic scales for Albertson’s x=1 to 7, starting from the left-side of scale
location (18) layout (27)
prices (19) shopping experience (28)
atmosphere (20) reputation (29)
quality of products (21) service (30)
modern (22) helpfulness of clerks (31)
friendliness of clerks (23) dull (32)
customers (24) selection of products (33)
cluttered (25) dirty (34)
check-out (26) like (35)
6 36-53 Semantic scales for Smith’s X=1 to 7, starting from the left-side of the scale
location (36) layout (45)
prices (37) shopping experience (46)
atmosphere (38) reputation (47)
quality of products (39) service (48)
modern (40) helpfulness of clerks (49)
friendliness of clerks (41) dull (50)
customers (42) selection of products (51)
cluttered (43) dirty (52)
check-out (44) like (53)
7 54 Gender 1=female 2=male
8 55 Marital status 1=single 2=married 3=divorced/separated/widowed 4=other
9 56 Age xx=actual number
10 57 Employment status 1=full time 2=part time 3=not employed
Tabulation: Cleaning the Data
The purpose of the initial data cleaning tabulation is to identify outliers, missing data and
other indications of data, coding, transcription, or entry errors. The tabulation of responses will
invariably reveal codes that are out of range or otherwise invalid. For example, one tabulation
might reveal 46 males (category 1), 54 females (category 2), and one category 5 response, which
is obviously an error. Some errors, such as the preceding one, represent entry of values that are
out-of-range or wild codes (Lewis-Beck, 1995, p. 7). That is, the value is not one that has been
264
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
assigned to a possible response to a question. A miscoding error that is more difficult to detect is
one where an erroneous recording of a response category is made using a number that is assigned
to another response category. That is, in the coding shown in Figure 11.1 a response of self to
question number 3 (code = 1) might have been coded as spouse (code = 2). Hopefully, not too
many errors of this type occur.
Table 11.1 Illustration of Data Matrix
Variable
Object 1 2 3 … j … m
1
2
3
.
.
i
.
.
n
.
.
.
.
.
.
.
.
.
.
.
.
…
…
…
…
…
.
.
.
.
…
…
…
…
…
.
.
.
.
An aspect of cleaning the data is dealing with missing data. That is, some respondents
may not provide responses for all the questions. One way of handling this is to use statistical
imputation. This involves estimating how respondents who did not answer particular questions
would have answered if they had chosen to. Researchers are mixed in their views about this
process. A much safer way to handle a nonresponse situation is to treat the nonresponse as
missing data in the analysis. Statistical software programs can handle this either question-byquestion
or by deleting the respondent with a missing value from all analyses. Also, the
researcher can choose to eliminate a respondent from the data set if there is too much missing
data. Yet another way is to simply assign the group’s mean value to the missing items. Or, when
an item is missing from a multi-item measure, a respondent’s mean value for the rest of the items
can be used for the missing value.
Another issue that can arise is how to deal with outliers (Fink, 2003, pp. 22–23). Outliers
are respondents whose answers appear to be inconsistent with the rest of the data set. An easy
way to check for outliers is by running frequency analyses, or counts, of responses to questions.
Regression analysis also can be used to detect outliers. This is discussed in Chapter 13. Outliers
can be discarded from the analysis, but one must be careful to not throw out important and useful
information as would be the case when the outliers belong to a unique category of respondents
heretofore unidentified. If an outlier is retained then it may be best to use the median rather than
the mean as the measure of central tendency when such a measure is part of the analysis.
Short of having two or more coders create the data file independently of each other and
then assessing intercoder reliability, there is not much that can be done to prevent coder error
except to impress upon coders the necessity of accurate data entry. Multiple coders can be very
time-consuming and costly, particularly for large data files. Each error that is identified should
be traced back to the questionnaire to determine the proper code. The cleaning process is
complete when either the data file has been edited to correct the errors or the corrections have
been made in the analysis program.
265
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Tabulation: Basic Analysis
Tabulation may be thought of as the final step in the data collection process and the first
step in the analytical process. Tabulation is simply the counting of the number of responses in
each data category (often a single column of the data matrix contains the responses to all
categories).
The most basic is the simple tabulation, often called the marginal tabulation and familiar
to all students of elementary statistics as the frequency distribution. A simple tabulation or
distribution consists of a count of the number of responses that occur in each of the data
categories that comprise a variable. A cross-tabulation involves the simultaneous counting of the
number of observations that occur in each of the data categories of two or more variables. An
example is given in Table 11.2. We shall examine the use of cross-tabulations in detail later in
the chapter. A cross-tabulation is one of the more commonly employed and useful forms of
tabulation for analytical purposes.
The flexibility and ease of conducting computer analysis increases the importance of
planning the tabulation analysis. There is a common tendency for the researcher to decide that,
because cross-tabulations (and correlations) are so easily obtained, large numbers of tabulations
should be run. Not only is this methodologically unsound, but in commercial applications it is
often costly in analyst time as well. For 50 variables, for example, there are 1,225 different twovariable
cross-tabulations that can be made. Only a few of these are potentially of interest in a
typical study.
Table 11.2 Cross-Tabulation: Energy Drink Purchases by Income Classes of Respondents*
Number of liters purchased
Income Class Zero One Two Three or more Total
Less than $15,000
$15,000 - $34,999
$35,000 - $54,999
$55,000 - $74,999
$75,000 and over
Total
160
120
60
5
5
350
25
15
20
10
5
75
15
10
15
5
5
50
0
5
5
5
10
25
200
150
100
25
25
500
*Hypothetical Data
FORMULATING HYPOTHESES
As a beginning point in the discussion of hypotheses testing, we ask: what is a
hypothesis? A hypothesis is an assertion that variables (measured concepts) are related in a
specific way such that this relationship explains certain facts or phenomena. From a practical
standpoint, hypotheses may be developed to solve a problem, answer a question, or imply a
possible course of action. Outcomes are predicted if a specific course of action is followed.
Hypotheses must be empirically testable. A hypothesis is often stated as a research question
when reporting either the purpose of the investigation or the findings. The hypothesis may be
stated informally as a research question, or more formally as an alternative hypothesis, or in a
testable form known as a null hypothesis. The null hypothesis makes a statement that no
difference exists (see Pyrczak, 1995, pp. 75-84).
Research questions state in layman's terms the purpose of the research, the variables of
interest, and the relationships to be examined. Research questions are not empirically testable,
but aid in the important task of directing and focusing the research effort. To illustrate, a sample
research question is developed in the following scenario:
266
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 11.1 Development of a Research Question for Mingles
Mingles is an exclusive restaurant specializing in seafood prepared with a light Italian flair. Barbara
C., the owner and manager, has attempted to create an airy contemporary atmosphere that is
conducive to conversation and dining enjoyment. In the first three months, business has grown to
about 70 percent of capacity during dinner hours.
Barbara wants to track customer satisfaction with the Mingles concept, the quality of the
service, and the value of the food for the price paid. To implement the survey, a questionnaire was
developed using a five-point expectations scale with items scaled as values from -2 to +2. The
questionnaire asks, among other things:
“How would you rate the value of Mingles food for the price paid”?
The response format for the five answers appeared as:
When tabulated, the average response was found to be +0.89 with a sample standard deviation of
1.43. The research question asks if Mingles is perceived as being better than average when
considering the price and value of the food.
Additional questions measure customer satisfaction by addressing “How satisfied Mingles customers
are with the concept, service, food, and value”.
Null Hypotheses (H0) are statements identifying relationships that are statistically testable
and can be shown not to hold (nullified). The logic of the null hypothesis is that if we
hypothesize no difference, and we “reject” the hypotheses if a difference is found. If, however
we confirm that no difference exists, then we "tentatively accept" the null hypothesis. We may
only accept the null on a "tentative" basis because another testing of the null hypothesis using a
new sample may reveal that sampling error was present and that the null hypothesis should be
rejected.
For example, to compare the population and the sample, the null hypothesis might be:
"There is no difference between the perceived price-value of Mingles food and what is expected
on average. In this example, the difference between the population average, which is assumed to
be the middle scale value of 0 = "about average" and the sample's mean evaluation of Mingles
can be tested using the z distribution.
A null hypothesis may also be used to specify other types of relationships that are being
tested, such as the difference between two groups, or the ability of a specific variable to predict a
phenomenon such as sales or repeat business. Two examples:
1: Comparing two sample groups: H0: There is no difference in the value of the food for the
price paid as perceived by first time patrons and repeat patrons. This is tested by a t-test of the
difference in means between two patron groups.
2: Predicting intention to return to Mingles: H0: The perceived quality of service is not related
to the likelihood of returning to Mingles. This is a regression analysis problem that uses quality
of service to predict likelihood of returning to Mingles.
Alternative hypotheses may be considered to be the opposite of the null hypotheses. The
alternative hypothesis makes a formal statement of expected difference, and may state simply
267
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
that a difference exists or that a directional difference exists, depending upon how the null
hypothesis is stated. Because population differences may exist, even if not verified by the
current sample data, the alternative form is considered to be empirically non-testable. The
relationship between hypothesis and research questions is summarized in Table 11.3.
Table 11.3 Hypotheses and Research Questions
Purpose Example Decision
Research
Question
Express the purpose of the research What is the perception
of Mingles customers
regarding the pricevalue
of the food?
None used
Alternative
Hypothesis
The alternative hypothesis states the
specific nature of the hypothesized
relationship. i.e., that there is a
difference. The alternative hypothesis is
the opposite of the null hypothesis. The
alternative hypothesis cannot be
falsified because a relationship
hypothesized to exist may not have
been verified, but may in truth exist in
another sample. (You can never reject
an alternative hypothesis unless you test
the population on all possible samples.
Mingles is perceived as
having superior food
value for the price
when compared to the
average evaluation.
Not tested because we
cannot reject. We may
only accept that a
relationship exists.
Null
Hypothesis
The null hypothesis is testable in the
sense that the hypothesized lack of
relationship can be tested. If a
relationship is found, the null hypothesis
is rejected. The Null hypothesis states
that there is no difference between
groups (with respect to some variable)
or that a given variable does not predict
or otherwise explain an observed
phenomena, effect or trend.
There is no difference
in perceived food value
for the price for
Mingles and the
average evaluation.
We may reject a null
hypothesis (Find a
relationship). We may
only tentatively accept
that no relationship
exists.
The objectives and hypotheses of the study should be stated as clearly as possible and
agreed upon at the outset. Objectives and hypotheses shape and mold the study; they determine
the kinds of questions to be asked, the measurement scales for the data to be collected, and the
kinds of analyses that will be necessary. However, a project will usually turn up new
hypotheses, regardless of the rigor with which it was planned and developed. New hypotheses
are continually suggested as the project progresses from data collection through the final
interpretation of the findings.
In Chapter 2 it was pointed out that when the scientific method is strictly followed,
hypothesis formulation must precede the collection of data. This means that according to the
rules for proper scientific inquiry, data suggesting a new hypothesis should not be used to test it.
New data must be collected prior to testing a new hypothesis.
In contrast to the strict procedures of the scientific method, where hypotheses formulation
must precede the collection of data, actual research projects almost always formulate and test
new hypotheses during the project. It is both acceptable and desirable to expand the analysis to
examine new hypotheses to the extent that the data permit. At one extreme, it may be possible to
268
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
show that the new hypotheses are not supported by the data and that no further investigation
should be considered. At the other extreme, a hypothesis may be supported by both the specific
variables tested and by other relationships that give similar interpretation. The converging
results from these separate parts of the analysis strengthen the case that the hypothesized
relationship is correct. Between these extremes of nonsupport-support are outcomes of
indeterminacy: the new hypothesis is neither supported nor rejected by the data. Even this result
may indicate the need for an additional collection of information.
In a position yet more extreme from scientific method, Selvin and Stuart (1966)
convincingly argue that in survey research, it is rarely possible to formulate precise hypotheses
independently of the data. This means that most survey research is essentially exploratory in
nature. Rather than having a single pre-designated hypothesis in mind, the analyst often works
with many diffuse variables that provide a slightly different approach and perspective on the
situation and problem. The added cost of an extra question is so low that the same survey can be
used to investigate many problems without increasing the total cost. However, researchers must
resist the syndrome of “just one more question”. Often, the one more question escalates into
many more questions of the type “it would be nice to know”, which can be unrelated to the
research objectives.
In a typical survey project, the analyst may alternate between searching the data
(analyzing) and formulating hypotheses. Obviously, there are exceptions to all general rules and
phenomena. Selvin and Stuart (1966), therefore, designate three practices of survey analysts:
1. Snooping. The process of searching through a body of data and looking at many relations
in order to find those worth testing (that is, there are no pre-designated hypotheses)
2. Fishing. The process of using the data to choose which of a number of pre-designated
variables to include in an explanatory model
3. Hunting. The process of testing from the data all of a pre-designated set of hypotheses
This investigative approach is reasonable for basic research but may not be practical for
decisional research. Time and resource pressures seem to require that directed problem solving
be the focus of decision research. Rarely can the decision maker afford the luxury of dredging
through the data to find all of the relationships that must be present. Again, it simply reduces to
the question of cost versus value.
MAKING INFERENCES
Testing hypotheses is the broad objective that underlies all decisional research.
Sometimes the population as a whole can be measured and profiled in its entirety. Often,
however, we cannot measure everyone in the population but instead must estimate the population
using a sample of respondents drawn from the population. In this case we estimate the
population “parameters” using the sample “statistics”. Thus, in both estimation and hypothesis
testing, inferences are made about the population of interest on the basis of information from a
sample.
We often will make inferences about the nature of the population and ask a multitude of
questions, such as: Does the sample's mean satisfaction differ from the mean of the population
of all restaurant patrons? Does the magnitude of observed the differences between categories
indicate that actual differences exist, or are they the result of random variations in the sample?
269
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In other studies, it may be sufficient to simply estimate the value of certain parameters of
the population, such as the amount of our product used per household, the proportion of stores
carrying our brand, or the preferences of housewives concerning alternative styles or package
designs of a new product. Even in these cases, however, we would want to know about the
underlying associated variables that influence preference, purchase, or use (color, ease of
opening, accuracy in dispensing the desired quantity, comfort in handling, etc.), and if not for
purposes of the immediate problem, then for solving later problems. In yet other case studies, it
might be necessary to analyze the relationships between the enabling or situational variables that
facilitate or cause behavior. Knowledge of these relationships will enhance the ability to make
reliable predictions, when decisions involve changes in controllable variables.
The Relationship Between a Population, a Sampling Distribution, and a Sample
In order to simplify the example, suppose there is a population consisting of only five
persons. On a specific topic, these five persons have a range of opinions that are measured on a
7-point scale ranging from very strongly agree to very strongly disagree. The frequency
distribution of the population is shown in the bar chart of Figure 11.2.
Figure 11.2 Population Frequency Distribution
The parameters describe this population as having a mean of µ = 4 and standard deviation = 2.
Now that we know the “parameters” of the population, we will consider the sampling
distribution for our example data. Assume for a moment that like most populations, ours is so
large that we are not able to measure all persons in this population, but must rely instead on a
sample. In our example, the population is not large, and we will assume a sample of size n = 2.
The sampling distribution is the distribution if sample means from all possible samples of
size n=2. The sampling distribution of means and standard errors are shown in Table 11.4.
270
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 11.4 Computation of Sampling Distribution, Mean, and Standard Error
The mean of all possible two-member sample means is
and summing the standard errors of the mean for the sampling distribution gives
which gives a standard deviation of
Understand that the sampling distribution becomes more normal as the sample size
increases and that even in this simple case, we observe a somewhat normal shape. Also
understand that the population mean 4=µ is always equal to the mean of the sampling
distribution of all possible sample means ( = 4 ).
The Relationship Between the Sample and the Sampling Distribution
When we draw a sample, we rarely know anything about the population, including its
shape, µ, or σ. We must, therefore compute statistics from the sample ( x and s) and make
inferences about the population µ, and σ using the sample information.
Suppose we were to repeatedly draw samples of n = 2 (without replacement). The
relevant statistics for the first of these samples having the values of (1,2) are:
271
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Given this x and , we can now estimate with a given probability, the intervals that
give a range of possible values that could include µ, the population mean. For this single
sample, they are:
68% = 1.5 ± 6.31 (.5) or -1.655 to 4.655
95% = 1.5 ± 12.71 (.5) or -4.855 to 7.855
99% = 1.5 ± 31.82 (.5) or -14.41 to 17.41
Thus, we could state that we are 99% confident that the population mean would fall
within the interval –14.41 to 17.41. We note that this range is very wide and does include our
population mean of 4.0. The size of the range is large because the small sample size (n=2). As
the sample size increases, the numbers become larger, and gradually approximate a standard
normal distribution.
The above discussion is summarized graphically in Exhibit 11.2. Part I of Exhibit 11.2
shows the relationship between the population, the sample, and the sampling distribution while
Part II illustrates the impact of sample size on the shape of the sampling distribution for
differently shaped population distributions.
EXHIBIT 11.2 Population, Sample, and Sampling Distribution
Part I
272
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Part II
Acceptable Error in Hypothesis Testing
A question that continually plagues analysts is, what significance level should be used in
hypothesis testing? The significance level refers to the amount of error we are willing to accept
in our decisions that are based on the hypothesis test. Hypotheses testing involves specifying the
value α, which is the allowable amount of Type I error.
In hypothesis testing the sample results sometimes lead us to reject H0 when it is true.
This is a Type I error. On other occasions the sample findings may lead us to accept H0 when it
is false. This is a Type II error. The nature of these errors is shown in Exhibit 11.3.
The amount of type I error, α, we are willing to accept should be set after considering (a)
how much it costs to make such an error and (b) the decision rule used. A ”classic" paper
dealing with criteria relevant for this question is Labovitz (1968). There are substantial
differences in costs of errors in research conducted for exploratory purposes and research
conducted to make a decision where large financial investments are made. The acceptable levels
of significance (error) used in a basic research project may be entirely inappropriate for a
managerial decision dealing with the same problem, but where millions of dollars, or a
company’s market strategy is involved, managerial decisions are rarely simple or based on a
single piece of research.
A tradition of conservatism exists in basic research and has resulted in the practice of
keeping the Type I error at a low level (.05 or .01). The Type I error has been traditionally
considered to be more important than the Type II error and, correspondingly, that it is more
important to have a low α than a low β. The basic researcher typically assigns higher costs to a
Type I than to a Type II error.
273
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 11.3 Types of Error in Making a Wrong Decision
There are two types of error that result from a mismatch between the conclusion of a research study
and reality. In the null-hypothesis format, there are two possible research conclusions, to retain H0 and
to reject H0. There are also two possibilities for the true situation: H0 is true or H0 is false.
These outcomes provide the definitions of Type I and Type II errors and the confidence level and power
of the test:
1. A Type I error occurs when we incorrectly conclude that a difference exists. This is
expressed as α, the probability that we will incorrectly reject H0, the null hypothesis, or
sometimes called the hypothesis of no difference.
2. A Type II error occurs when we accept a null hypothesis when it is in reality false (we
find no difference when a difference really does exist).
3. Confidence level: we correctly retain the null hypothesis (we could also say tentatively
accept or it could not be rejected). This is equal to the area under the normal curve less the
area occupied by α, the significance level.
4. The power of the test is the ability to reject the null hypothesis when it should be
rejected (when false). That is, the power to not make an error. Because power increases as α
becomes larger, researchers may choose an α of .10 or even .20 to increase power.
Alternatively, sample size may be increased to increase power. Increasing sample size is the
preferred option for most market researchers.
The four possible combinations are shown in the following diagram:
In decisional research, the costs of an error are a direct result of the consequences of the
errors. The cost of missing a market entry by not producing a product and foregoing gain (Type
II error) may be even greater than the loss from producing a product when we should not (Type I
error). The cost depends on the situation and the decision rule being used. Of course not all
decision situations have errors leading to such consequences. In some situations, making a Type
I error may lead to an opportunity cost (for example, a foregone gain) and a Type II error may
create a direct loss.
In decisional research, the acceptable level of significance (i.e., the specification of a
significance level or α) should be made by the client or the manager for whom the study has
been conducted. This allows the assessment of risk by the person who will use the results as a
274
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
basis for decisions (Semon, 2000). This means that the researcher should merely report the level
at which significance occurs, letting the manager decide what this means. For basic research
situations, a typical finding is often reported as significant or not as tested against a specified
level, often .05. Again, it seems sensible that the researcher reports for each finding, the level at
which significance occurs, letting the reader of a study decide the meaning of the alpha level.
Power of a Test
The power of a hypotheses test is defined as 1-β, or 1 minus the probability of a Type II
error. This means that the power of a test is the ability to reject the null hypothesis when it is
false (or to find a difference when one is present).
The power of a statistical test is determined by several factors. The main factor is the
acceptable amount of discrepancy between the tested hypothesis and the true situation. This can
be controlled by increasing α. Power is also increased by increasing the sample size (which
decreases the confidence interval).
In Figure 11.3, we observe two sampling distributions, the first having a mean µ and the
second having a mean µ0 + 1α. In testing the equality of the two distributions, we first identify
α, which is here set to .05. (The one tail probably is .95 for 1.65σ). In Figure 11.2, we observe
that for the second sampling distribution, power is the area to the right of µ0 + 1.65 σ, i.e., the
area in which a difference between means from the two sampling distributions exists and was
found to exist.
Figure 11.3 Two Sampling Distributions
Frequency
µo o
µ +σ
x
µo + 1.65 σ
σ
I II
Type II - Error
ß
Power
(1-ß)
α
x x
x
If σ were decreased to, say, µ0 + 1.5 σ, the shaded area would move to the left, and the
area defining power for curve II would increase.
The second method of increasing power is to increase sample size. Because a sample
size increase directly reduces the standard error (σ = σ / ) any increase in n decreases the
absolute width of the confidence interval and narrows sampling distribution I.
In decisions that imply a greater cost for a type II error (missing a difference that does in
fact exist), the analyst may want to consider expanding the probability of a type I error (σ) to .10,
.15, or even .20, especially when sample sizes are small. Exhibit 11.2 discusses some
ramifications of ignoring statistical power.
SELECTING TESTS OF STATISTICAL SIGNIFICANCE
Up to this point, we have considered data analysis at a descriptive level. It is now time to
introduce ways to test whether the association observed is statistically significant. In many cases
this involves testing hypotheses concerning tests of group means. These types of tests are
275
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
performed on interval or ratio data using what is known as "parametric tests" and include such
techniques as the F, t, and z tests. Often however, we have only nominal or loosely ordinal data
and we are not able to meet the rigid assumptions of a parametric test. Cross tabulation analysis
with the χ2
test is often used for hypothesis testing in these situations. The χ2
statistic is from
the family of non-parametric methods.
Nonparametric methods are often called distribution-free methods because the inferences
are based on a test statistic whose sampling distribution does not depend upon the specific
distribution of the population from which the sample is drawn (Gibbons, 1993, p., 2). Thus, the
methods of hypothesis testing and estimation are valid under much less restrictive assumptions
than classical parametric techniques—such as independent random samples drawn from normal
distributions with equal variances, interval level measurement. These techniques are appropriate
for many Marketing applications where measurement is often at an ordinal or nominal level.
There are many parametric and nonparametric tests. The one that is appropriate for
analyzing a set of data depends on: (1) the level of measurement of the data, (2) the number of
variables that are involved, and for multiple variables, how they are assumed to be related.
As measurement scales become more restrictive and move from nominal to ordinal, and
interval levels of measurement, the amount of information and power to extract that information
from the scale increases. Corresponding to this spectrum of available information is an array of
non-parametric and parametric statistical techniques that focus on describing (i.e., measures of
central tendency and dispersion) and making inferences about the variables contained in the
analysis--i.e., tests of statistical significance (as shown in Table 11.5). A useful reference for
selecting an appropriate statistical technique is found in the Qualtrics Survey University.
Table 11.5 Selected Parametric and Nonparametric Univariate Analyses
What type of analysis do you want?
Level of Measurement
Nominal Ordinal Interval (Parametric)
Measure of central tendency Mode Median Mean
Measure of Dispersion None Percentile Standard Deviation
One-Sample test of statistical
significance
Binomial test
Π
2
one-sample test
Kolmogorov-Smirnov
One-sample test
One-sample runs test
t-test
Z-test
EXHIBIT 11.4 IGNORING STATISTICAL POWER
Much has been written about product failure, and about the failure of advertising campaigns. But,
except in rare instances, very little has been said about research failure, or research that leads to
incorrect conclusions. Yet research failure can occur even when a study is based on an expertly
designed questionnaire, good field work, and sophisticated analysis. The flaw may be inadequate
statistical power.
Mistaking chance variation for a real difference is one risk, called Type II error and the 95%
criterion largely eliminates it. By doing so, we automatically incur a high risk of Type II errormistaking
a real difference for chance variation. The ability of a sample to guard against Type II
error is called statistical power.
276
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Power was hot stuff in the 1980s. According to Semon (1990), however, but we continued to
ignore statistical power, for three reasons:
1. The concept is more complicated than statistical significance or confidence limits.
2. Consideration of statistical power would often indicate that we need larger (more
costly) samples than we are now using.
3. Numerical objectives must be specified before a research budget is fixed.
The last reason may be the worst of all. Management, even if it has a target in mind, is often
unwilling to do so.
The issue really is one of the degree of sensitivity required, the answer or the question,
"What is the smallest change or difference we need to measure, with a specified degree of
confidence?"
Suppose we have a pricing or a package test using two monadic samples for determining
which of two options to use. Suppose also that (unknown to us) Option X if superior to Y by 10
percentage points, say 30% strong buying interest vs. 20% interest.
If we use two samples of 200 each, how likely will our study identify X as being
significantly superior, using a 95% significance criterion? The likelihood of that correct result is only
68%, just a shade better than 2:1, in effect a 32% risk of research failure. If we want to reduce that
risk to 10% (that is, 90% statistical power), we need samples of 400 each.
Setting the desired levels of protection against the two types of error should not be a
rote process. It should take into consideration the relative business risks and, therefore, requires
management participation since researchers usually do not have access to enough information to
make these decisions
Another way of looking at statistical power is to suggest that the researcher look at
statistical insignificance rather than statistical significance. By “testing” for statistical insignificance
one asks, “Is this result so likely to occur by chance that we should ignore it?” Is statistical
insignificance high enough to dismiss the finding?
But, as pointed out by one observer, there is no scientific, objective way to determine
what is “high enough” (Semon, 1999). For example, if the significance level associated with the
difference in preference for two packages is 75%, the insignificance level is 25%. That’s the same
as saying the likelihood of a difference as large as observed occurring by chance is at odds of 1:3. Is
that “too high?” Clearly, 1:3 odds should deserve some consideration. But, the level that such
odds should have is a matter of personal preference and reflects one’s risk taking philosophy.
Perhaps we should review the specs. Why do we need 95% protection against Type I
error? If we lower our significance criterion to 90%, the sample size requirement drops to 300.
PARAMETRIC AND NON-PARAMETRIC ANALYSIS
Our discussion of statistical inference has emphasized that analytical procedures require
assumptions about the distribution of the population of interest. For example, for normally
distributed variables, we assume normality and homogeneity of variances. When a variable to be
analyzed conforms to the assumptions of a given distribution, the distribution of the variable can
be expressed in terms of its parameters (µ and σ). This process of making inferences from the
sample to the population's parameters is called parametric analysis.
Sometimes, however, problems occur: what if we cannot assume normality, or we must
question the measurement scale used. Parametric methods rely almost exclusively on either
277
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
interval or ratio scaled data. In cases where data can be obtained only using ordinal or
categorical scales the interpretation of the results may be questionable especially if the ordinal
categories are not of equal interval. When data do not meet the rigorous assumptions of
parametric method, we must rely on non-parametric methods which free us of the assumptions
about the distribution.
Whereas parametric methods make inferences about parameters of the population (µ and
s), non-parametric methods may be used to compare entire distributions that are based on
nominal data. Other non-parametric methods that use an ordinal measurement scale test for the
ordering of observations in the data set.
Problems that may be solved with parametric methods may often be solved by a nonparametric
method designed to address a similar question. Often times, the researcher will find
that the same conclusion regarding significance is made when data are analyzed by a parametric
method and by its “corresponding” non-parametric method. We will now discuss a univariate
parametric and non-parametric analyses. In the next chapter bi-variate parametric and nonparametric
analyses are presented. Additional non-parametric analyses are presented in the
appendix to Chapter 12.
Univariate Analyses of Parametric Data
Marketing researchers are often concerned with estimating parameters of a population.
In addition, many studies go beyond estimation and compare population parameters by testing
hypotheses about differences between them. Very often, the means, proportions and variances
are the summary measures of concern. Our concern at this point is with differences between the
means and proportions of the sample and those of the population as a whole. These comparisons
involve a single variable. In the following sections, we will demonstrate three important
concepts: (1) how to construct and interpret a confidence interval; (2) how to perform a
hypothesis test, and (3) how to determine the power of a hypothesis test. These issues are
discussed in more depth by Mohr (1990).
The Confidence Interval
The concept of a confidence interval is central to all parametric hypothesis testing. The
confidence interval is a range of values with a given probability (.95, .99, etc.) of covering the
true population parameter.
For example, assume we have a normally distributed population with population mean µ
and a known population variance σ2
. Suppose we sample one item from the population, X. This
single item is an estimate of µ, the population mean. Further, because the single item has been
drawn randomly from a normally distributed population, the possible distribution of x values is
the same as the population. This normal distribution permits us to estimate the probability
associated with various intervals of values of X. For example, p(-1.96 ≤ z ≥ +1.96) = .95. Or,
about 95% of the area under the normal probability curve is within this range. The confidence
interval shows both the z values and the values that are included in the confidence interval. We
compute this range for the sample problem that follows.
Suppose that it is a well-known fact that the average supermarket expenditure on laundry
and paper products is normally distributed, with a mean of µ=$32.00 per month, and the known
population standard deviation is σ =10.00. The 95% confidence interval about the population
mean is computed as:
278
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Lower limit ≤ population mean ≥ upper limit
µ - 1.96σ ≤ µ ≥ µ + 1.96 σ
32 - (1.96)(10) ≤ 32 ≥ 32 + (1.96)(10)
32 - 19.6 ≤ 32 ≥ 32 + 19.6
12.4 ≤ 32 ≥ 51.6
Thus, we expect that 95% of all household expenditures will fall within this range, as shown in
Figure 11.5.
Figure 11.5 Sampling Distribution of the Mean ( x
u )
.025
$12.40
-1.96
µ x = $32.00 $51.60
.025
0 +1.96
x
z
Reject H o Accept H o Reject Ho
If we expand this analysis to construct a confidence interval around the mean of a sample
rather than the mean of a population we must rely on a sampling distribution to define our
normal distribution. The sampling distribution is defined by the means of all possible samples of
size n. Recall that the population mean µ is also the mean of the normally distributed sampling
distribution, whereas μ ± zσ describes the confidence interval for a population, the value ( ± α
s/ )) describes the confidence interval for the sampling distribution. This is the probability that
this specified area around the sample mean covers the population mean. It is interesting to note
that because n, the sample size, is included in the computation of the "standard error," we may
estimate the population mean with any desired degree of precision, simply by having a large
enough sample size.
Univariate Hypothesis Testing of Means
Where the Population Variance is Known
Researchers often desire to test a sample mean to determine if it is the same as the
population mean. The z statistic describes probabilities of the normal distribution and is the
appropriate tool to test the difference between µ, the mean of the sampling distribution, and, the
sample mean when the population variance is known. The z statistic may, however, be used only
when the following conditions are met:
(1) Individual items in the sample must be drawn in a random manner.
(2) The population must be normally distributed. If this is not the case, the sample must be
large (>30), so that the sampling distribution is normally distributed.
(3) The data must be at least interval scaled.
(4) The variance of the population must be known.
279
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
1. When these conditions are met, or can at least be reasonably assumed to
exist, the traditional hypothesis testing approach is as follows:
(1) The null hypothesis (H0) is specified that there is no difference between µ and . Any
observed difference is due solely to sample variation.
(2) The alpha risk (Type I error) is established (usually .05).
(3) The z value is calculated by the appropriate z formula:
(4) The probability of the observed difference having occurred by chance is determined from
a table of the normal distribution (Appendix B, Table B-1).
(5) If the probability of the observed differences having occurred by chance is greater than
the alpha used, then H0 cannot be rejected and it is concluded that the sample mean is
drawn from a sampling distribution of the population having mean µ.
Returning to our supermarket example, suppose now that a random sample of 225
respondents was collected. The sample has a mean of =$30.80 and the population standard
deviation is known to be $10.00. We want to know if the population mean µ=$32.00 equals the
sample mean =30.80, given sample variation.
Steps:
The results are shown in Figure 11.6.
Figure 11.6 Hypothesis Test, Variance Known
280
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Population Variance is Unknown
Researchers rarely know the true variance of the population, and must therefore rely on
an estimate of σ2
, namely, the sample variance s2
. With this variance estimate, we compute the t
statistic.
Let's assume that for the supermarket example, we do not know the population values,
but we again want to test if the average monthly purchase of laundry and paper goods equals
$32.00, when the sample of 225 shoppers shows a mean, , of $30.80 and a standard deviations,
of $9.00. The t statistic is computed as:
With a probability: P (t=2.00, df = n-1=224) = .951 (2 tailed test)
1-P (t=2.00) = .049
Confidence Interval:
1.96 s/ )= 1.176
32 ± 1.176 = 30.824 + 32 + 33.176
We therefore reject the null hypothesis. Figure 11.7 shows these results.
Figure 11.7 Hypothesis Test, Variance Unknown
.05
30.824
-1.96
32
.05
33.176
t
x
0 +1.96
30.80
-2.00
A question may arise why we reject the null hypothesis for the t-test but do not reject it
for z-test. The answer is relatively simple. In the z-test, the numerator is a known constant
σ/ ). It is the same regardless of the sample that is drawn, as long as the sample is of size n.
The denominator of the t-test varies with the sample variance. Because the variance of the
sample was less than the variance of the population (9 vs. 10), the size of the confidence interval
was reduced, allowing us to reject the null hypothesis.
281
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Unlike the z-distribution, the t is a family of distributions, each having a different shape,
depending on the degrees of freedom (see Figure 11.8). The probability for a given value of t
varies across distributions.
The appropriate t-distribution to use in an analysis is determined by the available
"degrees of freedom." Unfortunately, the concept of degrees of freedom is inadequately defined.
Many statisticians use the term to describe the number of values that are free to vary. A second
definition is that “degrees of freedom” is a measure of how much precision an estimate of
variation has. A general rule is that the degrees of freedom decrease when more parameters have
to be estimated. This concept seems to something that “we all know what it is, but cannot
precisely define it.” It seems reasonable to accept the meaning in terms of “values that vary” by
specifying that degrees of freedom refers to a mathematical property of a distribution related to
the number of values in a sample that can be freely specified once one knows something about
the sample.
In univariate analyses, the available degrees of freedom are n-1. We lose one degree of
freedom for each population parameter that we estimate (µ). To explain further, suppose we
have a sample with size n = 5 in which the mean value of some measure is calculated to be zero.
If we select at random any 4 numbers from this sample, say -6, -4, 4, and 9, we know that the last
number is completely determined and must be -3. Each time we estimate a parameter we lose
one degree of freedom, while n - 1 are free to estimate that parameter.
Figure 11.8 The t-distributions for df = 1, df = 4 and df = 10
As n becomes large (≥ 30) the curve is approximately normal, with true normality
reached at the limit (n = ∞). The t-statistic is widely used in both univariate and bi-variate
market research analyses, due to the relaxed assumptions over the z-statistic, as follows:
(1) Individual items in the sample are drawn at random.
(2) The population must be normally distributed. If not, the sample must be
large (>30).
(3) The data must be at least interval scaled.
(4) The population variance is not known exactly, but is estimated by the
variance of the sample.
282
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Univariate Analysis of Categorical Data:
The Chi-Square Goodness of Fit Test
Chi-square analysis (χ2
) can be used when the data identifies the number of times or
frequency that each category of a tabulation or cross tabulation appears. Chi-square is a useful
technique for testing the following relationships:
1. Determining the significance of sample deviations from an assumed
theoretical distribution; that is, does a certain model fit the data. This is
typically called a goodness-of-fit test.
2. Determining the significance of the observed associations found in the
cross tabulation of two or more variables. This is typically called a test of
independence. (Discussed in Chapter 12).
The procedure involved in chi-square analysis is quite simple. We compare the observed
(frequency) data with another set of "data" based on a set of theoretical frequencies. These
theoretical frequencies may result from application of some specific model of the phenomenon
being investigated--relationship 1 above. Or we might specify that the frequency of occurrence
of two or more characteristics is mutually independent--relationship 2 above.
In either case we compute a measure (chi-square) of the variation between actual and
theoretical frequencies, under the null hypothesis that there is no difference between the model
and the observed frequencies. We may say that the model fits the facts. If the measure of
variation is "high," we reject the null hypothesis at some specified alpha risk. If the measure is
"low," we accept the null hypothesis that the model's output is in agreement with the actual
frequencies.
Single Classification
Suppose that we are interested in how frequently a sample of respondents selects each of
two test packages. Test package A contains an attached coupon involving so many cents off on a
subsequent purchase of the brand by the respondent. Test package B is the same package but,
rather than an attached coupon, it contains an attached premium (a ballpoint pen) that the
respondent may keep. The packages are presented simultaneously to a sample of 100
respondents and each respondent is asked to choose one of the two packages.
The frequency of choice is presented in the first column (labeled "Observed") of Table
11.6. We see that 63 respondents out of 100 select package A. Suppose that the researcher
believes the "true" probability of selecting A versus B to be 50-50 and that the observed 63-37
split reflects sampling variation. Given that the researcher's model would predict an estimated
frequency of 50-50, are the observed frequencies compatible with this theoretical prediction? In
chi-square analysis we set up the null hypothesis that the observed (sample) frequencies are
consistent with those expected under application of the model.
We use the following notation. Assume that there are k categories and a random sample
of n observations; each observation must fall into one and only one category. The observed
frequencies are
283
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The theoretical frequencies are:
In the problem above,
f1=63, f2=37, F1=50, F2=50, n=100
We compute the chi-square statistic:
Table 11.6 Observed vs. Theoretical Frequencies (Test-Package Illustration)
Package fi Observed Fi Predicted fi – Fi (fi – Fi)2
/ Fi
A 63 50 13 169/50 = 3.38
B 37 50 -13 169/50 = 3.38
Total 100 100 6.67
In the above one-way classification problem, the statistic is approximately distributed as
chi square with k-1 degrees of freedom. That is, we have only two categories and, hence, 1
degree of freedom. Table A.3 in the Appendix at the end of this book shows the appropriate
distribution.
In Table A.3 the tabular chi-square value for α = 0.05 and k-1=1 is 3.84. If the null
hypothesis is true, the probability of getting a chi-square value greater then 3.84 is 0.05. Since
our computed chi-square value is 6.76 (see Table 11.6), we reject the null hypothesis that the
output of the theoretical model corresponds with the observed frequencies (see Figure 11.9). In
using the chi-square table we note that only k, the number of categories, is pertinent, rather than
the sample size n. Sample size is important to the quality of the approximation and the power of
the test. A good rule of thumb, however, is that chi-square analysis should be used only when
the theoretical frequencies in each cell exceed five; otherwise, the distribution in Table A.3 will
not be a good approximation. Pragmatically, the "risk" is that with a theoretical frequency less
than five, a single cell's chi-square value may be unusually high and, thus, unduly influence the
overall value.
Figure 11.9 Chi Square Distribution
284
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Univariate Analysis: Test of a Proportion
The standard normal distribution may be used to test not only means, as explained above,
but also differences in proportions. The univariate test of proportions, like the univariate test of
means, compares the population proportion to the proportion observed in the sample. For a
sample proportion, p,
Where sp, the estimated standard error of the proportion,
z = standard normal value
p = the sample proportion of successes
q = (1-p) = the sample proportion of failures
n = sample size
In a simple example, suppose the marketing manager of a snack food company is
evaluating a new snack. 225 respondents are surveyed at a local shopping mall. The survey
indicates that 87% are favorable toward the snack. The manager needs a 90% favorability rate.
Is it safe to say that this is simply sampling variation?
given α = .05 with z = 1.96, we cannot reject the null hypothesis.
CI = .90 - 1.96(.022) < 90 < 90 + 1.96(.022)
CI = .856 < .90 < .944
In this example the manager could (in a statistical sense) claim that 87% approval is no
different that 90% approval with sampling variation. In terms of corporate policy, however, set
cut points may be more rigidly held.
SUMMARY
Chapter 11 has introduced the basic concepts of formulating hypothesis testing and
making statistical inference in the context of univariate analysis. In actual research, the analyst
may alternate between analyzing the data and formulating hypotheses.
A hypothesis is a statement that variables (measured constructs) are related in a specific
way. The null hypothesis, H0, is a statement that no relationship exists between the variables
tested or that there is no difference.
Statistics are based on making inferences from the sample of respondents to the
population of all respondents by means of a sampling distribution. The sampling distribution is a
distribution of the parameter values (means or variances) that are estimated when all possible
samples are collected.
285
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
When testing hypotheses, the analyst may correctly identify a relationship as present or
absent, or may commit one of two types of errors. A Type I Error occurs when a true H0 is
rejected (there is no difference, but we find there is). A Type II Error occurs when we accept a
false H0 (there is a difference, but we find that none exists). The power of a test was explained
as the ability to reject H0 when it should be rejected.
Selecting the appropriate statistical technique for investigating a given relationship
depends of the level of measurement (nominal, ordinal, interval) and the number of variables to
be analyzed. The choice of parametric vs. non-parametric analyses depends on the analyst's
willingness to accept the distributional assumptions of normality and homogeneity of variances.
Finally, univariate hypothesis testing was demonstrated using the standard normal
distribution statistic (z) to compare a mean and proportion to the population values. The t-test
was demonstrated as a parametric test for populations of unknown variance and samples of small
size. The chi-square goodness of fit test was demonstrated as a non-parametric test of nominal
data that make no distributional assumptions. The observed frequencies were compared to an
expected distribution.
In this chapter, we focused on univariate analyses. Chapter 12 expands this discussion to
include non-parametric and parametric analyses involving two variables. In Chapter 12, we also
focus on bivariate analyses involving Measures of Association. Finally, in Chapters 13 and 14,
we focus on multivariate analyses, involving dependence, and interdependence of associative
data.
In each chapter we will observe that the appropriate statistical technique is selected, in
part based on the number of variable and their relationship, when included in the analysis.
However, without higher levels of measurement and their associated characteristics of central
tendency dispersion and rates of change, the ability to investigate these more complex
relationships is not possible.
286
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
Andrews, F.M., Klem, L., Davidson, T.N., O’Malley, P.M., and Rodgers, W.L. (1981), A Guide
for Selecting Statistical Techniques for Analyzing Social Science Data. Ann Arbor, MI: Institute
for Social Research, University of Michigan, Second Edition
Fink, A. (2003). How to Manage, Analyze, and Interpret Survey Data. Thousand Oaks, CA: Sage
Publications, Second Edition
Gibbons, J.D. (1993). Nonparametric Statistics: An Introduction. Newbury Park, CA: Sage
Publications
Labovitz, S. (1968). “Criteria for Selecting A Significance Level: A Note on the Sacredness of
.05,” The American Sociologist, 3 (August), 220-222.
Lewis-Beck, M. S. (1995). Data analysis: An introduction. Thousand Oaks, CA: Sage.
Mohr, L.B. (1990). Understanding Significance Testing. Newbury Park, CA: Sage Publications
Pyrczak, F. (1995). Making Sense of Statistics: A Conceptual Overview. Los Angeles: Pyrczak
Publishing
Selvin, H. and Stuart, A. (1966). “Data-Dredging Procedures in Survey Analysis,” The American
Statistician, June, 20-23
Semon, T.T. (2000). “When Determining Significance, Let Clients Decide the Risks,” Marketing
News October 9, 15
Semon, T.T. (1999). “Consider A Statistical Insignificance Test,” Marketing News, 33, 3
(February 1), 9
Semon, T.T. (1990). :Keep Ignoring Statistical Power at Your Own Risk,” Marketing news, 24,
181 (September 30), 21
287
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 12
BIVARIATEDATA ANALYSIS
Analysis can be viewed as the categorization, aggregation, and the manipulation of data
to obtain answers to research question underlying the research project. The final aspect of
analysis is interpretation. The process of interpretation involves taking the results of
analysis, making inferences relevant to the research relationships studied, and drawing
managerially useful conclusions about these relationships.
BASIC CONCEPTSOF ANALYZING ASSOCIATIVEDATA
Our chapter begins with a brief discussion of cross-tabulation to mark the
beginning of a major topic of this chapter—the analysis of associative data. A large part
of the rest of the chapter will focus on methods for analyzing how the variation of one
variable is associated with variation in other variables.
The computation of row or column percentages in the presentation of crosstabulations
is taken up first. We then show how various insights can be obtained as one
goes beyond two variables in a cross-tabulation to three (or more) variables. In particular,
examples are presented of how the introduction of a third variable can often refine or
explain the observed association between the first two variables.
Bivariate Cross-Tabulation
Bivariate cross-tabulation represents the simplest form of associative data
analysis. At the minimum we can start out with only two variables, such as occupation
and education, each of which has a discrete set of exclusive and exhaustive categories.
Data of this type is called categorical, since each variable is assumed to be nominalscaled.
Bivariate cross-tabulation is widely used in marketing applications to analyze
variables at all levels of measurement. In fact, it is the single most widely used bivariate
technique in applied settings. Reasons for the continued popularity of bivariate crosstabulation
include the following (Feick, 1984, p. 376):
1. It provides a means of data display and analysis that is clearly interpretable even
to the less statistically inclined researcher or manager.
2. A series of bivariate tabulations can provide clear insights into complex
marketing phenomena that might be lost in a simultaneous analysis with many
variables.
3. The clarity of interpretation affords a more readily constructed link between
market research and market action.
4. Bivariate cross-tabulations may lessen the problems of sparse cell values that
often plague the interpretation of discrete multivariate analyses (bivariate crosstabulations
require the expected number of respondents in any table cell be 5).
288
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The entities being cross-classified are usually people, objects, or events. The
cross-tabulation, at its simplest, consists of a simple count of the number of entities that
fall into each of the possible categories of the cross-classification. Excellent discussions
of ways to analyze cross-tabulations can be found in Hellevik (1984) and Zeisel (1957).
However, we usually want to do more than show the raw frequency data. At the
very least, row or column percentages (or both) are usually computed.
Percentages
The simple mechanics of calculating percentages are known to all of us. In crosstabulation,
percentages serve as a relative measure indicating the relative size of two or
more categories.
The ease and simplicity of calculation, the general understanding of its purpose,
and the near universal applicability have made the percent statistic, or its counterpart the
proportion, the most widely used statistical tool in marketing research. Yet its simplicity
of calculation is sometimes deceptive, and the understanding of its purpose is frequently
insufficient to ensure sound application and interpretation. The result is that the percent
statistic is often the source of misrepresentations, either inadvertent or intentional.
The sources of problems in using percentages are largely in identifying the
direction in which percentages should be computed, and in knowing how to interpret
percentage of change.
Both these problems can be illustrated by a small numerical example that uses a
before and after with control experimental design. Let us assume that KEN’s Original, a
small regional manufacturer of salad dressings, is interested in testing the effectiveness of
spot TV ads in increasing consumer awareness of a new brand—called Life. Two
geographic areas are chosen for the test: (1) test area A and (2) control area B. The test
area receives five 15-second television spots per week over an eight-week period,
whereas the control area receives no spot TV ads at all. (Other forms of advertising were
equal in each area.)
Assume that four independent samples (of telephone interviews) were conducted
as before and after tests in each of the areas. Respondents were asked to state all the
brands of salad dressing they could think of, on an aided basis. If Life was mentioned, it
was assumed that this constituted consumer awareness of the brand. However, as it
turned out, sample sizes differed across all four sets of interviews. This common
occurrence in surveys (i.e., variation in sample sizes) increases the value of computing
percentages.
Table 12.1 shows a set of frequency tables that were compiled before and after a
TV ad for Life Salad Dressing was aired. Interpretation of Table 12.1 would be hampered
if the data were expressed as raw frequencies and different percentage bases were
reported. Accordingly, Table 12.1 shows the data, with percentages based on column and
row totals. Can you see why the row percentages are more useful for analytical purposes?
Direction in Which to Compute Percentages
In examining the relationship between two variables, it is often clear from the
context that one variable is more or less the independent or control variable and the other
is the dependent or criterion variable. In cases where this distinction is clear, the rule is to
compare percentages within levels of the dependent variable.
289
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In Table 12.1, the control variable is the experimental area (test versus control)
and the dependent variable is awareness. When comparing awareness in the test and
control areas, row percentages are preferred. We note that before the spot TV campaign
the percentage of respondents who are aware of Life is almost the same between test and
control areas: 42 percent and 40 percent, respectively.
However, after the campaign the test-area awareness level moves up to 66
percent, whereas the control-area awareness (42 percent) stays almost the same. The
small increase of 2 percentage points reflects either sampling variability or the effect of
other factors that might be serving to increase awareness of Life in the control area.
On the other hand, computing percentages across the independent variable
(column percent) makes little sense. We note that 61 percent of the aware group (before
the spot TV campaign) originates from the test area; however, this is mainly a reflection
of the differences in total sample sizes between test and control areas.
After the campaign we note that the percentage of aware respondents in the
control area is only 33 percent, versus 39 percent before the campaign. This may be
erroneously interpreted as indicating that spot TV in the test area depressed awareness in
the control area. But we know this to be false from our earlier examination of raw
percentages.
It is not always the case that one variable is clearly the independent or control
variable and the other is the dependent or criterion variable. This should pose no
particular problem as long as we agree, for analysis purposes, which variable is to be
considered the control variable. Indeed, cases often arise in which each of the variables in
turn serves as the independent and dependent variable.
Table 12.1 Aware of Life Salad Dressing—Before and After Spot TV
Before Spot TV After Spot TV
Aware
Not
Aware Total Area Aware
Not
Aware
Total
Area
Test Area
Freq.
Col %
Row %
250
61%
42%
350
59%
58%
600
60%
Test Area
Freq.
Col %
Row %
330
67%
66%
170
44%
34%
550
57%
Control Area Control Area
Freq.
Col %
Row %
160
39%
40%
240
41%
60%
400
40%
Freq.
Col %
Row %
160
33%
42%
220
56%
58%
380
43%
Total
Before TV Spot
410
41%
590
59%
1000
100%
Total
After TV Spot
490
56%
390
44%
880
100%
Interpretation of the Percentage Change
A second problem that arises in the use of percentages in cross-tabulations is the choice
of which method to use in measuring differences in percentages. There are three principal
ways to portray percentage change:
1. The absolute difference in percentages
2. The relative difference in percentages
3. The percentage of possible change in percentages
The same example can be used to illustrate the three methods.
290
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Absolute Percentage Increase
Table 12.2 shows the percentage of respondents who are aware of Life before and after
the spot TV campaign in the test and control areas. First, we note that the test-area
respondents displayed a greater absolute increase in awareness. The increase for the testarea
respondents was 24 percentage points, whereas the control-area awareness increased
by only 2 percentage points.
Table 12.2 Aware of Life—Percentages Before and After the Spot TV Campaign
Before the
Campaign
After the
Campaign
Test area
Control area
42%
40%
66%
42%
Relative Percentage Increase
The relative increase in percentage is [(66 – 42)]/42] × 100 = 57 percent and [(42
– 40)/40] × 100 = 5 percent, respectively, for test- and control-area respondents.
Percentage Possible Increase
The percentage of possible increase for the test area is computed by first noting
that the maximum percentage-point increase that could have occurred is 100 – 42 = 58
points. The increase actually registered is 24 percentage points, or 100(24/58) = 41
percent of the maximum possible. That of the control area is 100(2/60) = 3 percent of the
maximum possible.
In terms of the illustrative problem, all three methods give consistent results in the
sense that the awareness level in the test area undergoes greater change than that in the
control area. However, in other situations conflicts among the measures may arise.
The absolute-difference method is simple to use and requires only that the
distinction between percentage and percentage points be understood. The relativedifference
method can be misleading, particularly if the base for computing the
percentage change is small. The percentage-of-possible-difference method takes
cognizance of the greater difficulty associated with obtaining increases in awareness as
the difference between potential-level and realized-level decreases. In some studies all
three measures are used, inasmuch as they emphasize different aspects of the relationship.
Introducing a Third Variable into the Analysis
Cross-tabulation analysis to investigate relationships need not stop with two
variables. Often much can be learned about the original two-variable association through
the introduction of a third variable that may refine or explain the original relationship. In
some cases, it may show that the two variables are related even though no apparent
relationship exists before the third variable is introduced. These ideas are most easily
explained by example.
Consider the hypothetical situation facing MCX Company, a specialist in
telecommunications equipment for the residential market. The company has recently testmarketed
a new service for online recording of cable or satellite programs without a
storage box. Several months after the introduction, a telephone survey was taken in which
respondents in the test area were asked whether they had adopted the innovation. The
total number of respondents interviewed was 600 (Table 11.3).
291
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 12.3 Adoption—Percentage by Gender and Age
Men Women
Frequency <35 Yrs ≥35 Yrs Total % <35 Yrs ≥35 Yrs Total %
Adopters
Number of Cases
Column %
Row %
100
50%
62.5%
60
30%
37.5%
160
40%
11
11%
55%
9
9%
45%
20
10%
Non Adopters
Number of Cases
Column %
Row %
100
50%
41.7%
140
70%
58.3%
240
60%
89
89%
49.4%
91
91%
50.6%
180
90%
Total
Percentage
200
50%
200
50%
400
100%
100
50%
100
50%
200
100%
One of the variables of major interest in this study was the age of the respondent.
Based on earlier studies of the residential market, it appeared that adopters of the firm’s
new products tended to be less than 35 years old. Accordingly, the market analyst decides
to cross-tabulate adoption and respondent age. Respondents are classified into the
categories “under 35 years” (<35) and “equal to or greater than 35 years” (>35) and then
cross-classified by adoption or not. Table 12.3 shows the full three-variable crosstabulation.
It seems that the total sample of 600 is split evenly between those who are
under 35 years of age and those who are 35 years of age or older. Younger respondents
display a higher percentage of adoption (37 percent = 100 + 11)/300) than older
respondents (23 percent = (60 + 9)/300).
Analysis and Interpretation
The researcher is primarily interested in whether this finding differs when gender
of the respondent is introduced into the analysis. As it turned out, 400 respondents in the
total sample were men, whereas 200 were women.
Table 12.3 shows the results of introducing gender as a third classificatory
variable. In the case of men, 50 percent of the younger men adopt compared with only 30
percent of the older men. In the case of women, the percentages of adoption are much
closer. Even here, however, younger women show a slightly higher percentage of
adoption (11 percent) than older women (9 percent).
The effect of gender on the original association between adoption and age is to
refine that association without changing its basic character; younger respondents show a
higher incidence of adoption than older respondents. However, what can now be said is
the following: If the respondent is a man, the differential effect of age on adoption is
much more pronounced than if the respondent is a woman.
This pattern is even easier to identify when we show this information graphically
(Figure 12.1). The height of the bars within each rectangle represents the percentage of
respondents who are adopters. The relative width of the bars denotes the relative size of
the categories—men versus women—representing the third variable, gender. The shaded
portions of the bars denote the percentage adopting by gender, the dashed line represents
the weighted average percentage adopting by gender, and the dashed line represents the
weighted average percentage across the genders.
292
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 12.1 Adoption—Percentage by Age and Gender
It is easy to see from the figure that adoption differs by age group (37 percent
versus 23 percent). Furthermore, the size of the difference depends on the gender of the
respondent: Men display a relatively higher rate of adoption, compared with women, in
the younger age category.
Recapitulation
Representatives of three-variable association can involve many possibilities that could be
illustrated by the preceding adoption-age-gender example:
1. In the example presented, adoption and age exhibit initial association; this association
is still maintained in the aggregate but is refined by the introduction of the third variable,
gender.
2. Adoption and age do not appear to be associated. However, adding and controlling on
the third variable, gender, reveals suppressed association between the first two variables
within the separate categories of men and women. In the two-variable cases, men and
women exhibit opposite patterns, canceling each other out.
Although the preceding example was contrived to illustrate the concepts, the
results are not unusual in practice. It goes almost without saying that the introduction of a
third variable can often be useful in the interpretation of two-variable cross-tabulations.
However, the reader should be aware of the fact that we have deliberately used
the phrase associated with rather than caused by. Association of two or more variables
does not imply causation, and this statement is true regardless of our preceding efforts to
refine some observed two-variable association through the introduction of a third
variable.
In principle, of course, we could cross-tabulate four or even more variables with
the possibility of obtaining further insight into lower-order (e.g., two-variable)
associations. However, somewhere along the line, a problem arises in maintaining an
adequate cell size for all categories. Unless sample sizes are extremely large in the
aggregate and the number of categories per variable is relatively small, cross-tabulations
rarely can deal with more than three variables at a time. A further problem, independent
of sample size, concerns the high degree of complexity of interpretation that is introduced
by cross-tabulations involving four or more variables. In practice, most routine
applications of cross-tabulation involve only two variables at a time.
293
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
As noted in Table 12.3, there are definite advantages associated with having a two
category criterion variable, such as adoption versus non-adoption. In many applications,
however, the criterion variable will have more than two categories. Cross-tabulations can
still be prepared in the usual manner, although they become somewhat more tedious to
examine.
BIVARIATEANALYSIS: DIFFERENCEBETWEEN SAMPLEGROUPS
Marketing activities largely focus on the identification and description of market
segments. These segments may be defined demographically, attitudinally, by the quantity
of the product used, by activities participated in or interests, by opinions, or by a
multitude of other measures. The key component of each of these variables is the ability
to group respondents into market segments. Often this segmentation analysis involves the
identification of differences and the asking of questions about the marketing implications
of those differences: Do differences in satisfaction exist for the two or more groups that
are defined by age categories?
Bivariate statistical analysis refers to the analysis of relationships between two
variables. These analyses are often differences between respondent groups. In the
following discussion, we explore bivariate statistical analysis and focus on the twovariable
case as a bridge between the comparatively simple analyses already discussed
and the more sophisticated techniques that will command our attention in later chapters.
We begin with what is perhaps the most used test of market researchers: cross-tabulation.
Next, we consider analysis of differences in group means. First, we discuss the t-test of
differences in means of two independent samples, and then we look at one-way analysis
of variance (ANOVA) that tests for differences in means for k groups. Finally, we
provide a discussion of some of the more widely used nonparametric techniques. These
are but a few of the possible parametric and nonparametric analyses that could be
discussed (see Table 12.4).
Bivariate Cross-Tabulation
The chi-square statistic in a contingency table analysis is used to determine if the
observed associations between the variables in the cross-tabulation are statistically
significant.
Often called chi-square analysis, this technique is used when the data consist of
counts or frequencies within categories of a tabulation or cross-tabulation table. In
conjunction with the cross-tabulation we will introduce the chi-square statistic, χ2
, to
determine the significance of observed association in cross-tabulations involving two or
more variables. This is typically called a X2
test of independence.
Cross-tabulation represents the simplest form of associative data analysis. At the
minimum we can start out with a bivariate cross-tabulation of two variables, such as
occupation and education, each of which identifies a set of exclusive and exhaustive
categories. We know that such data are called categorical because each variable is
assumed to be only nominal-scaled. Bivariate cross-tabulation is widely used in
marketing applications to analyze variables at all levels of measurement. In fact, it is the
single most widely used bivariate technique in applied settings.
294
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 12.4 Selected Nonparametric Statistical Tests for Two-Sample Cases
What type of analysis
do you want?
Level of Measurement
Nominal Ordinal Interval / Ratio
Two-sample related
samples
McNemar test for the
significance of changes
Sign test
Wilcoxon matched-pairs
signed-ranks test
Independent samples Fischer exact probability
test
X
2
test for two
independent samples
Median test
Mann-Whitney U test
Kolmogorov-Smirnov
two-sample test
Wald-Wolfowitz runs test
t-test
One-way ANOVA
In marketing research, observations may be cross-classified, as when we are
interested in testing whether occupational status is associated with brand loyalty.
Suppose, for illustrative purposes, that a marketing researcher has assembled data on
brand loyalty and occupational status—white collar, blue collar, and unemployed or
retired—that describes consumers of a particular product class. The data for our
hypothetical problem appear in Table 12.5.
A total of four columns, known as banner points, are shown. Four rows or stubs
are also shown. Professional cross-tabulation software will output multiple side by side
tables that join multiple variables on the column banner points such that loyalty and
another variable such as usage occasion could be analyzed simultaneously.
In a study of 230 customers, we are interested in determining if occupational
category may be associated with the characteristic loyalty status. The data suggests that a
relationship exists, but is the observed association a reflection of sampling variation, or
can we conclude that a true relationship exists?
Table 12.5 Contingency Table of Observed versus Theoretical Frequencies
N, Expected n
X2
Contribution
Occupation Highly Loyal Moderately Loyal Brand Switchers
Total Number
(percent
distribution)
White-collar
Blue Collar
Unemployed/
retired
Total
30 (30.5)
.01
14 (22.1)
2.93
34 (25.4)
2.88
78 (33.9%)
42 (34.1)
1.86
20(24.5)
.86
25 (28.4)
.40
87 (37.8%)
18 (25.4)
2.17
31 (18.4)
8.68
16 (21.2)
1.27
65 (28.3%)
90 (39.1%)
65 (28.3%)
75 (32.6%)
230 x2
=21.08
In analyzing the problem by means of chi-square, we make use of the marginal
totals (column and row totals) in computing theoretical frequencies given that we
hypothesize independence (no relationship) between the attributes loyalty status and
occupational status. For example, we note from Table 12.5 that 33.9 percent (78/320) of
the respondents are highly loyal. If possession of this characteristic is independent of
occupational status, we would expect that 33.9 percent (78/320) of the 90 respondents
classified as white-collar workers (i.e., 30.5) would be highly loyal. Similarly, 37.8
295
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
percent (87/320) of the 90 = 34.1 would be moderately loyal, and 28.3 percent (65/320)
of the 90 = 25.4 would be brand switchers. In a similar fashion we can compute
theoretical frequencies for each cell on the null hypothesis that loyalty status is
statistically independent of occupational status. (It should be noted that the frequencies
are the same, whether we start with the percentage of the row or the percentage of the
column.)
The theoretical frequencies (under the null hypothesis) are computed and appear
in parentheses in Table 12.5. The chi-square statistic is then calculated (and shown in the
table) for each of the data cells in the table using the observed and theoretical
frequencies:
where fi = actual observed frequency, Fi = theoretical expected frequency, and k = number
of cells (r x c).
The appropriate number of degrees of freedom to use in this example is four. In
general, if we have R rows and C columns, the degrees of freedom associated with the
chi-square statistic are equal to the product
If we use a significance level of 0.05, the probability table value of chi-square is
11.488. Hence, because the computed χ2
of 21.08 is greater than the table value of 11.488,
we reject the null hypothesis of independence between the characteristics loyalty status
and occupational status.
A correction factor must be applied to the formula for chi-square when the
number of observations in a cell is less than 10 (or where a 2 × 2 contingency table is
involved),. The numerator within the summation sign becomes ( | fi − Fi | − 1/2)2
where
the value 1/2 is the Yates continuity correction. This correction factor adjusts for the use
of a continuous distribution to estimate probability in a discrete distribution.
Chi-square analysis of independence can be extended to deal with more than two
variables. No new principles are involved. Three characteristics of the technique should
be borne in mind, however. First, chi-square analysis deals with counts (frequencies) of
data. If the data are expressed in percentage form, they should be converted to absolute
frequencies. Second, the technique assumes that the observations are drawn
independently. Third, the chi-square statistic cannot describe the relationship; it only
gauges its statistical significance, regardless of logic or sense (Semon, 1999).
To assess the nature of the relationship, the researcher must look at the table and
indicate how the variables appear to be related—a type of eyeball approach. This may
involve examining any of the following combinations: (a) the variable combinations that
produce large χ2
values in the cells; (b) those with a large difference between the
observed and expected frequencies; or (c) those where the cell frequency count,
expressed as a percentage of the row total, is most different from the total column
percentage (marginal column %). When variables are ordered or loosely ordered, a
pattern can sometimes be observed by marking cells with higher than expected observed
frequencies with a (+) and those with lower than expected observed frequencies with a (–
), or even graphing the deviations from expected values, or cell X2
values using a 3dimensional
contour map.
296
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Bivariate Analysis: Differences in Means and Proportions
A great amount of marketing research is concerned with estimating parameters of
one or more populations. In addition, many studies go beyond estimation and compare
such population parameters by testing hypotheses about differences between them.
Means, proportions, and variances are often the summary measures of concern. Our
concern at this point is with differences in means and proportions. Direct comparisons of
variances are a special case of the more general technique of analysis of variance, which
is covered later in this chapter.
Standard Error of Differences
Here we extend the topic of sampling distributions and standard errors as they
apply to a single statistic - to cover differences in statistics and show a traditional
hypothesis for differences.
Standard Error of Difference of Means
For two samples, A and B, that are independent and randomly selected, the
standard error of the differences in means is calculated by
This estimate of the standard error is appropriate for use in the denominator of the
z test formula. If the population standard deviations, σi, are not known, then the estimated
standard error becomes
For relatively small samples the correction factor Ni/ni–1 is used and the resulting
formula for the estimated standard error is
Of course, these formulas would be appropriate for use in the denominator of the
t-test.
Standard Error of Differencesof Proportions
Turning now to proportions, the derivation of the standard error of the differences
is somewhat similar. Specifically, for large samples,
For small samples, the correction factor is applied, resulting in
297
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This estimate would again be appropriate for use in the denominator of the Z-test
of proportions.
Testing of Hypotheses
When applying the standard error formulas for hypotheses testing concerning
parameters, the following conditions must be met:
1. Samples must be independent.
2. Individual items in samples must be drawn in a random manner.
3. The population being sampled must be normally distributed (or the sample
must be of sufficiently large size).
4. For small samples, the population variances must be equal.
5. The data must be at least interval scaled.
When these five conditions are met, or can at least be reasonably assumed to
exist, the traditional approach is as follows.
1. The null hypothesis (H0) is specified such that there is no difference between
the parameters of interest in the two populations (e.g., H0: µA − µB = 0); any
observed difference is assumed to occur solely because of sampling variation.
2. The alpha risk is established (α= .05 or other value).
3. A Z value is calculated by the appropriate adaptation of the Z formula. For
testing the difference between two means, Z is calculated in the following
way:
and for proportions
For unknown population variance and small samples, the student t distribution must be
used, and for a test of the differences in means, t is calculated from
4. The probability of the observed difference of the two sample statistics having
occurred by chance is determined from a table of the normal distribution (or
the t distribution, interpreted with [nA + nB – 2] degrees of freedom).
5. If the probability of the observed difference having occurred by chance is
greater than the alpha risk, the null hypothesis of no difference is not rejected;
it is concluded that the parameters of the two universes are not significantly
different. If the probability of the observed difference having occurred by
chance is less than the alpha risk, the null hypothesis is rejected; it is
concluded that the parameters of the two populations differ significantly. In an
applied setting, there are times when the level at which significance occurs
(the alpha level) is reported and management decides whether to accept or
reject.
298
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
An example will illustrate the application of this procedure. Let us assume we
have conducted a survey of detergent and paper goods purchases from supermarkets
among urban (population A) and rural (population B) families (see Table 12.6).
Table 12.6 Sample Group and Average Expenditure
Family Type Sample
Average
Amount Spent
Standard
Deviation
Urban
(Sample A)
Rural
(Sample B)
NA = 400
NB = 225
A - $32.00
B = $30.80
SA = $10.00
SB = $9.00
The question facing the researcher is the following: “Do urban families spend
more on these items, or is the $1.20 difference in means caused by sampling variations?”
We proceed as follows. The hypothesis of no difference in means is established. We
assume the alpha risk is set at .05. Since a large sample test is called for, the Z value is
calculated using the separate variances estimate of the standard error of differences in
means:
The Z value is then determined to be
The probability of the observed difference in the sample means having been due
to sampling is specified by finding the area under the normal curve that falls to the right
of the point Z = + 1.54. Consulting a table of the Cumulative Normal Distribution, we
find this area to be 1.0 − .9382 = 0.0618. Since this probability associated with the
observed difference ( p = 0.06) is greater than the preset alpha, a strict interpretation
would be that there is no difference between the two types of families concerning the
average expenditure on nonfood items. In a decision setting, however, the manager would
have to determine whether this probability (0.06) is low enough to conclude, on
pragmatic grounds, that the families do not differ in their behavior.
As stated previously, and discussed in the previous chapter, often there is no
preset alpha and decisional considerations require that the manager determines the
meaning of the reported alpha.
To illustrate the small sample case, let us assume that we obtain the same mean
values and get values for sA and sB such that sxA -sxB =$0.78 from samples nA =15 and nB
=12. With these data we calculate t as follows:
299
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The critical value of t is obtained from a table of percentiles for the t distribution.
For, say, α= .05, we determine the critical value of t for (nA nB − 2) = 25 degrees of
freedom to be 1.708 (one-tailed test).
Since the calculated t of 1.54 < 1.708, we cannot reject the hypothesis of no difference in
average amount spent by the two types of families. This result is shown in Figure 12.4.
When samples are not independent, the same general procedure is followed. The
formulas for calculating the test statistics differ, however.
Figure 12.2 t-Distribution
Testing the Means of Two Groups: The Independent Samples t-Test
The t-distribution revolutionized statistics and the ability to work with small
samples. Prior statistical work was based largely on the value of Z, which was used to
designate a point on the normal distribution where population parameters were known.
For most market research applications it is difficult to justify the Z-test’s assumed
knowledge of µ and σ. The t-test relaxes the rigid assumptions of the Z-test by focusing
on sample means and variances (X and s). The t-test is a widely used market research
statistic to test for differences between two groups of respondents.
In the previous section, the t statistic was described. Most computer programs
recognize two versions of this statistic. In the previous section, we presented what is
called the separate variance estimate formula:
This formula is appropriate where differences in large samples are tested.
The second method, called the pooled variance estimate, computes an average for
the samples that is used in the denominator of the statistic:
The pooling of variances is a simple averaging of variances that is required when
(1) testing for the same population proportion in two populations or (2) testing the
difference in means between two small samples.
Table 12.7 shows a sample output of t-tests from SPSS. Two respondent groups
were identified for the supermarket study:
300
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
1. Respondents who were males
2. Respondents who were females
The analysis by gender is shown in Part A of the output for the attitudinal
dimensions, friendliness of clerks and helpfulness of clerks. It will be noted that the
sample sizes are approximately equal. To show a contrast we have included in Part B of
the output a t-test from another study where the number of males and females varies
widely. In all three examples, the differences between separate-variance and pooledvariance
analysis are very small. Table 12.9 also shows an F statistic. This is Levene’s
test for equality of variances. Using this test the researcher knows which t-test output to
use—equal variances assumed or equal variances not assumed.
Table 12.7 Selected Output From PASW (SPSS) t-Test
Testing of Multiple Group Means: Analysis of Variance
Analysis of variance (ANOVA) is a logical extension of the independent groups ttest
methodology. Rather than test differences between two group means, we test the
overall difference in k group means, where the k groups are thought of as levels of a
treatment or control variable(s) or factor(s). ANOVA is a general set of methodologies
that handle many different types of research and experimental designs. Traditional use of
analysis of variance has been to isolate the effects of experimental variables that are
related in various experimental situations. The respondent receives a combination of the
treatment levels from the different factors and the response is measured. More
specifically, ANOVA is used to test the statistical significance of differences in mean
responses given the introduction of one or more treatment effects.
Much experimental work has been conducted in medicine and agriculture. In
pharmaceutical drug testing, positive and negative effects of dosage and formulation are
measured over time and for different types of ailments and patient illness levels. The
variables influencing the results are called experimental factors. In agriculture, crop
yields are measured for plots of land, each of which receives a different treatment level
301
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
that is defined by a factor or control variable. Control variables in this application might
include seed type, fertilizer type, fertilizer dosage, temperature, moisture, and many other
variables thought to influence production. In each of these plots, the average yield is
measured and analyzed to determine the effect of the specific measured levels of the
factors being evaluated. Marketing research experiments have control variables that are
certainly different from agricultural experiments, but the principles are the same.
The proper planning and design of the relationships between the experimental
factors may require the use of experimental design methodology such as the completely
randomized, randomized block, Latin square, and factorial designs. Here we discuss the
two most basic forms of the ANOVA methodology: the one-way ANOVA and the twofactor
ANOVA designs (see Exhibit 12.1).
EXHIBIT 12.1 ANOVA Designs
Example: It is well known that interest ratings for TV ads are related to the advertising message.
A simple one-way ANOVA to investigate this relationship might compare three messages:
Advertising message A
A
Advertising message B
B
Advertising message
C
C
Two-factor ANOVA includes a second factor, possibly the type of advertisement (magazine or
TV):
Message (A) Message (B) Message (C)
Magazine Ad (M) MA MB MC
TV Ad (T) TA TB TC
Each of the cells in this matrix would contain an average interest rating for the measures taken for
the particular combination of message and media.
This brief introduction to the idea behind an ANOVA will hopefully reduce the
impression that the technique is used to test for significant differences among the
variances of two or more sample universes. This is not strictly the case. ANOVA is used
to test the statistical significance of differences in mean responses given the introduction
of one or more treatments effects.
The ANOVA Methodology
The appropriateness of the title analysis of variance comes from the method’s
ability to explain the variation in responses to the various treatment combinations. The
methodology for explaining this variation is explained in Exhibit 12.2, which presents an
example regarding responses to messages.
302
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 12.2 Example of ANOVA Methodology
Using the single-factor ANOVA design for the problem described in Exhibit 12.1, the actual
interest rating data for 12 respondents (12 = 4 respondents for each of the 3 treatments) appears in
tabular and graphical format as follows:
It is apparent that the messages differ in terms of their interest ratings, but how do we
perform ANOVA to determine if these differences are different statistically? There are three
values that must be computed in order to analyze this pattern of values.
1. Compute the total sum of squares: The grand mean of the 12 observations is computed,
followed by the variance of the individual observations from this mean.
Total Sum of Squares:
Grand mean = 4.0 = G
Computation of
Total Sum of Squares
Σ( ij - G ) =
(6-4)
2
+ (4-4)
2
+ … + (1-4)
2
= 110
2. Between-treatment sum of squares: The means of the factor levels (messages A, B, and C) are
computed, followed by the deviation of the factor level means from the overall mean, weighted
by the number of observations Σ n( j - )2
3. Compute the within-treatment sum of squares: The means of the factor levels are computed,
followed by the deviation of the observations within each factor level from that factor level mean.
303
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Thus, an observation may be decomposed into three terms that are additive, and each
explains a particular type of variance:
The overall mean is constant and common to all observations: The deviation of a group
mean from the overall mean represents the effect on each observation of belonging to that
particular group; the deviation of an observation from its group mean represents the effect on that
observation of all variables other than the group variable.
The basic idea of ANOVA is to compare the between-treatment-groups sum of
squares (after dividing by degrees of freedom to get a mean square) with the withintreatment-group
sum of squares (also divided by the appropriate number of degrees of
freedom). This is the F statistic that indicates the strength of the grouping factor.
Conceptually,
The larger the ratio of between to within, the more we are inclined to reject the
null hypothesis that the group mean µ1 = µ2= µ3. Conversely, if the three sample means
were very close to each other, the between-samples sum of squares would be close to
zero and we would conclude that the population means are the same, once we consider
the variability of individual cases within each sample group.
304
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
However, to make this comparison, it is necessary to assume that the error-term
distribution has constant variance over all observations. This is exactly the same
assumption as was made for the t-test.
In the next section we shall (a) use more efficient computational techniques, (b)
consider the adjustment for degrees of freedom to obtain mean squares, and (c) show the
case of the F ratio in testing significance. Still, the foregoing remarks represent the basic
ANOVA idea for comparing between- with within-sample variability.
One-Way (Single Factor) Analysis of Variance
One-way ANOVA is analysis of variance in its simplest (single-factor) form.
Suppose a new product manager for the hypothetical Friskie Corp. is interested in the
effect of shelf height on supermarket sales of canned dog food. The product manager has
been able to secure the cooperation of a store manager to run an experiment involving
three levels of shelf height (knee level, waist level, and eye level) on sales of a single
brand of dog food, which we shall call Snoopy. Assume further that our experiment must
be conducted in a single supermarket and that our response variable will be sales, in cans,
of Snoopy dog food for some appropriate unit of time. But what shall we use for our unit
of time? Sales of dog food in a single store may exhibit week-to-week variation, day-today
variation, and even hour-to-hour variation. In addition, sales of this particular brand
may be influenced by the price or special promotions of competitive brands, the store
management’s knowledge that an experiment is going on, and other variables that we
cannot control at all or would find too costly to control.
Assume that we have agreed to change the shelf-height position of Snoopy three
times per day and run the experiment over eight days. We shall fill the remaining sections
of our gondola with a filler brand, which is not familiar to customers in the geographical
area in which the test is being conducted. Furthermore, since our primary emphasis is on
explaining the technique of analysis of variance in its simplest form (analysis of one
factor: shelf-height), we shall assign the shelf heights at random over the three time
periods per day and not design an experiment to explicitly control and test for within-day
and between-day differences. Our experimental results are shown in Table 12.8.
Here, we let Xij denote the sales (in units) of Snoopy during the ith day under the
jth treatment level. If we look at mean sales by each level of shelf height, it appears as
though the waist-level treatment, the average response to which is X2 = 90.9, results in the
highest mean sales over the experimental period. However, we note that the last
observation (93) under the eye-level treatment exceeds the waist-level treatment mean. Is
this a fluke observation? We know that these means are, after all, sample means, and our
interest lies in whether the three population means from which the samples are drawn are
equal or not.
Now we shall show what happens when one goes through a typical one-way
analysis-of variance computation for this problem.
305
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 12.8 Sales of Snoopy Dog Food (in Units) by Level of Shelf Height
Table 12.9 shows the calculations for the among-treatments, within-treatments,
and total sums of squares, the mean squares, and the F ratio. Had the experimenter used
an alpha risk of 0.01, the null hypothesis of no differences among treatment levels would
have been rejected. A table of F ratios is found in most statistics and research texts.
Table 12.9 Analysis of Variance – Snoopy Dog Food Experiment
Note that Table 12.9 shows shortcut procedures for finding each sum of squares.
For example, the total sum of squares is given by
306
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This is the same quantity that would be obtained by subtracting the grand mean of
85.5 from each original observation, squaring the result, and adding up the 24 squared
deviations. This mean-corrected sum of squares is equivalent to the type of formula used
earlier in this chapter.
The interpretation of this analysis is, like the t-test, a process of comparing the F
value of 14.6 (2, 21 df) with the table value of 4.32 (p = .05). Because (14.6 > 4.32), we
reject the null hypothesis that treatments 1, 2, and 3 have equivalent appeals.
The important consideration to remember is that, aside from the statistical
assumptions underlying the analysis of variance, the variance of the error distribution
will markedly influence the significance of the results. That is, if the variance is large
relative to differences among treatments, then the true effects may be swamped, leading
to an acceptance of the null hypothesis when it is false. As we know, an increase in
sample size can reduce this experimental error. Though beyond the scope of this chapter,
specialized experimental designs are available, the objectives of which are to increase the
efficiency of the experiment by reducing the error variance.
Follow-up Tests of Treatment Differences
The question that now must be answered is the following: Which treatments
differ? The F-ratio only provides information that differences exist. The question of
where differences exist is answered by follow-up analyses, usually a series of
independent samples t-tests, that compare the treatment level combinations ((1,2), (1,3),
and (2,3)). Because of our previous discussion of the t-test, we will not discuss these tests
in detail. We will only allude to the fact that there are various forms of the t-statistic that
may be used when conducting a series of two group tests. These test statistics (which
include techniques known as the LSD (Least Significant Difference), Bonnferoni’s test,
Duncan’s multiple range tests, Scheffe’s test, and others) control the cumulative
probability that a Type I error will occur when a series of statistical tests are conducted.
Recall that if in a series of statistical tests, each test has a .05 probability of a Type I
error, then in a series of 20 such tests we would expect one (20 * .05 = 1) of these tests
would report a significant difference that did not exist (Type I error). These tests typically
are options provided by the standard statistical packages, such as the PASW (SPSS)
program Oneway.
BIVARIATEANALYSIS: MEASURESOF ASSOCIATION
Bivariate measures of association include the two-variable case in which both
variables are interval or ratio scaled. Our concern is with the nature of the associations
between the two variables and the use of these associations in making predictions.
Correlation Analysis
When referring to a simple two-variable correlation, we refer to the strength and
direction of the relationship between the two variables. As an initial step in studying the
relationship between the X and Y variables, it is often helpful to graph this relationship in
a scatter diagram (also known as an X-Y plot). Each point on the graph represents the
appropriate combination of scale values for the associated X and Y variables, as shown in
Figure 12.3.
307
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The objective of correlation analysis, then, is to obtain a measure of the degree of
linear association (correlation) that exists between the two variables. The Pearson
correlation coefficient is commonly used this purpose and is defined by the formula
where n pairs of (Xi, Yi) values provide a sample size n, and X, Y, SX, and sY represent the
sample means and sample standard deviations of the X and Y variables. The values of the
correlation coefficient may range from +1 to –1.
These extreme values indicate perfect positive and negative linear correlations.
Other relationships may appear to be curvilinear or even random plots and have
coefficients between zero and the extreme values.
The alternate formulation shows the correlation coefficient to be the product of
the Z scores for the X and Y variables. In this method of computing the correlation
coefficient, the first step is to convert the raw data to a Z score by finding the deviation
from the respective sample mean. The Z scores will be centered as a normally distributed
variable (mean of zero and standard deviation of one).
The transformation of the X and Y variables to Z scores means that the scale measuring
the original variable is no longer relevant, as a Z-score variable originally measured in
dollars can be correlated with another Z-score variable originally measured on a
satisfaction scale. The original metric scales are replaced by a new abstract scale (called
correlation) that is the product of the two Z distributions.
Figure 12.3 Scatter Diagrams
308
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Alternative formula:
Where
By continuing our digression one step further, we can show how the correlation
coefficient becomes positive or negative.
We know that the ZX and ZY values will generally fall in the range –3 to +3. When
both ZX and ZY are positive or both are negative, rXY is positive, as shown in Figure 12.6.
When ZX is positive and ZY negative (or the opposite), a negative correlation will exist. Of
course, we are talking of the individual respondent’s pairs of the ZX and ZY variables,
which when summed produce the overall correlation coefficient.
Figure 12.4 Bivariate Products of Standard Z-Scores
To summarize, the sign (+ or −) indicates the direction of the correlation and the
absolute value of the coefficient indicates the degree of correlation (from 0 to 1). Thus,
correlations of + .7 and − .7 are of exactly the same strength, but the relationships are in
opposite directions.
We are cautioned that near-perfect positive or negative linear correlations do not
mean causality. Often, other factors underlie and are even responsible for the relationship
between the variables. For example, at one time Kraft Foods reported that sales of its
macaroni and cheese product were highly correlated (negatively) with national indices of
the health of the economy. We may not imply that Kraft sales directly caused fluctuations
in the national economy, or vice versa. Consumer expectations and possibly personal
income vary as a function of the national economy. In times of reduced family income,
macaroni and cheese is a low-price dietary substitute for more expensive meals.
To demonstrate, we will consider a brief example of a correlation analysis that
examines the relationships between (1) family income and (2) family consumption
expenditures. The data and plot appear in Figure 12.5.
309
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In order to calculate the correlation coefficient, we will reduce the previously
identified alternative formula to the basic computational formula consisting of sums for
the X and Y variables. This equation looks formidable but allows for easy computation by
simply entering the appropriate summation values from the bottom of Table 12.10.
Figure 12.5 Consumption Expenditure and Income Data
Table 12.10 Family Income and Family Consumption Expenditures
Needless to say, hand computations such as this are rarely done today.
Researchers routinely perform their analyses using Excel spreadsheets or computer
packages. However, our brief discussion is included to provide understanding of
underlying processes and the computational formula:
=
Introduction to Bivariate Regression Analysis
In the analysis of associative data the marketing researcher is almost always
interested in problems of predictions:
310
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Can we predict a person’s weekly fast food and restaurant food purchases from
that person’s gender, age, income, or education level?
• Can we predict the dollar volume of purchase of our new product by industrial
purchasing agents as a function of our relative price, delivery schedules, product
quality, and technical service?
The list of such problems is almost endless. Not surprisingly, the linear regression
model—as applied in either the bivariate (single predictor) or multivariate form (multiple
predictors)—is one of the most popular methods in the marketing researcher’s tool kit.
The bivariate form is also known as simple regression.
The regression model has been applied to problems ranging from estimating sales
quotas to predicting demand for new shopping centers. As an illustration, one of the
leading ski resorts in the United States used a regression model to predict the weekend
ticket sales, based on variables including the following:
• Highway driving conditions
• Average temperature in the three-day period preceding the weekend
• Local weather forecast for the weekend
• Amount of newspaper space devoted to the resort’s advertisements in the
surrounding city newspapers
• A moving average of the three preceding weekends’ ticket sales
The model’s accuracy was within ± 6 percent of actual attendance throughout the season.
Regression analysis in its simplest bivariate form involves a single dependent
(criterion) and a single independent (predictor) variable. In its more advanced multipleregression
form, a set of predictors are used to form an additive linear combination that
predicts the single dependent variable. In considering the use of either simple or multiple
regressions, the researcher is interested in four main questions:
1. Can we find a predictor variable (or a linear composite of the predictor
variables in the multiple case) that will parsimoniously express the relationship
between a criterion variable and the predictor (set of predictors)?
2. If we can, how strong is the relationship; that is, how accurately can we predict
values of the criterion variable from values of the predictor (linear composite)?
3. Is the overall relationship statistically significant?
4. Which predictor is most important in accounting for variation in the criterion
variable? (Can the original model be reduced to fewer variables, but still provide
adequate prediction of the criterion?)
The basic ideas of bivariate regression are most easily explained by a numerical
example. We proceed one step at a time.
311
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Suppose that a marketing researcher is interested in consumers' attitudes toward
nutritional additives in ready-to-eat cereals. Specifically, a set of written concept
descriptions of a children's cereal is prepared that vary on
X1: the amount of protein (in grams) per 2-ounce serving.
The researcher obtains consumers' interval-scaled evaluations of ten concept
descriptions using a preference rating scale that ranges from 1, dislike extremely, up to 9,
like extremely well.
TABLE12.11 Preference Ratings of Ten Cereal Concepts Varying in Protein
(1) (2) (3) (4) (5) (6)
Consumer Preference Rating,Y Protein, X1 Y
2
X YX1
1
2
3
4
5
6
7
8
9
10
3
7
2
1
6
2
8
3
9
2
4
9
3
1
3
4
7
3
8
1
9
49
4
1
36
4
64
9
81
4
16
81
9
1
9
16
49
9
64
1
12
63
6
1
18
8
56
9
72
2
Total
Mean
Std. Dev.
43
4.3
2.908
43
4.3
2.791
261 255 247
Figure 12.6 Scatter Diagram and Least-Squares Regression Line--Preference Rating versus
Grams of Protein
One of the first things that is usually done in examining two-variable relationships
is to prepare a scatter diagram in which the ten values of Y are plotted against their X1
counterparts. Figure 12.6 shows this plot. It appears that there is a direct relationship
between Y and X1. Moreover, it would seem that a linear or straight-line relationship
might be an appropriate model for describing the functional form. A scatter diagram is a
useful tool for aiding in model specification. The value of using a scatter diagram is
illustrated in Exhibit 12.3.
312
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Exhibit 12.3 Look at Your Data Before You Analyze
In deciding which type of regression approach to use, it is important that the
researcher know the shape of the interrelationship. The shape of the interrelationship is
easy to see on a scatter diagram. Looking at this visually helps decide whether the
relationship is, or approximates being, linear or whether it has some other shape which
would require a transformation of the data—by converting to square roots or logarithms--
or treatment as nonlinear regression (Semon, 1993).
Examination of the data by scatter diagrams also allows the researcher to see if
there are any “outliers”—i.e., cases where the relationship is unusual or extreme as
compared to the majority of the data points. A decision has to be made whether to retain
such outliers in the data set for analysis.
When the regression line itself is included on the scatter diagram, comparisons
between actual values and the values estimated by the regression formula can be
compared and used to assess the estimating error. Of course, what the analyst is seeking
is a regression function that has the best fit for the data, and this is typically based on
minimizing the squares of the distances between the actual and estimated values—the socalled
least-squares criterion.
_________________________________________________________________
The equation for a linear model can be written bx+a=Yˆ , where Ŷ denotes
values of the criterion that are predicted by the linear model; a denotes the intercept, or
value of Ŷ when X is zero; and b denotes the slope of the line, or change in Ŷ per unit
change in X.
But how do we find the numerical values of a and b? The method used in this
chapter is known as least squares as discussed in Exhibit 12.3. As the reader will recall
from introductory statistics, the method of least squares finds the line whose sum of
squared differences between the observed values Yi and their estimated counterparts Ŷi
(on the regression line) is a minimum.
Parameter Estimation
To compute the estimated parameters (a and b) of the linear model, we return to
the data of Table 12.11. In the two-variable case, the formulas are relatively simple:
0.886=
)10(4.3-255
3)10(4.3)(4.-247
=
Xn-X
XYn-YX=b
2
22
1
∑
∑
where n is the sample size and Y and X denote the mean of Y and X, respectively.
Having found the slope b, the intercept a is found from
313
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
0.491=
0.886(4.3)-4.3=
Xb-Y=a
leading to the linear function 0.886X+0.491=Yˆ
This function is drawn with the scatter plot of points in Figure 12.6. It appears to
fit the plotted points rather well, and the model seems to be well specified (a linear rather
than arc linear or other form seems to fit).
Assumptions of the Model
Underlying least-squares computations is a set of assumptions. Although leastsquares
regression models do not need to assume normality in the (conditional)
distributions of the criterion variable, this assumption is made when we test the statistical
significance of the contribution of the predictor variable in explaining the variance in the
criterion (does it differ from zero?) With this in mind the assumptions of the regression
model are as follows (the symbols α and β are used to denote population counterparts of a
and b:
1. For each fixed value of X we assume a normal distribution of Y values exists.
Our particular sample we assume that each y value is drawn independently of all
others. What is being described is the “classical” regression model. Modern
versions of the model permit the predictors to be random variables, but their
distribution is not allowed to depend on the parameters of the regression
equation.
2. The means of all of these normal distributions of Y lie on a straight line with
slope β.
3. The normal distributions of Y all have equal variances. This (common) variance
does not depend on values assumed by the variable X.
Because all values rarely fall on the regression line (only when the correlation is
1.0), there is unexplained error in predicting Y. This error is shown in Figure 12.3 as the
difference between the values Yi and the regression line Ŷ. For the population, our
model is expressed algebraically as:
εβα +X+=Y 1
where α = mean of Y population when X1 = 0
β = change in Y population mean per unit change in X1
ε = error term drawn independently from a normally distributed universe with
mean μ(ε); the error term is independent of X1
The nature of these assumptions is apparent in Figure 12.7. The reader should
note that each value of X has associated with it a normal curve for Y (assumption 1). The
means of all these normal distributions lie on the straight line shown in the figure
(assumption 2).
314
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
What if the dependent variable is not continuous? Exhibit 12.4 gives an
alternative when the dependent variable can be viewed as a categorical dichotomous
variable—use logistic regression (also known as logit). The analysis proceeds generally
as we are discussing it—the major change is that a transformation has been applied to the
dependent variable values.
Exhibit 12.4 When to Use Logistic Regression
Data collected for customer satisfaction research provides a good illustration of when the
researcher should consider transformation of data. Typically multi-point rating scales are
used to obtain customer satisfaction data. Many believe that customer satisfaction ratings
obtained on rating scales are not normally distributed, but are skewed toward higher scale
values (Dispensa, 1997). Thus, in practice customers do not really view customer
satisfaction ratings as continuous.
Ultimately, overall a customer is either satisfied or not satisfied. This creates a dichotomous
dependent variable. Typically those customers who rate at the upper end of the scale, say 9
or 10 on a 10-point scale, are considered satisfied while all others are considered to be not
satisfied. If this is so, normal regression analysis is not the proper technique to use as the
dependent variable is binary not continuous.
A binary overall customer satisfaction variable follows the logistic distribution, thus allowing
for the use of logistic regression. With one or more independent variables, this technique
allows a researcher to determine the extent to which an independent variable affects the
prediction of a satisfied customer through the logistic regression coefficients and their
associated log-odds (Dispensa, 1997). Log-odds specify the direct association between the
independent variable and the dependent variable. In addition, logistic regression calculates
the probability of each customer being satisfied or not.
Typically, logistic regression is used for multiple regression situations where there are two or
more independent variables. But, it is suitable for bivariate situations as well. The key to its
being of value is the nature of the dependent variable, not the independent variable(s).
Figure 12.7 Two-Variable Regression Model--Theoretical
315
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In constructing the estimating equation by least squares we have computed a
regression line for a sample, and not for the population: bX+a=Y 1
ˆ where Ŷ is the
estimated mean of Y, given X, and a, b are the sample estimates of α and β in the
theoretical model. As already noted, this line appears in Figure 12.6 for the specific
bivariate problem of Table 12.11.
However, functional forms other than linear may be suggested by the preliminary
scatter plot. Figure 12.8 shows various types of scatter diagrams and regression lines for
the two-variable case. Panel I shows the ideal case in which all the variation in Y is
accounted for by variation in X1. We note that the regression line passes through the
mean of each variable and that the slope b happens to be positive. The intercept a
represents the predicted value of Y when X1 = 0. In Panel II we note that there is residual
variation in Y, and, furthermore, that the slope b is negative. Panel III demonstrates the
case in which no association between Y and X1 is found. In this case the mean of Y is as
good a predictor as the variable X1 (the slope b is zero). Panel IV emphasizes that a
linear model is being fitted. That is, no linear association is found (b = 0), even though a
curvilinear relationship is apparent from the scatter diagram. Figure 12.8 illustrates the
desirability of plotting one's data before proceeding to formulate a specific regression
model.
Figure 12.8 Illustrative Scatter Diagrams and Regression Lines
Strength of Association
It is one thing to find the regression equation (as shown in Figure 12.6), but at this
point we still do not know how strong the association is well. Does the regression line Ŷ
= a + bx (which uses X as a predictor) explain the variation in Y (predict Y)? To answer
this question, consider that the total variation in the Y variable may be divided into two
component parts: (1) variance that is explained by the regression line, and (2)
unexplained variation (residual). This may be expressed as and is viewed graphically in
Figure 12.9.
316
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 12.9 Scatter Diagram and Regression Line for Cereal Problem
=−Σ 2
)( YYi
2
)( YY i −Σ
∧
+ 2
)(
∧
−Σ YY i
Total variation or
Total Sum of
Squares Deviation
from the Mean
Variation Explained
by the Regression or
Sum of Squares Due to
Regression
Variation
Unexplained by the Regression.
Sum of Squares Deviation from
Regression (Error)
The measure of strength of association in bivariate regression is denoted by r2
and
is called the coefficient of determination. This coefficient varies between 0 and 1 and
represents the proportion of total variation in Y (as measured about its own mean Y) that
is accounted for by variation in X1. For regression analyses it can also be interpreted as a
measure of substantive significance, as we have previously defined this concept.
If we were to use the average of the Y values (Y) to estimate each separate value
of Y, then a measure of our inability to predict Y would be given by the sum of the
squared deviations )Y-Y( 2
i
n
=1i
∑ . On the other hand, if we tried to predict Y by
employing a linear regression based on X, we could use each Ŷi to predict its counterpart
Yi. In this case a measure of our inability to predict Yi is given by )Y-Y( 2
Ii
n
=1i
ˆ∑ . We
can define r2
as a function of these two qualities:
)Y-Y(
)Y-Y(
-1=r
2
i
n
=1i
2
ii
n
=1i2
∑
∑
If each Ŷi predicts its counterpart Yi perfectly, then r2
= 1, since the numerator of
the variance unaccounted for is zero. However, if using the regression equation does no
better than Y alone, then the total variance equals the variance unaccounted for and
r2
= 0, indicating no ability to predict Yi (beyond the use of Y itself). The use of x1 in a
317
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
linear regression can do no worse than Y. Even if b turns out to be zero, the predictions
are Ŷi = a = Y, which are the same as using the mean of criterion values in the first place.
Table 12.12 shows the residuals obtained after using the regression equation to
predict each value of Yt via its counterpart Ŷi. We then find rx (where we now show the
explicit subscripts) by computing from the table:
21.09=
)(0.623+...+)(-1.464+)(-1.034=)Y-Y( 2222
ii
n
=1i
ˆ∑
Table 12.12 Actual Yi, Predicted Ŷi, and Residuals Yi -Ŷi
Rater
Actual,
Yi
Predicted,*
Ŷi
Residuals
Yi -Ŷi
1
2
3
4
5
6
7
8
9
10
Mean
3
7
2
1
6
2
8
3
9
2
4.3
4.034
8.464
3.148
1.377
3.148
4.034
6.692
3.148
7.579
1.377
4.3
-1.034
-1.464
-1.148
-0.377
2.852
-2.034
1.308
-0.148
1.422
0.623
0
*From the equation Ŷi = 0.491 + 0.886Xi1.
This is the sum of squared errors in predicting Yi from Ŷi. Next, we find:
76.10=
)4.3-(2+...+)4.3-(7+)4.3-(3=)Y-Y( 2222
i
n
=1i
∑
This is the sum of squared errors in predicting Yi from Y . Hence,
0.723=
76.10
21.09
-1=r
2
yx1
and we say that 72% of the variation in Y has been accounted for by variation in X1. As
might also be surmised, there is one more quantity of interest:
55.01=
)4.3-(1.377+...+)4.3-(8.464+)4.3-(4.034=)Y-Y( 2222
i
n
=1i
ˆ∑
which is the accounted-for sum of squares due to the regression of Y on X1.
Figure 12.10 (and 12.9 as well) put all these quantities in perspective by first
showing deviation of Yi -Y . As noted above, the sum of these squared deviations is
76.10. Panel II shows the counterpart deviations of Yi from Ŷi; the sum of these squared
deviations is 21.09. Panel III shows the deviations of Ŷi from Y i ; the sum of these
squared deviations is 55.01. We note that the results are additive: 21.09 + 55.01 = 76.10.
318
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 12.10 Breakdown of Deviations Yi - Y into Two Additive Parts
Interpretation of Bivariate Regression
The sample data of Table 12.11 were analyzed using a standard linear regression
analysis routine. Table 12.13 shows the output provided. The coefficients shown earlier
appear here, along with some other measures as well.
Table 12.13 Summary Output of Regression Analysis of Table 12.11 Sample data
Variables Entered/Removed
b
Model Variables
Entered
Variables
Removed
Method
1 PROTEIN
a
Enter
a. All requested variables entered.
b. Dependent Variable: PREFEREN
Model Summary
Model R R Square Adjusted R Square Std. Error of the
Estimate
1 .850
a
.723 .688 1.62354
a. Predictors: (Constant), PROTEIN
ANOVA
b
Model Sum of
Squares
df Mean Square F Sig.
1 Regression
Residual
Total
55.013
21.087
76.100
1
8
9
55.013
2.636
20.871 .002
a
a. Predictors: (Constant), PROTEIN
b. Dependent Variable: PREFEREN
Coefficients
a
Model
Unstandardized Coefficients Standardized
Coefficients
t Sig.
B Std. Error
1 (Constant)
PROTEIN
.491
.886
.979
.194 .850
.501
4.568
.630
.002
a. Dependent Variable: PREFEREN
Overall, the linear regression equation is:
Y = 0.491 + 0.886 X1
The standardized coefficient for the independent variable (beta or β) is more
meaningful for multiple regression as it is a measure of the change in Y due to a unit
change in Xi when all independent variables have been “transformed” to the same units.
The F ratio is the appropriate test for the hypotheses that the regression
coefficient, b1 = 0 and r2
= 0. This test was discussed in more detail in Chapter 11, in the
context of analysis of variance. However, as recalled from basic statistics, the F
319
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
distribution is the distribution followed by the ratio of two independent, unbiased
estimates of the normal population variance σ2
. If r2
is zero, then the sample r2
reflects
only sampling error and the F ratio will tend to be equal to 1.0. We first obtain the mean
squares, 55.013 and 2.636, by dividing the corresponding sums of squares by their
respective degrees of freedom. Then we find the F ratio of 20.871. This value, with 1
degree of freedom for numerator and 8 degrees of freedom for denominator, is compared
with a tabular F (see Table A-4 in Appendix A) of 3.46 with an (illustrative) significance
level of 0.1. We reject the preceding null hypotheses and conclude that the overall
regression equation is statistically significant.
Rank Correlation
The previous discussion of correlation and bivariate regression were based on the
premise that the dependent variable is at least interval scaled or can be treated as such
with little error. There are marketing problems, however, where the dependent and
independent variables are rank orders or are best transformed into such rankings. In this
situation the Spearman r’s rank correlation can be used to estimate the association
between sets of data.
We show the use of this measure by an example. Suppose a sales manager ranks
salespersons by two different methods (performance index and a new subjective method).
Since the new method is easier to use, the manager wants to know if it will yield the same
relative results as the proven existing method. The scores have been transformed into
rankings so that each salesperson has two rankings. Table 12.14 shows the rankings.
Table 12.14 Ranks on Two Methods of Salesperson Evaluation
To measure the extent of rank correlation we use the statistic
where N is the number of pairs of ranks and d is the difference between the two rankings
for an individual (that is, X − Y). Applying this formula to our example, we get
320
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
If the subjects whose scores were used in computing rs were randomly drawn
from a population, we can test the significance of the obtained value. The null hypothesis
is that the two variables are not associated, and thus the true value of ρ is zero. Under H0
any observed value would be due to chance. When N ≥ 10, significance can be tested
using the statistic
which is interpreted from a table of t values with (N - 2) degrees of freedom. For our
example, we calculate
Looking at the table of critical values of t, we find that p > .10 (two-tailed test) for
(10 − 2 = 8) degrees of freedom. Thus, if a strict α level is to be adhered to (.10 or less),
we tentatively accept H0 and conclude that it is unlikely that a correlation exists between
the scores from the two evaluation methods.
One final point concerning the use of the Spearman rank correlation coefficient is
warranted. At times, tied observations will exist. When this happens, each of them is
assigned the average of the ranks λ minute. If large, however, a correction factor must be
applied (see Siegel, 1956, pp. 206–10).
The Spearman rank correlation coefficient is equivalent to the Pearson productmoment
correlation coefficient with ranks substituted for the measurement observations,
X and Y. The Spearman and other measures are discussed more fully by Siegel (1956,
Chapter 9) and Gibbons (1993).
Finally, when the variables are nominally scaled, ordinal measures such as tau and
rs are not appropriate measures. Nominal variables lack the ordering property. One
measure that can be used is Goodman and Kruskal’s lambda (λ ). Lambda is a measure of
association whose calculation and interpretation are straightforward. Lambda tells us how
much we can reduce our error in predicting Y once we know X, and is shown as
This measure, and others as well, are discussed by Lewis-Beck (1995, Chap. 4).
Lambda is an option provided by most Crosstab programs. We discuss lambda further in
the next section.
NONPARAMETRIC ANALYSIS
One reason for the widespread use of chi-square in cross-tabulation analysis is
that most computer computational routines show the statistic as part of the output, or at
least it is an option that the analyst can choose. Sometimes ordinal data are available and
as such are stronger than simple nominal measurement. In this situation other tests are
more powerful than chi-square. Three regularly used tests are the Wilcoxon Rank Sum
(T), the Mann-Whitney U, and the Kolmogorov-Smirnov test. Siegel (1956) and Gibbons
(1993) provide more detailed discussions of these techniques.
321
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The Wilcoxon T test is used for dependent samples in which the data are collected
in matched pairs. This test takes into account both the direction of differences within
pairs of observations and the relative magnitude of the differences. The Wilcox matchedpairs
signed-ranks test gives more weight to pairs showing large differences between the
two measurements than to a pair showing a small difference. Again, to use this test,
measurements must at least be ordinal scaled within pairs. In addition, ordinal
measurement must hold for the differences between pairs.
This test has many practical applications in marketing research. For instance, an
ordinal scaling device, such as a semantic differential, can be used to measure attitudes
toward, say, a bank. Then, after a special promotional campaign, the same sample would
be given the same scaling device. Changes in values of each scale could be analyzed by
this Wilcoxon test.
With ordinal measurement and two independent samples, the Mann-Whitney U
test may be used to test whether the two groups are from the same population. This is a
relatively powerful nonparametric test, and is an alternative to the Student t test when the
analyst cannot meet the assumptions of the t test or when measurement is at best ordinal.
Both one- and two tailed tests can be conducted. As indicated earlier, results of U and t
tests often are similar, leading to the same conclusion.
The Kolmogorov-Smirnov two-sample test is a test of whether two independent
samples come from the same population or from populations with the same distribution.
This test is sensitive to any kind of difference in the distributions from which the two
samples were drawn—differences in location (central tendency), dispersion, skewness,
and so on. This characteristic of the test makes it a very versatile test. Unfortunately, the
test does not by itself show what kind of difference exists. There is a KolmogorovSmirnov
one-sample test that is concerned with the agreement between an observed
distribution of a set of sample values and some specified theoretical distribution. In this
case it is a goodness-of-fit test similar to single-classification chi-square analysis.
Indexes of Agreement
Chi-square is appropriate for making statistical tests of independence in crosstabulations.
Usually, however, we are interested in the strength of association as well as
the statistical significance of association. This concern is for what is known as
substantive or practical significance. An association is substantively significant when it is
statistically significant and of sufficient strength. Unlike statistical significance, however,
there is no simple numerical value to compare with and considerable experimental
research judgment is necessary. Although such judgment is subjective, it need not be
completely arbitrary. The nature of the problem can offer some basis for judgment, and
common sense can indicate that the degree of association is too low in some cases and
high enough in others (Gold, 1969, p. 44).
Statisticians have devised a large number of indexes—often called indexes of
agreement—for measuring the strength of association between two variables in a crosstabulation.
The main descriptors for classifying the various indexes are
1. Whether the table is 2 x 2 or larger, R x C
2. Whether one, both, or neither of the variables has categories that obey some
natural order (e.g., age, income level, family size)
322
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
3. Whether association is to be treated symmetrically or whether we want to
predict membership in one variable’s categories from (assumed known)
membership in the other variable’s categories
Space does not permit coverage of even an appreciable fraction of the dozens of
agreement indexes that have been proposed. Rather, we shall illustrate one commonly
used index for 2 x 2 tables and two indexes that deal with different aspects of the larger R
x C (row-by-column) tables.
The 2 × 2 Case
The phi correlation coefficient is a useful agreement index for the special case of
2 x 2 tables in which both variables are dichotomous. Moreover, an added bonus is the
fact that phi equals the product-moment correlation—a cornerstone of multivariate
methods—that one would obtain if he or she correlated the two variables expressed in
coded 0 – 1 form.
To illustrate, consider the 2 x 2 cross-tabulation in Table 12.15, taken from a
study of shampoos. We wish to see if inclusion of the shampoo benefit “body” in the
respondent’s ideal set is associated with the respondent’s indication that her hair lacks
natural “body.” We first note from the table that high frequencies appear in the cells: (a)
“body” included in ideal set and “no” to the question of whether her hair has enough
(natural) body; and (b) “body” excluded from the ideal set and “yes” to the same
question.
Table 12.15 Does Hair Have Enough Body Verses Body Inclusion in Ideal Set
Hair Have Enough Body?
No Yes Total
Body included in ideal set
Body excluded from ideal set
Total
26 (A)
17 (C)
43
8 (B)
33 (D)
41
34
50
84
Before computing the phi coefficient, first note the labels, A, B, C, and D
assigned to the four cells in Table 12.15. The phi coefficient is defined as
The value 0.417 is also what would be found if an ordinary product-moment
correlation, were computed across the 84 pairs of numbers where the following
dichotomous code values are used to identify the responses:
• Body in ideal set: Included ⇒ 1, Excluded ⇒ 0
• Hair have enough body? No ⇒ 1, Yes ⇒ 0
323
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The phi coefficient can vary from −1 to 1 (just like the ordinary product-moment
correlation). However, in any given problem the upper limit of phi depends on the
relationships among the marginals. Specifically, a phi coefficient of –1 (perfect negative
association) or 1 (perfect positive association) assumes that the marginal totals of the first
variable are identical to those of the second. Looking at the letters (A, B, C, D) of Table
12.15, assume that the row marginals equaled the column marginals: then, Φ = 1 if B = C
= 0; similarly, Φ = − 1 if A = D = 0. The more different the marginals, the lower the
upper limit that the (absolute) value of phi can assume.
The phi coefficient assumes the value of zero if the two variables are statistically
independent (as would be shown by a chi-square value that is also zero). Indeed, the
absolute value of phi is related to chi-square by the expression
where n is the total frequency (sample size). This is a nice feature of phi, in the sense that
it can be computed quite easily after chi-square has been computed. Note, however, that
phi, unlike chi-square, is not affected by total sample size because we have the divisor n
in the above formula to adjust for differences in sample size.
The R x C Case
One of the most popular agreement indexes for summarizing the degree of
association between two variables in a cross-tabulation of R rows and C columns is the
contingency coefficient. This index is also related to chi-square and is defined as
where n is again the total sample size. From Table 12.15 we can first determine that chisquare
is equal to 14.61, which, with 1 degree of freedom, is significant beyond the 0.01
level.
We can then find the contingency coefficient C as the following:
As may be surmised, the contingency coefficient lies between zero and 1, with
zero reserved for the case of statistical independence (a chi-square value of zero).
However, unlike the phi coefficient, the contingency can never attain a maximum value
of unity. For example, in a 2 × 2 table, C cannot exceed 0.707. As might be noticed by
the reader, there is an algebraic relationship between phi and the contingency coefficient
(if the latter is applied to the 2 × 2 table):
In a 4 X 4 table its upper limit is 0.87. Therefore, contingency coefficients
computed from different-sized tables are not easily comparable.
324
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
However, like phi, the contingency coefficient is easy to compute from chisquare;
moreover, like phi, its significance has already been tested in the course of
running the chi-square test.
Both phi and the contingency coefficient are symmetric measures of association.
Occasions often arise in the analysis of R X C tables (or the special case of 2 ⋅ 2 tables)
where we desire to compute an asymmetric measure of the extent to which we can reduce
errors in predicting categories of one variable from knowledge of the categories of some
other variable. Goodman and Kruskal’s lambda-asymmetric coefficient can be used for
this purpose (Goodman & Kruskal, 1954).
To illustrate the lambda-asymmetric coefficient, let us return to the crosstabulation
of Table 12.15. Suppose that we wished to predict what category—no versus
yes—a randomly selected person would fall in when asked the question, “Does your hair
have enough body?” If we had no knowledge of the row variable (whether that person
included “body” in her ideal set or not), we would have only the column marginal
frequencies to rely on.
Our best bet, given no knowledge of the row variable, is always to predict “no,”
the higher of the column marginal frequencies. As a consequence, we shall be wrong in
41 of the 84 cases, a probability error of 41/84 = 0.49 Can we do better, in the sense of
lower prediction errors, if we utilize information provided by the row variable?
If we know that “body” is included in the ideal set, we shall predict “no” and be
wrong in only 8 cases. If we know that “body” is not included in the ideal set, we shall
predict “yes” and be wrong in 17 cases. Therefore, we have reduced our number of
prediction errors from 41 to 8 + 17 = 25, a decrease of 16 errors. We can consider this
error reduction relatively:
In other words, 39 percent of the errors in predicting the column variable are eliminated
by knowing the individual’s row variable.
A less cumbersome (but also less transparent) formula for lambda-asymmetric is
Where f kr is the maximum frequency found within each subclass of the row variable, Fc is
the maximum frequency among the marginal totals of the column variable, and n is the
total number of cases.
Lambda-asymmetric varies between zero, indicating no ability at all to eliminate
errors in predicting the column variable on the basis of the row variable, and 1, indicating
an ability to eliminate all errors in the column variable predictions, given knowledge of
the row variable. Not surprisingly, we could reverse the role of criterion and predictor
variables and find lambda-asymmetric for the row variable, given the column variable. In
the case of Table 12.15, this results in λ=0.26.
325
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Note that in this case we simply reverse the roles of row and column variables.
Finally, if desired, we could find a lambda-symmetric index via a weighted
averaging of λC | R and λR | C. However, in the authors’ opinion, lambda-asymmetric is of
particular usefulness to the analysis of cross-tabulations because we often want to
consider one variable as a predictor and the other as a criterion. Furthermore, lambdaasymmetric
h is a natural and useful interpretation as the percentage of total prediction
errors that are eliminated in predicting one variable (e.g., the column variable) from
another (e.g., the row variable).
SUMMARY
We began by stating that data can be viewed as recorded information useful in
making decisions. In the initial sections of this chapter, we introduced the basic concepts
of transforming raw data into data of quality. The introduction was followed by a
discussion of elementary descriptive analyses through tabulation and cross-tabulation.
The focus of this discussion was heavily oriented toward how to read the data and how to
interpret the results. The competent analysis of research-obtained data requires a blending
of art and science, of intuition and informal insight, and of judgment and statistical
treatment, combined with a thorough knowledge of the context of the problem being
investigated.
The first section of the chapter dealt with cross-tabulation and chi-square analysis.
This was followed by discussing bivariate analysis of differences in means and
proportions. We next focused on the necessary statistical machinery to analyze
differences between groups: t-test, and one-factor and two-factor analysis of variance.
These techniques are useful for both experimental and nonexperimentally obtained data.
We then looked at the process of analysis of variance. A simple numerical example was
used to demonstrate the partitioning of variance into among- and within-components. The
assumptions underlying various models were pointed out and a hypothetical data
experiment was analyzed to show how the ANOVA models operate.
We concluded by examining bivariate analyses of associations for interval- or
ratio scaled data. The concept of associations between two variables was introduced
through simple two-variable correlation. We examined the strength and direction of
relationships using the scatter diagram and Pearson correlation coefficient. Several
alternative (but equivalent) mathematical expressions were presented and a correlation
coefficient was computed for a sample data set.
Investigations of the relationships between variables almost always involve the
making of predictions. Bivariate (two-variable) regression was discussed as the
foundation for the discussion of multivariate regression in the next chapter.
We ended the chapter with a discussion of the Spearman rank correlation as an
alternative to the Pearson correlation coefficient when the data is of ordinal measurement
and does not meet the assumptions of parametric methods. Also, the Goodman and
Kruskal lambda measure for nominal measurement was briefly introduced, as were other
nonparametric analyses.
There is a wide array of statistical techniques (parametric and non parametric) that
focus on describing and making inferences about the variables being analyzed. Some of
these were shown in Table 12.4. Although somewhat dated, a useful reference for
selecting an appropriate statistical is the guide published by the Institute for Social
326
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Research at the University of Michigan (Andrews, et. al., 1981 and its corresponding
software Statistical Consultant. Fing (2003, pp. 78-80) presents a summary table of
which technique to use under which condition.
REFERENCES
Andrews, F. M., Klem, L., Davidson, T. N., O’Malley, P. M., & Rodgers, W. L. (1981).
A guide for selecting statistical techniques for analyzing social science data (2nd ed.).
Ann Arbor: Institute for Social Research, University of Michigan.
Feick, L. F. (1984, November). Analyzing marketing research data with associated
models. Journal of Marketing Research, 21, 376–386.
Fink, A. (2003). How to manage, analyze, and interpret survey data. Thousand Oaks,
CA: Sage.
Gibbons, J. D. (1993). Nonparametric statistics: An introduction. Newbury Park, CA:
Sage.
Gold, D. (1969, February). Statistical tests and substantive significance. American
Sociologist, 4, 44.
Goodman, L. A., & Kruskal, W. H. (1954, December). Measures of association for cross
classification. Journal of the American Statistical Association, 49, 732–764.
Hellevik, O. (1984). Introduction to causal analysis: Exploring survey data. Beverly
Hills, CA: Sage.
Lewis-Beck, M. S. (1995). Data analysis: An introduction. Thousand Oaks, CA: Sage.
Semon, T. T. (1999, August 2). Use your brain when using a chi-square. Marketing
News, 33, 6.
Siegel, S. (1956). Nonparametric statistics for the behavioral sciences. New York:
McGraw-Hill.
Smith, S. and Albaum, G. (2005). Fundamentals of marketing research. Thousand Oaks,
CA: Sage Publications.
Zeisel, H. (1957). Say it with figures (4th ed.). New York: Harper & Row.
327
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 13
MULTIVARIATE STATISTICAL ANALYSIS I
Analyzing Criterion-Predictor Association
In the previous chapters that discussed analysis of differences and association for two
variables, we determined that selecting the appropriate method of analysis requires evaluating
the number of dependent and independent variables and their level of measurement. The same
process applies to multivariate analysis, as is illustrated in Figure 13.1.
Association implies only that two or more variables tend to change together to a greater
or lesser degree, depending upon the degree of association involved. If we measure the amount
of mutual change and find it to be persistent in both direction and degree, we may not conclude
that there is necessarily a causal relationship, such that one variable is dependent (the effect) and
the other variable (or variables) is independent (the deterministic or probabilistic cause). It
should be understood, however, that association does not imply causation. However, if a set of
variables are causally related, they will be associated in some way. Analytical techniques such as
causal-path analysis and structural equation models have been developed to aid in examining
possible causal relations in correlational data. These are discussed in depth by Blalock (1962),
Bagozzi (1980), and Monroe and Petroshius (n.d.).
In this and the next chapter, we present a series of brief discussions that focus on
individual multivariate techniques. Multivariate analyses arise when more than two variables are
to be analyzed at the same time. We begin by extending simple bivariate regression discussed in
the last chapter to include a second predictor variable. This use of multiple predictors allows the
researcher to conduct both multiple and partial regression.
We next focus on two-group discriminant analysis. Like multiple regression, discriminant
analysis is a predictor model. However, the dependent variable is categorical in measurement
and defines two or more groups. In discriminant analysis, group membership (such as purchaser
and non-purchaser groups) is predicted using the multiple independent variables.
Finally, a brief conceptual introduction is given to a diverse group of techniques,
including canonical correlation, correspondence analysis, Chi-Square Automatic Interaction
Detection (CHAID), and probit and logit analyses. We start with a broad overview of
multivariate procedures.
AN OVERVIEW OF MULTIVARIATE PROCEDURES
The Data Matrix
The raw input to any analysis of associative data consists of the data matrix, whose
informational content is to be summarized and portrayed in some way. For example, the
computation of the mean and standard deviation of a single variable is often done simply to
obtain a summary of the meaning of the entire set of values. In so doing, we choose to forgo the
full information provided by the data in order to understand some of its basic characteristics,
such as central tendency and dispersion.
In virtually all marketing research studies we are concerned with variation in some
characteristic, be it per capita consumption of soft drinks, TV viewing frequency, or customer
intention to repurchase. Our objective now, however, is to concentrate on explaining the
variation in one variable or group of variables in terms of covariation with other variables. When
328
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
we analyze associative data, we hope to explain variation, achieving one or more of the
following:
1. Determine the overall strength of association between the criterion and predictor variables
(often called dependent and independent variables, respectively)
2. Determine a function or formula by which we can estimate values of the criterion variable(s)
from values of the predictor variable(s)
3. Determine the statistical confidence in either or both of the above
In some cases of interest, however, we may have no criterion (dependent) variable. We
may still be interested in the interdependence of a group of variables as a whole and the
possibility of summarizing information provided by this interdependence.
Figure 13.1 Process for Selecting Appropriate Method of Analysis
A Classification of Techniques for Analyzing Associative Data
The field of associative data analysis is vast; hence, it seems useful to enumerate various
descriptors by which the field can be classified. A conceptual illustration is shown in Table 13.1.
The key notion underlying our classification is the data matrix. We note that a data matrix is
often thought of as being in spreadsheet form (referred as a flat data matrix) and consists of a set
of objects (the n rows) and a set of measurements on those objects (the m columns). The row
objects may be people, things, concepts, or events. The column variables are characteristics of
the objects. The cell values represent the state of object i with respect to variable j. Cell values
may consist of nominal-, ordinal-, interval-, or ratio-scaled measurements, or various
combinations of these as we go across columns. When using data analysis software, the data file
created is structured exactly like the basic data matrix, which is, in effect, a type of spreadsheet.
Note however that some analysis software also has the ability to read relational databases that are
329
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
not in the traditional row x column spreadsheet format. However for relational database format
data, the concepts of objects and variables still apply.
Table 13.1 Illustrative Basic Data Matrix
Variable
Object 1 2 3 . . . j . . . m
1 X11 X12 X12 . . . X1j . . . X1m
2 X21 X22 X23 . . . X2j . . . X2m
3 X31 X32 X33 . . . X3j . . . X3m
. . . . . .
. . . . . .
. . . . . .
i Xi1 Xi2 Xi3 . . . Xij . . . Xim
. . . . . .
. . . . . .
. . . . . .
n Xn1 Xn2 Xn3 . . . Xnj . . . Xnm
By reflecting on the steps in the research process discussed in Chapter 2, we can gain a
perspective on the many available approaches to analysis of associate data. Although not
exhaustive (or exclusive), the following represent the more common perspectives by which
analysis activities can be directed:
1. Purpose of the study and types of assertions desired by the researcher: What kinds of
statements—descriptive or inferential—does the researcher wish to make?
2. Focus of research: Is the emphasis on the objects (the whole profile or “bundle” of
variables), the variables, or both?
3. Nature of the prior judgments assumed by the researcher as to how the data matrix
should be partitioned (subdivided) into a number of subsets of variables.
4. Number of variables in each of the partitioned subsets: How many criterion versus
predictor variables?
5. Type of association under study: Is the association linear, transformable to linear, or
nonlinear?
6. Scales by which variables are measured: Are the scales nominal, ordinal, interval,
ratio, or mixed?
All of these descriptors require certain decisions of the researcher, three of which are of
immediate interest in determining the form of analysis:
1. Whether the objects or variables of the data matrix are partitioned into subsets
or kept intact
2. If the matrix is partitioned into subsets of criterion and predictor variables, the
number of variables in each subset
3. Whether the variables are nominal-scaled or interval-scaled
4. The relationships between the variables (linear, transformation to linear, or
non-linear).
Most decisions about associative data analysis are based on the researcher’s private
model of how the data are interrelated and what features are useful for study. The choice of
330
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
various public models for analysis (multiple regression, discriminant analysis, etc.) is predicated
upon prior knowledge of both the characteristics of the statistical universe from which the data
were obtained and the assumption structure and objectives associated with each statistical
technique.
Fortunately, we can make a few simplifications of the preceding descriptors. First,
concerning types of scales, all the multivariate techniques of this book require no stronger
measurement than interval scaling. Second, except for the ordinal scaling methods of
multidimensional scaling and conjoint analysis (to be discussed in Chapter 13), we shall assume
that (a) the variables are either nominal-scaled or interval-scaled, and (b) the functional form is
linear in the parameters. While the original data may have been transformed by some nonlinear
transformation (e.g., logarithmic), linear in the parameters means that all computed parameters
after the transformation are of the first degree. For example, the function Y = b0 + b1X1 + b2X2 + .
. . + bmXm is linear in the parameters. Note that b0 is a constant, while the other parameters, b1, b2,
. . ., bm are all of the first degree. Moreover, none of the bj’s depends on the value of either its
own Xj or any other Xk (k ≠ j). Even with these simplifying assumptions, we shall be able to
describe a wide variety of possible techniques, and these may vary from being relatively simple,
(as shown in Exhibit 13.1), to complex.
Analysis of Dependence
If we elect to partition the data matrix into criterion and predictor variables, the problem
becomes one of analyzing dependence structures. This, in turn, can be broken down into
subcategories:
1. Single criterion/multiple predictor association
2. Multiple criterion/multiple predictor association
Multivariate techniques that deal with single criterion/multiple predictor association
include multiple regression, analysis of variance and covariance, two-group discriminant
analysis, and CHAID (chi-square automatic interaction detection).
Multivariate techniques that deal with multiple criterion/multiple predictor association
include canonical correlation, multivariate analysis of variance and covariance, and multiple
discriminant analysis.
Analysis of Interdependence
In some cases, we may not wish to partition the data matrix into criterion and predictor
subsets. If so, we refer to this case as the analysis of interdependence structures. Techniques
such as factor analysis are used if the focus of interest is on the variables of the (intact) data
matrix. Cluster analysis is relevant if we wish to focus on the grouping of objects in the data
matrix, as based on their profile similarities.
Selecting a Technique
How does a researcher select the multivariate technique to use in a given situation? In
addition to the research question underlying the analysis, we suggested in Figure 13.1 that the
number of dependent and independent variables and their levels of measurement are two of the
major criteria. Four major sets of factors influence the choice of an appropriate statistical
technique:
331
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Research parameters (analysis purpose and specifications)
• Known characteristics of the data
• Properties of the statistical technique
• Researcher characteristics
Research parameters include the purpose of analysis and the kind of analysis required.
Before the analysis begins, certain characteristics of the data are known, including the level of
measurement, distribution of the data, and the nature and size of the sample. This information is
matched with the properties of a particular statistical technique (robustness, assumptions
required, and design purpose). Finally, a researcher’s individual philosophy, expertise, and
experience will influence the choice of statistical technique.
Operationally, the researcher can be assisted by any of numerous printed or online
guides, such as that provided at http://www.qualtrics.com/wiki/index.php/Analysis.
MULTIPLE AND PARTIAL REGRESSION
Multiple regression extends the concepts of bivariate regression by simply adding
multiple predictor (independent) variables. To illustrate this extension, we will introduce the
second predictor X2. Table 13.2 presents a set of example data with two predictor variables.
Table 13.2 Consumer Preference Ratings of Ten Cereals Varying in Nutritional Level
Rater Preference Protein, X1 Vitamin D, X2
1 3 4 2
2 7 9 7
3 2 3 1
4 1 1 2
5 6 3 3
6 2 4 4
7 8 7 9
8 3 3 2
9 9 8 7
10 2 1 3
Total 43 43 40
Mean 4.3 4.3 4.0
Standard deviation 2.908 2.791 2.708
EXHIBIT 13.1 Multivariate Analysis Can Be Simple
Not all multivariate analyses need to involve
complex and sophisticated statistical techniques. A
relatively simple analysis is importanceperformance
analysis. In the B2B area, for
example, this could be done by asking customers
to rate a supplier’s performance on a set of
attributes and characteristics, and then asking
those customers how important each is to them. A
central tendency measure would be calculated for
each attribute’s importance and performance
measure and then plotted in two-dimensional
space, as illustrated in the following figure, which
shows a hypothetical grid for 14
attributes/characteristics.
332
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The theoretical model now becomes
with parameters estimated by
All assumptions previously discussed for the bivariate regression model continue to hold
in the present case. However, it is important to remember that, in general, the current b1— now
called a partial regression coefficient with respect to X1—will not equal its counterpart
coefficient (b) obtained from the bivariate regression. This is because X1 itself will usually be
correlated with X2. In the bivariate case, X2 was not entered into the analysis and any variation in
Y that was shared by X1 and X2 was credited solely to X1. Such will no longer be the case. When
there are only two predictors, X1 and X2, it is still possible to prepare a scatter plot.
However, now the plot appears in terms of a three-dimensional space. In this space we fit
a plane whose sum of squared deviations of the Yi from their counterparts (on the plane) is a
minimum. Figure 13.3 shows the original scatter plot and the fitted plane for the sample problem
of Table 13.2. Using regression analysis programs, the estimated (least-squares) equation is
Y = 0.247 + 0.493 X1 + 0.484 X2
Figure 13.3 Scatter Plot and Fitted Regression Plane
Table 13.3 Single Correlation Between X1 (Protein) and X2 (Vitamin)
Protein Vitamin
Protein Pearson Correlation
Sig. (2-tailed)
N
1
.
10
.838**
.002
10
Vitamin Pearson Correlation
Sig. (2-tailed)
N
.838**
.002
10
1
.
10
**. Correlation is significant at the 0.01 level (2-tailed).
The first slope coefficient of 0.493 denotes the change in per unit change in X1 when X2
is held constant. Similarly, 0.484 denotes the change in per unit change in X2 when X1 is held
constant. We note that 0.493 ≠ 0.886, the slope obtained earlier in the bivariate case discussed in
Chapter 11. This is because X1 and X2 are themselves correlated, and X1 is now forced to share
some of its Y-variable association with the second predictor, X2.
333
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Parameter Estimation
Finding the partial regression coefficients b1 = 0.493 and b2 = 0.484 and the intercept a
= 0.247 is considerably more complicated than finding a single regression coefficient in bivariate
regression. Exhibit 13.2 shows the statistical analysis for a regression analysis. This result can
be replicated with any statistical package or a spreadsheet multiple regression function.
______________________________________________________________________________
EXHIBIT 13.2 Summary Output of Regression Analysis (All Variables): Sample Data
Variables Entered/Removed
Model Variables Entered Variables
Removed
Method
1 VITAMIN,
PROTEIN
a
. Enter
All requested variables entered.
Dependent Variable: PREFERENCE
Model Summary
Model R R Square Adjusted R Square Std. Error of the Estimate
1 .885
a
.783 .721 1.53482
Model
Change Statistics
R Square
Change F Change df1 df2 Sig. F Change
1 .783 12.652 2 7 .005
Predictors: (Constant), VITAMIN, PROTEIN
ANOVA
Model Sum of Squares df Mean Square F Sig.
1 Regression
Residual
Total
59.610
16.490
76.100
2
7
9
29.805
2.356
12.652 .005
a
Predictors: (Constant), VITAMIN, PROTEIN
Dependent Variable: PREFERENCE
Coefficients
Model
Unstandardized
Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant)
PROTEIN
VITAMIN
.247
.493
.484
.942
.336
.346
.473
.450
0.262
1.466
1.397
.801
.186
.205
A. Dependent Variable: PREFERENCE
______________________________________________________________________________
The Coefficient of Multiple Determination, R2
Finding the coefficient of multiple determination proceeds in the same way as the
bivariate case. Exhibit 13.2 shows this value as: = 0.783.
We note that the multiple regression equation does quite well at prediction, since 78
percent of the variation in Y is accounted for by variation in X1 and X2 ( ). However, we
recall from Chapter 12 that if only X1 is employed, the (simple) coefficient of determination is 72
percent. Apparently, X2 (vitamin) and X1 (protein) are so highly correlated—their simple
334
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
correlation is 0.838 (see Table 13.3)—that once X1 is in the regression equation there is little
need for X2 as well.
What about the square root of ?
This is called the coefficient of multiple correlation: and is interpreted as the simple correlation
(i.e., bivariate correlation) between Yi and where, as we know, the are the predicted values
obtained by the best linear composite of X1 and X2 in the least-squares sense; this composite is
given by the multiple regression equation.
Standard Errors and t-Tests
While the F-test indicates that the overall regression model is significant, it does not
follow that both b1 and b2 contribute significantly to overall accounted-for variance. It may be
the case that a simpler model involving only X1 (or only X2) would be sufficient. The standard
error of each individual regression coefficient provides the basis for a t-test of this simpler
model.
The standard error is a measure of dispersion about the average partial regression
coefficient over repeated samplings of Y for a fixed set of values on each of the predictors. The
larger the standard error is, the less reliable our estimate of b1 is (across repeated samplings from
the same universe). Conceptually, this statement means that the standard error of b1 increases as
X1 becomes more completely accounted for by the remaining predictor variables.
The t-value of 1.466 in Exhibit 13.2 is simply the ratio of b1 to its own standard error,
SE(b1) = 4.73/3.36. The test of significance of t1 is carried out by finding the tabular value of the
t-distribution (Table A.2 in Appendix A) for seven degrees of freedom. If we continue to use a
significance level of 0.1, then the tabular value of t is 1.415 and b1 is significant. However, b2
(whose t-value is 1.397) is not significant at the 0.1 level. Note that in Exhibit 13.2, the two
variable (protein and vitamin) regression analysis reports protein to be nonsignificant with a
probability of .186. This is due to the degrees of freedom issue used in the two variable
regression. When considered alone, the t-test for protein is significant at the .1 level.
The t-test for each individual partial regression coefficient tests whether the increment in
R2
produced by the predictor in question is significant when a model including the predictor (and
all other predictors) is compared with a model including all predictors but the one being tested.
Contribution to Accounted-for Variation
The coefficient of multiple determination in this example indicates that the variation of
both independent variables accounted for 78.3 percent of the variation in the dependent variable.
Since b2 is not significant, we can presume that X2 contributes little to the dependent variable’s
variation. In fact, this is the case. When a stepwise procedure was used for the regression
analysis, X1 was the only variable included and R2
= 0.723. Thus X2’s contribution to the
coefficient of multiple determination is only 0.06 (0.783−0.723).
In summary, Exhibit 13.2 shows the major output measures of interest in applied multiple
regression studies:
1. The regression equation
2. R2
—both the sample-determined and the population-adjusted values
3. An F-test for testing the significance of the overall regression (involving both X1 and X2)
335
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
4. Individual t-tests and standard errors for testing each specific partial regression coefficient
5. Partial correlation coefficients
6. The accounted-for variance contributed by each predictor, where any shared variance of each
predictor (beyond the first) is credited to the predictors that precede it, based on the researcher’s
order for including the predictor variables
In the sample problem R2
was highly significant, but only the first variable X1 was needed
to account for most of the variance in Y. That is, the addition of X2 accounted for an incremental
variance of only six percentage points; moreover, the t-test for b2 was not significant at the 0.1
level. Therefore, in practice we would employ a simple regression of Y on X1 alone. This has
already been computed in Chapter 11 as 1886.0491.0ˆ X=Y + .
Other Forms of Regression
With the large variety of computer programs available, the computation of multiple
regression analysis is both fast and easy for the user. However, decisions must still be made by
the user. Within a package, individual programs may differ with respect to the criterion for
including the independent variables in the regression, the ability to repeat the analysis on
subgroups of respondents and compare the subgroups, the types of residual analysis available,
and the linearity assumption used in the analysis.
The two most widely used approaches are all-variable regression (demonstrated above)
and stepwise regression, as described further in Exhibit 13.3 and shown in Exhibit 13.4. Many
items shown in an output are common to both procedures, though not necessarily in the same
format or sequence. Similarly, the results of the regression analyses are often similar for a small
number of variables, becoming increasingly disparate as more variables are introduced. This
disparity is explained by the fact that the rationale for including variables differs greatly. For
more detail contrasting the difference between the all-variable and stepwise regression
methodologies, see Green, Tull, and Albaum (1988, pp. 443–55).
______________________________________________________________________________
EXHIBIT 13.3 Stepwise Regression Can Be Hazardous
Avoid using stepwise regression whenever possible, particularly when dealing with sample data.
The theoretical and statistical shortcomings are inherent in the process and difficult to overcome (Curtis,
1990). If you do use a stepwise approach, however, certain precautions should be taken:
1. Limit the number of variables initially included in the predictor pool, eliminating some with
high intercorrelations.
2. Use as large a sample as possible, ideally with 40–50 cases for each variable in the pool.
3. Validate the model through a follow-up sample or by splitting the original sample.
4. When multicollinearity is a problem, consider using factor analysis as a data reduction
technique.
Stepwise regression is a computer-driven statistical procedure in which a subset of independent
variables is selected from a larger pool of variables through a series of repetitive steps to arrive at the
best-fitting model for the data. “Best” typically refers to maximizing the explained variance (R2
) of the
dependent variable.
Stepwise regression begins with a dependent variable and a pool of independent variables. It
starts building a model by selecting the variable most highly correlated with the dependent variable. The
program then looks for the next variable in the pool that offers the largest increment in R2
when added
to the model. As new variables are added, others may be dropped in a repetitive process until the bestfitting
model is produced.
336
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
But the method has serious implications. First, from a substantive perspective, the best-fitting
model may not be the best—or even good. There are other variable selection techniques in the stepwise
family, such as forward entry and backward elimination, which also are designed to obtain the bestfitting
model. It would not be unusual to obtain three completely different bestfitting models from the
same data set using these methods. The only time this is unlikely to happen is in the unusual situation
when all predictor variables are uncorrelated with one another.
Another problem occurs when high levels of multicollinearity exist in the pool of independent
variables. A single variable may be included, while two or more others that combine to make a more
accurate predictive model may be excluded. The greater the multicollinearity, the more likely an
unreasonable and unstable best-fitting model is selected.
Stepwise methods also may cause a problem when used for explanatory model building, which
focuses on the causal relationships between the dependent and independent variables. The problem is
that stepwise methods are purely data-driven, after-the-fact techniques not based on a prior theoretical
model. Because data are typically nothing more than a snapshot of reality at any given time, and subject
to significant fluctuation, theory is ultimately important to understand the true, underlying relationships
at work.
______________________________________________________________________________
______________________________________________________________________________
EXHIBIT 13.4 Summary Output of Stepwise Regression Analysis
Sample Data from Table 13.2
Variables Entered/Removed
Model
Variables
Entered
Variables
Removed
Method
Stepwise (Criteria: Probability of F to enter
1 VAR00002 . <=.050, Probability of F to remove>=.100).
Dependent Variable:VAR00001
Model Summary
Model R R Square Adjusted R Square Std. Error of the Estimate
1 .850
a
.723 .688 1.62354
Model
Change Statistics
R Square Change F Change df1 df2 Sig. F Change
1 .723 20.871 1 8 .002
Predictors: (Constant), VAR00002
ANOVA
Model Sum of Squares df Mean Square F Sig.
1 Regression
Residual
Total
55.013
21.087
76.100
1
8
9
55.013
2.636
20.871 .002
a
Predictors: (Constant), VAR00002
Dependent Variable: VAR00001
Coefficients
Model
Unstandardized
Coefficients
Standardized
Coefficients
t Sig.B Std. Error Beta
1 (Constant)
VAR00002
.491
.886
.979
.194 .850
.501
4.568
.630
.002
Dependent Variable: VAR00001
Excluded Variables
Model Beta In t Sig.
Partial
Correlation
Collinearity Statistics
Tolerance
1 VAR00003 .450
a
1.397 .205 .467 .298
Predictors in the Model: (Constant), VAR00002
Dependent Variable: VAR00001
______________________________________________________________________________
337
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Multicollinearity
Put simply, multicollinearity refers to the problem in applied regression studies in which
the predictor variables exhibit very high correlation among themselves. This condition distorts
the value of the estimated regression coefficients, inflates the standard error of beta, and thereby
makes it more difficult to determine which predictor variable is having an effect.
Unless one is dealing with experimental design data, it is almost always the case that
predictor variables in multiple regression will be correlated to some degree. The question is how
much multicollinearity can be tolerated without seriously affecting the results? Unfortunately,
there is no simple answer to this question.
The study of multicollinearity in data analysis revolves around two major problems: (a)
How can it be detected; and (b) what can be done about it? These problems are particularly
relevant to marketing research, where one often faces the dilemma of needing a large number of
variables to achieve accuracy of predictors, and yet finds that as more predictors are added to the
model, their intercorrelations become larger.
As indicated above, what constitutes serious multicollinearity is ambiguous. Some
researchers have adopted various rules-of-thumb: For example, any pair of predictor variables
must not correlate more than 0.9; if so, one of the predictors is discarded. While looking at
simple correlations between pairs of predictors has merit, it can miss more subtle relationships
involving three or more predictors. The above rule can be extended, of course, to the
examination of multiple correlations between each predictor and all other predictors. Usually one
would want to guard against having any of these multiple correlations exceed the multiple
correlation of the criterion variable with the predictor set.
Essentially there are three procedures for dealing with multicollinearity:
1. Ignore it.
2. Delete one or more of the offending predictors.
3. Transform the set of predictor variables into a new set of predictor-variable
combinations that are mutually uncorrelated.
Ignoring multicollinearity need not be as cavalier as it might sound. First, one can have
multicollinearity in the predictor variables and still have strong enough effects that the estimating
coefficients remain reasonably stable. Second, multicollinearity may be prominent in only a
subset of the predictors, a subset that may not contribute much to accounted-for variance
anyway. A prudent procedure in checking one’s predictor set for multicollinearity is to examine
the standard errors of the regression coefficients (which will tend to be large in the case of high
multicollinearity). Second, one may randomly drop some subset of the cases (perhaps 20 percent
or so), rerun the regression, and then check to see if the signs and relative sizes of the regression
coefficients are stable. Third, most computer regression routines incorporate checks for serious
multicollinearity (see SPSS Linear Regression Analysis Statistics Help for a list of diagnostic
analyses); if the program does not indicate this condition, the researcher can generally assume
that the problem is not acute.
If multicollinearity is severe, one rather simple procedure is to drop one or more predictor
variables that represent the major offenders. Usually, because of their high intercorrelations with
the retained predictors, the overall fit will not change markedly. Pragmatically, if a particular
pair of predictors is highly collinear, one would retain that member of the pair whose
338
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
measurement reliability or theoretical importance is higher in the substantive problem under
study.
Methods also exist for transforming the original set of predictors to a mutually
uncorrelated set of linear composites—for example, principal components analysis (which is
discussed in Chapter 14). If these components (linear composites) are interpretable in
themselves, the researcher may use these in the regression analysis rather than the original
variables. If all components are retained, the predictive accuracy will be precisely the same as
that obtained from the original set of predictors. However, the problem here is that the
components may not be interpretable in their own right. Another possibility, of course, is to use
only one of the variables for each component to represent that component in the regression
analysis.
Cross-validation
Probably the safest procedure for dealing with the variety of problems in multiple
regression, including multicollinearity, is to use cross-validation. We have commented in this
chapter on the tendency of regression models (and the same is true of other multivariate
techniques) to capitalize on chance variation in the sample data. Since these techniques are
optimization methods, they find the best possible fit of the model to the specific data, though one
can also almost invariably find a poorer fit.
Cross-validation is a simple procedure for examining whether the regression equation
holds up beyond the data on which its parameters are based. The researcher simply takes part of
the data (perhaps a quarter to a third) and puts it aside. The regression equation is then computed
from the remaining data. Following this, the researcher takes the “hold-out” data and computes a
set of Yi using the earlier-computed regression equation and the predictor-variable values of the
hold-out sample. The researcher then finds the simple coefficient of determination between the
Yi in the hold-out sample and their predicted Yi counterparts. This coefficient is compared with
the R2
obtained from the original analysis to find the degree of shrinkage. Doing this requires a
larger sample size.
An even better procedure is to double cross-validate, using the following steps:
1. Split the cases randomly into halves.
2. Compute separate regression equations for each half.
3. Use the first-half equation to predict the second-half Yi values.
4. Use the second-half equation to predict the first-half Yi values.
5. Examine each partial regression coefficient across split halves to see if agreement is
obtained in both directions (algebraic sign) and in magnitude.
6. Compute a regression equation for the entire sample, using only those variables that
show stability in the preceding step.
Since high multicollinearity will make sample-to-sample regression coefficients unstable,
double cross-validation can help the researcher determine which coefficients exhibit stability
across split halves.
Depending on the sample size, of course, one could split the sample into thirds, quarters,
and so on. Usually, however, there is sufficient constraint on sample size, relative to the number
of predictors, that split-half testing is about all that gets done. Even so, single and double crossvalidation
are extremely useful undertakings, and, with a few clicks of a mouse, are easy to
implement.
339
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Steckel and Vanhonacker (1991) developed a formal test for the cross-validation of
regression models under the simple random splitting framework. Their results indicated that
splitting the data into halves is suboptimal. They recommend that more observations should be
used for estimation than validation.
TWO-GROUP DISCRIMINANT ANALYSIS
When dealing with associative data, the marketing researcher may encounter cases where
the criterion variable is categorical, but where the predictor variables involve interval-scaled
data. For example, one may wish to predict whether sales potential in a given marketing territory
will be good or bad, based on certain measurements regarding the territory’s personal disposable
income, population density, number of retail outlets, and the like. Other potential applications
also come to mind:
Two Groups:
• How do consumers who are heavy users of my brand differ in their demographic profiles
from those who are nonusers?
• How do respondents who show high interest in a new set of concept descriptions differ in
their readership levels of certain magazines from those who show low interest?
• How do homeowners who select a variable rate mortgage differ in their demographic profiles,
mortgage shopping behavior and attitudes, and preferences for mortgage features from
homeowners selecting a conventional fixed-rate mortgage?
Multiple Groups:
• Are significant demographic differences observed among purchasers of Sears, Goodyear,
Goodrich, and Michelin tires?
• How do doctors, lawyers, and bankers differ in terms of their preference ratings of eight
different luxury automobiles?
• How can loyal shoppers of Nordstrom’s, Macy’s, and Neiman-Marcus be distinguished on
the basis of their attitudes about each retailer?
Many other such problems could be added to the list. However, each one has a common
structure in which we assume that some test object (usually a person) falls into one of a set of
categories. It is also assumed we know that person’s profile on a set of (assumed or otherwise)
interval-scaled predictor variables, such as age, income, years of education, or other background
variables.
The problem is to predict a person’s category from some function of the predictor
variables. Here we shall assume that the function is linear. If only two categories are involved,
the problem is a two-group discriminant case; if three or more categories are involved, we are
dealing with multiple (group) discriminant analysis.
Objectives of Two-Group Discriminant Analysis
Two-group discriminant analysis (and classification) involves four main objectives:
1. Finding linear composites of the predictor variables that enable the analyst to separate the
groups by maximizing among-groups (relative to within-groups) variation
2. Establishing procedures for assigning new individuals, whose profiles (but not group identity)
are known, to one of the two groups
340
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
3. Testing whether significant differences exist between the mean predictor-variable profiles of
the two groups
4. Determining which variables account most for intergroup differences in mean profiles
These objectives are the bases for very distinct, purposes and procedures for conducting
discriminant analysis (Smith, 2004). The first procedure, discriminant predictive (or
explanatory) analysis, is used to optimize the predictive functions.
The second procedure, discriminant classification analysis, uses the predictive functions
derived in the first procedure to either classify fresh sets of data of known group membership,
thereby validating the predictive function, or if the function has previously been validated, to
classify new sets of observations of unknown group membership. The differences in the
application and requirements for each are summarized in Table 13.4.
Table 13.4 Stages of Discriminant Analysis
Stages
Predictive
discriminant
analysis
Classification
analysis of
initial data
set of known
groupings
Classification
analysis of
new data
set of known
groupings
Classification
analysis of
new data
set of unknown
groupings
Purpose
Derive discriminant
function using initial
data set. No
classification involved
Determine
how well
discriminant
function
classifies
(biased)
(1) Classify data using
classification rule derived
from predictive function
(2) May be part of
validation analysis of
initial predictive function
(1) Classify data using
classification rule derived
from predictive function
(2) May be part of
validation analysis of initial
predictive function
Assumptions of linear
Discriminant model.
Requirements
No validation
required
No validation
required Validation required
Initial predictive function
must have been previously
validated
Geometric Representation
If we have n persons measured on m variables, the profile of each person can be
portrayed as a point in m dimensions. If we also know the group to which each person belongs,
and the groups differ in terms of average profiles (often called centroids), we might expect to
find different groups occupying different regions of the space. The less overlap noted among
intergroup profiles in that space, the more likely it is that discriminant analysis can help us
separate the groups.
The scatter diagram and projection in Figure 13.4 provides one way to show what
happens when a two-group discriminant function is computed. Suppose that we had two groups,
A and B, and two measures, X1 and X2, on each member of the two groups. We could plot in the
scatter diagram the association of variable X2 with X1 for each group, maintaining group identity
by the use of filled-in dots or open circles. The resultant ellipses enclose some specified
proportion of the points, say 95 percent, in each group. If a straight line is drawn through the two
points where the ellipses intersect and is then projected to a new axis Z, we can say that the
overlap between the univariate distributions A´ and B´ (represented by the shaded area) is
smaller than would be obtained by any other line drawn through the ellipses representing the
scatter plots. The shaded region under the curves can also be interpreted as representing the
probabilities of mistakes when classification is done on the basis of likelihoods only.
341
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 13.4 Graphical Illustration of Two-Group Discriminant Analysis
The important thing to note about Figure 13.4 is that the Z-axis expresses the twovariable
profiles of groups A and B as single numbers. That is, by finding a linear composite of
the original profile scores we can portray each profile as a point on a line. Thus, the axis Z
condenses the information about group separability (shown in the bivariate plot) into a set of
points on a single axis. Here, Z is the discriminant axis.
In most problems, we have more than two predictor variables. If so, each predictor can
represent a separate dimension (although we would be limited to three predictor variables if we
wished to plot the data). In any case, the basic objective is still to find one axis in the mdimensional
space that maximally separates the centroids of the two groups after the points are
projected onto this new axis.
In our discussion of multiple regression analysis we noted that one finds a linear
composite that maximizes the coefficient of multiple determination, R2
. Analogously, in twogroup
discriminant analysis we try to find a linear composite of the original variables that
maximizes the ratio of among-to-within groups variability. It should be noted that if m, the
number of predictor variables, is quite large, we shall be parsimonious by portraying among-towithin
groups variation in many fewer dimensions (actually a single dimension in the two group
case) than found originally.
A Numerical Example
Let us return to the example involving ready-to-eat cereals that was first presented in
Table 13.2 in the context of multiple regression. As recalled, we wished to see if amount of
protein and vitamin D influenced consumers’ evaluations of the cereals.
In the present case we shall assume that each of the ten consumer raters is simply asked
to classify the cereal into one of two categories: like or dislike. The (hypothetical) data appear in
Table 13.5, which has two predictor variables (again):
X1: the amount of protein (in grams) per two-ounce serving
X2: the percentage of minimum daily requirements of vitamin D per two-ounce serving
We first note in the table that the two groups are much more widely separated on X1
(protein) than they are on X2 (vitamin D). If we were forced to choose just one of the axes, it
would seem that X1 is a better bet than X2. However, there is information provided by the group
342
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
separation on X2, so we wonder if some linear composite of both X1 and X2 could do better than
X1 alone.
Table 13.5 Consumer Evaluations (Like Versus Dislike) of Ten Cereals Varying in Nutritional
Content
Person Evaluation
Protein
X1
Vitamin
D, X2
1
2
3
4
5
Mean
Dislike
Dislike
Dislike
Dislike
Dislike
2
3
4
5
6
4
4
2
5
4
7
4.4
6
7
8
9
10
Mean
Like
Like
Like
Like
Like
7
8
9
10
11
9
6
4
7
6
9
6.4
Grand Mean
Std Deviation
6.5
3.028
5.4
2.011
Figure 13.5 shows a scatter plot of the X1 and X2 data of Table 13.5. We note that perfect
discrimination can be achieved with X1 if we erect a line perpendicular to the horizontal axis
between the scale values of 6 and 7. On the other hand, there is no way that the use of X2 alone
would enable us to separate the groups. Given this picture, we would not be surprised if the best
linear composite places a considerably larger weight on X1 than on X2.
Figure 13.5 Scatter Plot of Two-Group Sample Data of Table
Why not use X1 alone, rather than a composite of X1 and X2? First, the data of Table 13.5
represent only a sample; it is quite possible that additional observations would show that X1
alone would not effect perfect discrimination between the two groups. Second, we have not
explicitly taken into consideration either the variability of X1 versus X2 or their correlation. One
343
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
of the nice features of discriminant analysis is that all three aspects of the data— centroid,
variance, and correlation—are considered when developing the linear composite that maximally
separates the groups.
As noted earlier, the key problem of two-group discriminant analysis is to find a new axis
so that projections of the points onto that axis exhibit the property of maximizing the separation
between group means relative to their within-groups variability on the composite.
This discriminant axis can be defined in terms of a set of weights—one for each predictor
variable axis—so that we have the following linear function:
where d1 and d2 are the weights that we seek.
Using a discriminant analysis program, we solve for the discriminant weights d1 and d2
that maximize the separation between the groups. This involves a procedure where two
simultaneous equations are solved, leading to the desired discriminant function:
The weights d1 and d2 are known and labeled on computer output as unstandardized
discriminant function coefficients. The general form for a linear discriminant function is Z = d0 +
d1X1 + d2X2, where d0 is the intercept or constant term in the function. Since the example used
deviations to calculate the weights d1 and d2, the function goes through the origin and d0 = 0.
Having found the discriminant function, it is a straightforward procedure to find discriminant
scores for the centroids of the two groups and the grand mean:
We may also apply this procedure to each of the 10 pairs of X1, X2 values in Table 13.5 to get the
linear composite. For example, the discriminant score for the first case in the disliker group is
Z = 0.368(2) - 0.147(4) = 0.148
Plotting the Discriminant Function
The original scatter plot of the ten observations is reproduced in Figure 13.6. However,
this time we also show the discriminant axis (linear composite) by passing a straight line through
the point (0.368, – 0.147) and the intersection of the original axes (i.e., the origin).
Alternatively, one could use any point with coordinates proportional to (0.368, −.147).
However, it should be noted that the scale unit on the discriminant axis in Figure 13.6 differs
from the original unit in which X1 and X2 are expressed. To maintain the original unit, d1 and d2
would have to be normalized. The original points can then be projected onto this new axis.
To illustrate, we show projections of the grand centroid and the centroids of the dislikers
and likers, respectively. (Similarly, all ten original points could be projected onto the
344
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
discriminant axis as well.) We note that the discriminant axis favors X1 (as we guessed it would)
by giving about 2.5 times the (absolute-value) weight (d1 = 0.368 versus d2 = −0.147 to X1) as is
given to X2.
Figure 13.6 Plot of the Discriminant Axis and Point Projections
The analysis output is shown in Exhibit 13.5. Note that the discriminant function
coefficients are different from those shown earlier, meaning that the discriminant scores for each
person also will differ. This is because the discriminant axis is not run through the origin, but
rather there is an intercept term. The unstandardized discriminant function coefficients evaluated
at group means sum to zero. The analysis is not affected by this, and all conclusions to be made
will not differ whether or not the axis is through the origin.
EXHIBIT 13.5 Discriminant Analysis for Cereal Evaluations
Group Statistics
EVALUATE Mean Std. Deviation
Valid N (listwise)
Unweighted Weighted
GRP 1 PROTEIN
VITAMIN
4.0000
4.4000
1.58114
1.81659
5
5
5.000
5.000
GRP 2 PROTEIN
VITAMIN
9.0000
6.4000
1.58114
1.81759
5
5
5.000
5.000
Total PROTEIN
VITAMIN
6.5000
5.4000
3.02765
2.01108
10
10
10.000
10.000
Tests of Equality of Group Means
Wilks’
Lambda F df1 df2 Sig.
PROTEIN
VITAMIN
.242
.725
25.000
3.030
1
1
8
8
.001
.120
Summary of Canonical Discriminant Functions
Eigenvalues
Function Eigenvalue % of Variance Cumulative %
Canonical
Correlation
1 3.860
a
100.0 100.0 .891
First 1 canonical discriminant functions were used in the analysis.
Wilks’ Lambda
Test of
Function(s)
Wilks’
Lambda Chi-Square df Sig.
1 .206 11.068 2 .004
345
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Standardized Canonical Discriminant Function Coefficients
Function
1
PROTEIN
VITAMIN
1.323
–.608
Structure Matrix
Function
1
PROTEIN
VITAMIN
.900
.313
Pooled within-groups correlations between discriminating variables and standardized canonical discriminant
functions. Variables ordered by absolute size of correlation within function.
Canonical Discriminant Function Coefficients
Function
1
PROTEIN
VITAMIN
(Constant)
.837
–.335
–3.632
Unstandardized coefficients
Functions at Group Centroids
EVALUATE
Function
1
1.00
2.00
–1.757
1.757
Unstandardized canonical discriminant functions evaluated at group means
Prior Probabilities for Groups
EVALUATE Prior
Cases Used in Analysis
Unweighted Weighted
1.00
2.00
Total
.500
.500
1.000
5
5
10
5.000
5.000
10.000
Classification Function Coefficients
EVALUATE
1.00 2.00
PROTEIN
VITAMIN
(Constant)
1.035
.706
–4.317
3.976
–.471
–17.081
Fisher’s linear discriminant functions
Classification Tables
b,c
EVALUATE Predicted Group
Membership
Total
1.00 2.00
Original Count 1.00
2.00
5
0
0
5
5
5
% 1.00
2.00
100.0
.0
.0
100.0
100.0
100.0
Cross-validated
a
Count 1.00
2.00
5
0
0
5
5
5
% 1.00
2.00
100.0
.0
.0
100.0
100.0
100.0
Cross validation is done only for those cases in the analysis. In cross validation, each case is classified by the
functions derived from all case other than that case.
100.0% of original grouped cases correctly classified.
100.0% of cross-validated grouped cases correctly classified.
346
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Classifying the Persons
It is well and good to find the discriminant function, but we are still interested in how
well the function classifies the ten cases. The classification problem, in turn, involves two
additional questions: (a) how well does the function assign the known cases in the sample; and
(b) how well does it assign new cases not used in computing the function in the first place?
To answer these questions we need an assignment rule. One rule that seems intuitively
plausible is based on Figure 13.6. A classification boundary between the two groups, Zcrit, can be
identified as being midway between the means of the function for each of the two groups. To
classify an individual, if Zi > Zcrit then the individual belongs in one group, while if Zi < Zcrit then
the individual goes into the other group. As can be seen from Figure 13.6, no misassignments
will be made if we adopt the rule:
• Assign all cases with discriminant scores that are on the left of the midpoint (1.596)
to the disliker group.
• Assign all cases with discriminant scores that are on the right of the midpoint (1.596)
to the liker group.
That is, all true dislikers will be correctly classified as such, as will all true likers. This
can be shown by a 2 ⋅ 2 table, known as a classification or confusion matrix, as shown later in
Table 13.6. We see that all entries fall along the main diagonal. For example, had any of the five
true dislikers been called likers, the first row and second column would contain not a zero but,
rather, the number of such misassignments.
The application of this rule can be stated in equivalent terms:
• Substitute the centroid of each group in the discriminant function and find the
respective group scores (in our case, 0.824 for dislikers and 2.368 for likers).
• For any new case, compute the discriminant score and assign the case to that group
whose group score is closer.
This rule makes two specific assumptions: (a) the prior probability of a new case falling
into each of the groups is equal across groups; and (b) the cost of misclassification is equal
across groups.
If a higher probability existed for likers, we could reduce the expected probability of
misclassification by moving the cutting point of 1.596 to the left (closer to 0.824, the mean score
for dislikers) so as to give a wider interval for the larger (likers) group. Similarly, if the cost of
misclassifying a liker is higher than that for misclassifying a disliker, the cutting point would
also be moved closer to 0.824.
A second point of interest concerns the tendency for classification tables (i.e., the
confusion matrix) based on the sample used to develop the classification function to show better
results than would be found upon cross-validation with new cases. That is, some capitalization
on chance takes place in discriminant analysis, and one needs a way to measure this bias of an
inflated percentage of correctly classified observations, just as cross-validation should be
employed in multiple regression.
347
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Procedures are available to develop a truer summary of the degree of correct assignments
than would be obtained from fresh data (Lachenbruch, 1975; Klecka, 1980; Frank, Massy, &
Morrison, 1965).
The most frequently suggested validation approach is the holdout method, in which the
data set is randomly split into two subsamples. One subsample is used to develop the
discriminant function and the other subsample is used as “fresh data” to test the function. It is
suggested that this split-sample validation be replicated by using a different convention for the
random assignment of the observations. Whether a researcher uses only one validation sample or
replications, there may be problems associated with the sampling procedures used to arrive at the
analysis and validation subsamples. These problems relate to the number of variables included in
the discriminant function (when a stepwise procedure is used) and to inferences that the
researcher might make concerning the classificatory power of resulting discriminant functions.
There may be differing numbers of variables included on any given replication. Also,
there can be a wide range in the correct classification performance of the validation replications,
although the analysis functions are quite stable.
There is another dimension to the validation process discussed above. For the analysis
and validation samples there is a measure of the percentage of cases that would have been
correctly classified based on chance alone. Obviously, the researcher is interested in whether the
discriminant function is a significant improvement over chance classification.
Because our “Cereal Data” achieved 100 percent correct classification, we will now
switch to another example to demonstrate these measures of chance classification. The
classification or “confusion” matrix shown at the end of Table 13.6 was derived from a
discriminant analysis where accident insurance purchasers and non-purchasers were predicted
using demographic variables.
Overall, 74.16 percent of all subjects surveyed were correctly classified as purchasers and
non-purchasers of accident insurance. This observed classification was found to be significant at
the .001 level (χ2 = 24.59, df = 1) and (Q = 69.58, df = 1). Thus, observed classification is
significantly different from expected chance classification.
Tests of Group Differences
Morrison (1969) considered the question of how well variables discriminate by
formulating a likelihood ratio to estimate chance classification. This likelihood analysis provides
a criterion that may be used to compare the proportion of correctly classified observations with
the proportion expected by chance. This proportion, designated the proportional chance criteria,
or Cpro (Morrison, 1969), is expressed as where
α = the proportion of customers in the sample categorized as purchasers
p = the true proportion of purchasers in the sample
(1 − α) = the proportion of the sample classified as non-purchasers
(1 − p) = the true proportion of non-purchasers in the sample
This likelihood analysis states that 65.94 percent of the overall sample is expected to
receive correct classification by chance alone. The proportional chance criterion, Cpro, has been
used mainly as a point of reference for subjective evaluation (Morrison, 1969), rather than the
basis of a statistical test to determine if the expected proportion differs from the observed
proportion that is correctly classified.
348
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
This relationship between chance and observed proportions can be tested using a Z
statistic of the form
where Pcc is the percent of observations correctly classified, Cpro = pα + (1 − p)(1 − α).
Thus for the example problem, the difference between expected and actual overall correct
classification is significantly different at the .01 level. This overall test of significance suggests
that further analysis should be conducted to determine the source of the deviation from chance
expectations.
Table 13.6 Classification Table for Accident Insurance Purchasers
Fequency, Row %
Chi-Square Contribution
Predicted Purchase Predicted
Non-Purchase
Row Total
Row Percentage
Actual
Purchase
n=22
66.7%
15.41
n=11
33.3%
6.46
n=33
11.1%
Actual
Non-Purchase
N=66
24.9%
1.92
N=199
75.1%
.80
n=265
88.9%
Column Totals
Column Percentage
n=88
29.5%
n=21
70.5%
n=298
100%
Percent of Cases Correctly Classified = 221/298 = 74.16%
Chi-Square = 24.599 df=1, Significance < .001
Divergence may be present in any of the classification matrix cells (i.e., purchasers or
non-purchasers, that are either correctly or incorrectly categorized), and thus each may be tested
to determine whether its proportion differs from chance.
Classification and Misclassification Within Groups
The analysis to determine the source of deviation is conducted using the maximum
chance criterion, designated Cmax (Morrison, 1969). Cmax is the maximum expected correct
classification for a selected group of interest. The computation of Cmax is based on the
assumption that all observations are categorized as coming from that group: for example, given
that all 298 purchasers and non-purchasers were classified as purchasers, then the maximum
correct classification, Cmax, would be expressed
Because we are interested in the correct classification of insurance purchasers, the test of
classification involves asking if the 66.67-correct insurance purchaser classification differs
significantly from the 11.1-percent maximum expected chance classification. The Z statistic is
used to test this relationship as shown for the example analysis.
349
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
* Significant at the .001 level.
This test may also be conducted for the other cells in the classification matrix:
Thus cell Z11 shows that observed classification is significantly greater than is expected
to occur by chance classification alone. The analysis of cells (1,2) and (2,1) shows that observed
and expected misclassification results differ in that purchasers are misclassified into cell (1,2)
less often than expected by chance, and non-purchasers are misclassified into cell (2,2) more
often than expected by chance. Thus the discriminant functions appear to shift the classification
of subjects toward the purchaser categories, as demonstrated by significantly greater than
expected classification in the upper and left portions of the classification matrix.
At the most basic level, the validity of discriminant function analysis lies in the stability
of the coefficients derived. These coefficients are the basis of classifying, profiling, and
evaluating the underlying discriminant dimensions. It would be valuable to have a validation
procedure that uses all the sample data for evaluating the stability of parameter estimates while
allowing unbiased estimation of error rates. The jackknife statistic and the U-method have been
proposed as such procedures. It is beyond the scope of this book to discuss these specialized
techniques, and the reader is referred to the excellent references already cited.
When a discriminant analysis is run using a computer program, there is another option for
validation, especially when a small sample is involved. The leave-one-out procedure operates by
having each case (or observation) classified into a group according to classification functions
computed from all the data except the case being classified. While we do not deal specifically
with the classification of cases other than those from which the discriminant function is
developed, it should be recognized that once the discriminant function is developed and validated
for the population it may be used to classify other groups within the population.
Testing Statistical Significance
While the discriminant function perfectly classifies the ten cases of the calibration sample
in our cereal likes and dislikes illustration, we still have not tested whether the group centroids
differ significantly. This is analogous to testing for the significance of R2
in multiple regression.
Tests of the equality of group centroids can also proceed on the basis of an F-ratio that, in turn, is
calculated from a variability measure known as Mahalanobis squared distance. We do not delve
into the technical details of Mahalanobis squared distance, other than to say that it is like
ordinary (Euclidian) squared distance that is computed between two centroids in a space with
correlated axes and different measurement units.
Two other measures are widely used for testing overall statistical significance related to a
discriminant function. The canonical correlation coefficient is a measure of the association that
350
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
summarizes how related the discriminant function is to the groups. We discuss canonical
correlation analysis later in this chapter. An indirect, and most widely used, approach to test for
the statistical significance of the discriminant function examines the ability of the variables to
discriminate among the group beyond the information that has been extracted by the previously
computed functions. This is known as residual discrimination, and is measured by the statistic
Wilks’ lambda (also called the U-statistic). Wilks’ lambda is a multivariate measure of group
differences over discriminating variables and can be calculated in several ways. In general, it is
calculated such that values of lambda near zero indicate high discrimination, and when it equals
its maximum value of 1.0 the group centroids are equal and there is no discrimination (Klecka,
1980).
The statistical significance of a discriminant analysis merely provides answers to the
questions:
• Is there a relationship?
• Do the predictor variables discriminate among the groups?
• Does this particular discriminant function contribute to the relationship?
Statistical significance says nothing about how strong a relationship is, how much
difference exists between the groups, or to what extent a function contributes to overall
discrimination. Moreover, tests of statistical significance are sensitive to sample size. For
example, with a large sample size it is not difficult to get a significant F-ratio, even though
classification accuracy is poor. Finally, there is a need to go beyond statistical significance and
test for the practical (or substantive) significance. There is no single index of practical
significance that is widely accepted.
Relative Importance of Predictor Variables
Because the original variables X1 and X2 were expressed in different units and display
different standard deviations as well, the analyst generally standardizes the discriminant weights
before examining their relative importance. A simple standardization procedure multiplies each
discriminant weight (unstandardized) by the total sample standard deviation of that variable:
Standardized coefficients allow only an ordinal interpretation of variable importance.
These coefficients are not appropriate to assess the relative discriminatory power of the variables
included in the analysis. Mosteller and Wallace (1963) offer an appropriate measure of relative
discriminating power:
where
Ij = the importance value of the jth variable
dj = unstandardized discriminant coefficient for the jth variable
Xjk = mean of the jth variable for the kth group
The relative importance weights may be interpreted as the portion of the discriminant
score separation between the groups that is attributable to the jth variable. Since a relative
importance value shows the value of a particular variable relative to the sum of the importance
values of all variables, the relative importance of a variable (Rj) is given by Aw and Waters
(1974):
351
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The end result of using this procedure is shown in Table 13.7, which is taken from a
study of teenage smoking designed to see if consumption values (as represented by underlying
had smoked and those who had never smoked (Albaum, Baker, Hozier, & Rogers, 2002). As
these data show, although nine variables were included in the discriminant function, slightly
more than 88 percent of the total discrimination was accounted for by only three variables.
Table 13.7 Relative Importance of Consumption Values for Teenage Smokers and Nonsmokers
Variables
(Factors)
Standardized
Coefficients
Unstandardized
Coefficients (kj)
Have
Smoked
Mean (X1)
Never
Smoked
Mean (X2)
[kj(Xj1-Xj2)]
(Ij)
Relative
Importance
Weight (Rj)
1 -0.298 -0.299 -0.0989 0.0908 0.0567 7.9%
2 0.091 0.091 0.0301 -0.0277 0.0053 0.7%
3 0.832 0.864 0.2861 -0.2627 0.4741 66.4%
4 0.344 0.345 0.1144 -0.1050 0.0757 10.6%
5 0.364 0.366 0.1211 -0.1112 0.0850 11.9%
6 -0.018 -0.018 -0.0061 0.0056 0.0002 0.0%
7 -0.004 -0,004 -0.0014 0.0013 0.0000 0.0%
8 -0.007 -0.007 -0.0023 0.0021 0.0000 0.0%
9 0.161 0.161 0.0533 -0.0490 0.0165 2.3%
0.7135 100.0%
Correctly Classified 60.2% (cross-validated grouped cases); Cpro – 51.8%
Wilk’s Lambda - .886 (prob. < .01)
Canonical Correlation - .337
Source: Albaum, Baker, Hozier and Rogers, 2002, p.68.
Determining relative importance of the predictor variables in discriminant analysis
becomes increasingly difficult when more than two groups are involved. Although various
coefficients and indices can be determined, interpretation becomes critical since more than one
discriminant function may be involved.
Multiple Discriminant Analysis
All of the preceding discussion regarding objectives and assumption structure applies to
multiple discriminant analysis as well. Accordingly, discussion of this section will be
comparatively brief. What primarily distinguishes multiple discriminant analysis from the twogroup
case is that more than one discriminant function may be computed. For example, if we
have three groups we can compute, in general, two nonredundant discriminant functions (as long
as we also have at least two predictor variables). In general, with G groups and m predictors we
can find up to the lesser of G − 1, or m, discriminant functions.
Not all the discriminant functions may be statistically significant, however. Moreover, it
turns out to be a characteristic of multiple discriminant analysis that the first function accounts
for the highest proportion of the among-to within-groups variability; the second function, the
next highest; and so on. Accordingly, we may want to consider only the first few functions,
particularly when the input data are rather noisy or unreliable to begin with. There remains the
problem of interpretation of the functions.
352
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Multiple discriminant analysis is considerably more detailed than might be surmised by
this brief overview. Interested readers should consult more extensive and advanced discussions
of the topic such as Johnson and Wichen (2002), Hair, Tatham, Anderson, and Black (1998), and
Stevens (1996).
OTHER CRITERION-PREDICTOR ASSOCIATION MULTIVARIATE
TECHNIQUES
Chi-Square Automatic Interaction Detection (CHAID) Analysis
One of today’s key tools in database management and data mining is CHAID. This tool
allows the researcher to explore relationships where both the number of respondents and the
number of variables is very large. Inherent in these relationships, and in the resulting predictive
models, is the phenomenon of interaction in which the response to changes in the level of one
predictor variable depends on the level of some other predictor (or predictors). When interaction
effects exist, the simple additive property of individual predictor-variable contributions to
changes in the criterion no longer holds.
The major problem is that when mining data using exploratory analyses of observational
and survey data, one does not ordinarily know which predictors are interactive (and how they
interact). CHAID is insensitive to various forms of interaction.
CHAID is a sequential merging routine that merges nonsignificant predictor variable
categories by analyzing a series of two-way cross-tabulations of the predictor and dependent
variable.
While the predictor variables might originally be (a) nominal-, (b) ordinal-, or (c) interval
scaled, these predictors are all recorded into categorical (nominal-scaled) variables with one of
two characteristics:
1. Order across classes can be disregarded
2. Order across classes can be retained
For example, if one of the original predictors is employment status (white-collar, bluecollar,
unemployed) one may wish to treat this variable as an unordered polytomy. Some other
variable such as age may be recorded into 18–20; 21–23; 24–26 years, and so forth. In this case
one would probably wish to maintain order across classes. Each predictor variable can be
designated by the researcher as unordered or ordered, independently of the rest. It has been
suggested that the primary goals of CHAID are segmentation analysis and an exploration of the
relationships between variables. CHAID is often an appropriate technique in the following
circumstances:
• We do not want to make strong assumptions about relationships.
• There are more than just a few potential explanatory variables.
• There are a large number of observations.
• AID also offers a variety of uses:
• Identification of demographic variables that discriminate and predict the dependent variable
category, such as heavy users
• Identification of interaction effects for further analysis, such as logistic regression
• Data mining to find significance among the hundreds of possible crosstabulation
combinations
• Identification of predictor variables that contribute little to explaining the dependent variable
353
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Basically, CHAID performs a series of contingency table analyses of the data (similar in
spirit to the bivariate cross-tabulation analyses discussed in Chapter 12). After splitting the initial
sample on the basis of the “best” predictor, the process is repeated on each of the two
subsamples, and so on. At each step of the sequential splitting and merging process, CHAID
looks for the best available split, not the best set of some number of final groups. The main result
of this is a tree structure that shows three statistics at each stage:
1. The predictor variable leading to the best binary split and how that predictor is split
2. The number of persons assigned to each of the two subgroups
3. The criterion-variable splits on each of the two subgroups
An Example of CHAID
Advertising executives have conducted research to identify profitable segments for a
sweepstakes promotion.
Profit levels associated with responses are found to be $35 when the respondent pays for
a subscription; −$7 when a respondent does not subscribe, but receives an introductory issue and
follow-up sales materials; −$.15 for nonrespondents for mailing costs.
Of particular interest is the extent to which a respondent is profitable. That is, when asked
to consider the subscription to magazines through the associated sweepstakes plan, which market
segments are most likely to subscribe?
A large database (n = 81,040) was available that contained information about their
response to the sweepstakes offer and known information about eight different demographic and
credit card ownership predictors. All respondents had recently received a sweepstakes offer and
had a known response that was recorded in the criterion variable to indicate that he or she was a
subscriber (coded “Paid Respondent”), respondent-nonsubscriber (coded “Unpaid Respondent”),
or a “nonrespondent.”
CHAID was applied to this large data bank, and the results are shown in the tree diagram
of Figure 13.7. At the top of the diagram we note that the total-sample probability of being a
subscriber is 0.0059. That is, 478 out of the total of 81,040 recipients of the sweepstakes offer
paid for a subscription. The first variable on which the total sample is split involves the size of
the respondent’s household, followed by age, income, and presence of a bankcard in the
household.
Figure 13.8 shows a summary map of the CHAID tree that is numbered by the split order.
These splits are detailed in Table 13.8, the gain summary, that shows the key segments for
sweepstakes profitability based on the measure of profitability gain.
We observe that Node 8 provides the most gain (profitability) and is defined by the
predictor variables as a two-person household, headed by an individual aged 55–64 with income
of $15,000 or more. If we were to label this profile, it is the classic “empty-nester” whose
children are no longer at home and now has a little extra money and time for their own pursuits,
one of which is reading.
354
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 13.7 CHAID Sweepstakes Tree
355
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 13.8 CHAID Tree Summary Map
Table 13.8 CHAID Gain
Gain Summary
Target variable: Response to Sweepstakes Promotion
Node-by-Node Cumulative
Node Node: n Node:% Gain Index (%) Node: n Node:% Gain Index (%)
8
12
4
14
5
9
13
1
18
7
619
650
1191
1482
6800
2359
2283
25384
3011
271
0.76
0.80
1.47
1.83
8.39
2.91
2.82
31.32
3.72
0.33
0.69
0.49
0/45
0.19
0.11
0.06
0.05
0.03
-0.05
-0.15
3628.50972
2572.90367
2381.18629
1008.12275
554.14567
309.87554
271.75404
168.00454
-283.25238
-787.99468
619
1269
2460
3942
10742
13101
15384
40768
79938
80209
0.76
1.57
3.04
4.86
13.26
16.17
18.98
50.31
98.64
98.97
0.69
0.59
0.52
0.40
0.21
0.19
0.17
0.08
0.02
0.02
3628.50972
3087.81316
2745.70235
2092.45705
1118.66098
973.02898
868.95912
432.51311
115.27446
122.22261
Node 12 is a 3–4 person household with a young to middle-age head of household who
holds a bankcard.
The process of identifying profitable segments continues until we observe that Nodes 18
and 7 offer a negative gain (meaning they are unprofitable). We conclude that we should not
target the large family segment having five or more family members (Node 18), or the poor
empty-nester segment that is identical to our most profitable segment (Node 8), except without
the income base.
The upshot of all of this is that by data mining with CHAID, we can identify 50.31
percent of our database that promises to be profitable and another 49.69 percent (39,441
individuals) that may be unprofitable. We further observe that the average gain (profit) drops
from 0.17 to 0.08 with the inclusion of Node 1, which has an average gain of 0.03 per person for
that segment. The successive targeting of the next most profitable segment can do much to
maximize profits, as can eliminating the negative gain segments. Simply refining the database by
deleting nonconforming profiles often does much to improve profitability of the marketing
effort.
In short, this kind of information underscores the importance of certain types of
demographic variables in this case, and leads to a more detailed understanding of the kinds of
segments that responded by subscribing to this sweepstakes magazine offer.
While other summary statistics (e.g., probabilities, chi-square) can also be shown, the
above outputs represent the principal ones. It should be emphasized that a basic assumption
underlying the application of CHAID is that variables useful in explaining one part of the
database are not necessarily those most effective for explaining another part.
356
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
A few other key considerations, in the nature of restrictions or criteria for the stopping of
the splitting sequence, enter into the actual application of CHAID:
1. All partitionings of the sample may be subject to a requirement that the proportionate
reduction in the criterion-variable probability exceed some level (specified by the
researcher). This is to guard against partitionings that do not appreciably reduce
variation in the criterion variable.
2. To qualify for further splitting, a group must have a probability greater than some
level (again specified by the researcher). This is to guard against splits that,
pragmatically speaking, are not worth the effort (for example, where the group is
already quite homogeneous).
3. In addition to the control parameters above, the researcher may place an upper limit
on the total number of groups formed and/or the minimum number of persons (or
objects) in each group.
A few caveats regarding the application of CHAID should be mentioned at this point.
First, CHAID is generally designed for really large samples, on the order of 1,000 or more. Since
many versions of CHAID will take many predictors, the program has ample opportunity to
capitalize on chance variation in the data. Moreover, no statistical inferential apparatus is
associated with the approach. This suggests the value of cross-validating results on new data or,
possibly, double cross-validating by applying CHAID to separate halves of the sample.
Second, CHAID, being a sequential search procedure, does not specify an explicit model
in the way, for example, that Discriminant Analysis does. In this regard it is often useful to use
CHAID as an initial screening device to find those predictors that appear to be most prominent in
accounting for criterion dependence. This can then be followed by the formulation of an explicit
logit regression model that includes main effect and interaction terms of specific interest to the
research. The joint use of CHAID with other statistical techniques, such as CART models
(classification and regression trees), provides additional useful features for exploratory data
analysis.
Third, despite the explicit concern for interaction, CHAID is found to be insensitive to
various forms of interaction. Since CHAID only examines the immediate effect of a predictor as
a two-way table and not future possible splits, any interactions that are not one-stage will not be
identified.
One of the nice characteristics of CHAID is its simplicity of output, which facilitates an
ease of understanding by researchers and managers alike. The output takes the form of a tree
diagram, in which one can follow the result of each split as it takes place. This is illustrated in
the brief case study analyzed using the SPSS program AnswerTree, a suite of programs that
includes CHAID and CART models. The data is that supplied by Magidson (1993) in the
original SPSS CHAID 6.0 program. AnswerTree is compatible with SPSS and all current
Windows operating systems.
357
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Canonical Correlation Analysis
Canonical correlation is a generalization of multiple correlation analysis to two or more
criterion (dependent) variables. To illustrate the technique of canonical correlation, let us
consider a radial tire study which has three main predictor variables:
X1: General interest in the product class of steel-belted radials
X2: Whether the firm’s old brand was chosen as the respondent’s last purchase of
replacement tires
X3: Pre-exposure (before seeing the commercials) interest in the firms’ new brand of
steel-belted radials
Now let us assume that two criterion variables are involved:
Y1: Believability of the claims made in the firm’s new TV commercial;
Y2: Post-exposure interest in the claims made in the firm’s new brand (as before).
What we would like to find out is how highly correlated the group of two criterion
variables is with the group of three predictors. Moreover, we would like to find a linear
composite of the Y-variable set and a (different) linear composite of the X-variable set that will
produce a maximal correlation.
This is what canonical correlation is all about. Canonical correlation deals with (a) both
description and statistical inference of (b) a data matrix partitioned into at least two predictors
where (c) all variables are interval-scaled and (d) the relationships are assumed to be linear.
Thompson (1984, p. 10) notes that, in more general terms, canonical correlation analysis can be
used to investigate the following research questions:
• To what extent can one set of two or more variables be predicted or explained by
another set of two or more variables?
• What contribution does a single variable make to the explanatory power of the set of
variables to which the variable belongs?
• To what extent does a single variable contribute to predicting or explaining the
composite of the variables in the variable set to which the variable does not belong?
• What different dynamics are involved in the ability of one variable set to explain, in
different ways, different portions of the other variable set?
• What relative power do different canonical functions have to predict or explain
relationships?
• How stable are canonical results across samples or sample subgroups?
• How closely do obtained canonical result conform to expected canonical results?
Table 13.9 shows the results of the radial tire problem. In general, with p criteria and q
predictors, one can obtain more than a single pair of linear composites—up to a maximum of the
smaller of p and q. Thus, in our case we would obtain two pairs of linear composites,
uncorrelated across pairs, with the first pair exhibiting maximum correlation.
In general, the canonical correlation of successive pairs decreases; that is, the first pair
displays the highest correlation, the second pair the next highest, and so on. All composites are
mutually uncorrelated across pairs. However, as it turned out, only the first pair of linear
composites was statistically significant at the 0.05 alpha level; hence, only this pair of weights is
shown.
358
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table 13.9 Result of Canonical Correlation
Canonical Weights Structure Correlations
Criterion set
Y1 -0.110 0.594
Y2 1.069 0.997
Predictor set
X1 0.346 0.539
X2 0.141 0.382
X3 0.817 0.930
Canonical correlation 0.582
Input Correlations For Canonical Analysis
Y1 Y2 X1 X2 X3
Y1 1.000
Y2 0.659 1.000
X1 0.202 0.315 1.000
X2 0.097 0.218 0.086 1.000
X3 0.321 0.539 0.226 0.258 1.000
Table 13.9 shows that the canonical correlation between the two batteries is 0.582. As is
the case with multiple correlations, this measure varies between zero (no correlation) and one
(perfect correlation). Since mutual association between the pair of batteries is involved, we can
say that the pair of linear composites account for (0.582)2
, or 34 percent of the shared variation
between the two batteries.
The canonical weights for the criterion set show that Y2 (post-exposure interest) is the
dominant variable in the criterion set; its canonical weight is 1.069. The dominant variable in the
predictor set is X3; its weight is 0.817. Since all variables are standardized to zero mean and unit
standard deviation before the analysis, the weights are already in standardized form.
Table 13.9 really says that if we formed a linear composite of the criterion variables using
the canonical weights:
Tc = −0.110Ys1 + 1.069Ys2
and another linear composite of the predictors, using the canonical weights:
Tp = 0.346Xs1 + 0.141Xs2 + 0.817Xs3
and took these two columns of numbers (the canonical weight scores) and correlated them, the
result would be a simple correlation of 0.582 (the canonical correlation).
The structure correlations are also shown in Table 13.9. These are the simple correlations
of each original variable with the canonical scores of its own battery’s linear composite. Again
we note that Y2 is the most important variable (structure correlation of 0.997) in the criterion set
and X3 the most important variable (structure correlation of 0.930) in the predictor set. Indeed, as
noted in the input correlation matrix, the simple correlation between Y2 and X3 is 0.539, almost as
high as the correlation between the full batteries.
Since this is a rather sophisticated technique, we refer the reader to other sources for
further discussion of the procedure, such as Thompson (1984) and Levine (1977).
Probit and Logit Analysis
At the beginning of this chapter we emphasized that multiple regression is a useful technique for
estimating a dependent variable from a set of independent variables. A major assumption of
standard regression analysis is that the dependent variable is continuous and at least intervalscaled.
When the dependent variable is dichotomous, binary regression can be used.
359
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
What can the researcher do when the problem requires estimating relationships where the
dependent variables that are nonmetric (i.e., nominal- or ordinal-scaled)? When regression
analysis is misapplied, we often see results with an unnecessarily high proportion of unexplained
variance (i.e., a lower R2
), misleading estimates of the effects of the predictor variables, and an
inability to make statements about the probability of given responses. To overcome these
problems, probit and logit have been developed.
Probit and logit deal with the same type of problem as regression, the problem of
predicting a dependent variable that is nominal or ordinal scaled. They differ solely in the
assumption made about the frequency distribution of the response:
• In probit, the response is assumed to be normally distributed.
• In logit, a logistic distribution is assumed.
Reported research about marketing applications seems to indicate that logit is the favored
technique of the two (Malhotra, 1984). Logit analysis is often used in choice based conjoint
analysis where the dependent variable is the brand chosen (discussed in Chapter 14).
Explanations of how these techniques work and the underlying theories behind them tend
to be technical. For a more detailed explanation we refer the reader to the references already
cited and to the excellent works of Aldrich and Nelson (1984), Liao (1994), and Menard (2001).
Path Analysis/Causal Modeling
In recent years, marketing researchers have become increasingly interested in the useful
application of path analysis and structural equation modeling as an approach to causal modeling.
Causal modeling provides the researcher with a systematic methodology for developing and
testing theories in marketing. From all this interest has emerged the development of structural
equation modeling procedures. These procedures blend two basic techniques:
• Factor analysis
• Simultaneous equation regression
Both of these techniques use multiple measures and simultaneous equation models to
estimate the path of causation.
The essence of what happens with this type of analysis can be seen from a simple
illustration. In Figure 13.10 we show a system with four unobservable constructs, each of which
has more than one observable measure. In some situations the measures themselves may be
correlated, which precludes finding a strong reliability from either simple indices or regressions
that relate the indices themselves as means of showing relations between the unobserved
constructs.
To overcome this problem and better estimate the relevant parameters, some rather
complicated procedures have been developed which estimate both between the constructs (bij)
and their measures and the links between the constructs themselves (aij) simultaneously. One
widely used technique to estimate these parameters is LISREL (structural equation software).
360
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 13.10 Hypothetical Causal Structure
SOURCE: From Albaum, G., Baker, K., Hozier, G., and Rogers, R., “Smoking Behavior, Information Sources, and Other Intervention
Activities,” in Journal of Consumer Affairs, 36/1:68. Copyright © 2002. Reprinted by permission of the University of Wisconsin Press.
Another technique is PLS (partial least squares). A third technique, available from SPSS,
is AMOS (structural equation and path analysis software). These techniques, and structural
equation modeling in general, are highly technical methods of analyses and are recommended
only for experienced users.
SUMMARY
Chapter 13 has focused on a number of multivariate techniques. Our discussions of
multiple regression and discriminant analysis have been more detailed than the descriptions of
other techniques. We have deliberately focused on those techniques that are most often used in
market research.
We started the chapter with a discussion of multiple regression and described such
measures as regression coefficients, the coefficient of determination, and the product-moment
coefficient.
Next we discussed discriminant analysis, one of the most widely used criterion-predictor
methods. We described a sample problem and solved it numerically. Discriminant analysis was
also discussed as a tool for prediction and classification.
This chapter concluded with a brief introduction to some less well-known techniques for
analyzing between-set dependence: CHAID, canonical analysis, correspondence analysis, and
probit and logit analysis. At this point, it might appear that an almost bewildering array of
techniques has been paraded before the reader.
We have tried to discuss the principal assumption structure of each technique, appropriate
problems for applying it, and sufficient numerical applications to give the reader a feel for the
kind of output generated by each program. Further discussion of the use of multivariate analysis
in marketing can be found in Hair, Anderson, Tatham.
Our coverage of so vast and complex a set of methods is limited in depth as well as
breadth. The fact remains, however, that marketing researchers of the future will have to seek
grounding in multivariate methodology, if current research trends are any indication.
This grounding will probably embrace three facets: (a) theoretical understanding of the
techniques; (b) knowledge of the details of appropriate computer algorithms to implement the
techniques; and (c) a grasp of the characteristics of substantive problems in marketing that are
relevant for each of the methods. Fortunately, most statistical analysis packages have programs
for using these techniques.
361
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
Aaker, D. A. (1981). Multivariate analysis in marketing. In Multivariate analysis in marketing
(2nd ed.). Palo Alto, CA: Scientific Press.
Albaum, G., Baker, K. G., Hozier, G. C., Jr., & Rogers, R. D. (2002). Smoking behavior,
information sources, and consumption values of teenagers: Implications for public policy and
other invention failures. Journal of Consumer Affairs, 36(1), 50–76.
Aldrich, J. H., & Nelson, F. D. (1984). Linear probability, logit, and probit models. Beverly
Hills, CA: Sage.
Aw, R. W., & Waters, D. (1974). A discriminant analysis of economic, demographic, and
attitudinal characteristics of bank charge-card customers. Journal of Finance, 29, 973–980.
Bagozzi, R. (1980). Causal models in marketing. New York: Wiley.
Berk, K. N., & Carey, P. (2000). Data analysis with Microsoft Excel. Pacific Grove, CA:
Duxbury Press.
Blalock, H. M., Jr. (1962). Causal inferences in nonexperimental research. Chapel Hill:
University of North Carolina Press.
Carroll, J. D., Green, P. E., & Schaffer, C. M. (1987). Comparing interpoint distances in
correspondence analysis. Journal of Marketing Research, 24, 445–450.
Carroll, J. D., Green, P. E., & Schaffer, C. M. (1986). Interpoint distance comparisons in
correspondence analysis. Journal of Marketing Research, 23, 271–280.
Clausen, S. E. (1998). Applied correspondence analysis: An introduction. Thousand Oaks, CA:
Sage.
Crask, M. R., & Perrault,W. D., Jr. (1979, February). Validation of discriminant analysis in
marketing research. Journal of Marketing Research, 14, 60–68.
Curtis, M. E. (1990, Sept. 3). Avoid hazards of stepwise regression. Marketing News, 24, 26.
Frank, R. E., Massy,W. F., & Morrison, D. G. (1965, August). Bias in multiple discriminant
analysis. Journal of Marketing Research, 2, 250–258.
Green, P. E., Tull, D. S., & Albaum, G. (1988). Research for marketing decisions (5th ed.).
Englewood Cliffs, NJ: Prentice Hall.
Greenacre, M. J. (1993). Correspondence analysis in practice. London: Academic Press.
Hair, J. F., Tatham, R. L., Anderson, R. E., & Black, W. (1998). Multivariate data analysis (5th
ed.). Upper Saddle River, NJ: Prentice Hall.
362
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Johnson, R. A., & Wichen, D. W. (2002). Applied multivariate statistical analysis (5th ed.).
Upper Saddle River, NJ: Prentice Hall.
Klecka,W. R. (1980). Discriminant analysis. Beverly Hills, CA: Sage.
Lachenbruch, P. A. (1975). Discriminant analysis. New York: Hafner Press.
Levine, M. S. (1977). Canonical analysis and factor comparison. Beverly Hills, CA: Sage.
Liao, T. F. (1994). Interpreting probability models: Logit, probit, and other generalized linear
models. Thousand Oaks, CA: Sage.
Magidson, J., & SPSS, Inc. (1993). SPSS for Windows CHAID Release 6.0. Chicago: SPSS, Inc.
Malhotra, N. K. (1984, February). The use of linear logit models in marketing. Journal of
Marketing
Research, 21, 20–31. Menard, S. (2001). Applied logistic regression analysis (2nd ed.).
Thousand Oaks, CA: Sage.
Monroe, K. B., & Petroshius, S. M. (n.d.). Developing causal priorities in marketing. Working
Paper, Virginia Polytechnic Institute and State University, Blacksburg, VA.
Morrison, D. G. (1969, May). On the interpretation of discriminant analysis. Journal of
Marketing Research, 6, 156–163.
Mosteller, F., & Wallace, D. L. (1963, June). Influence in an authorship problem. Journal of the
American Statistical Association, 55, 275–309.
Smith, S. M. (2004). A note on the interpretation and analysis of the linear discriminant model
for prediction and classification. Retrieved May 18, 2004, from
http://marketing.byu.edu/htmlpages/ tutorials/discriminant.htm
Steckel, J. H., & Vanhonacker, W. R. (1991). Crossvalidating regression models in marketing
research. (Working Paper # MARK-91-10.) Stern School of Business, New York University.
Stevens, J. (1996). Applied multivariate statistics for the social sciences. Mahwah, NJ: Lawrence
Erlbaum.
Thompson, B. (1984). Canonical correlation analysis: Uses and interpretations. Beverly Hills,
CA: Sage.
Whitlark, D., & Smith, S. (2001, Summer). Using correspondence analysis to map relationships:
It’s time to think strategically about positioning and image data. Marketing Research, 2, 23–27.
Whitlark, D., & Smith, S. (2004). Measuring brand performance using online questionnaires:
Advantages and issues with “pick any” data. Provo, UT: Marriott School, Brigham Young Univ.
363
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 14
MULTIVARIATE ANALYSIS II
FACTOR ANALYSIS, CLUSTERING METHODS,
MULTIDIMENSIONAL SCALING AND CONJOINT ANALYSIS
In this chapter we discuss two techniques that do not require data to be partitioned into
criterion and predictor variables. Rather, it is the entire set of interdependent relationships that
are of interest. We discuss factor analysis as a methodology that identifies the commonality
existing in sets of variables. This methodology is useful to identify consumer lifestyle and
personality types.
Continuing with analyses that do not partition the data, a second set of methods is
effective in clustering respondents to identify market segments. The fundamentals of cluster
analysis are described using examples of respondents and objects grouped because they have
similar variable scores.
Third, we discuss two sets of multivariate techniques, multidimensional scaling and
conjoint analysis, that are particularly well suited (and were originally developed) for measuring
human perceptions and preferences. Multidimensional scaling methodology is closely related to
factor analysis, while conjoint analysis uses a variety of techniques (including analysis of
variance designs and regression analysis) to estimate parameters, and both techniques are related
to psychological scaling (discussed in Chapter 9). The use of both multidimensional scaling and
conjoint analysis in marketing is widespread.
AN INTRODUCTION TO THE BASIC CONCEPTS OF FACTOR ANALYSIS
Factor analysis is a generic name given to a class of techniques whose purpose often
consists of data reduction and summarization. Used in this way, the objective is to represent a set
of observed variables, persons, or occasions in the form of a smaller number of hypothetical,
underlying, and unknown dimensions called factors.
Factor analysis operates on the data matrix. The form of the data matrix can be flipped
(transposed) or sliced to produce different types, or modes, of factor analysis. The most widely
used mode of factor analysis is the R-technique (relationships among items or variables are
examined), followed distantly by the Q-technique (persons or observations are examined). These,
together with other modes, are identified in Exhibit 14.1. “Creative” marketing researchers may
find S- and T-techniques helpful when analyzing purchasing behavior or advertising recall data.
The P- and O-techniques might be appropriate for looking at the life cycle of a product class, or
perhaps even changes in demographic characteristics of identified market segments.
EXHIBIT 14.1 Modes of Factor Analysis
Six distinct modes of factor analysis have been identified (Stewart, 1981, p. 53). The alternative modes of
factor analysis can be portrayed graphically. The original data set is viewed as a variables/persons/occasions
matrix (a). R-type and Q-type techniques deal with the variables/persons dichotomy (b). In contrast, P-type
and O-type analyses are used for the occasions/ variables situation and S-type and T-type are used when the
occasions/persons relationship is of interest (c).
364
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Factor analysis does not entail making predictions using criterion and predictor variables;
rather, interest is centered on summarizing the relationships involving a whole set of variables.
Factor analysis has three main qualities:
1. The analyst is interested in examining the strength of the overall association among variables,
in the sense that a smaller set of factors (linear composites of the original variable) may be
able to preserve most of the information in the full data set. Often one’s interest will stress
description of the data rather than statistical inference.
2. No attempt is made to divide the variables into criterion versus predictor sets.
3. The models typically assume that the data are interval-scaled.
The major substantive purpose of factor analysis is to search for (and sometimes test)
structure in the form of constructs, or dimensions, assumed to underlie the measured variables.
This search for structure is accomplished by literally partitioning the total variance associated
with each variable into two components: (a) common factors and (b) unique factors. Common
factors are the underlying structure that contributes to explaining two or more variables. In
addition, each variable is usually influenced by unique individual characteristics not shared with
other variables, and by external forces that are systematic (non-random) and not measured
(possibly business environment variables). This non-common factor variance is called a unique
factor and is specific to an individual variable. Graphically, this may appear as diagrammed in
Figure 14.1, in which four variables are reduced to two factors that summarize the majority of
the underlying structure, and four unique factors containing information unique to each variable
alone.
365
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
The structure of the factors identified by the analysis will, of course, differ for each data
set analyzed. In some applications the researcher may find that the variables are so highly
correlated that a single factor results. In other applications the variables may exhibit low
correlations and result in weak or ill-defined factors. In response to these eventualities, the
researcher may revise the variable list and add additional variables to the factor analysis. The
process of adding and eliminating variables is common in factor analysis when the objective of
the analysis is to identify those variables most central to the construct and to produce results that
are both valid and reliable. Behavioral and consumer researchers have employed these methods
to develop measurement instruments such as personality profiles, lifestyle indexes, or measures
of consumer shopping involvement. Thus, in addition to serving as a data reduction tool, factor
analysis may be used for to develop behavioral measurement scales.
We use a numerical example to illustrate the basic ideas of factor analysis. A grocery
chain was interested in the attitudes (in the form of images) that customers and potential
customers had of their stores. A survey of 169 customers was conducted to assess images. The
information obtained included 14 items that were rated using a seven-category semantic
differential scale. These items are shown in Table 14.1. The resulting data set would then be a
matrix of 169 rows (respondents) by 14 columns (semantic differential scales). These data will
be analyzed as R-type factor analysis.
Figure 14.1 The Concept of Factor Analysis
Table 14.1 Bipolar Dimensions Used in Semantic Differential Scales for Grocery Chain Study
Inconvenient location—Convenient location
Low-quality products—High-quality products
Modern—Old-fashioned
Unfriendly clerks—Friendly clerks
Sophisticated customers—Unsophisticated customers
Cluttered—Spacious
Fast check-out—Slow check-out
Unorganized layout—Organized layout
Enjoyable shopping experience—Unenjoyable shopping experience
Bad reputation—Good reputation
Good service—Bad service
Unhelpful clerks—Helpful clerks
Good selection of products—Bad selection of products
Dull—Exciting
366
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
IDENTIFYING THE FACTORS
If we now input the raw data into a factor analysis program, correlations between the
variables are computed, as is the analysis. Some relevant concepts and definitions for this type of
analysis are presented in Exhibit 14.2.
A factor analysis of the 14 grocery-chain observed variables produces a smaller number
of underlying dimensions (factors) that account for most of the variance. It may be helpful to
characterize each of the 14 original variables as having an equal single unit of variance that is
redistributed to 14 underlying dimensions or factors. In every factor analysis solution, the
number of input variables equals the number of common factors plus the number of unique
factors to which the variance is redistributed. In factor analysis, the analysis first determines how
many of the 14 underlying dimensions or factors are common, and then the common factors are
interpreted.
EXHIBIT 14.2 Some Concepts and Definitions of R-Type Factor Analysis
Factor Analysis: A set of techniques for finding the underlying relationships between many
variables and condensing the variables into a smaller number of dimensions
called factors.
Factor: A variable or construct that is not directly observable, but is developed as a
linear combination of observed variables.
Factor Loading: The correlation between a measured variable and a factor. It is computed by
correlating factor scores with observed manifest variable scores.
Factor Score: A value for each factor that is assigned to each individual person or object from
which data was collected. It is derived from a summation of the derived weights
applied to the original data variables.
Communality (h
2
): The common variance of each variable summarized by the factors, or the
amount (percent) of each variable that is explained by the factors. The
uniqueness component of a variable’s variance is 1-h
2
.
Eigenvalue: The sum of squares of variable loadings of each factor. It is a measure of the
variance of each factor, and if divided by the number of variables (i.e., the total
variance), it is the percent of variance summarized by the factor.
Table 14.2 identifies the proportion of variance associated with each of the 14 factors
produced by the analysis where the factors were extracted by Principal Component analysis.
Principal Components, one of the alternative methods of factor analysis, is a method of factoring
which results in a linear combination of observed variables possessing such properties as being
orthogonal to each other (i.e., independent of each other), and the first principal component
represents the largest amount of variance in the data, the second representing the second largest,
and so on. It is the most conservative method. For a more detailed discussion of the alternative
methods, see Kim and Mueller (1978a, 1978b). In column two, the eigenvalues are reported.
Computed as the sum of the squared correlations between the variables and a factor, the
eigenvalues are a measure of the variance associated with the factor. The eigenvalues reported in
Table 14.2 are a measure of the redistribution of the 14 units of variance from the 14 original
variables to the 14 factors. We observe that factors 1, 2, 3, and 4 account for the major portion
367
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
(66.5 percent) of the variance in the original variables. In Figure 14.2, a scree plot depicts the
rapid decline in variance accounted for as the number of factors increase. This chart graphs the
eigenvalues for each factor. It is a useful visual tool for determining the number of significant
factors to retain. The shape of this curve suggests that little is added by recognizing more than
four factors in the solution (the additional factors will be unique to a single variable).
Figure 14.2 Scree Plot for Grocery Chain Data Factors
An accepted rule-of-thumb states that if a factor has an associated eigenvalue greater than
or equal to 1.0, then the factor is “common” and a part of the solution. This rule-of-thumb is
closely aligned with the intuitive decision rules associated with the scree chart. When we
observe an eigenvalue less than 1.0, the factor accounts for less variance than was input by a
single input variable.
Table 14.2 Factor Eigenvalues and Variance Explained for Grocery Chain Study
Initial eigenvalues
Factor Total % Variance Cumulative %
1 5.448 38.917 38.917
2 1.523 11.882 49.799
3 1.245 8.89 58.689
4 1.096 7.827 66.516
5 0.875 6.247 72.763
6 0.717 5.12 77.883
7 0.62 4.429 82.312
8 0.525 3.747 86.059
9 0.478 3.413 89.472
10 0.438 3.131 92.603
11 0.326 2.329 94.932
12 0.278 1.985 96.917
13 0.242 1.729 98.646
14 0.19 1.354 100
Table 14.3 shows the matrix of factor loadings, or correlations of the variables with the
factors. If each factor loading in each column were squared, the sum would equal the eigenvalue
shown in Table 14.2. Squaring the loadings (h2
) and summing across the columns results in the
amount of variance in the variables that is to be explained by the factors. These values are known
as communalities.
368
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Interpreting the Factors
The interpretation of the factors is subjectively based on the pattern of correlations
between the variables and the factors. The factor loadings provide the basis for interpreting the
factors; those variables having the highest loading contribute most to the factor and thereby
should receive the most weight in interpreting of the factor.
In factor analysis two solutions typically are obtained. The initial solution is based on
certain restrictions: (a) there are k common factors; (b) underlying factors are orthogonal (i.e.,
uncorrelated or independent) to each other; and (c) the first factor accounts for as much variance
as possible, the second factor for as much of the residual variance as possible left unexplained by
the first factor, and so on (Kim & Mueller, 1978a). The second solution is accomplished through
rotation aimed at getting loadings for the variables that are either near one or near zero for each
factor. The most widely used rotation is called varimax, a method of rotation which leaves the
factors uncorrelated. This rotation maximizes the variance of a column of the factor loading
matrix, thus simplifying the factor structure and making the factors more interpretable.
In Table 14.3, Factor 1 is identified by four variables. The major contributions are made
by the variables “Quality of products,” “Reputation,” Selection of products,” and “Modernism.”
We might interpret this factor as the construct up-to-date quality products.
Factor 2 is identified by three variables: “Sophistication of customers,” “Speed of
checkout,” and “Dull/Exciting.” This factor might be interpreted as the fast and exciting for
sophisticated customers. Factor 3 is explained by the variables “Friendliness of clerks,”
Cluttered/ Spacious,” and “Layout.” One interpretation of this factor is that it represents the
construct of friendliness of store. Finally, the last factor is defined by five variables. These all
might be a reflection of satisfaction with the shopping experience.
Table 14.3 Varimax Rotated Factor Loading Matrix for Grocery Chain Data Factor
Factor Loadings (Sorted)
Variable 1 2 3 4 Communalities (h2
)
Selection of products 0.799 -0.066 0.184 0.126 0.692
Quality of products 0.789 0.033 0.237 -0.081 0.687
Reputation 0.724 -0.283 0.096 0.296 0.702
Modern -0.665 0.216 -0.071 -0.221 0.543
Customers -0.235 0.781 -0.139 0.069 0.689
Check-out 0.170 0.720 -0.040 -0.326 0.655
Dull -0.284 0.668 -0.227 0.046 0.581
Cluttered 0.070 -0.162 0.894 0.052 0.834
Layout 0.323 -0.058 0.742 0.150 0.681
Friendliness of clerks 0.199 -0.298 0.606 0.433 0.683
Location 0.013 0.218 0.011 0.735 0.587
Service -0.257 0.339 -0.393 -0.588 0.680
Helpfulness of clerks 0.281 -0.338 0.290 0.597 0.634
Shopping experience -0.353 0.448 -0.183 -0.552 0.664
The example of Table 14.3 depicts a set of factors with loadings that are generally high or
low. However, the loadings are often in the .4 to .8 range, questioning at what level the variables
make significant enough input to warrant interpretation in the factor solution. A definitive
369
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
answer to this question cannot be given; it depends on sample size. If the sample size is small,
correlations should be high (generally .6 and above) before the loadings are meaningful. But as
the sample size increases, the meaning of correlations of lower value may be considered
(generally .4 and above).
Overall, it should be obvious that more than one interpretation may be possible for any
given factor. Moreover, it may be that a factor may not be interpretable in any substantive sense.
This may or may not be a problem, depending upon the objective of the factor analysis. If done
for data-reduction purposes, and the results will be used in a further analysis (such as multiple
regression or discriminant analysis), being unable to interpret substantively may not be critical.
One use of factor analysis is to identify those variables that reflect underlying dimensions or
constructs. Once identified, the researcher can select one or more original variables for each
underlying dimension to include in a subsequent multivariate analysis. This ensures that all
underlying or latent dimensions are included in the analysis.
Factor Scores
Once the underlying factors are identified, the resulting factors or constructs are often
interpreted with respect to the individual respondents. Simply stated, we would like to know how
each respondent scores on each factor. Does the respondent have high scores on the up-to-datequality
products and friendliness of store constructs? In general, since a factor is a linear
combination (or linear composite) of the original scores (variable values), it can be shown as
where Fi is the factor score for the ith factor, the an are weights (factor loadings for the n
variables) and the Xn are respondent i’s standardized variable scores.
Most factor analysis computer programs produce these factor score summates and merge
them with the original data file. Augmenting the data set with factor scores enables the analyst to
easily prepare descriptive or predictive analyses that segment respondents scoring high on a
given factor. In short, factor scores (rather than original data values) can be used in subsequent
analysis.
Correspondence Analysis
Correspondence analysis can be viewed as a special case of canonical correlation
analysis that is analogous to a principal components factor analysis for nominal data. Canonical
correlation, as we have just seen, examines the relations between two sets of continuous
variables; correspondence analysis examines the relations between the categories of two discrete
variables. Correspondence analysis can be applied to many forms of contingency table data,
including frequency counts, associative data (pick k of n), or dummy variables. This analysis
develops ratio-scaled interpoint distances between the row and column categories that depict
accurate and useful positioning maps.
Correspondence analysis is often used in positioning and image studies where the
researcher wants to explore the relationships between brands, between attributes, and between
brands and attributes. In strategic terms, the marketing researcher may want to identify (a)
closely competitive brands, (b) important attributes, (c) how attributes cluster together, (d) a
brand’s competitive strengths, and most importantly (e) ideas for improving a brand’s
competitive position (Whitlark & Smith, 2001).
370
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
According to Clausen (1998, p. 1), the main purpose of correspondence analysis is
twofold:
1. To reveal the relationships in a complex set of variables by replacing the data with a simpler
data matrix without losing essential information.
2. To visually display the points in space. This helps interpretation. Correspondence analysis
analyzes the association between two or more categorical variables and represents the
categorical marketing research data with a two- or three-dimensional map.
Figure 14.3 Example Correspondence Analysis Map of Logistical Services Providers
Another way of thinking about correspondence analysis is as a special type of crosstabulation
analysis of contingency tables. Categories with similar distributions will be placed in
close proximity to each other, while those with dissimilar distributions will be farther apart. The
technique is capable of handling large contingency tables.
An example from Whitlark and Smith (2004) will be used to illustrate this technique.
When administering long and complicated surveys, it is sometimes impractical to collect
attribute ratings for all brands and products the researcher is interested in. The respondent may
have little or no knowledge of some brands, and providing many ratings can be an onerous task
that produces poor quality data. Obviously the respondent should not be asked to evaluate
unfamiliar brands. If the list of brands is large, the respondent will be asked to only evaluate a
subset of the list that contains familiar brands. But more to the point of using correspondence
analysis, in this situation the researcher will often simplify the data collection task by not using
rating scales, but will instead give the respondent a list of attributes and ask them to check off the
ones they feel best describe a particular brand that they are familiar with. This type of question
produces “pick k of n” associative data, where k represents the number of attributes a respondent
associates with a brand and n represents the total number of descriptive attributes included in the
survey.
The correspondence analysis map shown in Figure 10.9 describes 12 companies
providing communications, logistics consulting, and software support to a group of international
freight handlers and shippers. Three well-known companies in the United States—Oracle, Nokia,
and FedEx—are labeled using their names, and a series of nine less-familiar companies are
labeled using letters of the alphabet. The 12 companies were evaluated by nearly 800 freight
371
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
handlers, who indicated which attributes (pick k of n) best described the companies. The twodimensional
map of the companies accounts for nearly 90 percent of the variance in the data.
The results of this analysis show that Oracle and Nokia are perceived as being the most
innovative and as industry leaders, while FedEx offers a relevant and total solution.
Correspondence analysis is a very helpful and interesting analysis tool that provides
meaning and interpretation to large, complex data sets that contain this type of data. A more
detailed explanation of this technique will be found in the excellent works by Clausen (1998),
Greenacre (1993), and Carroll, Green, and Schaffer (1986; 1987).
BASIC CONCEPTS OF CLUSTER ANALYSIS
Like factor analysis, clustering methods are most often applied to object ⋅ variable
matrices. However rather than focus on the similarity of variables as in factor analysis, the usual
objective of cluster analysis is to separate objects (or people) into groups such that we maximize
the similarity of objects within each group, while maximizing the differences between groups.
Cluster analysis is thus concerned ultimately with classification, and its techniques are part of a
field of study called numerical taxonomy (Sokal & Sneath, 1963; Sneath & Sokal, 1973). Cluster
analysis can also be used to (a) investigate useful conceptual schemes derived from grouping
entities; (b) generate a hypothesis through data exploration; and (c) attempt to determine if types
defined through other procedures are present in a data set (Aldenderfer & Blashfield, 1984).
Thus, cluster analysis can be viewed as a set of techniques designed to identify objects, people,
or variables that are similar with respect to some criteria or characteristics. As such, it seeks to
describe so-called natural groupings, as described in Exhibit 14.3.
EXHIBIT 14.3 Clustering for Segmentation
From a marketing perspective, it should be made clear that a major application of cluster analysis is for
segmentation. To illustrate, consider a financial services company that wanted to do a segmentation study among
its sales force of dealers/agents (Swint, 1994/1995). The objective was to identify the characteristics of “high
producers” and “mediocre producers” of sales revenue. The desire was to profile the dealers/agents and segment
them with respect to motivations, needs, work styles, beliefs, and behaviors. The data were analyzed using cluster
analysis, and six cluster solutions emerged. The six clusters were then subject to discriminant analysis to how well
the individual clustering attributes actually discriminated between the segments. The end result of all these analyses
was six well defined clusters that identified the producer segments.
The type of clustering procedure that we shall discuss assigns each respondent (object) to
one and only one class. Objects within a class are usually assumed to be indistinguishable from
one another. Thus in cluster analysis, we assume here that the underlying structure of the data
involves an unordered set of discrete classes. In some cases we may also view these classes as
hierarchical in nature, where some classes are divided into subclasses.
Primary Questions
Clustering procedures can be viewed as preclassificatory in the sense that the analyst has
not used prior information to partition the objects (rows of the data matrix) into groups. We note
that partitioning is performed on the objects rather than the variables; thus, cluster analysis deals
with intact data (in terms of the variables). Moreover, the partitioning is not performed a priori
but is based on the object similarities themselves. Thus, the analyst is assuming that clusters
372
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
exist. This type of presupposition is different from the case in discriminant analysis (discussed in
Chapter 13), where groups of objects are predefined by a variable:
• Most cluster-analysis methods are relatively simple procedures that are usually not supported
by an extensive body of statistical reasoning.
• Cluster-analysis methods have evolved from many disciplines, and the inbred biases of these
disciplines can differ dramatically.
• Different clustering methods can (and do) generate different solutions from the same data set.
• The strategy of cluster analysis is structure-seeking, although its operation is structure
imposing.
Given that no information on group definition in advance, we can identify four important
considerations in selecting (or developing) cluster analysis algorithms. We must decide:
1. What measure of inter-object similarity is to be used, and how is each variable to be
weighted in the construction of such a summary measure?
2. After inter-object similarities are obtained, how are the classes of objects to be formed?
3. After the classes have been formed, what summary measures of each cluster are appropriate
in a descriptive sense—that is, how are the clusters to be defined?
4. Assuming that adequate descriptions of the clusters can be obtained, what inferences can be
drawn regarding their statistical reliability?
Choice of Proximity Measure
The choice of a proximity, similarity, or resemblance measure (all three terms will be
used synonymously here) is an interesting problem in cluster analysis. The concept of similarity
always raises the question: Similarity with respect to what? Proximity measures are viewed in
relative terms—two objects are similar, relative to the group, if their profiles across variables are
close or if they share many aspects in common, relative to those which other pairs share in
common.
Most clustering procedures use pairwise measures of proximity. The choice of which
objects and variables should be included in the analysis, and how they should be scaled, is
largely a matter of the researcher’s judgment. The possible measures of pairwise proximity are
many. Generally speaking, these measures fall into two classes: (a) distance-type measures
(Euclidean distance); and (b) matching-type measures. A simple application illustrating the
nature of cluster analysis using a distance measure is shown in Exhibit 14.4.
EXHIBIT 14.4 A Simple Example of Cluster Analysis
373
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
We can illustrate cluster analysis by a simple example. The problem is to group a set of twelve branches of
a bank into three clusters of four branches each. Groups will be formed based on two variables, the number of
men who have borrowed money (X1) and the number of women who have borrowed money (X2). The branches
are plotted in two dimensions in the figure.
We use a proximity measure, based on Euclidean distances between any two branches. Branches 2 and 10
appear to be the closest together. The first cluster is formed by finding the midpoint between branches 2 and 10
and computing the distance of each branch from this midpoint (this is known as applying the nearest-neighbor
algorithm). The two closest branches (6 and 8) are then added to give the desired-size cluster. The other clusters
are formed in a similar manner. When more than two dimensions (that is, characteristics) are involved, the
algorithms become more complex and a computer program must be used for measuring distances and for
performing the clustering process.
Selecting the Clustering Methods
Once the analyst has settled on a pairwise measure of profile similarity, some type of
computational routine must be used to cluster the profiles. A large variety of such computer
programs already exist, and more are being developed as interest in this field increases. Each
clustering program tends to maintain a certain individuality, although some common
characteristics can be drawn out. The following categories of clustering methods are based, in
part, on the classification of Ball and Hall (1964):
1. Dimensionalizing the association matrix. These approaches use principal-components or other
factor-analytic methods to find a dimensional representation of points from interobject association
measures. Clusters are then developed on the basis of grouping objects according to their pattern
of component scores.
2. Nonhierarchical methods. The methods start right from the proximity matrix and can be
characterized in three ways:
a. Sequential threshold. In this case a cluster center is selected and all objects within a
prespecified distance threshold value are grouped. Then a new cluster center is selected and the
process is repeated for the unclustered points, and so on. (Once points enter a cluster, they are
removed from further processing.)
b. Parallel threshold. This method is similar to the preceding method, except that several cluster
centers are selected simultaneously and points within a distance threshold level are assigned to
the nearest center; thresholds can then be adjusted to admit fewer or more points to clusters.
c. Optimizing partitioning. This method modifies categories (a) or (b) in that points can later be
reassigned to clusters on the basis of optimizing some overall criterion measure, such as
average within-cluster distance for a given number of clusters.
3. Hierarchical methods. These procedures are characterized by the construction of a hierarchy or
tree-like structure. In some methods each point starts out as a unit (single point) cluster. At the
next level the two closest points are placed in a cluster. At the following level a third point joins
the first two, or else a second two-point cluster is formed based on various criterion functions for
assignment. Eventually all points are grouped into one larger cluster. Variations on this procedure
involve the development of a hierarchy from the top down. At the beginning the points are
partitioned into two subsets based on some criterion measure related to average within-cluster
distance. The subset with the highest average within-cluster distance is next partitioned into two
subsets, and so on, until all points eventually become unit clusters.
374
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
While the above classes of programs are not exhaustive of the field, most of the more
widely used clustering routines can be classified as falling into one (or a combination) of the
above categories. Criteria for grouping include such measures as average within-cluster distance
and threshold cutoff values. The fact remains, however, that even the optimizing approaches
achieve only conditional optima, since an unsettled question in this field is how many clusters to
form in the first place.
A Product-Positioning Example of Cluster Analysis
Cluster analysis can be used in a variety of marketing research applications. For example,
companies are often interested in determining how their products are positioned in terms of
competitive offerings and consumers’ views about the types of people most likely to own the
product.
For illustrative purposes, Figure 14.4 shows the result of a hypothetical study conducted
for seven sport cars, six types of stereotyped owners, and 13 attributes often used to describe
cars.
Figure 14.4 Complete-Linkage Analysis of Product-Positioning Data
The inter-object distance data were based on respondents’ degree-of-belief ratings about
which attributes and owner types described each car. In this case, a complete-linkage algorithm
was also used to cluster the objects (Johnson, 1967). The complete linkage algorithm starts by
finding the two points with the minimum Euclidean distance. However, joining points to clusters
is accomplished by maximizing the distance from a point in the first cluster to a point in the
second cluster. Looking first at the four large clusters, we note the car groupings:
• Mazda Miata, Mitsubishi Eclipse
• VW Golf
• Mercedes 600 SL, Lexus SC, Infinity G35
• Porsche Carrera
375
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In this example, the Porsche Carrera is seen as being in a class by itself, with the
attributes high acceleration and high top speed. Its’ perceived (stereotyped) owners are rally
enthusiast and amateur racer.
Studies of this type enable the marketing researcher to observe the interrelationships
among several types of entities—cars, attributes, and owners. This approach has several
advantages. For example, it can be applied to alternative advertisements, package designs, or
other kinds of communications stimuli. That is, the respondent could be shown blocks of
advertising copy (brand unidentified) and asked to provide degree-of-belief ratings that the brand
described in the copy possesses each of the n features.
Similarly, in the case of consumer packaged goods, the respondent could be shown
alternative package designs and asked for degree-of-belief ratings that the contents of the
package possess various features. In either case one would be adding an additional set (or sets) of
ratings to the response sets described earlier. Hence, four (or more) classes of items could be
represented as points in the cluster analysis.
Foreign Market Analysis
Companies considering entering foreign markets for the first time, as well as those
considering expanding from existing to new foreign markets, have to do formal market analysis.
Often a useful starting point is to work from a categorization schema of potential foreign
markets. Cluster analysis can be useful in this process.
To illustrate, we use the study by Green and Larsen (1985). In this study, 71 nations were
clustered on the basis of selected economic characteristics and economic change. The specific
variables used were (a) growth in Gross Domestic Product; (b) literacy rate; (c) energy
consumption per capita; (d) oil imports; and (e) international debt. Variables a, d, and e were
measured as the change occurring during a specified time period.
Table 14.4 Composition of Foreign Market Clusters
CLUSTER 1 CLUSTER 3 CLUSTER 5
Belgium Finland Ethiopia Colombia Venezuela
Canada Norway Ghana Costa Rica Yugoslavia
Denmark Switzerland India Ecuador El Salvador
Sweden New Zealand Liberia Greece Iran
USA France Libya Hong Kong Tunisia
Germany Ireland Madagascar Jordan Indonesia
Netherlands Italy Mali Mexico Nigeria
UK Austria Senegal Paraguay Malawi
Australia Portugal
CLUSTER 2 CLUSTER 4
Cameroon Honduras Brazil Uruguay
Central African Republic Nicaragua Chile Spain
Egypt Morocco Israel Sri Lanka
Somalia Ivory Coast Japan Thailand
Togo Tanzania Korea Turkey
Zaire Pakistan Peru Argentina
Zambia Philippines Guatemala
Singapore Kenya
SOURCE: From Green, R. T. & Larsen, T., Export Markets and Economic Change, 1985 (working paper).Reprinted with permission.
Clustering was accomplished by use of a K-means clustering routine. This routine is a
nonhierarchical method that allocates countries to the group whose centroid is closest, using a
376
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Euclidean distance measure. A total of five clusters was derived based on the distance between
countries and the centers of the clusters across the five predictor variables. The number of
clusters selected was based on the criteria of total within-cluster distance and interpretability. A
smaller number of clusters led to a substantial increase of within-cluster variability, while an
increase in the number of clusters resulted in group splitting with a minimal reduction in
distance. The composition of the clusters is shown in Table 14.4.
Computer Analyses
There are many computer programs available for conducting cluster analysis. Most
analysis packages have one or more routines. Smaller, more specialized packages (such as PCMDS)
for cluster routines are also available. Finally, some academicians have developed their
own cluster routines that they typically make available to other academicians for no charge.
MULTIDIMENSIONAL SCALING (MDS) ANALYSIS
Multidimensional scaling is concerned with portraying psychological relations among
stimuli—either empirically-obtained similarities, preferences, or other kinds of matching or
ordering—as geometric relationships among points in a multidimensional space. In this approach
one represents psychological dissimilarity as geometric distance. The axes of the geometric
space, or some transformation of them, are often (but not necessarily) assumed to represent the
psychological bases or attributes along which the judge compares stimuli (represented as points
or vectors in his or her psychological space).
Figure 14.5 Nonmetric MDS of 10 U.S. Cities
I. Geographic locations of ten U.S. cities II. Airline distance between ten U.S. cities
III. Nonmetric (Rank Order) Distance Data IV. Original (•) and recovered (o) city locations via nonmetric MDS
377
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
In this section we start with an intuitive introduction to multidimensional scaling that
uses a geographical example involving a set of intercity distances. In particular, we show how
MDS takes a set of distance data and tries to find a spatial configuration or pattern of points in
some number of dimensions whose distances best match the input data.
Let us start by looking at Panel I of Figure 14.5. Here we see a configuration of ten U.S.
cities, whose locations have been taken from an airline map (Kruskal & Wish, 1978). The actual
intercity distances are shown in Panel II. The Euclidean distance between a pair of points i and j,
in any number of r dimensions, is given by
In the present case, r = 2 and only two dimensions are involved. For example, we could
use the map to find the distance between Atlanta and Chicago by (a) projecting their points on
axis 1 (East-West), finding the difference, and squaring it; (b) projecting their points on axis 2
(North-South) and doing the same; and then (c) taking the square root of the sum of the two
squared differences.
In short, it is a relatively simple matter to go from the map in Panel I to the set of
numerical distances in Panel II. However, the converse is not so easy. And that is what MDS is
all about.
Suppose that we are shown Panel II of Figure 14.4 without the labels so that we do not
even know that the objects are cities. The task is to work backward and develop a spatial map.
That is, we wish to find, simultaneously, the number of dimensions and the configuration (or
pattern) of points in that dimensionality so that their computed interpoint distances most closely
match the input data of Panel II. This is the problem of metric MDS.
Next, suppose that instead of the more precise air mileage data, we have only rank order
data. We can build such a data matrix by taking some order-preserving transformation in Panel II
to produce Panel III. For example, we could take the smallest distance (205 miles between New
York and Washington) and call it 1. Then we could apply the same rules and rank order the
remaining 44 distances up to rank 45 for the distance (2,734 miles) between Miami and Seattle.
We could use a nonmetric MDS program to find the number of dimensions and the configuration
of points in that dimensionality such that the ranks of their computed interpoint distances most
closely match the ranks of the input data.
In this example where the actual distance data is considered, it turns out that metric MDS
methods can find, for all practical purposes, an exact solution (Panel I). However, what is rather
surprising is that, even after downgrading the numerical data to ranks, nonmetric methods can
also achieve a virtually perfect recovery as well.
Panel IV shows the results of applying a nonmetric algorithm to the ranks of the 45
numbers in Panel III. Thus, even with only rank-order input information, the recovery of the
original locations is almost perfect.
We should quickly add, however, that neither the metric nor nonmetric MDS procedures
will necessarily line up the configuration of points in a North-South direction; all that the
methods try to preserve are relative distances. The configuration can be arbitrarily rotated,
translated, reflected, or uniformly stretched or shrunk by so-called configuration congruence or
378
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
matching programs, so as to best match the target configuration of Panel I. None of these
operations will change the relative distances of the points.
Psychological Versus Physical Distance
The virtues of MDS methods are not in the scaling of physical distances but rather in
their scaling of psychological distances, often called dissimilarities. In MDS we assume that
individuals act as though they have a type of “mental map”, (not necessarily visualized or
verbalized), so that they view pairs of entities near each other as similar and those far from each
other as dissimilar. Depending on the relative distances among pairs of points, varying degrees
of dissimilarity could be imagined.
We assume that the respondent is able to provide either numerical measures of his or her
perceived degree of dissimilarity for all pairs of entities, or, less stringently, ordinal measures of
dissimilarity. If so, we can use the methodology of MDS to construct a physical map in one or
more dimensions whose interpoint distances (or ranks of distances, as the case may be) are most
consistent with the input data.
This model does not explain perception. Quite the contrary, it provides a useful
representation of a set of subjective judgments about the extent to which a respondent views
various pairs of entities as dissimilar. Thus, MDS models are representations of data rather than
theories of perceptual processes.
Classifying MDS Techniques
Many different kinds of MDS procedures exist. Accordingly, it seems useful to provide a
set of descriptors by which the methodology can be classified. These descriptors are only a
subset of those described by Carroll and Arabie (1998); and Green, Carmone and Smith (1989)
1. Mode: A mode is a class of entities, such as respondents, brands, use occasions, or attributes of a
multiattribute object.
2. Data array: The number of ways that modes are arranged. For example, in a two-way array of
single mode dissimilarities, the entities could be brand-brand relationships, such as a respondent’s
rating of the ijth brand pair on a 1–9 point scale, ranging from 1 (very similar) to 9 (very
different). Hence, in this case, we have one mode, two-way data on judged dissimilarities of pairs
of brands.
3. Type of geometric model: Either a distance model or a vector or projection model (the latter
represented by a combination of points and vectors).
4. Number of different sets of plotted points (or vectors): One, two, or more than two.
5. Scale type: Nominal-, ordinal-, interval-, or ratio-scaled input data.
Data Mode/Way
In marketing research most applications of MDS entail either single mode, two-way data,
or two-mode, two-way data. Single mode, two-way data are illustrated by input matrices that are
square and symmetric, in which all distinct pairs of entities (e.g., brands) in a I ⋅ I matrix are
often judgment data expressing relative similarity/dissimilarity on some type of rating scale. The
data collection instructions can refer to the respondent making judgments that produce data
representing pair-wise similarity, association, substitutability, closeness to, affinity for,
congruence with, co-occurrence with, and so on. Typically, only I(I – 1)/2 pairs are evaluated
and entered into the lower- or upper-half of the matrix. This is the case because of the
symmetric relationship that is assumed to exist. MDS solutions based on single mode, two-way
input data lead to what are often called simple spaces—that is, a map that portrays only the set of
379
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
I points, as was shown in Figure 14.4. Pairs of points close together in this geometric space are
presumed to exhibit high subjective similarity in the eyes of the respondent.
Another popular form of marketing research data entails input matrices that represent
two-mode, two-way relationships, such as the following six examples:
1. A set of I judges provide preference ratings of J brands
2. Average scores (across respondents) of J brands rated on I attributes
3. The frequency (across respondents) with which J attributes are assumed to be associated with I
brands
4. The frequency (across respondents) with which respondents in each of I brand-favorite groups
pick each of J attributes as important to their brand choice
5. The frequency (across respondents) with which each of J use occasions is perceived to be
appropriate for each of I brands
6. The frequency (across respondents) with which each of J problems is perceived to be associated
with using each of I brands.
These geometric spaces are often called joint spaces in that two different sets of points
(e.g., brands and attributes) are represented in the MDS map. In Figure 14.6 we observe that
brands are positioned in 3-dimensional space as points and the attributes are vectors extending a
distance of 1 unit away from the origin. The brands project onto each attribute vector to define
their degree of association with that attribute. The further out on the vector the brand projects,
the stronger the association. In some cases three or more sets of entities may be scaled.
Figure 14.6 MDPREF Joint Space MDS Map
Type of Geometric Model
In applications of single-mode, two-way data the entities being scaled are almost always
represented as points (as opposed to vectors). However, in the case of two-mode, two-way data,
the two sets of entities might each be represented as points or, alternatively, one set may be
represented as points while the other set is represented as vector directions. In the latter case the
termini of the vectors are often normalized to lie on a common circumference around the origin
of the configuration.
The point-point type of two-mode, two-way data representation is often referred to as an
unfolding model (Coombs, 1964). If the original matrix consists of I respondents’ preference
evaluations of J brands, then the resulting joint-space map has I respondents’ ideal points and J
brand points. Brand points that are near a respondent’s ideal point are assumed to be highly
380
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
preferred by that respondent. Although the original input data may be based on between-set
relationships, if the simple unfolding model holds, one can also infer respondent to- respondent
similarities in terms of the closeness of their ideal points to each other. Brand to- brand
similarities may be analogously inferred, based on the relative closeness of pairs of brand points.
The point-vector model of two-mode, two-way data is a projection model in which one
obtains respondent i’s preference scale by projecting the J brand points onto respondent i’s
vector (Figure 14.5). Point-vector models also show ideal points or points of “ideal preference”.
This ideal point is located at the terminus or end of the vector. Projections are made by drawing a
line so that it intersects the vector at a 90-degree angle. The farther out (toward vector i’s
terminus) the projection is, the more preferred the brand is for that respondent.
Collecting Data for MDS
The content side of MDS—dimension interpretation, relating physical changes in
products to psychological changes in perceptual maps—poses the most difficult problems for
researchers. However, methodologists are developing MDS models that provide more flexibility
than a straight dimensional application. For example, recent models have coupled the ideas of
cluster analysis and MDS into hybrid models of categorical-dimensional structure. Furthermore,
conjoint analysis, to be discussed next, offers high promise for relating changes in the physical
(or otherwise controlled) aspects of products to changes in their psychological imagery and
evaluation. Typically, conjoint analysis deals with preference (and other dominance-type)
judgments rather than similarities. However, more recent research has extended the methodology
to similarities judgments.
On the input side, there are issues that arise concerning data collection methods. In MDS
studies, there are four most commonly used methods of data collection:
• Sorting task. Subjects are asked to sort the stimuli into a number of groups, according to
similarity. The number of groups is determined by the subject during the judgment task.
• Paired comparison task. Stimuli are presented to subjects in all possible pairs of two stimuli.
Each subject has to rate each pair on an ordinal scale (the number of points can vary) where the
extreme values of the scale represent maximum dissimilarity and maximum similarity.
• Conditional ranking task. Subjects order stimuli on the basis of their similarity with an anchor
stimulus. Each of the stimuli is in turn presented as the anchor.
• Triadic combinations task. Subjects indicate which two stimuli of combinations of three stimuli
form the most similar pair, and which two stimuli form the least similar pair.
When subjects perform a similarity (or dissimilarity) judgment they may experience increases in
fatigue and boredom (Bijmolt and Wedel,1995, p. 364).
Bijmolt and Wedel examined the effect of the alternative data collection methods on
fatigue, boredom and other mental conditions. They showed that when collecting data,
conditional rankings and triadic combinations should be used only if the stimulus set is relatively
small, and in situations where the maximum amount of information is to be extracted from the
respondents. If the stimulus set is relatively large, sorting and paired comparisons are better
suited for collecting similarity data. Which of these two to use will depend on characteristics of
the application, such as number of stimuli and whether or not individual-level perceptual maps
are desired.
Marketing Applications of MDS
381
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
MDS studies have been used in a variety of situations to help marketing managers see
how their brand is positioned in the minds of consumers, vis-à-vis competing brands.
Illustrations include (a) choosing a slogan for advertising a soft drink, (b) the relationship
between physical characteristics of computers and perceptions of users and potential users, (c)
effectiveness of a new advertising campaign for a high-nutrition brand of cereal, (d) positioning
in physicians’ minds of medical magazines and journals, and (e) positioning of new products and
product concepts. There is no shortage of applications in real-world marketing situations.
Current research activity in MDS methods, including the increasing use of
correspondence analyses for representing nominal data (Hoffman & Franke, 1986; Carroll,
Green, & Schaffer, 1986; Whitlark & Smith, 2003), shows few signs of slowing down. In
contrast, industry applications for the methods still seem to be emphasizing the graphical display
and diagnostic roles that characterized the motivation for developing these techniques in the first
place. The gap between theory and practice appears to be widening. A comprehensive overview
of the developments in MDS is provided by Carroll and Arabie (1998).
FUNDAMENTALS OF CONJOINT ANALYSIS
Conjoint analysis is one of the most widely used advanced techniques in marketing
research. It is a powerful tool that allows the researcher to predict choice share for evaluated
stimuli such as competitive brands. When using this technique the researcher is concerned with
the identification of utilities—values used by people making tradeoffs and choosing among
objects having many attributes and/or characteristics.
There are many methodologies for conducting conjoint analysis, including two-factor-at
a- time tradeoff, full profile, Adaptive Conjoint Analysis (ACA), choice-based conjoint, self
explicated conjoint, hybrid conjoint, and Hierarchical Bayes (HB). In this chapter, two of the
most popular methodologies are discussed: the full-profile and self-explicated models.
Conjoint analysis, like MDS, concerns the measurement of psychological judgments,
such as consumer preferences. The stimuli to be presented to the respondent are often designed
beforehand according to some type of factorial structure. In full-profile conjoint analysis, the
objective is to decompose a set of overall responses to a set of stimuli (product or service
attribute descriptions) so that the utility of each attribute describing the stimulus can be inferred
from the respondent’s overall evaluations of the stimuli. As an example, a respondent might be
presented with a set of alternative product descriptions (automobiles). The automobiles are
described by their stimulus attributes (level of gas mileage, size of engine, type of transmission,
etc.). When choice alternatives are presented, choice or preference evaluations are made. From
this information, the researcher is able to determine the respondent’s utility for each stimulus
attribute (i.e., what is the relative value of an automatic versus a five-speed manual
transmission). Once the utilities are determined for all respondents, simulations are run to
determine the relative choice share of a competing set of new or existing products.
Conjoint analysis models are constrained by the amount of data required in the data
collection task. Managers demand models that define products with increasingly more stimulus
attributes and levels within each attribute. Because more detail increases the size, complexity,
and time of the evaluation task, new data collection methodologies and analysis models are
continually being developed.
One early conjoint data collection method presented a series of attribute-by-attribute (two
attributes at a time) tradeoff tables where respondents ranked their preferences of the different
combinations of the attribute levels. For example, if each attribute had three levels, the table
382
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
would have nine cells and the respondents would rank their tradeoff preferences from 1 to 9. The
two-factor-at-a-time approach makes few cognitive demands of the respondent and is simple to
follow . . . but it is both time-consuming and tedious. Moreover, respondents often lose their
place in the table or develop some stylized pattern just to get the job done. Most importantly,
however, the task is unrealistic in that real alternatives do not present themselves for evaluation
two attributes at a time.
For the last 30 years, full-profile conjoint analysis has been a popular approach to
measure attribute utilities. In the full-profile conjoint task, different product descriptions (or even
different actual products) are developed and presented to the respondent for acceptability or
preference evaluations. Each product profile is designed as part of a fractional factorial
experimental design that evenly matches the occurrence of each attribute with all other attributes.
By controlling the attribute pairings, the researcher can estimate the respondent’s utility for each
level of each attribute tested.
A third approach, Adaptive Conjoint Analysis, was developed to handle larger problems
that required more descriptive attributes and levels. ACA uses computer-based interviews to
adapt each respondent’s interview to the evaluations provided by each respondent. Early in the
interview, the respondent is asked to eliminate attributes and levels that would not be considered
in an acceptable product under any conditions. ACA next presents attributes for evaluation and
finally full profiles, two at a time, for evaluation. The choice pairs are presented in an order that
increasingly focuses on determining the utility associated with each attribute.
A fourth methodology, choice-based conjoint, requires the respondent to choose a
preferred full-profile concept from repeated sets of 3–5 concepts. This choice activity simulates
an actual buying situation, thereby giving the respondents a familiar task that mimics actual
shopping behavior.
The self-explicated approach to conjoint analysis offers a simple but robust approach that
does not require the development or testing of full-profile concepts. Rather, the conjoint factors
and levels are presented to respondents for elimination if not acceptable in products under any
condition. The levels of the attributes are then evaluated for desirability. Finally, the relative
importance of attributes is derived by dividing 100 points between the most desirable levels of
each attribute. The respondent’s reported attribute level desirabilities are weighted by the
attribute importances to provide utility values for each attribute level. This is done without the
regression analysis or aggregated solution required in many other conjoint approaches. This
approach has been shown to provide results equal or superior to full-profile approaches, and
requires less rigorous evaluations from respondents.
Most recently, academic researchers have focused on an approach called Hierarchical
Bayes (HB) to estimate attribute level utilities from choice data. HB uses information about the
distribution of utilities from all respondents as part of the procedure to estimate attribute level
utilities for each individual. This approach again allows more attributes and levels to be
estimated with smaller amounts of data collected from each individual respondent.
An Example of Full-Profile Conjoint Analysis
In metric conjoint analysis, the solution algorithm involves a dummy variable regression
analysis in which the respondent’s preference ratings of the product profile (service or other
item) being evaluated serve as the dependent (criterion) variable, and the independent (predictor)
variables are represented by the various factorial levels making up each stimulus. In the
nonmetric version of conjoint analysis, the dependent (criterion) variable represents a ranking of
383
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
the alternative profiles and is only ordinal-scaled. The full-profile methods for collecting
conjoint analysis data will be illustrated to show how conjoint data are obtained.
The multiple-factor approach illustrated in Figure 14.7 consists of sixteen cards, each
made up according to a special type of factorial design. The details of each card are shown on the
left side of Figure 14.6.
Figure 14.7 Multiple-Factor Evaluations (Sample Profiles)
Advertising Appeal Study for Allergy Medication
4
4
Fractional Factorial Design (4 Levels
4 Factors
)
0 , 0 , 0 , 0
1 , 0 , 1 , 2
2 , 0 , 2 , 3
3 , 0 , 3 , 1
0 , 1 , 1 , 1
1 , 1 , 0 , 3
2 , 1 , 3 , 2
3 , 1 , 2 , 0
0 , 2 , 2 , 2
1 , 2 , 3 , 0
2 , 2 , 0 , 1
3 , 2 , 1 , 3
0 , 3 , 3 , 3
1 , 3 , 2 , 1
2 , 3 , 1 , 0
3 , 3 , 0 , 2
Legend:
Profile 1 has levels 0,0,0,0:
No Medication is More Effective
Most Recommended by Allergists
Less Sedating than Benadryl
Won’t Quit on You
Column 1: "Efficacy", 4 Levels
0="No med more effective"
1="No med works faster"
2="Relief all day"
3="Right Formula"
Column 2: "Endorsements",4 Levels
0="Most recommended by allergists"
1="Most recommended by pharmacist"
2="National Gardening Association."
3="Professional Gardeners (Horticulture.)"
Column 3: "Superiority",4 Levels
0="Less sedating than Benadryl"
1="Rec. 2:1 over Benadryl"
2="Relief 2x longer than Benadryl"
3="Leading long acting Over The Counter "
Column 4: "Gardening",4 Levels
0="Won't quit on you"
1="Enjoy relief while garden"
2="Brand used by millions"
3="Relieves allergy symptoms"
All card descriptions differ in one or more attribute level(s).1 The respondent is then asked to
group the 16 cards (Figure 14.8) into three piles (with no need to place an equal number in each
pile) described in one of three ways:
• Definitely like
• Neither definitely like nor dislike
• Definitely dislike
The criterion variable is usually some kind of preference or purchase likelihood rating.
Following this, the respondent takes the first pile and ranks the cards in it from most to least
liked, and similarly so for the second and third piles. By means of this two-step procedure, the
full set of 16 cards is eventually ranked from most liked to least liked.
While it would be easier for the respondent to rate each of the 16 profiles on, a 1-10
rating scale, the inability or unwillingness of the respondent to conscientiously differentiate
between all of the profiles considered typically results in end-piling of ratings where many
profiles incorrectly receive the same score values.
Again, the analytical objective is to find a set of part-worths or utility values for the
separate attribute (factor) levels so that, when these are appropriately added, one can find a total
384
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
utility for each combination or profile. The part-worths are chosen so as to produce the highest
possible correspondence between the derived ranking and the original ranking of the 16 cards.
While the two-factor-at-a-time and the multiple-factor approaches, as just described, assume
only ranking-type data, one could just as readily ask the respondent to state his or her preferences
on (say), an 11-point equal-interval ratings scale, ranging from like most to like least. Moreover,
in the multiple-factor approach, a 0-to-100 rating scale, representing likelihood of purchase, also
could be used.
Figure 14.8 Product Descriptions for Conjoint Analysis (Allergy Medication)
Card 1
Efficacy
No med more effective
Endorsements
Most recom. by allergists
Superiority
Less sedating than Benadryl
Gardening
Won't quit on you
Card 2
Efficacy
No med works faster
Endorsements
Most recom. by allergists
Superiority
Rec. 2:1 over Benadryl
Gardening
Brand used by millions
Card 3
Efficacy
Relief all day
Endorsements
Most recom. by allergists
Superiority
Relief 2x longer than Benadryl
Gardening
Relieves allergy symptoms
Card 4
Efficacy
Right Formula
Endorsements
Most recom. by allergists
Superiority
Leading long acting OTC
Gardening
Enjoy relief while garden
Card 5
Efficacy
No med more effective
Endorsements
Most recom. by pharmacist
Superiority
Rec. 2:1 over Benadryl
Gardening
Enjoy relief while garden
Card 6
Efficacy
No med works faster
Endorsements
Most recom. by pharmacist
Superiority
Less sedating than Benadryl
Gardening
Relieves allergy symptoms
Card 7
Efficacy
Relief all day
Endorsements
Most recom. by pharmacist
Superiority
Leading long acting OTC
Gardening
Brand used by millions
Card 8
Efficacy
Right Formula
Endorsements
Most recom. by pharmacist
Superiority
Relief 2x longer than Ben
Gardening
Won't quit on you
Card 9
Efficacy
No med more effective
Endorsements
Nat. Gardening Assoc.
Superiority
Relief 2x longer than Ben
Gardening
Brand used by millions
Card 10
Efficacy
No med works faster
Endorsements
Nat. Gardening Assoc.
Superiority
Leading long acting OTC
Gardening
Won't quit on you
Card 11
Efficacy
Relief all day
Endorsements
Nat. Gardening Assoc.
Superiority
Less sedating than Benadryl
Gardening
Enjoy relief while garden
Card 12
Efficacy
Right Formula
Endorsements
Nat. Gardening Assoc.
Superiority
Rec. 2:1 over Benadryl
Gardening
Relieves allergy symptoms
Card 13
Efficacy
No med more effective
Endorsements
Prof. Gardeners (Horticul.)
Superiority
Leading long acting OTC
Gardening
Relieves allergy symptoms
Card 14
Efficacy
No med works faster
Endorsements
Prof. Gardeners (Horticul.)
Superiority
Relief 2x longer than Ben
Gardening
Enjoy relief while garden
Card 15
Efficacy
Relief all day
Endorsements
Prof. Gardeners (Horticult.)
Superiority
Rec. 2:1 over Benadryl
Gardening
Won't quit on you
Card 16
Efficacy
Right Formula
Endorsements
Prof. Gardeners (Horticul.)
Superiority
Less sedating than Benadryl
Gardening
Brand used by millions
As may be surmised, the multiple-factor evaluative approach makes greater cognitive
demands on the respondent, since the full set of factors appears each time. In practice, if more
than six or seven factors are involved, this approach is often modified to handle specific subsets
of interlinked factors across two or more evaluation tasks.
Consider the situation in which a manufacturer of over-the-counter allergy medication is
interested in measuring consumers’ tradeoffs among the four attributes identified in Figure 14.6.
Figure 14.9 shows a table of the resulting utility values for each of the attribute levels
derived for one respondent. These values can be obtained from an ordinary multiple regression
program using dummy-variable coding. All one needs to do to estimate the respondent’s utility
score for a given concept profile is to add each separate value (the regression coefficient) for
each component of the described combination. (The regression’s intercept term may be added in
385
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
later if there is interest in estimating the absolute level of purchase interest.) For example, to
obtain the respondent’s estimated evaluation of card 1, one sums the part-worths:
Value for: Efficacy “No med more effective” = 0.00
Value for: Endorsements “Most recom. by allergists” = 3.00
Value for: Superiority “Less sedating than Benadryl” = 0.00
Value for: Gardening “Won’t quit on you” = 6.00
Total = 9.00
Figure 14.9 Part-Worth Functions Obtained From Conjoint Analysis of One Respondent
In this instance we obtain an almost perfect prediction of a person’s overall response to
card one. Similarly, we can find the estimated total evaluations for the other 15 options and
compare them with the respondent’s original evaluations. The regression technique guarantees
that the (squared) prediction error between estimated and actual response will be minimized. The
information in Figure 14.8 also permits the researcher to find estimated evaluations for all
combinations, including the 256 – 16 = 240 options never shown to the respondent. Moreover,
all respondents’ separate part-worth functions (as illustrated for the average of all respondents in
Table 14.6) can be compared in order to see if various types of respondents (e.g., high-income
versus low-income respondents) differ in their separate attribute evaluations. In short, while the
respondent evaluates complete bundles of attributes, the technique solves for a set of partworths—one
for each attribute level—that are imputed from the overall tradeoffs.
Table 14.6 Average Utilities of All Respondents
LEVEL Importance % 1 2 3 4
Efficacy
Importance 14.63
Effective
2.48
Faster
2.37
Relief
2.10
Formula
2.18
Endorsements
Importance 54.24
Allergist
3.83
Pharmacist
3.18
Nat. Garden
2.57
Prof. Garden
2.41
386
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Superiority
Importance 17.76
Less
2.78
Recommend
2.63
Relief
2.71
Leading
2.31
Gardening
Importance 13.37
Quit
2.39
Enjoy
2.74
Millions
2.58
Relief
2.54
These part-worths can then be combined in various ways to estimate the evaluation that a
respondent would give to any combination of interest. It is this high leverage between options
that are actually evaluated and those that can be evaluated (after the analysis) that makes conjoint
analysis a useful tool. It is clear that the full-profile approach requires much sophistication in
developing the profiles and performing the regression analyses to determine the utilities. We will
now consider the self-explicated model as an approach that provides results of equal quality, but
does so with a much easier design and data collection task.
Self-Explicated Conjoint Analysis
The development of fractional factorial designs and required dummy regression for each
respondent places a burden on the researcher and respondent alike, especially when the number
of factors and levels require that a large number of profiles be presented to the respondent.
The self-explicated model provides a simple alternative producing utility score estimates
equal to or superior to that of the ACA or full-profile regression models. The self-explicated
model is based theoretically on the multi-attribute attitude models that combine attribute
importance with attribute desirability to estimate overall preference. This model is expressed as
where Ij is the importance of attribute j and Djk is the desirability of level k of attribute j. In this
model, Eo, the evaluation of profile for product or service o, is formed by summing the
importance weighted desirabilities of the attributes and attribute levels that make up the profile.
The Self-Explicated Data Collection Task
Initially, all attribute levels are presented to respondents for evaluation to eliminate any
levels that would not be acceptable in a product under any conditions. Next, the list of attribute
levels is presented and each level is evaluated for desirability (0 –10 scale). Finally, based on
these evaluations, the most desirable levels of all attributes are presented in a constant sum
question where the relative importances of the attributes are evaluated. Using this information,
the attribute importance scores are used to weight the standardized attribute level scores, thereby
producing self-explicated utility values for each attribute level. This is done for each respondent
and does not require a fractional factorial designs or regression analysis.
As with the full-profile model, these scores can be summed and simulations run to obtain
a score for any profile of interest. This simple self-reporting approach is easier for the respondent
to complete and straightforward in terms of determining the importance or desirability of
attributes and attribute levels (Srinivasan, 1997). An easy to use online implementation of the
self-explicated model is found at www.qualtrics.com. For this implementation, the conjoint
analysis is automatically developed after the attribute level descriptors are entered into the
question builder.
Conjoint Reliability and Validity Checks
387
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Irrespective of the method used to carry out a conjoint analysis, it is useful to include the
following ancillary analyses: (a) test-retest reliability; (b) a comparison of actual utilities with
those of random respondents; and (c) an internal validity check on model-based utilities.
The test-retest reliability can be conducted by including a few replicate judgments (drawn
from the original set of 16) at a later stage in the interview. The purpose here is to see if the
judgments are highly enough correlated, on a test-retest basis, to justify the analysis of the
respondent’s data.
An internal validity check could, in the case of the allergy medication example, be carried
out by collecting a few new evaluations (drawn randomly from the 240 stimulus combinations
not utilized in Figure 14.8). These constitute a hold-out sample. Their rank order is to be
predicted by the part-worths developed from the calibration sample of 16 combinations. Internal
validity checks could be conducted in similar manner for other conjoint methods.
Other Models
So far our discussion has centered on the most widely applied conjoint models—a main
effects model using rankings or ratings. Other models are available that permit some or all twofactor
interactions to be measured (as well as main effects). Interactions occur when attribute
levels combine to provide a differential effect. For example, this often happens in food products
where combinations of attribute levels seemingly produce more acceptable combinations than
would be predicted by the individual attribute levels when considered alone (oatmeal-raisin
cookies vs. oatmeal cookies or raisin cookies). These models again make use of various types of
fractional factorial designs or combine attributes. Specialized computer programs have been
designed to implement them. In short, the users of conjoint analysis currently have a highly
flexible set of models and data collection procedures to choose from.
OTHER ASPECTS OF CONJOINT ANALYSIS
The typical sequence that one goes through to implement a conjoint study involves four steps:
1. Using one of a variety of data collection procedures just described, obtain sufficient data at the
individual respondent level to estimate the part-worths of each person’s utility function.
2. Relate the respondent’s attribute-level part-worth data to other subject background data in an
effort to identify possible market segments based on similarities in part-worth functions.
3. Compose a set of product configurations that represent feasible competitive offerings. These
product profiles are entered into a consumer choice simulator, along with the earlier computed
individual utility functions.
4. Use the respondent’s individual part-worth function to compute the utility for each of the
competing profiles in the choice simulator. The respondent is then assumed to choose that profile
with the highest utility (i.e., the choice process is deterministic).
Use of Visual Aids in Conjoint analysis
Another problem in the application of conjoint measurement is the pragmatic one of
getting fairly complex concepts across to the respondent. Verbal descriptions of the type covered
in Figure 14.8 are not only difficult for the respondent to assimilate, but also introduce unwanted
perceptual differences. As an example, if conjoint analysis was used to test designs for
automobiles; two respondents may have quite different perceptions of the car length and carroominess
if verbalizations were used.
388
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Wherever possible, visual props can help to transmit complex information more easily
and uniformly than verbal description. As an illustration of the value of visual props, mention
can be made of a study involving styling designs for future compact cars. In the course of
preparing the questionnaire, rather complex experimental factors such as overall size and interior
layout, trunk size and fuel-tank capacity, exterior and interior width, and interior spaciousness
and visibility had to be considered. To provide quick and uniform treatment of these style
factors, visual props were prepared, as illustrated for two of the attributes in Figure 14.10. (These
can be projected on screens in full view of the respondents during the interview or made part of
the questionnaire itself.)
Figure 14.10 Illustrations of Visual Props Used on Conjoint Analysis
Visual props work particularly well for the multiple-factor approach, since a relatively
large amount of information can be communicated realistically and quickly.
Strategic Aspects of Conjoint Analysis
The output of conjoint analysis is frequently employed in additional analyses. Since most
conjoint analysis studies collect full sets of data at the individual respondent level, individual
utility functions and importance weights can be computed. This fosters two additional types of
analyses: (1) market segmentation and (2) strategic simulation of new factor-level combinations
(most often used to test the viability of new product design configurations). Frequently, both
kinds of analyses are carried out in the same study.
In segmentation studies, the respondents are usually clustered in terms of either their
commonality of utility functions or their commonality of importance weights. Having formed the
segments in one of these ways, the analyst can then determine how the segments differ with
regard to other background data—product-class usage, brand-selection behavior, demographics,
and so on.
389
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Strategic simulations are also relatively easy to construct from conjoint analysis data by
simply including each individual respondent’s utility function in a computerized-choice model.
Various combinations of factor levels can then be tried out to see what their share of choices
would be under different assumptions regarding competitive offerings and total market demand.
The simulators can employ a variety of consumer-choice procedures, ranging from
having each consumer simply select the alternative with the highest utility to more elaborate
probability of choice rules, where probability is related to utility differences in the set of
alternatives under evaluation.
Applications of Conjoint Analysis
Conjoint analysis is most conducive to predicting choice behavior when product or
service involves a relatively high resource commitment and tends to be “analyzable” by the
purchaser (e.g., banking or insurance services, industrial products). Conjoint analysis has
already been applied to a wide variety of problems in product design, price elasticity of demand,
transportation service design, and the like. Table 14.6 shows a representative list of applications.
As can be noted, areas of application cover the gamut—products and services, as well as
consumer, industrial, and institutional markets.
Table 14.6 Sample List of Conjoint Applications
Consumer Durables Other Products Industrial Goods
1. Automotive styling
2. Automobile and truck tires
3. Car batteries
5. Toaster ovens
6. Cameras
1. Bar soaps
2. Hair shampoos
3. Carpet cleaners
4. Clothing: sweatshirts, bras
5. Gasoline pricing
1. Copying machines
2. Printing equipment
3. Facsimile transmission
4. Data transmission
5. Personal computer design
Financial Services Other Services Transportation
1. Branch bank services
2. Auto insurance policies
3. Health insurance policies
4. Credit card features
5. Consumer discount cards
1. Car rental agencies
2. Telephone services and pricing
3. Medical laboratories
4. Hotel design
5. Toll collection on toll ways
1. Domestic airlines
2. Transcontinental airlines
3. Passenger train operations
4. Freight train operations
5. Electric car design
Recent Developments in Conjoint Analysis
Conjoint analysis has become a highly popular technique in a relatively short time.
Researchers estimate that business firms’ use of conjoint analysis entails several thousand studies
each year. With statistical software and conjoint data collection algorithms built into online
survey tools (Qualtrics.com), conjoint methodology is easily accessed by any interested user.
Software developments in data collection and analysis likewise make it easy to find
orthogonal main effects plans. Conjoint methodology has also been extended to encompass use
occasion and situation dependence in a series of dual-conjoint designs, called componential
segmentation.
Perhaps the most interesting extension of the methodology, however, is the recent
application of conjoint to the design of “optimal” products and product lines. Thus, it is feasible
to extend conjoint beyond the simulation stage (where one finds the best of a limited set of
options) to encompass the identification of the best product (or line) over the full set of
possibilities. These may number in the hundreds of thousands or even the millions. In sum,
390
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
conjoint methodology, like MDS, appears to be moving into the product-design-optimization
arena, a most useful approach from a pragmatic managerial viewpoint.
Still, conjoint analysis, like MDS, has a number of limitations. For example, the approach
assumes that the important attributes of a product or service can all be identified and that
consumers behave rationally as though all tradeoffs are being considered. In some products
where imagery is quite important, consumers may not evaluate a product analytically, or, even if
they do, the tradeoff model may be only a gross approximation to the actual decision rules that
are employed.
In short, MDS and conjoint are still maturing—both as techniques that provide
intellectual stimulation and as practical tools for product positioning, segmentation, and strategic
planning.
SUMMARY
Chapter 14 has focused on four multivariate techniques: factor analysis, cluster analysis,
multidimensional scaling, and conjoint analysis.
The factor-analytic method stressed in this chapter was principal-components analysis.
This procedure has the property of selecting sets of weights to form linear combinations of the
original variables such that the variance of the obtained component scores is (sequentially)
maximal, subject to each linear combination’s being orthogonal to previously obtained ones.
The principal-components model was illustrated on a set of data from a study conducted
by a grocery chain.
Cluster analysis was described in terms of three general questions: (a) selecting a
proximity measure; (b) algorithms for grouping objects; and (c) describing the clusters. In
addition, an application of clustering was briefly described.
MDS methods are designed to portray subjective similarities or preferences as points (or
vectors) in some multidimensional space. Psychological distance is given a physical distance
representation. We discussed metric and nonmetric MDS methods, and ideal-point and vector
preference models. A variety of applications were described to give the reader some idea of the
scope of the methodology.
Conjoint analysis was described along similar lines. We first discussed the primary ways
of collecting tradeoff data and then showed how such data are analyzed via multiple regression
with dummy predictor variables. The importance of fractional factorial designs was discussed, as
well as other practical problems in the implementation of conjoint analysis. We next turned to
some illustrative applications of conjoint analysis, including the design of new products and
services. We then presented a brief description of future developments that could serve to
increase the flexibility of conjoint methodology.
This chapter, together with Chapter 13, covers the major multivariate analysis techniques
and has included brief discussions of lesser-used techniques. We have not discussed such
extensions as canonical correlation of three or more sets of variables or tests for the equality of
sums of squares and cross-products matrices. Other advanced, but related procedures such as
moderated regression, multiple-partial correlation, discriminant analysis with covariate
adjustment, factorial discriminant analysis, to name a few, have been omitted from discussion.
We have discussed the principal assumption structure of each technique, appropriate
problems for applying it, and sufficient numerical applications to give the reader a feel for the
kind of output generated by each program.
Our coverage of this vast and complex a set of methods is limited in depth as well as
breadth. The fact remains, however, that marketing researchers of the future will have to seek
391
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
grounding in multivariate methodology, if current research trends are any indication. This
grounding will probably embrace three facets: (a) theoretical understanding of the techniques; (b)
knowledge of the details of appropriate computer algorithms for implementing the techniques;
and (c) a grasp of the characteristics of substantive problems in marketing that are relevant for
each of the methods.
REFERENCES
Aldenderfer, M. S., & Blashfield, R. K. (1984). Cluster analysis. Beverly Hills, CA: Sage.
Ball, G. H., & Hall, D. J. (1964, July). Background information on clustering techniques.
(Working paper.) Menlo Park, CA: Stanford Research Institute.
Bijmolt, T. H. A., & Wedel, M. (1995). The effects of alternative methods of collecting
similarity data for multidimensional scaling. International Journal of Research in Marketing, 12,
363–371.
Carroll, J. D., & Arabie, P. (1980). Multidimensional scaling. In M. R. Rosenzweig & L. W.
Porter (Eds.), Annual review of psychology. Palo Alto, CA: Annual Reviews.
Carroll, J. D., & Arabie, P. (1998). Multidimensional scaling. In M. H. Birnbaum (Ed.),
Handbook of perception and cognition. Volume 3: Measurement, judgment and decision-making
Carroll, J. D., Green, P. E., & Schaffer, C. M. (1986). Interpoint distance comparisons in
correspondence analysis. Journal of Marketing Research, 23, 271–280.
Clausen, S. E. (1998). Applied correspondence analysis: An introduction. Thousand Oaks, CA:
Sage.
Coombs, C. H. (1964). A theory of data. New York: Wiley.
Green, R. T., & Larsen, T. L. (1985, May). Export markets and economic change. (Working
paper 84/85-5-2). Austin: Department of Marketing Administration, University of Texas at
Austin.
Green, P. E., Carmone, F. J. and Smith , S. M. (1989). Multidimensional Scaling: Concepts and
Applications, Boston: Allyn and Bacon.
Greenacre, M. J. (1993). Correspondence analysis in practice. London: Academic Press.
Hoffman, D. L., & Franke, G. R. (1986, August). Correspondence analysis: Graphical
representation of categorical data in marketing research. Journal of Marketing Research, 23,
213–217.
Johnson, S. C. (1967, September). Hierarchical clustering schemes. Psychometrika, 32, 241–154.
Kim, J.-O., & Mueller, C. W. (1978b). Factor analysis: Statistical methods and practical issues.
Beverly Hills, CA: Sage.
392
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Kim, J.-O., & Mueller, C. W. (1978b). Factor analysis: Statistical methods and practical issues.
Beverly Hills, CA: Sage.
Kruskal, J. B., & Wish, M. (1978). Multidimensional scaling. Beverly Hills, CA: Sage.
PC-MDS Statistical Software, Scott M. Smith, Brigham Young University.
Sneath, P. H. A., & Sokal, R. R. (1973). Numerical taxonomy. San Francisco: W.H. Freeman.
Sokal, R. R., & Sneath, P. H. A. (1963). Principles of numerical taxonomy. San Francisco: W.H.
Freeman.
Srinivasan, V. (1997, May). Surprising robustness of the self-explicated approach to customer
preference structure measurement. Journal of Marketing Research, 34, 286–291.
Whitlark, D., & Smith, S. (2001, Summer). Using correspondence analysis to map relationships:
It’s time to think strategically about positioning and image data. Marketing Research, 2, 23–27.
Whitlark, D., & Smith, S. M. (2003). How many attributes does it take to describe a brand? What
is the right question when collecting associate data? Unpublished Paper, Graduate School of
Management, Brigham Young University, Provo, UT.
Whitlark, D., & Smith, S. (2004). Measuring brand performance using online questionnaires:
Advantages and issues with “pick any” data. (Research Paper.) Provo, UT: Graduate School of
Management, Brigham Young University.
393
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Chapter 15Chapter 20
PREPARING THE RESEARCH REPORT
Chapter 2 identified the research report as the culmination of the research process.
Whatever is to be included in the report—the study’s purpose, methodology, results, conclusions,
or recommendations for management—should be presented clearly, accurately, and honestly.
The key attributes of a report are completeness and conciseness. We can go further and state that,
when managing a research project by objectives, the researcher should focus on the last step—
the research report and the information needs of the user or client (Semon, 1998). The report
should be designed and outlined in detail at the outset. Since the report needs to focus on
information needs, this outline can be an invaluable aid in planning the other stages, such as the
plans for analysis, measurement, and so forth.
This chapter provides information and guidelines that may be invaluable in preparing and
presenting reports, both written and oral. All too often the research report is the only aspect of
the project that others will ever see. Consequently, if the project is not effectively presented,
everything done up to that point will be wasted. A research report must be read (or heard), and its
results, conclusions, and recommendations evaluated, if the resources devoted to the project are
to have been well spent.
COMMUNICATION
The importance of the research report derives from its purpose of communicating to
others the objectives, processes and results of the research. It is important to realize that the
research report and presentation may be the only measure of the quality of the research efforts.
A poor quality presentation of excellent research leaves the impression of poor quality research.
Effective communication has certain characteristics (Rockey, 1984; Vardaman, 1981).
First, the communication is has an objective and a purpose. Second, it is an interchange
of symbols—verbal and graphic representations of a sender’s ideas—between people. Third, it
seeks workable understanding and agreement between the parties. This means that the sender of
a report (the researcher) and the receiver (the client and stakeholders) have enough common
understanding about, and willingness to accept or at least evaluate, the ideas presented in the
report to use them in whatever manner is best. Finally, there is implicit recognition of the need
for feedback. In order to realize satisfaction, there needs to be an opportunity to respond to a
draft report. In short, communication is like playing catch with a baseball: it’s not how many
times the ball is thrown, but how many times it is caught. To be effective, communication must
ultimately be two-way.
One view of the communication process involved in presenting a research report to
management or an outside client is illustrated in Figure 15.1, where five basic steps of the
communication process are identified to include context, formality, organization, message, and
feedback.
Following this process will help you to communicate effectively and avoid a minefield of
obstacles like multiple objectives, political agendas, and special interests. Further, reports are
presented to groups with unique dynamics and individuals with personalities that guide, direct,
and even dominate the project, process, or presentation. Each of these factors must be considered
when determining how to best communicate the results of your research. Without considering
these factors, your research report’s effectiveness will be limited.
394
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 15.1 Preparing to Communicate the Research Process
SOURCE: From Bienvenu, S., & Timm, P. R., Business Communications: Discovering Strategy, Developing Skills, 1st edition, copyright
© 2002. Reprinted by permission of Pearson Education, Inc., Upper Saddle River, NJ.
THE RESEARCH REPORT
Written and oral reports can come in all sizes, shapes, and lengths. The nature of the
problem studied, how it was studied, and the client’s identity will influence the type of report to
be prepared. The length of a report on qualitative research, for example, which can run from as
short as eight pages to more than 100 pages, will be influenced by learning garnered, client
requirements, writing style, and formatting (Greenberg, 1999). Also, it is important to remember
that for many projects two reports are prepared. There is always need for a report to
management. This so-called popular report minimizes technical details and emphasizes
simplicity. Frequently, the simpler reports written in the marketer’s language are most
understood, read, and, thus, acted upon. However, there also may be need for a more detailed
technical report. Such a report emphasizes the methods used and underlying assumptions, and
presents findings in a detailed manner. This type of report can be a valuable resource for
researchers, as well as managers, for future studies.
There are two key dimensions to the report-writing process: (a) analysis of the data; and
(b) the actual writing. Analysis is the interpretation of the data obtained to provide useful and
meaningful insights into the problem at hand. Writing takes the learning and expresses it in a
coherent, logical, and succinct way (Greenberg, 1999). For both analysis and writing there are
many ways in which data can be presented visually. According to Jacoby (1997, pp. 2–4),
graphical statistical methods can be used to:
1. Define the Context
A. Define the research problem
1. Limit the problem
2. Evaluate the problem within the external
environment
3. Evaluate the problem within the corporate
culture
B. Define the audience
1. Decision makers
2. Presentation audience
C. Define the research objectives
2. Consider Formality
of the Report
A. Media options
B. Detail required
C. Timing (length, scheduling)
3. Select and Organize Your Information
A. Audience, objectives, political climate
B. Select most effective points
C. Limit major points
D. Enhance message with support materials
(PowerPoint, visual aids, numbers, examples)
5. Feedback for Continued Success
A. Give, solicit, receive feedback
B. Evaluate credibility of feedback
1. Goodwill, focus, concurrence
2. Expertise: education, knowledge,
experience
3. Power, status, success
4. Confidence: presence at presentation
4. Develop Your Message
A. Work with your key contact on draft
versions
B. Develop the presentation
C. Prepare through rehearsal and editing
D. Present with confidence and knowledge
E. Be yourself, adapted for your audience
and situation
395
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• Explore the contents of a data set
• Find structure in data
• Check assumptions in statistical models
• Communicate the results of an analysis
The requirements to both effectively analyze data and write a report may be beyond the abilities
of a one person. Thus, someone other than the analyst often does the report writing.
Criteria for a Good Report
Before turning to what should be included in the report and what the format for
presentation should be, we turn to a brief discussion of the criteria for a good report. The most
basic criterion is how well the report communicates with the reader. Was the communication
effective, in the way defined previously? Practitioners believe that reports written with clarity,
brevity, and concreteness get action and commitment. The major concerns are that a research
report be complete, accurate, concise, and clear in what is being said. All these are a reflection
of writing style. These topics are discussed in more detail in Rehart (1994), Fink (2003), and
Bailey (1990).
Completeness
A report is complete when it gives all the needed information in a language understood
by the intended audience. The writing must be done on the level of the reader’s understanding;
this is fundamental! A report may be incomplete because it is too long or too short. Not all
information obtained is significant or relevant enough to be reported. In the end, completeness is
always defined by the person(s) who will be reading the report and asked to act upon it.
Accuracy
Obviously, accuracy is related to completeness. But it goes further. A report may not be
completely accurate because the data (or the information in general) upon which it is based are
not accurate. There may have been flaws in research design, measurement and scaling, sampling,
and analysis that led to inaccurate results being presented. In addition, carelessness in handling
data, interpretation of analyses, or writing style may also lead to inaccuracy. The preparation of a
research report calls for attention to detail, including attention to the meaning of every word
used, punctuation, and so forth.
Conciseness
Being concise means being selective. Thus, as we said above, not all information
obtained needs to be reported. When writing a report, one should write not only so the reader can
understand, but can also do so as quickly and easily as possible. Conciseness refers to what is
included and how it is included. Sentences should be kept as short as possible. A concise report
is not the same as a brief report. A brief report usually contains only the highlights or base
essentials of what the researcher has to report. In contrast, a concise report may contain any
amount of detail and be very long. A concise report may be very complete. But it is efficient; that
is, it conveys all the researcher wants to present in the shortest, most direct way.
Clarity
Clarity may be the most misunderstood criterion for evaluating a research report. Clarity
derives from clear and logical thinking and organization. Clarity involves connections between
396
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
words, sentences, paragraphs, topics, or ideas. There is always some logical connection between,
say, a topic or idea and the one that preceded it. For example, there are different structural ways
to organize material in a logical framework, as illustrated in the list below:
1. Time order. Ideas are presented in chronological time, either forward or backward.
2. Space order. The relationship among places and locations determines organization.
3. Cause and effect. Cause may precede effect or the opposite may hold. A report may stress one or
the other.
4. Increasing difficulty. Structure involves going from the simple to the complex or from the
familiar to the unfamiliar, useful when the audience lacks expertise in the topic.
5. Established category. When a basic framework is understood, content can be organized in
recognized categories. In a market segmentation report, using accepted categories for any of the
demographic bases is useful.
6. Comparison or contrast. Readers grasp differences or similarities more easily when this
structural organization is used.
7. Pro and con order. Presenting arguments for and against something, usually without favoring
one or the other is another common structure.
A most important rule is that the report be well organized. At the same time, it must be
written clearly. This means that researchers who prepare reports should develop a writing style
conducive to clarity in presentation. Many different sets of rules for clear writing are available.
Most say the same things, but perhaps in different ways. One such set is shown in Exhibit 15.1.
EXHIBIT 15.1 Some Rules for Developing a Writing Style
1. Use concrete words. Such words are clear and specific. Be willing to specify, itemize, give
details or examples, define, and illustrate.
2. Keep sentences short. Shorter sentences usually are more readable, all other things being
equal. Lengthy sentences are more likely to cause confusion or complexity.
3. Vary sentence types and structures. Use declarative (assertion), interrogative (question),
imperative (command), and exclamatory sentences. Do not limit a report, but use all types
where appropriate. Also, vary structure among simple, compound, complex, and compoundcomplex
sentences. Varying sentence types and structure increases interest and reduces
monotony.
4. Maintain unity. Try to build each paragraph around one idea or topic.
5. Use active verbs. Active sentences tend to be more forceful and more direct. This does not
mean that the passive voice should never be used. Use it sparingly, however.
6. Avoid wordiness. Say what is to be said in as few words as possible.
7. Use varying means of emphasis. There are different ways to emphasize certain points made
in the report: space, repetition, position, and mechanical means (e.g., arrows, color, or
underscoring).
8. Write and speak naturally. Often it is difficult to identify what is natural. A helpful guide is to
use conversational language, but this is only a guide.
9. Write on the level of reader’s understanding. Do not overestimate the reader’s
understanding and confuse him or her with highly specialized technical jargon. Also, do not
underestimate the reader by using overly simplistic and childish terms.
10. Watch the pace. Avoid trying to say too much in too few words. In the same light, it is bad to
stretch out an idea by using too many words.
397
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
11. Keep the tone appropriate. The way in which words are put together says a great deal about
the writer. Thus, tone of writing implies something about the personality of the writer and of
the writer’s organization or company. While it is rare for a person to have complete control of
the tone of his or her writing, it is important to acknowledge that it has many shades: positive or
negative, helpful or indifferent, courteous or impertinent, humble or arrogant, and so forth.
SOURCE: Rockey, 1984, pp. 101–106.
Practitioners’ Views
Many of the ideas presented in the last few pages are widely shared by research practitioners.
Four perspectives of report preparation are presented in Exhibit 15.2.
EXHIBIT 15.2 What the Research Practitioner Has to Say
Different researchers have different philosophies regarding what is important in report
preparation. All agree that poorly communicated research is useless. All too often, the written report
ends up as a second-class citizen to the demonstration of a new or repositioned statistical technique.
Perspective #1: Strengthen Your Communications
In order to strengthen written communication, Teresa Wrobel (1990), Vice-President of
Research 100 in Princeton, New Jersey, suggests six things:
1. Do the readers’ work for them. Presume that a reader feels uncomfortable manipulating or
interpreting numbers. The text relates the major issues, while the numerical data serve as supporting
information. Numbers should not interrupt the flow of the text, and the reader should not be forced
to perform mental calculations in mid-sentence. Calculations should already be completed and
presented in an easily to understand form.
2. You don’t have to analyze everything. Not every attitude or behavior that has been quantified is
worthy of discussion. Even statistically significant results do not always merit discussion. Select only
those findings that illustrate the main issues.
3. Write in plain English (or whatever language is being used). Avoid industry jargon. Jargon tends to
conceal the main ideas being conveyed and may be incomprehensible to the intended readers, who
are often not researchers. The use of jargon can alienate or confuse the reader.
4. Convey the main ideas quickly. Most decision-makers don’t have time to plow through pages of text
to unearth the main findings. The front section of a research document should reduce the findings to
their skeletal elements, and include the conclusions and recommendations supported by key findings.
5. Did you answer the question at hand? Conclusions should address the initial objectives with clarity
and precision.
6. Keep the report interesting. Take advantage of the richness and breadth of language available. Keep
searching for new ways of expressing points that convey ideas most effectively. While it is easiest to
rely on language that is familiar and tested, expanding one’s use of language can encourage idea
generation and provide a more enjoyable document for the reader.
7. Some format limitations must be followed. In addition, using esoteric language for its own sake may
lose the reader’s attention. However, such limitations should not stifle creativity. Diversity in
communication styles should be accepted and encouraged.
Perspective #2 Make It Readable
398
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
A slightly different view is that of Howard Gordon (1998), a principal of George R. Frerichs &
Associates of Chicago. Mr. Gordon feels the following seven points can contribute to preparing
more readable, usable, and action-oriented reports:
1. Present tense works better. Use present tense. Results and observations expressed in “now“ terms
sound better and are easier to read. Don’t say: “The test panel liked the taste of the juice.” Say:
“People like the taste of the juice.” We do studies for clients to help them make decisions today and
tomorrow—not yesterday. Clients want to know what people think now.
2. Use active voice. Use active voice where possible, which is in most cases. There is nothing wrong with
saying: “We believe . . .” rather than: “It is believed that . . . .” Passive voice is stilted. Use first person
plural, not third person singular. Present tense and active voice make reports sound action oriented
and businesslike. And active is easy to read. Passive voice makes the reader work harder.
3. Findings—don’t use the word findings. This word makes your report sound as if some archaeologist
just came across some old bones from the Paleolithic age. Marketers should use results, conclusions,
or observations—not findings. Conclusions is the take-away section. What is it you want the reader to
take away from your report? If somebody could spend only five minutes with your report, what
should that person know?
4. Use informative headlines, not label headlines. Your reader will welcome this headline:
“Convenience is the packaging’s major benefit.” Your reader will not generally welcome this headline:
“Analysis of packaging.” Nor this: “How people feel about packaging.”
5. Let your tables and charts work. Help them with words when needed. Don’t take a paragraph or a
page to describe what a table already says. If there is nothing to say about a table or a chart, don’t
say it. The purpose of a table or a chart is to simplify. Use your words to point out significant items in
the table—something that is not readily clear. Or use your words to offer interpretive comments. Use
verbatims from sample respondents to support a point.
6. Use the double-sided presentation whenever possible. This format will reduce the verbiage in your
report. It simply presents the table on the left side of the open report. Your informative headline and
interpretive comments are on the right-hand page. The double-sided presentation is one of the most
readable report formats. And it is much easier to write a report that uses it. The double-sided system
also allows you to use white space effectively. White space helps readers get through your report—
especially those readers in executive suites.
7. Make liberal use of verbatims. Great nuggets of marketing wisdom have come from people’s
comments. Use verbatims if you have them. Sprinkle the verbatims throughout your report in
appropriate places as they make research reports interesting and readable
Perspective #3 Make Your Results Talk
Yet another approach views things differently. James Nelems (1997), President of The Marketing
Workshop in Norcross, Georgia, cautioned us more than 20 years ago on what should be
avoided. He believed that research reports have more faults beyond improper analysis!
1. The longer the report, the better. Research isn’t bought or sold by the pound. Any research report—
regardless of sample size, multivariate techniques, or number of questions— should be reducible to a
one-page executive overview. The report itself also should be clearly written and easy to read,
including marketing conclusions and implications—the indications for future action—drawn from the
data. Management has both the time and desire to read such reports—and the confidence to accept.
2. Indiscriminate use of fancy tools. Some, particularly the multivariate ones, may confuse rather than
enlighten.
3. Letting questions do all the work. Research is more than asking questions. How simple it is in a
package test to ask if people like the package. Or if someone doesn’t use a product, just ask him why
he doesn’t use it. Even though these are the wrong questions, they are often asked, and, of course,
when questions are asked, answers are received. But wrong answers to wrong questions hardly
constitute analysis.
399
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
4. Reporting what happened, rather than what it means. Too much research text simply repeats
numbers in the accompanying table. What is the significance of the numbers? How are we better off
by knowing the answers?
5. The fallacy of “single-number research.” This refers to any kind of research, the outcome of which is
a “single number” that purports to provide an easy decision, such as an on-air recall score, a positive
buying intention, or the overall preference in a product test.
6. Failure to relate findings to objectives or reality. Low brand awareness should be improved, of
course, but what is unaided brand awareness per $1,000 of advertising spent vs. competition? In a
product test, preferences can vary between current users and nonusers. Yet the brand objectives—
gain new users or hold current ones—can produce opposite results with the exact same preference
data.
7. Spurious accuracy. This includes summary statistics to one decimal place. More than likely, they’re
only accurate within 10 percentage points. Yet we’ve actually seen in print a comment on “20% of the
sample of five respondents . . . .”
Perspective #4 Write With Style
A final example of the thinking of practitioners is provided by Lynn Greenberg (1999, pp. 39–41),
a principal of Lynn Greenberg Associates, a Scarsdale, New York, research consulting company.
She suggests the following 11 tips for qualitative research reports, particularly focus groups:
1. Capture your initial thoughts in a research journal as you work. Some of your best insights may come
at this time.
2. Have a flexible plan, one that initially has been developed before starting the writing.
3. Break the task into small pieces and start with the easiest ones. Once the basic sections have been
written, it is easier to deal with the more challenging areas.
4. Be succinct. The reader should be able to clearly and quickly come away with two or three key points
after reading the report.
5. Provide overviews to promote understanding of the learning. The first statement of a section should
be an interpretation of what was learned, followed by more details.
6. Turns negatives into constructive learning and be clear about their meaning.
7. Integrate information from other sources where appropriate to provide more relevant meaning.
8. Feature action-oriented conclusions and implications that address all of the study’s objectives, plus
any important ancillary issues.
9. Use a visually appealing format that allows the readers to skim relevant issues. Use italics, boldface,
underlining, bullet points, and so forth, wherever possible.
10. Edit, edit, and re-edit with independent proofing for clarity, grammar, and typos.
11. Do it your way! Write with your natural style.
REPORT FORMAT AND ORGANIZATION
Every report prepared for a research project is unique and custom-made. Although
reports can be customized, there also is a degree of standardization to report format and
organization.
Many techniques people use to organize research reports are based on placement of
information to gain emphasis. A piece of information at the beginning—of a report itself, of a
section in a report, or even in the first paragraph of a section—tends to get a disproportionate
share of the reader’s attention. Thus, something to be emphasized should be presented at the
beginning of the message. In a similar manner, the end of a message also can be used as a point
of strong emphasis, while the middle of a message is a point of minimum emphasis. The
executive (or management) summary is placed at the front of the report and the
conclusions/recommendations at the end of the report.
400
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Parts of a Formal Report
A formal report consists of a number of components that can be organized into three
major parts. Exhibit 15.3 shows all the components that would be included in a complete formal
report. Research reports, however, vary in formality; not all components will always be included.
Figure 15.2 illustrates how the components are affected by a decreased need for formality.
Typically, marketing research project reports would follow formats II or III. There are many
instances where another version of format IV may be used, in which an executive summary will
be substituted for the table of contents.
Figure 15.2 Components of a Report and Formality
I II III IV V VI
Title Fly Page
Title Page
Letter of Transmittal
Letter of Authorization
Table of Contents
Summary
Report Body
Appendix
Title Page
Letter of
Transmittal
Table of Contents
Summary
Report Body
Appendix
Title Page
Table of Contents
Summary
Report Body
Appendix
Title Page
Executive Summary
Report Body
Appendix
Title Page
Report
Body
Report Body
DIMINISHING NEED FOR FORMALITY
SOURCE: Robinson, 1969, p. 295.
Prefatory Pages
Every research report must have a title page, and may have a title fly page preceding it
(only the title appears on a fly page). Any report longer than a couple of pages should have a
table of contents. The table of contents can be as thin as having only the major sections of the
report listed, or as detailed as including all headings and subheadings. Lists of tables, charts,
figures, and so forth, will be included.
Some research reports may include letters of transmittal and authorization. A letter of
transmittal is the means by which a report is released to a particular person or group. Often, such
a letter contains material usually found in a preface or a foreword. If there is no formal summary
included, then a synopsis of the findings should be included in the letter of transmittal. Other
times, the letter of transmittal is relatively short. A letter of authorization is a letter to the
researcher approving the project and specifying some of the details. Except in the most formal of
marketing research projects, a letter of authorization need not be included in the report. An
illustration of letters of authorization and transmittal used for a project awarded to an outside
firm is given in Figure 15.3.
From a manager’s (or client’s) perspective, the most important component of the research
report is the executive (management) summary. This summary reduces the essence of the
study—the why, what, how, conclusions, and recommendations—to one or two pages. Often,
this is the only part of a project report that the manager reads. Exhibit 15.4 shows examples of
management summaries.
401
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
EXHIBIT 15.3 Components of a Research Report
A formal report consists of the prefatory section, the report body, and the appended materials. A
number of components can be organized and presented within these sections. This rather complete
outline provides a general framework for organizing your research report.
PREFATORY PAGES
I. Title Page
(a) Title
(b) Author’s name
(c) Documentation numbers and
project identification numbers
(d) Classification
(e) Circulation test
(f) Issue date and destroy date
II. Table of Contents
(a) Section subtitles and pages
(b) Illustration titles and pages
(c) Graph titles and pages
(d) Figure titles and pages
III. Letters of Transmittal and
Authorization
IV. Summary of Report
(a) Problem definition and date
(b) What was researched
(c) When
(d) Where
(e) How and with what techniques
(f) Major findings
(g) Recommendations
REPORT BODY
V. Introduction
(a) What prompted the
undertaking of the project
(b) Who prompted the project
(c) How the problem was defined
VI. Statement of Objectives
(a) Problem resolution
(b) The research objective
VII. Research Methods
(a) Research design
(b) Data instruments
(questionnaire, records, etc.)
(c) Data collection methods
(d) Sampling technique
(e) Field work
VIII. Methodological Limitations
(a) Weaknesses in research design
(b) External events that may have
influenced findings
(c) Errors in research methods
(d) Alternative causes for findings
IX. Analysis of Findings
(a) Items of significance
(b) Items of insignificance
(c) Interpretations of findings
X. Conclusions and Recommendations
(a) What research findings show
(b) What actions should be taken
(or not taken)
APPENDED PARTS
XI. Technical Appendix
(a) Informational tables,
graphs, illustrations
(b) Technical discussion of
research methodology
and sample
(c) Sample validation (if
relevant)
XII. General Appendix
(a) Selected portions of
preliminary interviews
(b) Project diary: dates,
places, names, events
(c) Copies of forms,
questionnaires,
records, and data
instruments
XIII. Acknowledgments
(a) Names, titles, and
affiliations of
contributors
(b) Contribution of each
contributor
XIV. References, Bibliography
(a) Names, titles, and
source-related
research
402
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 15.3 Illustrations of Letters of Authorization and Transmittal
A. Letter of Authorization
Lane County / Private Industry Council
May 13, 2010
JVJ Research Associates
554 Pinto Way
Eugene, OR 97401
Dear Sirs:
Thank you for your response to our recent solicitation for proposals to conduct public opinion research for the
Lane County Private Industry Council and Department of Employment & Training. I am pleased to inform you that
your firm has been chosen as the contractor for this study, pending contract negotiation and discussion of certain
aspects of your proposal.
Per recent discussions with Dan Cudaback of my staff, we will be meeting with XXX on Monday, May 18, at 4:00
p.m. in our offices to discuss your proposal further. Please call Mr. Cudaback with any questions between now and
then.
I am looking forward to working with you.
Sincerely,
Steven J. Ickes, Director
EMPLOYMENT & TRAINING DEPARTMENT
B. Letter of Transmittal
JVJ Research Associates
554 Pinto Way
Eugene, Oregon 97401
August 21, 2010
Mr. Steven J. Ickes, Director
Lane County Department of Employment and Training
Eugene, Oregon 97401
Dear Mr. Ickes:
This letter accompanies our final report for the Lane County Public Opinion Study.
The findings contained in this report provide an insightful view of the opinions and desires of the residents of Lane
County. We trust that this report will provide meaningful direction to the community as a whole and policy makers
within Lane County.
Very truly yours,
JVJ Research Associates
John Researcher
Senior Research Associate
Body of the Report
This is the main component of the marketing research project report from the point of
view of size or bulk. Included in the body is the introduction, which deals with why the project
was undertaken and just what the problem was. Also included are research methods used, their
strengths and limitations, and all definitions, results, and analyses. The main body of the report
usually ends with the conclusions and any recommendations to be made.
403
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Appended Materials
Several types of materials may be appended to the marketing research report. These may
include detailed tables that provide more depth than the data presented in the body of the report,
and all forms and measurement instruments used in data collection. If appropriate to the project,
a bibliography and list of relevant references should be added to the end of the report.
EXHIBIT 15.4 Project Summaries Come in All Sizes and Shapes
The executive summary is an abstract that should be the last section of the report prepared.
Typically, it will be a one- or two-page exposition. Example A is a one-page report summary of a study
done for a pipe manufacturer that examined market potential for epoxy-coated pipe. Example B is
drawn from a study of market needs and potential for a local motel. Example C is a two-page summary
of a study done for a supplier of janitorial supplies.
A. Epoxy-Coated Pipe Study
The focus of this report is the market potential for epoxy-coated pipe. A survey of three
separate samples was conducted to explore the market penetration potential of the product. These
samples were: irrigators, municipalities, and industries. On the basis of data collected from the irrigator
sample, it was concluded that the demand for pipe in 2010 would not be greater than that purchased in
2009. Respondents would not be willing to pay more for ECP than for PVC. There is some indication that
a higher price would be paid for ECP relative to CTP. The irrigation market appears to be tightly
controlled by PVC, decreasing the opportunity for market penetration by ECP.
On the basis of data collected from the municipality sample, it was concluded that the demand
for pipe in 2010 would decrease from levels demanded in 2009. This expectation is moderated by the
fact that demand in this market fluctuates yearly. The respondents would not be willing to pay more for
ECP than for other types of pipe. The municipality market also appears to be tightly controlled by PVC,
decreasing the opportunity for market penetration by ECP.
On the basis of data collected from the industry sample, it was concluded that demand for pipe
in 2010 would increase relative to demand in 2009. Respondents would be willing to pay more for ECP
than for other types of pipe. The industrial market is not tightly controlled by PVC. While other types of
pipe dominate, however, ECP does not appear to be perceived as a viable substitute for types of pipe
already in use.
Research was conducted on the future price levels of steel and oil to enable a comparison
between plastic pipe (PVC) and steel pipe (ECP). Steel prices are extremely difficult to forecast. In
general, they are expected to increase. Forecasts for oil prices (in order to predict prices) are also
expected to show increase over time.
In general, it is recommended that ECP not be introduced into any of the markets at this time.
B. Motel
The focus of this report is the potential for increase in the occupancy at the Duck Inn Motel. To
determine this, three different potential target markets were defined and examined. These markets
include bus tourism, university departments, and general tourism. Three separate methodological tools
were used and were assigned to the target markets as follows:
TARGET MARKET METHODOLOGICAL TOOL
BUS TOURISM . . . . . . . . . . . . . . . . . . . . FOCUS GROUP/MAIL SURVEY
UNIVERSITY DEPARTMENTS . . . . . . . . PERSONAL INTERVIEWS
GENERAL TOURISM. . . . . . . . . . . . . . . . MAIL SURVEY
On the basis of data collected for the bus tourism market group, the most important thing to
this segment is the attitude of the motel staff towards those traveling in a large group.
The results gathered for the university departments group indicate that the Duck Inn has many
qualities attractive to campus departments, but the motel needs to increase its exposure to increase
potential customer awareness.
And finally, for the general tourism target market, the target market is generally happy with the
Duck Inn on most of the important choice criteria, but improvements can be made in advertising,
404
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
appearance, and cohesion with the on-site restaurant to increase occupancy rates further.
C. Janitorial Supplies
Objectives
This study attempts to achieve several objectives.
There are two primary concerns the research design intends to determine:
1. The need for a distribution center in Eugene
2. The need for a product showroom, if a distribution outlet is established
Four support questions are also included:
1. Current market share of leading competitors
2. Degree of satisfaction with current suppliers
3. Degree of loyalty to current suppliers
4. Customer base and trends, including demand for Saturday access to supplies and the
potential for telemarketing
Findings
The study shows that there is a substantial lack of confidence in the current suppliers operating
in the Eugene/Springfield metropolitan area. Common complaints include poor credibility, insufficient
product knowledge, and high-pressure sales tactics. Paulsen & Roles Laboratories (P&R) exhibited none
of these traits. Quite to the contrary, genuine compliments for P&R’s business tactics flowed freely from
respondents associated with P&R.
Conclusions
The Eugene/Springfield metropolitan area is the second-largest statistical metropolitan area in
Oregon. For this reason alone, it is critically important that Paulsen & Roles actively participate in this
market. The foundation is in place, as exhibited by both current market share and current consumer
awareness of Paulsen & Roles in the Eugene/Springfield area.
In order to further penetrate this market, it is necessary to increase local representation. P&R’s
current lack of sufficient local representation was the only significant complaint lodged against Paulsen
& Roles during the entire interviewing process.
Recommendations
Survey respondents offered recommendations to Paulsen & Roles Laboratories:
• Aggressively pursue service as your primary competitive advantage. Develop a “full service”
image.
• Continue with competitive pricing. Fully pursue the advantages of P&R’s current level of vertical
integration (i.e., both the manufacturer and distributor).
• Establish a distribution center in Eugene with a showroom, thus enabling a more complete
utilization of P&R’s competitive advantage of complete service to the consumer.
• Maintain Saturday business hours.
• Develop a seminar program for end-users in the Eugene/Springfield metropolitan area.
• Incorporate telemarketing techniques enabling (a) a reduction in the cost per sales call, and
(b) expansion of coverage per sales representative.
• Establish a direct-mail campaign aimed primarily toward the janitorial contractor segment. The
intent will be to minimize the cost associated with informing this market segment of Paulsen &
Roles’s new services in the Eugene area.
• Attempt to acquire the Johnson Wax franchise from Scot Supply.
405
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
USE OF GRAPHICS
Graphical tools are used to show the structure of data. They are relevant for both data
analysis and the communication of data of results. When used properly, graphic material can
enhance a research report by emphasizing the important points and more clearly showing
complex relationships. However, when graphical material is used improperly or is of poor
quality, it can be distracting, misleading, incomplete, confusing and erroneous (Wainer, 1984).
Many types of graphic aids can be used in a marketing research report, including tables, charts,
figures, maps, diagrams (e.g., a flow diagram), and other devices. Our brief discussion will
concentrate mainly on tables and charts. Statistical packages and online survey research tools
such as Qualtrics.com can prepare report-ready tables and charts directly from the data file. More
complete discussion is given by Jacoby (1997, 1998) and Witzling and Greenstreet (1989).
Tables
There are many different ways to show numerical information in tabular form. Creating a
good table of numerical data is an art as well as a science, and it goes far beyond simply putting
data in columns, rows, or both. Using tables allows the researcher to minimize the amount of
narrative material in the report. Only relatively short, perhaps even only summary, tables should
be included in the report’s body. More detailed and comprehensive tables should be placed in an
appendix.
Each table should include a number of items:
1. Table number
2. Title. This should indicate by itself the contents of the table.
3. Bannerhead and stubhead. These are the headings for the columns and rows.
4. Footnotes. Explanations for particular items are placed here.
5. Source note. This acknowledges the source(s) of material used.
Figure 15.4 is taken from a study done for a chamber of commerce to examine
involvement with business firms and is useful to illustrate the various parts of a table.
Not only are there different ways to show numerical data in a table, there also are
different ways to present these data. Table 15.1 shows a typical way to present survey response
data. These data were generated in a study for a manufacturer of business forms. Multiple
responses from a respondent were possible. It is not always necessary to show the specific
question asked in a table. If the title cannot be short and descriptive as well, then it may be
necessary to show the question asked.
Table 15.1 Presenting Survey Response Data
“What role do you play in the purchase of business forms?”
Role played
Market segments Specify Recommend Purchase None
Total sample
size
Hospital 25% 44% 59% 17% 259
Finance 29% 32% 62% 11% 164
Manufacturing 18% 27% 81% 3% 205
Retail 10% 0% 52% 37% 52
Government 39% 48% 4% 8% 213
Transportation 28% 28% 74% 6% 109
406
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 15.4 Table Example
When using statistical analysis software, a researcher may receive report-ready tables.
This is quite common for the many specialized cross-tabulation packages, and even in online
packages like Qualtrics where, for example, one dependent variable could allow for simultaneous
cross-tabulations and drill downs of the data for multiple demographic variables. Such a table is
known as a stub-and-banner table.
Exhibit 15.5 Principles of Graph Construction
Clear Vision
• Make the data stand out. Avoid superfluity.
• Use visually prominent graphical elements to show the data.
• Use a pair of scale lines for each variable. Make the data region the interior of the rectangle formed by
the scale lines. Put tick marks outside the data region.
• Do not overdo the number of tick marks.
• Use a reference line when an important value must be seen across the entire graph, but do not let the line
interfere with the data.
• Do not allow data labels in the data region to interfere with the quantitative data or to clutter the graph.
• Avoid putting notes, keys, and markers in the data region. Put keys and markers just outside the data
region and put notes in the legend or in the text.
• Overlapping plotting symbols must be visually distinguishable.
• Superposed data sets must be readily visually discriminated.
• Visual clarity must be preserved under reduction and reproduction.
Clear Understanding
• Put major conclusions into graphical form. Make legends comprehensive and informative.
• Error bars should be clearly explained.
• When logarithms of a variable are graphed, the scale label should correspond to the tick mark labels.
• Proofread graphs.
• Strive for clarity.
Scales
• Choose the range of the tick marks to include or nearly include the range of data.
• Subject to the constraints that scales have, choose the scales so that the data fill up as much of the data
region as possible.
407
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
• It is sometimes helpful to use the pair of scale lines for a variable to show two different scales.
• Choose appropriate scales when graphs are compared.
• Do not insist that zero always be included on a scale showing magnitude.
• Use a logarithmic scale when it is important to understand the percent change or multiplicative factors.
• Showing data on a logarithmic scale can improve resolution.
• Use a scale break only when necessary. If a break cannot be avoided, use a full scale break. Do not
connect numerical value on two sides of a break.
General Strategy
• A large amount of quantitative information can be packed into a small region.
• Graphing data should be an iterative, experimental process.
• Graph data two or more times when it is needed.
• Many useful graphs require careful, detailed study.
SOURCE: Cleveland, 1985, pp. 100–101. Reprinted with kind permission of Kleuwér Publications.
Charts and Graphs
Charts and graphs are a popular way to present numerical data. In one way or another, the
information contained in a table can be presented with the help of a chart. The overall advantages
are that relationships and structure can be more easily seen and the narrative can be minimized.
Indeed, charts and graphs are often constructed with the premise “a picture is worth a thousand
words,” and as such make results more easily understandable (Gutsche, 2001). There are many
different kinds of charts. These include line charts, bar charts, pie charts, and dot charts. At
times, some combinations may be used in the form of dashboards that provide a real-time
overview of a market as the data is continually updated.
Line Chart
Also known as a line graph, the line chart is useful to show the relationship between
variables. A simple line chart relates two variables, a multiple line chart shows the relationship
between the independent variable and more than one dependent variable, and a stratum line chart
(or stacked line chart) consists of a set of line charts whose quantities are grouped together.
Figures 15.5a–d illustrates horizontal and vertical line charts using categorical and ratio scales,
as well as in combination with vertical and horizontal bar charts. It is not necessary that the
independent variable be continuous or ordered (The Faneuil Group, Inc., 1999).
Figure 15.5a-d Example Horizontal Line Charts
(a) Multiple Horizontal Line Chart: Percentage Change Over Time (c) Single Horizontal Line Chart: Attributes x Performance Rating
408
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
(b) Multiple Horizontal Line Chart: Switching Percent x Percent Savings (d) Multiple Horizontal Line Chart with Horizontal Bars
Bar Chart
A bar chart is another useful type of graph. Such a chart shows the relationship between
a dependent variable and some independent variable at discrete levels. The stacked vertical bar
chart shown in Figure 15.6a is used to show the relative satisfaction given on-time performance
(x-axis). Figure 15.6b shows percentages associated with frequency counts of three different
market segments by age categories.
Figure 15.6a-d Example Bar Charts
(a) Stacked Vertical Bar Chart (b) Vertical Bar Chart: Competitive Market Share by Customer Age
(c) Stacked Vertical Bar Chart: Loyalty x Satisfaction (d) Multivariable Horizontal Bar Chart with Threshold Line
409
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
There are times when a researcher may feel the need to draw special attention to a
particular result and so may add multiple variables with demarcation lines or special cables.
When using this type of notation, the researcher should make sure that he or she has not added
too much information that confuses the reader. See Figures 15.6c and 15.6d for more detail (The
Faneuil Group, Inc., 1999).
Specialty Charts
Other widely used chart formats include the quadrant, high-low, hierarchical tree, and
surface charts (see Figures 15.7a-d). These charts are particularly useful to display a quantity that
is subdivided into parts. When more categories are analyzed, tree diagrams may be used. Care
should be exercised when too many subdivisions are shown and some portions are so small and
narrow it is difficult to see them.
Dot Charts
A simple, yet powerful, type of chart is the dot chart. This is a way to display measurements of
one or more quantitative variables in which each value may have a label associated with it. We
have already shown such charts by the scatter plots (diagrams) in Chapter 13.
Figure 15.7 Example Specialty Charts
(a) Quadrant Chart with Threshold Lines (b) High-Low Chart: Satisfaction by Channel
(c) Heat Map: Analysis of Ad Interest Areas (d) Surface Map of Neural Network Analysis
Dot charts are used as output from multidimensional scaling (MDS) analyses (see
Chapter 14). Figure 15.8a shows a MDS result for a computer manufacturer where attributes of
computer manufacturers are plotted as dots. A dot chart may be used in a similar manner; Figure
15.8c presents utility functions obtained using conjoint analysis of a 33
fractional factorial design
410
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
examining product attributes (see Chapter 14). When they appear, the solid lines (they could be
dotted as well) should be made light so that they do not detract from the dots themselves.
Although a literal use of this type of chart would show only dots in the body of graph, this need
not be the case. Either letters or numbers also can be used, as illustrated by the
Importance/Performance analysis of Figure 15.8b. The numbers are a result of plotting both
mean ratings of importance and performance.
Figure 15.8a-c Example Dot Charts
(a) Dot Chart of MDS Analysis
(b) Dot Chart Using Numbers
(c) Dot Chart for Conjoint-Analysis Derived Utility Functions
Another useful modification of the dot chart is one that essentially uses tick marks, each
of which is labeled. A beer company was interested in knowing something about preferences for
its beer and a number of competitive brands. Preference rankings for 12 brands were obtained
and analyzed by Thurstonian Case V scaling (see Chapter 10). Prior to applying this procedure,
each respondent was classified in one of three ways:
411
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
1. Sponsor brand-loyal—drinking the sponsor’s brand regularly for at least a year
2. Other brand-loyal—drinking a particular (competitive) brand regularly for at least one year
3. Switcher—others not meeting either of the above conditions
A Thurstonian scale of preferences was then developed for each of the preceding segments
and appears in Figure 15.9.
Figure 15.9 Example Thurstone Case V Scale
Other Graphic Aids
Sometimes key results from a study can be presented by use of maps. Often we see a map
of the United States or another political entity used to display census data. Maps are used to
enhance the meaning of data. For example, the situation facing the plastic irrigation pipe
manufacturer regarding PIP in different selling areas of the country is shown in Figure 15.10a,
which shows the geographic boundaries of selling areas. Figure 15.10b shows a map of the eight
western states and reports the percent of respondent pipe sellers who sell PIP in each area. This
type of map gives some life to the data beyond listing market area and showing percentages.
Another device that can enhance the presentation is a pictorial chart or pictogram. A
pictogram depicts data with the help of symbols, which can be anything—stars, stacks of coins,
trees, castles, facial expressions, money, caricatures of people, and so forth. Each symbol
represents a specific and uniform amount or value. An example is shown in Figure 15.10c. There
are situations where the size of the pictorial symbol is made proportional to the values that are to
be shown, as illustrated in Figure 15.10d. One must be careful when doing this so as to not
confuse the reader by distorting symbols. For example when doubling the size of a threedimensional
symbol, if each dimension is doubled the volume is increased eight times.
412
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Figure 15.10a-e Use of Maps and Pictograms in Displaying Research Results
Selling Area
Percentage of All
PIP Respondents
Selling in the Area
Percentage of PIP
Sellers With More
Than 5% Sales in the
Area
Area 1 78% 43%
Area 2 71% 71%
Area 3 57% 57%
Area 4 14% 0%
Area 5 57% 57%
Area 6 43% 29%
(a) Use of Maps in Plastic Irrigation Pipe Study: Selling Areas of PIP
Manufacturers
(b) Use of Maps in Plastic Irrigation Pipe Study: PIP Respondent Selling
Areas as a Percentage of All Respondents Reporting Sales in Area
(c) Pictogram Illustration of Number of Customers by Sales Area (d) Pictogram Illustration: Symbol Proportional to Value
To
Preference
Without Price
Most Prefer 8 6 124 82
2nd Most Prefer 12 16 75 116
3rd Most Prefer 47 152 9 12
Least Prefer 152 46 12 10
(e) Frequencies of Preference for New Package Designs
Pictures or drawings themselves may be useful to show what something looks like. A
manufacturer of carpet cleaning agents developed a new type of foam cleaner called Lift Away,
made especially for spot cleaning. A study was done to test four alternative aerosol packages,
which used a sample of 220 housewives. Figure 15.10e shows sketches of the four package
designs and frequencies of choice. It would have been difficult to describe the packages in words
only. The sketches greatly aid understanding of the differences between the test items.
413
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
A Warning
When developing graphical material one must be very careful. It is not too difficult to get
different impressions from graphs of identical data, each of which is technically correct. Figure
15.11 illustrates this point.
Figure 15.11 Effects of Changing an Axis
By stretching or compressing the horizontal or vertical axis, as the case may be, one can
convey different meanings.
A researcher needs to follow sound principles when constructing graphical material. Such
principles as those discussed in this chapter are focused on obtaining a clear vision, clear
understanding, sound scales, and sound overall strategy for presentation.
THE ORAL REPORT AND PRESENTATION
In general, the written report will be supplemented by an oral/PowerPoint presentation.
The objective behind this oral report is to identify and emphasize the major findings of the study
and to allow the client or manager to ask questions about matters not clear to him or her. In some
instances, however, the oral presentation may be more important than the written report. When
this is the case, the written report consists of an extended management summary (3–4 pages) and
copies of all overhead projector transparencies, photocopies, slides, or PowerPoint slides
presented in the oral report.
When planning for the oral presentation, the researcher should consider the basic
ingredients of such a presentation: (a) the target (purpose of the report); (b) the receiver of the
report (the willingness and capacity of the listeners to understand and accept the report); (c) the
impact needed; and (d) the methods that must be used to achieve the desired impact, given the
receiver and target. The presenter must adapt to the audience, and this may require having to
defend the results being presented. This is not the same as being in an adversarial position. It
does mean, however, that a presenter should be prepared to deal with any and all questions in a
professional, expert, and competent manner.
414
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Each graphic needs to be of sufficient quality that it can be interpreted by the audience.
When the researcher wants to use visual aids, it is important to keep in mind the difference
between type of visual aid and medium of presentation. Any one of the types of visual aids (or
graphics) available from graphics programs can be adapted and presented through many of the
media available. In addition, actual actions and behaviors can be shown via videos, DVDs, and
personal computers.
Programs such as PowerPoint exist to incorporate text, graphics, video and statistical data
in the same report. Exhibit 15.6 presents some rules-of-thumb for using PowerPoint to give great
presentations.
EXHIBIT 15.6 Rules-of-Thumb for Presentation Graphics
Presenting with PowerPoint: 10 dos and don'ts
Jeff Wuorio
Cherie Kerr knows how PowerPoint can be both provocative and persuasive in a business meeting. She's
also aware that precisely the opposite can occur. "It can be the very best friend you have," says the Santa Ana,
Calif., public relations consultant. "But you have to use it right." Many embrace the values of PowerPoint as a
potent business tool, but there are others who contend that it's a drag on effective interaction — that it
confuses, distorts and even strangles communication. PowerPoint must be used to best advantage. Here are 10
ways to use PowerPoint to help make your business look brilliant, not brainless.
1. Hold up your end with compelling material.
In a way, PowerPoint's ease of use may be its own worst enemy. However simple and engaging it can be to
build eye catching slides and graphics, bear in mind that PowerPoint isn't autonomous. The audience has
come to hear you, not merely to stare at images tossed onto a screen. Build a strong PowerPoint program,
but make sure that your spoken remarks are no less compelling. "PowerPoint doesn't give presentations —
PowerPoint makes slides," says Matt Thornhill, president of Audience First. "Remember that you are
creating slides to support a spoken presentation."
2. Keep it simple.
The most effective PowerPoint presentations are simple — charts that are easy to understand, and graphics
that reflect what the speaker is saying. Some authorities suggest no more than five words per line and three
to five lines per individual slide. "Don't gum up the works with too many words and graphics," Kerr says. "Do
you really need to have everything up on the screen?"
3. Minimize numbers in slides.
PowerPoint's lure is the capacity to convey ideas and support a speaker's remarks in a concise manner.
That's hard to do through a haze of numbers and statistics. For the most part, most effective PowerPoint
displays don't overwhelm viewers with too many figures and numbers. Instead, leave those for a later, more
thorough digestion in handouts distributed at presentation's end. If you want to emphasize a statistic,
consider using a graphic or image to convey the point. "For instance, when I once was talking about the
prevalence of Alzheimer's patients, I used a photograph of an old woman rather than just throwing up a
number on the screen," Kerr says.
4. Don't parrot PowerPoint.
One of the most prevalent and damaging habits of PowerPoint users is to simply read the visual
presentation to the audience. Not only is that redundant — short of using the clicker, why are you even
there? — but it makes even the most visually appealing presentation boring to the bone. PowerPoint works
best with spoken remarks that augment and discuss, rather than mimic, what's on the screen. You've got to
make eye contact with your audience, so never turn your back to your audience.
5. Time your remarks.
A well-orchestrated PowerPoint program brings up a new slide and gives the audience a chance to read and
digest it, then follows up with remarks that broaden and amplify what's on the screen. "It's an issue of
timing," Kerr says. "Never talk on top of your slides." Plan on no more than 1 slide per minute.
415
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
6. Give it a rest.
PowerPoint is most effective as a visual accompaniment to the spoken word. Don’t be bashful about letting
the screen go blank on occasion. Not only can that give your audience a visual break, it's also effective to
focus attention on more verbally-focused give and take, such as a group discussion or question and answer
session.
7. Use vibrant colors.
A striking contrast between words, graphics and the background can be very effective in conveying both a
message and emotion. The standard theme colors are developed by graphic artists to be pleasing and not
overbearing.
8. Import other images and graphics.
Don't limit your presentation to what PowerPoint offers. Use outside images and graphics for variety and
visual appeal, including video. It can add humor, convey a message and get the audience on board.
9. Distribute handouts at the end — not during the presentation.
Some people may disagree with me here. But no speaker wants to be chatting to a crowd that's busy
reading a summation of her remarks. Unless it is imperative that people follow a handout while you're
presenting, wait until you're done to distribute them.
10. Edit ruthlessly before presenting.
Never lose the perspective of the audience. Once you're finished drafting your PowerPoint slides, assume
you're just one of the folks listening to your remarks as you review them. If something is unappealing,
distracting or confusing, edit ruthlessly. Chances are good your overall presentation will be the better for it.
A great presentation often takes several weeks to prepare.
Source: Jeff Wuorio, Presenting with PowerPoint: 10 dos and don'ts
http://www.microsoft.com/smallbusiness/resources/technology/business-software/presenting-with-powerpoint-10-dos-
and-donts.aspx#Powerpointtips
SUMMARY
This chapter has covered preparing the research report. In order to plan properly for the
report, the researcher must understand some basic concepts of communications.
We discussed the components of a formal research report in some depth. Reports need to
be complete, accurate, concise, and clear in presentation. There is no one correct way to prepare
a marketing research report, although we do discuss some general guidelines or rules for report
preparation. The intended audience must be kept in mind at all times. There are many degrees of
formality that may be used. Increasingly, marketing research reports seem to be written less
formally, with a major emphasis being placed on the oral presentation.
We discussed in some depth the use of graphics to show the structure of data. Tables,
charts, figures, maps, and other devices were presented. Numerous examples drawn from realworld
marketing research projects were given to illustrate the various graphic aids.
Next we turned to the oral report. In many situations, the oral report is used to
supplement a formal written report. This allows the potential user of the study results to ask
questions and seek clarification of aspects that may be confusing. In other situations, the oral
report is the major vehicle for the presentation of data and information. The written report then
consists of an extended management summary and copies of all overhead transparencies used in
the presentation.
416
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
REFERENCES
Bailey, E. P., Jr. (1990). The plain English approach to business report writing. New York: Oxford
University Press.
Cleveland, W. S. (1985). The elements of graphing data. Monterey, CA: Wadsworth Advanced Books
and Software.
Faneuil Group, Inc. (1999). Training Manual. Boston.
Fink, A. (2003). How to report on surveys. Thousand Oaks, CA: Sage.
Gordon, H. (1994, December), in S. J. Levy, G. R. Frerichs, & H. L. Gordon (Eds.) The Dartnell
marketing manager’s handbook. Chicago, IL: Dartnell Corporations.
Greenberg, L. (1999, December). The challenge of qualitative report writing. Quirk’s Marketing Research
Review, 36–41.
Gutsche, A. (2001, September 24). Visuals make the case. Marketing News, 21–23.
Jacoby, W. G. (1997). Statistical graphics for univariate and bivariate data. Thousand Oaks, CA: Sage.
Jacoby,W. G. (1998). Statistical graphics for visualizing multivariate data. Thousand Oaks, CA: Sage.
Nelems, J. H. (1979, January 12). Report results implications, not just pounds of data. Marketing News,
12, 7.
Rehart, M. J. V. (1994). Writing business research reports: A guide to scientific writing. Los Angeles:
Pyrczak Publishing.
Robinson, D. M. (1969). Writing reports for management decisions. Columbus, OH: Charles E. Merrill.
Rockey, E. H. (1984). Communicating in organizations. Lanham, MD: University Press of America, Inc.
Semon, T. T. (1998, November 23). To make research effective, start by drafting the report. Marketing
News, 32, 24, 17.
Timm, P. R., & Bienvenu, S. (2001). Business communications: Discovering strategy, developing skills.
Upper Saddle, NJ: Prentice Hall.
Vardaman, G. T. (1981). Making successful presentations. New York: AMACOM.
Wainer, H. (1984, May). How to display data badly. The American Statistician, 38, 137–147.
Witzling, L. P., & Greenstreet, R. C. (1989). Presenting statistics: A manager’s guide to the persuasive
use of statistics. New York: John Wiley & Sons.
Wrobel, T. K. (1990, September 3). Research is useless if it’s poorly communicated. Marketing News, 24,
18, 41ff.
Wuorio, J. Presenting with PowerPoint: 10 dos and don'ts. http://www.microsoft.com
A-1
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
APPENDIX A Statistical Tables
Table A.1 Cumulative Normal Distribution—Values of Probability
Z Values of P corresponding to Z for the normal curve. Z is the standard normal variable.
The value of P for −Z equals 1 minus the value of P for +Z, e.g., the P for –1.62 equals 1 – .9474 = .0526
Z 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.0 0.5000 0.5040 0.5080 0.5120 0.5160 0.5199 0.5239 0.5279 0.5319 0.5359
0.1 0.5398 0.5438 0.5478 0.5517 0.5557 0.5596 0.5636 0.5675 0.5714 0.5753
0.2 0.5793 0.5832 0.5871 0.5910 0.5948 0.5987 0.6026 0.6064 0.6103 0.6141
0.3 0.6179 0.6217 0.6255 0.6293 0.6331 0.6368 0.6406 0.6443 0.6480 0.6517
0.4 0.6554 0.6591 0.6628 0.6664 0.6700 0.6736 0.6772 0.6808 0.6844 0.6879
0.5 0.6915 0.6950 0.6985 0.7019 0.7054 0.7088 0.7123 0.7157 0.7190 0.7224
0.6 0.7257 0.7291 0.7324 0.7357 0.7389 0.7422 0.7454 0.7486 0.7517 0.7549
0.7 0.7580 0.7611 0.7642 0.7673 0.7704 0.7734 0.7764 0.7794 0.7823 0.7852
0.8 0.7881 0.7910 0.7939 0.7967 0.7995 0.8023 0.8051 0.8078 0.8106 0.8133
0.9 0.8159 0.8186 0.8212 0.8238 0.8264 0.8289 0.8315 0.8340 0.8360 0.8389
1.0 0.8413 0.8438 0.8461 0.8485 0.8508 0.8531 0.8554 0.8577 0.8599 0.8621
1.1 0.8643 0.8665 0.8686 0.8708 0.8729 0.8749 0.8770 0.8790 0.8810 0.8830
1.2 0.8849 0.8869 0.8888 0.8907 0.8925 0.8944 0.8962 0.8980 0.8997 0.9015
1.3 0.9032 0.9049 0.9066 0.9082 0.9099 0.9115 0.9131 0.9147 0.9162 0.9177
1.4 0.9192 0.9207 0.9222 0.9236 0.9251 0.9265 0.9279 0.9292 0.9306 0.9319
1.5 0.9332 0.9345 0.9357 0.9370 0.9382 0.9394 0.9406 0.9418 0.9429 0.9441
1.6 0.9452 0.9463 0.9474 0.9484 0.9495 0.9505 0.9515 0.9525 0.9535 0.9545
1.7 0.9554 0.9564 0.9573 0.9582 0.9591 0.9599 0.9608 0.9616 0.9625 0.9633
1.8 0.9641 0.9649 0.9656 0.9664 0.9671 0.9678 0.9686 0.9693 0.9699 0.9706
1.9 0.9713 0.9719 0.9726 0.9732 0.9738 0.9744 0.9750 0.9756 0.9761 0.9767
2.0 0.9772 0.9778 0.9783 0.9788 0.9793 0.9798 0.9803 0.9808 0.9812 0.9817
2.1 0.9821 0.9826 0.9830 0.9834 0.9838 0.9842 0.9846 0.9850 0.9854 0.9857
2.2 0.9861 0.9864 0.9868 0.9871 0.9875 0.9878 0.9881 0.9884 0.9887 0.9890
2.3 0.9893 0.9896 0.9898 0.9901 0.9904 0.9906 0.9909 0.9911 0.9913 0.9916
2.4 0.9918 0.9920 0.9922 0.9925 0.9927 0.9929 0.9931 0.9932 0.9934 0.9936
2.5 0.9938 0.9940 0.9941 0.9943 0.9945 0.9946 0.9948 0.9949 0.9951 0.9952
2.6 0.9953 0.9955 0.9956 0.9957 0.9959 0.9960 0.9961 0.9962 0.9963 0.9964
2.7 0.9965 0.9966 0.9967 0.9960 0.9969 0.9970 0.9971 0.9972 0.9973 0.9974
2.8 0.9974 0.9975 0.9976 0.9977 0.9977 0.9978 0.9979 0.9979 0.9980 0.9981
2.9 0.9981 0.9982 0.9982 0.9983 0.9984 0.9984 0.9985 0.9985 0.9986 0.9986
3.0 0.9987 0.9987 0.9987 0.9988 0.9988 0.9989 0.9989 0.9989 0.9990 0.9990
3.1 0.9990 0.9991 0.9991 0.9991 0.9992 0.9992 0.9992 0.9992 0.9993 0.9993
3.2 0.9993 0.9993 0.9994 0.9994 0.9994 0.9994 0.9994 0.9995 0.9995 0.9995
3.3 0.9995 0.9995 0.9995 0.9996 0.9996 0.9996 0.9996 0.9996 0.9996 0.9997
3.4 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9998
A-2
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
df = degrees of freedom
SOURCE: From Fisher, R. A., & Yates, F., Statistical Tables for Biological, Agricultural, and Medical Research, 6/e,
copyright © 1963. Reprinted with permission from Pearson Education, Ltd.
Table A.2 Upper Percentiles of the t Distribution
1– α df .75 .90 .95 .975 .99 .995 .9995
1 1.000 3.078 6.314 12.706 31.821 63.657 636.619
2 .816 1.886 2.920 4.303 6.965 9.925 31.598
3 .765 1.638 2.353 3.182 4.541 5.841 12.941
4 .741 1.533 2.132 2.776 3.747 4.604 8.610
5 .727 1.476 2.015 2.571 3.365 4.032 6.859
6 .718 1.440 1.943 2.447 3.143 3.707 5.959
7 .711 1.415 1.895 2.365 2.998 3.499 5.405
8 .706 1.397 1.860 2.306 2.896 3.355 5.041
9 .703 1.383 1.833 2.262 2.821 3.250 4.781
10 .700 1.372 1.812 2.228 2.764 3.169 4.587
11 .697 1.363 1.796 2.201 2.718 3.106 4.437
12 .695 1.356 1.782 2.179 2.681 3.055 4.318
13 .694 1.350 1.771 2.160 2.650 3.012 4.221
14 .692 1.345 1.761 2.145 2.624 2.977 4.140
15 .691 1.341 1.753 2.131 2.602 2.947 4.073
16 .690 1.337 1.746 2.120 2.583 2.921 4.015
17 .689 1.333 1.740 2.110 2.567 2.898 3.965
18 .688 1.330 1.734 2.101 2.552 2.878 3.922
19 .688 1.328 1.729 2.093 2.339 2.861 3.883
20 .687 1.325 1.725 2.086 2.528 2.845 3.850
21 .686 1.323 1.721 2.080 2.518 2.831 3.819
22 .686 1.321 1.717 2.074 2.508 2.819 3.792
23 .685 1.319 1.714 2.069 2.500 2.807 3.767
24 .685 1.318 1.711 2.064 2.492 2.797 3.745
25 .684 1.316 1.708 2.060 2.485 2.787 3.725
26 .684 1.315 1.706 2.056 2.479 2.779 3.707
27 .684 1.314 1.703 2.052 2.473 2.771 3.690
28 .683 1.313 1.701 2.048 2.467 2.763 3.674
29 .683 1.311 1.699 2.045 2.462 2.756 3.659
30 .683 1.310 1.697 2.042 2.457 2.750 3.646
40 .681 1.303 1.684 2.021 2.423 2.704 3.551
60 .679 1.296 1.671 2.000 2.390 2.660 3.460
120 .677 1.289 1.658 1.980 2.358 2.617 3.373
∞ .674 1.282 1.645 1.960 2.326 2.576 3.291
A-3
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table A.3 Percentiles of the χ 2
Distribution
df χ2 .005 χ2 .01 χ2 .025 χ2 .05 χ2 .10 χ2 .90 χ2 .95 χ2 .975 χ2 .99 χ2 .995
1 .000039 .00016 .00098 .0039 .0158 2.71 3.84 5.02 6.63 7.88
2 .0100 .0201 .0506 .1026 .2107 4.61 5.99 7.38 9.21 10.60
3 .0717 .115 .216 .352 .584 6.25 7.81 9.35 11.34 12.84
4 .207 .297 .484 .711 1.064 7.78 9.49 11.14 13.28 14.86
5 .412 .554 .831 1.15 1.61 9.24 11.07 12.83 15.09 16.75
6 .676 .872 1.24 1.64 2.20 10.64 12.59 14.45 16.81 18.55
7 .989 1.24 1.69 2.17 2.83 12.02 14.07 16.01 18.48 20.28
8 1.34 1.65 2.18 2.73 3.49 13.36 15.51 17.53 20.09 21.96
9 1.73 2.09 2.70 3.33 4.17 14.68 16.92 19.02 21.67 23.59
10 2.16 2.56 3.25 3.94 4.87 15.99 18.31 20.48 23.21 25.19
11 2.60 3.05 3.82 4.57 5.58 17.28 19.68 21.92 24.73 26.76
12 3.07 3.57 4.40 5.23 6.30 18.55 21.03 23.34 26.22 28.30
13 3.57 4.11 5.01 5.89 7.04 19.81 22.36 24.74 27.69 29.82
14 4.07 4.66 5.63 6.57 7.79 21.06 23.68 26.12 29.14 31.32
15 4.60 5.23 6.26 7.26 8.55 22.31 25.00 27.49 30.58 32.80
16 5.14 5.81 6.91 7.96 9.31 23.54 26.30 28.85 32.00 34.27
18 6.26 7.01 8.23 9.39 10.86 25.99 28.87 31.53 34.81 37.16
20 7.43 8.26 9.59 10.85 12.44 28.41 31.41 34.17 37.57 40.00
24 9.89 10.86 12.40 13.85 15.66 33.20 36.42 39.36 42.98 45.56
30 13.79 14.95 16.79 18.49 20.60 40.26 43.77 46.98 50.89 53.67
40 20.71 22.16 24.43 26.51 29.05 51.81 55.76 59.34 63.69 66.77
60 35.53 37.48 40.48 43.19 46.46 74.40 79.08 83.30 88.38 91.95
120 83.85 86.92 91.58 95.70 100.62 140.23 146.57 152.21 158.95 163.64
For large degrees of freedom. (approximately) where v = degrees of freedom and Z is given in Table A.1.
SOURCE: Adapted from Introduction to Statistical Analysis, 2nd ed., by W. J. Dixon and F. J. Massey, Jr.,
Copyright 1957, McGraw-Hill Book Company.
A-4
Table A.4 Percentiles of the F Distribution
F Table for alpha=.10
n
1= degrees of freedom for numerator ) α= 0.10
n
2=degreesoffreedomfordenominator)α=0.10
df2/df1 1 2 3 4 5 6 7 8 9 10 12 15 20 24 30 40 60 120 ∞
1 39.863 49.500 53.593 55.833 57.240 58.204 58.906 59.439 59.858 60.195 60.705 61.220 61.740 62.002 62.265 62.529 62.794 63.061 63.328
2 8.526 9.000 9.162 9.243 9.293 9.326 9.349 9.367 9.381 9.392 9.408 9.425 9.441 9.450 9.458 9.466 9.475 9.483 9.491
3 5.538 5.462 5.391 5.343 5.309 5.285 5.266 5.252 5.240 5.230 5.216 5.200 5.184 5.176 5.168 5.160 5.151 5.143 5.134
4 4.545 4.325 4.191 4.107 4.051 4.010 3.979 3.955 3.936 3.920 3.896 3.870 3.844 3.831 3.817 3.804 3.790 3.775 3.761
5 4.060 3.780 3.619 3.520 3.453 3.405 3.368 3.339 3.316 3.297 3.268 3.238 3.207 3.191 3.174 3.157 3.140 3.123 3.105
6 3.776 3.463 3.289 3.181 3.108 3.055 3.014 2.983 2.958 2.937 2.905 2.871 2.836 2.818 2.800 2.781 2.762 2.742 2.722
7 3.589 3.257 3.074 2.961 2.883 2.827 2.785 2.752 2.725 2.703 2.668 2.632 2.595 2.575 2.555 2.535 2.514 2.493 2.471
8 3.458 3.113 2.924 2.806 2.726 2.668 2.624 2.589 2.561 2.538 2.502 2.464 2.425 2.404 2.383 2.361 2.339 2.316 2.293
9 3.360 3.006 2.813 2.693 2.611 2.551 2.505 2.469 2.440 2.416 2.379 2.340 2.298 2.277 2.255 2.232 2.208 2.184 2.159
10 3.285 2.924 2.728 2.605 2.522 2.461 2.414 2.377 2.347 2.323 2.284 2.244 2.201 2.178 2.155 2.132 2.107 2.082 2.055
11 3.225 2.860 2.660 2.536 2.451 2.389 2.342 2.304 2.274 2.248 2.209 2.167 2.123 2.100 2.076 2.052 2.026 2.000 1.972
12 3.177 2.807 2.606 2.480 2.394 2.331 2.283 2.245 2.214 2.188 2.147 2.105 2.060 2.036 2.011 1.986 1.960 1.932 1.904
13 3.136 2.763 2.560 2.434 2.347 2.283 2.234 2.195 2.164 2.138 2.097 2.053 2.007 1.983 1.958 1.931 1.904 1.876 1.846
14 3.102 2.726 2.522 2.395 2.307 2.243 2.193 2.154 2.122 2.095 2.054 2.010 1.962 1.938 1.912 1.885 1.857 1.828 1.797
15 3.073 2.695 2.490 2.361 2.273 2.208 2.158 2.119 2.086 2.059 2.017 1.972 1.924 1.899 1.873 1.845 1.817 1.787 1.755
16 3.048 2.668 2.462 2.333 2.244 2.178 2.128 2.088 2.055 2.028 1.985 1.940 1.891 1.866 1.839 1.811 1.782 1.751 1.718
17 3.026 2.645 2.437 2.308 2.218 2.152 2.102 2.061 2.028 2.001 1.958 1.912 1.862 1.836 1.809 1.781 1.751 1.719 1.686
18 3.007 2.624 2.416 2.286 2.196 2.130 2.079 2.038 2.005 1.977 1.933 1.887 1.837 1.810 1.783 1.754 1.723 1.691 1.657
19 2.990 2.606 2.397 2.266 2.176 2.109 2.058 2.017 1.984 1.956 1.912 1.865 1.814 1.787 1.759 1.730 1.699 1.666 1.631
20 2.975 2.589 2.380 2.249 2.158 2.091 2.040 1.999 1.965 1.937 1.892 1.845 1.794 1.767 1.738 1.708 1.677 1.643 1.607
21 2.961 2.575 2.365 2.233 2.142 2.075 2.023 1.982 1.948 1.920 1.875 1.827 1.776 1.748 1.719 1.689 1.657 1.623 1.586
22 2.949 2.561 2.351 2.219 2.128 2.061 2.008 1.967 1.933 1.904 1.859 1.811 1.759 1.731 1.702 1.671 1.639 1.604 1.567
23 2.937 2.549 2.339 2.207 2.115 2.047 1.995 1.953 1.919 1.890 1.845 1.796 1.744 1.716 1.686 1.655 1.622 1.587 1.549
24 2.927 2.538 2.327 2.195 2.103 2.035 1.983 1.941 1.906 1.877 1.832 1.783 1.730 1.702 1.672 1.641 1.607 1.571 1.533
25 2.918 2.528 2.317 2.184 2.092 2.024 1.971 1.929 1.895 1.866 1.820 1.771 1.718 1.689 1.659 1.627 1.593 1.557 1.518
26 2.909 2.519 2.307 2.174 2.082 2.014 1.961 1.919 1.884 1.855 1.809 1.760 1.706 1.677 1.647 1.615 1.581 1.544 1.504
27 2.901 2.511 2.299 2.165 2.073 2.005 1.952 1.909 1.874 1.845 1.799 1.749 1.695 1.666 1.636 1.603 1.569 1.531 1.491
28 2.894 2.503 2.291 2.157 2.064 1.996 1.943 1.900 1.865 1.836 1.790 1.740 1.685 1.656 1.625 1.593 1.558 1.520 1.478
29 2.887 2.495 2.283 2.149 2.057 1.988 1.935 1.892 1.857 1.827 1.781 1.731 1.676 1.647 1.616 1.583 1.547 1.509 1.467
30 2.881 2.489 2.276 2.142 2.049 1.980 1.927 1.884 1.849 1.819 1.773 1.722 1.667 1.638 1.606 1.573 1.538 1.499 1.456
40 2.835 2.440 2.226 2.091 1.997 1.927 1.873 1.829 1.793 1.763 1.715 1.662 1.605 1.574 1.541 1.506 1.467 1.425 1.377
60 2.791 2.393 2.177 2.041 1.946 1.875 1.819 1.775 1.738 1.707 1.657 1.603 1.543 1.511 1.476 1.437 1.395 1.348 1.291
120 2.748 2.347 2.130 1.992 1.896 1.824 1.767 1.722 1.684 1.652 1.601 1.545 1.482 1.447 1.409 1.368 1.320 1.265 1.193
∞ 2.706 2.303 2.084 1.945 1.847 1.774 1.717 1.670 1.632 1.599 1.546 1.487 1.421 1.383 1.342 1.295 1.240 1.169 1.000
A-5
Table A.4 Percentiles of the F Distribution (continued)
F Table for alpha=.05
n
1= degrees of freedom for numerator ) α= 0.05
n
2=degreesoffreedomfordenominator)α=0.05
df2/df1 1 2 3 4 5 6 7 8 9 10 12 15 20 24 30 40 60 120 ∞
1 161.45 199.50 215.71 224.58 230.16 233.99 236.77 238.88 240.54 241.88 243.91 245.95 248.01 249.05 250.10 251.14 252.20 253.25 254.31
2 18.513 19.000 19.164 19.247 19.296 19.330 19.353 19.371 19.385 19.396 19.413 19.429 19.446 19.454 19.462 19.471 19.479 19.487 19.496
3 10.128 9.552 9.277 9.117 9.014 8.941 8.887 8.845 8.812 8.786 8.745 8.703 8.660 8.639 8.617 8.594 8.572 8.549 8.526
4 7.709 6.944 6.591 6.388 6.256 6.163 6.094 6.041 5.999 5.964 5.912 5.858 5.803 5.774 5.746 5.717 5.688 5.658 5.628
5 6.608 5.786 5.410 5.192 5.050 4.950 4.876 4.818 4.773 4.735 4.678 4.619 4.558 4.527 4.496 4.464 4.431 4.399 4.365
6 5.987 5.143 4.757 4.534 4.387 4.284 4.207 4.147 4.099 4.060 4.000 3.938 3.874 3.842 3.808 3.774 3.740 3.705 3.669
7 5.591 4.737 4.347 4.120 3.972 3.866 3.787 3.726 3.677 3.637 3.575 3.511 3.445 3.411 3.376 3.340 3.304 3.267 3.230
8 5.318 4.459 4.066 3.838 3.688 3.581 3.501 3.438 3.388 3.347 3.284 3.218 3.150 3.115 3.079 3.043 3.005 2.967 2.928
9 5.117 4.257 3.863 3.633 3.482 3.374 3.293 3.230 3.179 3.137 3.073 3.006 2.937 2.901 2.864 2.826 2.787 2.748 2.707
10 4.965 4.103 3.708 3.478 3.326 3.217 3.136 3.072 3.020 2.978 2.913 2.845 2.774 2.737 2.700 2.661 2.621 2.580 2.538
11 4.844 3.982 3.587 3.357 3.204 3.095 3.012 2.948 2.896 2.854 2.788 2.719 2.646 2.609 2.571 2.531 2.490 2.448 2.405
12 4.747 3.885 3.490 3.259 3.106 2.996 2.913 2.849 2.796 2.753 2.687 2.617 2.544 2.506 2.466 2.426 2.384 2.341 2.296
13 4.667 3.806 3.411 3.179 3.025 2.915 2.832 2.767 2.714 2.671 2.604 2.533 2.459 2.420 2.380 2.339 2.297 2.252 2.206
14 4.600 3.739 3.344 3.112 2.958 2.848 2.764 2.699 2.646 2.602 2.534 2.463 2.388 2.349 2.308 2.266 2.223 2.178 2.131
15 4.543 3.682 3.287 3.056 2.901 2.791 2.707 2.641 2.588 2.544 2.475 2.403 2.328 2.288 2.247 2.204 2.160 2.114 2.066
16 4.494 3.634 3.239 3.007 2.852 2.741 2.657 2.591 2.538 2.494 2.425 2.352 2.276 2.235 2.194 2.151 2.106 2.059 2.010
17 4.451 3.592 3.197 2.965 2.810 2.699 2.614 2.548 2.494 2.450 2.381 2.308 2.230 2.190 2.148 2.104 2.058 2.011 1.960
18 4.414 3.555 3.160 2.928 2.773 2.661 2.577 2.510 2.456 2.412 2.342 2.269 2.191 2.150 2.107 2.063 2.017 1.968 1.917
19 4.381 3.522 3.127 2.895 2.740 2.628 2.544 2.477 2.423 2.378 2.308 2.234 2.156 2.114 2.071 2.026 1.980 1.930 1.878
20 4.351 3.493 3.098 2.866 2.711 2.599 2.514 2.447 2.393 2.348 2.278 2.203 2.124 2.083 2.039 1.994 1.946 1.896 1.843
21 4.325 3.467 3.073 2.840 2.685 2.573 2.488 2.421 2.366 2.321 2.250 2.176 2.096 2.054 2.010 1.965 1.917 1.866 1.812
22 4.301 3.443 3.049 2.817 2.661 2.549 2.464 2.397 2.342 2.297 2.226 2.151 2.071 2.028 1.984 1.938 1.889 1.838 1.783
23 4.279 3.422 3.028 2.796 2.640 2.528 2.442 2.375 2.320 2.275 2.204 2.128 2.048 2.005 1.961 1.914 1.865 1.813 1.757
24 4.260 3.403 3.009 2.776 2.621 2.508 2.423 2.355 2.300 2.255 2.183 2.108 2.027 1.984 1.939 1.892 1.842 1.790 1.733
25 4.242 3.385 2.991 2.759 2.603 2.490 2.405 2.337 2.282 2.237 2.165 2.089 2.008 1.964 1.919 1.872 1.822 1.768 1.711
26 4.225 3.369 2.975 2.743 2.587 2.474 2.388 2.321 2.266 2.220 2.148 2.072 1.990 1.946 1.901 1.853 1.803 1.749 1.691
27 4.210 3.354 2.960 2.728 2.572 2.459 2.373 2.305 2.250 2.204 2.132 2.056 1.974 1.930 1.884 1.836 1.785 1.731 1.672
28 4.196 3.340 2.947 2.714 2.558 2.445 2.359 2.291 2.236 2.190 2.118 2.041 1.959 1.915 1.869 1.820 1.769 1.714 1.654
29 4.183 3.328 2.934 2.701 2.545 2.432 2.346 2.278 2.223 2.177 2.105 2.028 1.945 1.901 1.854 1.806 1.754 1.698 1.638
30 4.171 3.316 2.922 2.690 2.534 2.421 2.334 2.266 2.211 2.165 2.092 2.015 1.932 1.887 1.841 1.792 1.740 1.684 1.622
40 4.085 3.232 2.839 2.606 2.450 2.336 2.249 2.180 2.124 2.077 2.004 1.925 1.839 1.793 1.744 1.693 1.637 1.577 1.509
60 4.001 3.150 2.758 2.525 2.368 2.254 2.167 2.097 2.040 1.993 1.917 1.836 1.748 1.700 1.649 1.594 1.534 1.467 1.389
120 3.920 3.072 2.680 2.447 2.290 2.175 2.087 2.016 1.959 1.911 1.834 1.751 1.659 1.608 1.554 1.495 1.429 1.352 1.254
∞ 3.842 2.996 2.605 2.372 2.214 2.099 2.010 1.938 1.880 1.831 1.752 1.666 1.571 1.517 1.459 1.394 1.318 1.221 1.000
A-6
Table A.4 Percentiles of the F Distribution (continued)
F Table for alpha=.01
n
1= degrees of freedom for numerator ) α= 0.01
n
2=degreesoffreedomfordenominator)α=.01
df2/df1 1 2 3 4 5 6 7 8 9 10 12 15 20 24 30 40 60 120 ∞
1 4052.2 4999.5 5403.4 5624.6 5763.7 5859.0 5928.4 5981.1 6022.5 6055.8 6106.3 6157.3 6208.7 6234.6 6260.6 6286.8 6313.0 6339.4 6365.9
2 98.503 99.000 99.166 99.249 99.299 99.333 99.356 99.374 99.388 99.399 99.416 99.433 99.449 99.458 99.466 99.474 99.482 99.491 99.499
3 34.116 30.817 29.457 28.710 28.237 27.911 27.672 27.489 27.345 27.229 27.052 26.872 26.690 26.598 26.505 26.411 26.316 26.221 26.125
4 21.198 18.000 16.694 15.977 15.522 15.207 14.976 14.799 14.659 14.546 14.374 14.198 14.020 13.929 13.838 13.745 13.652 13.558 13.463
5 16.258 13.274 12.060 11.392 10.967 10.672 10.456 10.289 10.158 10.051 9.888 9.722 9.553 9.466 9.379 9.291 9.202 9.112 9.020
6 13.745 10.925 9.780 9.148 8.746 8.466 8.260 8.102 7.976 7.874 7.718 7.559 7.396 7.313 7.229 7.143 7.057 6.969 6.880
7 12.246 9.547 8.451 7.847 7.460 7.191 6.993 6.840 6.719 6.620 6.469 6.314 6.155 6.074 5.992 5.908 5.824 5.737 5.650
8 11.259 8.649 7.591 7.006 6.632 6.371 6.178 6.029 5.911 5.814 5.667 5.515 5.359 5.279 5.198 5.116 5.032 4.946 4.859
9 10.561 8.022 6.992 6.422 6.057 5.802 5.613 5.467 5.351 5.257 5.111 4.962 4.808 4.729 4.649 4.567 4.483 4.398 4.311
10 10.044 7.559 6.552 5.994 5.636 5.386 5.200 5.057 4.942 4.849 4.706 4.558 4.405 4.327 4.247 4.165 4.082 3.996 3.909
11 9.646 7.206 6.217 5.668 5.316 5.069 4.886 4.744 4.632 4.539 4.397 4.251 4.099 4.021 3.941 3.860 3.776 3.690 3.602
12 9.330 6.927 5.953 5.412 5.064 4.821 4.640 4.499 4.388 4.296 4.155 4.010 3.858 3.780 3.701 3.619 3.535 3.449 3.361
13 9.074 6.701 5.739 5.205 4.862 4.620 4.441 4.302 4.191 4.100 3.960 3.815 3.665 3.587 3.507 3.425 3.341 3.255 3.165
14 8.862 6.515 5.564 5.035 4.695 4.456 4.278 4.140 4.030 3.939 3.800 3.656 3.505 3.427 3.348 3.266 3.181 3.094 3.004
15 8.683 6.359 5.417 4.893 4.556 4.318 4.142 4.004 3.895 3.805 3.666 3.522 3.372 3.294 3.214 3.132 3.047 2.959 2.868
16 8.531 6.226 5.292 4.773 4.437 4.202 4.026 3.890 3.780 3.691 3.553 3.409 3.259 3.181 3.101 3.018 2.933 2.845 2.753
17 8.400 6.112 5.185 4.669 4.336 4.102 3.927 3.791 3.682 3.593 3.455 3.312 3.162 3.084 3.003 2.920 2.835 2.746 2.653
18 8.285 6.013 5.092 4.579 4.248 4.015 3.841 3.705 3.597 3.508 3.371 3.227 3.077 2.999 2.919 2.835 2.749 2.660 2.566
19 8.185 5.926 5.010 4.500 4.171 3.939 3.765 3.631 3.523 3.434 3.297 3.153 3.003 2.925 2.844 2.761 2.674 2.584 2.489
20 8.096 5.849 4.938 4.431 4.103 3.871 3.699 3.564 3.457 3.368 3.231 3.088 2.938 2.859 2.778 2.695 2.608 2.517 2.421
21 8.017 5.780 4.874 4.369 4.042 3.812 3.640 3.506 3.398 3.310 3.173 3.030 2.880 2.801 2.720 2.636 2.548 2.457 2.360
22 7.945 5.719 4.817 4.313 3.988 3.758 3.587 3.453 3.346 3.258 3.121 2.978 2.827 2.749 2.667 2.583 2.495 2.403 2.305
23 7.881 5.664 4.765 4.264 3.939 3.710 3.539 3.406 3.299 3.211 3.074 2.931 2.781 2.702 2.620 2.535 2.447 2.354 2.256
24 7.823 5.614 4.718 4.218 3.895 3.667 3.496 3.363 3.256 3.168 3.032 2.889 2.738 2.659 2.577 2.492 2.403 2.310 2.211
25 7.770 5.568 4.675 4.177 3.855 3.627 3.457 3.324 3.217 3.129 2.993 2.850 2.699 2.620 2.538 2.453 2.364 2.270 2.169
26 7.721 5.526 4.637 4.140 3.818 3.591 3.421 3.288 3.182 3.094 2.958 2.815 2.664 2.585 2.503 2.417 2.327 2.233 2.131
27 7.677 5.488 4.601 4.106 3.785 3.558 3.388 3.256 3.149 3.062 2.926 2.783 2.632 2.552 2.470 2.384 2.294 2.198 2.097
28 7.636 5.453 4.568 4.074 3.754 3.528 3.358 3.226 3.120 3.032 2.896 2.753 2.602 2.522 2.440 2.354 2.263 2.167 2.064
29 7.598 5.420 4.538 4.045 3.725 3.499 3.330 3.198 3.092 3.005 2.868 2.726 2.574 2.495 2.412 2.325 2.234 2.138 2.034
30 7.562 5.390 4.510 4.018 3.699 3.473 3.304 3.173 3.067 2.979 2.843 2.700 2.549 2.469 2.386 2.299 2.208 2.111 2.006
40 7.314 5.179 4.313 3.828 3.514 3.291 3.124 2.993 2.888 2.801 2.665 2.522 2.369 2.288 2.203 2.114 2.019 1.917 1.805
60 7.077 4.977 4.126 3.649 3.339 3.119 2.953 2.823 2.718 2.632 2.496 2.352 2.198 2.115 2.028 1.936 1.836 1.726 1.601
120 6.851 4.787 3.949 3.480 3.174 2.956 2.792 2.663 2.559 2.472 2.336 2.192 2.035 1.950 1.860 1.763 1.656 1.533 1.381
∞ 6.635 4.605 3.782 3.319 3.017 2.802 2.639 2.511 2.407 2.321 2.185 2.039 1.878 1.791 1.696 1.592 1.473 1.325 1.000
A-7
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table A.5 Short Table of Random Digits
9221 3452 9754 8813 6679 3081 3945 9982 1510 827
8127 9422 6665 2154 5450 7042 1481 8014 0978 865
4927 7883 9205 7590 3627 1888 1641 8781 4144 508
6213 5446 3717 0552 9623 0088 9843 0634 6831 832
5334 7757 4781 5395 3258 0296 7318 2323 2485 471
3716 5373 2907 9488 1166 1247 4451 6553 0183 394
5145 3765 9427 6725 4461 1222 8457 8517 5350 868
9851 0178 9628 7183 4597 7034 6444 4269 3567 546
3074 1224 2657 6255 2151 2313 6834 6522 7554 639
1588 1272 6511 3349 2966 8131 6845 8587 1967 646
6687 3025 2377 8365 8044 1782 0422 6664 3636 834
0637 9907 1325 1033 6821 3481 5749 6173 1494 546
7231 7647 3772 5057 0121 6369 2426 2618 2569 811
7484 4844 5451 1561 4101 5482 5315 5877 6473 786
7642 6822 8268 5671 6734 5773 6103 5138 2626 053
3371 6578 6175 6397 8748 0442 4084 6057 6865 695
2147 2406 3484 4852 3824 3382 4844 6785 9836 173
8599 7373 4168 6777 5902 3235 0784 0686 2108 841
0165 0783 5414 6605 8781 4367 4422 9073 1186 852
6838 7236 7321 7445 2493 4579 8267 1116 5165 726
6239 7121 3293 3523 8586 1748 6837 2329 6150 101
8630 9465 1743 7576 8735 7719 8041 4547 4376 674
8136 9768 0114 3089 7523 1724 3166 6376 8269 636
6457 8345 2083 1333 6147 8113 5458 2718 9540 360
4911 0984 1290 0011 7488 7126 8475 6076 7131 782
8326 4220 5646 2386 6453 8472 3784 5781 6127 346
7831 5687 6449 3576 9733 1519 7275 2332 5845 677
9599 9382 3147 0763 4663 7017 0611 4003 7556 261
1774 1331 7482 1843 7638 4427 6117 2794 2153 624
0715 6621 5781 3165 8465 2177 1975 3224 0256 993
3208 8414 9263 0753 1886 8883 8593 6527 8207 762
4730 3362 4449 4956 6728 5518 1642 4717 6252 366
7827 3312 2765 4981 7261 3083 9756 2406 5770 146
4137 1877 1339 1047 1243 6217 6452 2915 9888 582
6514 1122 3156 6239 3702 7011 7979 4711 5875 648
8394 5148 6209 8063 2389 1893 9125 6024 6337 845
5531 8886 8814 2202 2345 2212 3722 3606 5267 112
6271 8587 2617 8619 5073 3825 8246 5528 2517 770
2846 0853 9442 5267 4900 6451 1633 1359 4895 233
1996 0414 6748 2185 4977 3271 9411 4189 4257 087
8093 6724 3822 4761 9475 1245 3694 3453 1188 637
8688 7831 9877 4522 1423 1922 3852 1543 8419 081
1756 7154 3675 7702 0852 3859 5641 0651 9345 058
8112 8434 8807 0362 7829 1944 4644 5104 5348 422
1421 2538 3284 9826 9634 7782 8541 2381 6327 316
7298 6803 2634 8535 0604 7546 1468 5351 7794 206
2782 3515 8334 5127 4282 7832 2779 5837 3176 467
3209 9124 7177 1739 6871 7124 8845 8928 8434 900
9572 2166 0252 0571 2982 2319 3913 8976 2482 579
1877 3400 4822 3143 6211 3138 0253 4542 7575 530
0635 8728 9471 6782 8265 1065 7239 6273 6455 201
2282 4461 1485 2323 1556 2248 8216 7776 3554 655
3638 1484 5648 2673 5642 6048 9777 6035 6918 235
6938 4589 4821 5443 5515 3476 5885 2462 6393 943
7337 3261 6451 8698 0813 0249 5328 4866 1410 617
8782 5228 8387 7138 8954 5868 5456 5640 6977 857
5840 0548 4630 1764 5476 9654 7243 1354 9241 681
3307 5215 7673 3446 2861 1863 6363 4484 4684 676
1001 6776 0247 5929 9125 7368 3679 8355 3816 267
4877 4617 2257 0256 0135 4269 2567 7410 7322 877
5032 9802 9831 6230 3636 8888 2354 7093 7216 768
5230 3531 0676 9427 0702 7546 5664 9757 0353 110
3342 5442 6198 6621 7387 9691 1833 1387 1754 511
A-8
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Table A.6 Short Table of Random Normal Deviates
( 0, σ = 1 )
–0.670 0.518 0.387 0.523 0.641 1.243 0.322 –2.607 –1.097 –0.012
–2.912 1.448 1.343 –0.122 0.726 –0.617 0.609 2.319 –0.450 –1.197
–0.028 –0.790 0.057 1.425 1.940 1.161 –0.878 –0.716 –0.244 –1.151
–1.257 0.774 0.003 0.388 1.060 1.028 –0.236 1.172 0.442 –0.157
2.372 –1.376 –1.318 1.236 0.738 0.337 –0.534 0.090 0.886 0.676
–0.970 0.438 –0.672 –0.180 0.667 1.370 –0.481 0.329 0.842 0.449
–1.228 0.129 –0.426 –0.165 0.028 2.696 1.201 –1.351 0.724 –1.017
–0.369 0.310 0.432 0.237 0.884 –1.224 0.539 0.852 0.497 –0.283
1.161 1.219 1.615 0.336 1.100 –0.528 0.161 0.278 0.675 –1.143
–0.284 2.609 0.792 1.825 –0.249 1.654 0.621 0.979 –1.472 –1.173
–0.578 –0.789 0.106 0.832 –0.597 0.496 –0.561 –1.033 –0.578 –0.378
0.074 0.261 –0.766 –1.046 0.361 –0.043 –1.927 1.527 0.605 1.475
0.230 0.046 0.978 –1.901 1.162 –0.545 0.697 1.151 2.033 0.080
2.162 –0.562 1.190 0.925 –1.057 0.015 –1.371 1.067 –1.080 –1.129
–1.020 –1.130 –0.315 0.628 –0.140 2.050 –0.030 –0.629 0.128 –1.221
1.323 –0.836 –0.284 –0.249 –0.768 1.242 –1.879 –0.417 0.013 –0.502
2.329 1.884 0.033 0.598 –0.217 0.260 0.431 –1.914 0.205 1.155
2.761 1.800 –0.562 0.714 –0.407 0.009 –0.724 –1.168 0.247 1.166
–0.742 0.210 –0.741 –1.099 0.158 2.112 –0.765 –0.319 –0.247 0.345
–1.410 0.413 0.705 1.444 1.057 –0.843 0.043 –0.571 –0.001 0.203
2.272 –0.719 0.679 2.007 –0.180 0.698 –1.137 0.688 –0.571 –0.100
2.832 0.925 –1.350 1.529 –0.260 –1.007 –2.350 –1.501 0.289 1.522
–1.086 –0.558 –0.973 –1.285 –0.021 0.077 0.915 –0.241 –0.249 –0.529
0.134 1.815 0.313 1.571 –0.216 2.261 0.696 –0.130 0.393 0.017
0.783 0.600 –0.745 1.127 –0.684 –0.519 0.125 –0.499 1.543 –0.082
0.174 –0.897 0.575 –0.751 0.694 –2.959 0.529 1.587 0.339 –0.813
–1.319 0.556 2.963 1.218 1.199 –1.746 1.611 0.467 –0.490 0.202
1.298 –0.940 –1.143 –1.136 –1.516 0.548 0.629 0.250 –1.087 0.322
–0.676 –1.107 –1.483 0.278 0.493 –0.442 1.078 –0.336 –0.177 –0.057
–1.287 0.775 –1.095 1.161 –1.877 1.874 1.703 –1.619 –0.725 –1.407
0.260 –0.028 –1.982 0.811 0.999 1.662 0.908 1.476 –1.137 –0.945
0.481 1.060 1.441 0.163 0.720 1.490 –0.026 –0.502 0.427 –0.351
0.794 0.725 1.971 0.384 –0.579 –1.079 –1.440 –0.859 –0.346 0.007
0.584 –0.554 1.460 0.791 –0.426 –0.682 0.430 1.922 –2.099 0.221
–0.114 0.379 –0.698 1.570 –0.511 –0.725 0.680 –0.591 –1.091 0.357
–1.128 –1.707 0.921 –0.859 –1.566 1.523 –0.900 –0.988 0.264 0.282
0.691 0.153 0.076 1.691 0.553 0.457 –1.107 0.322 0.633 0.007
1.115 0.777 –0.738 0.868 1.484 –0.792 0.950 –0.842 –0.192 0.620
–0.389 0.559 0.670 –0.315 1.234 0.475 1.117 1.286 –0.649 –1.880
0.330 0.750 –0.642 0.148 –0.608 0.866 –1.720 0.653 –0.210 –0.959
–0.333 –0.084 1.239 –0.049 –0.095 –0.197 –0.213 –1.420 –0.491 0.102
1.718 1.111 –0.548 –0.653 1.534 –0.456 –0.395 1.614 –0.531 –0.785
–0.182 0.620 1.178 –1.071 0.444 –0.072 –1.001 1.325 –0.302 –1.119
1.260 –1.192 0.182 –0.397 –0.705 –1.085 –1.492 1.642 0.673 –0.707
–1.204 –1.725 1.695 1.473 0.665 –0.489 0.020 0.267 1.230 0.865
–0.619 0.307 –0.226 –0.096 0.987 –1.195 –1.412 0.433 2.052 0.022
–0.272 –0.096 0.137 –0.361 0.653 –0.156 1.309 –0.480 –0.397 1.302
0.245 –0.690 0.493 –1.123 1.465 0.132 0.582 –0.429 0.225 0.125
0.101 –0.855 0.782 –1.040 2.113 –1.423 –1.010 0.158 0.106 –1.232
G-1
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
GLOSSARY OF TERMS
Accuracy of information Degree to which information reflects reality.
Accuracy in sampling Is affected by nonresponse bias, memory error, misunderstanding of
questions, problem of definition of terms, and processing error,
A.C. Nielsen BASES Panel-based research conducted by the marketing research/consulting firm
A.C. Nielsen using the Internet for data collection.
Adequate sample A sample that is of sufficient size to provide confidence in the stability of its
characteristics.
After-only with control group design A true experiment requiring one treatment and an
“after” measurement of both the experimental and control groups.
All-variable regression Multiple regression with all independent variables included
simultaneously.
Alpha error See Type I error.
Alternate forms reliability The extent to which measurement scores from equivalent forms of
the measure administered to the same sample are the same.
Alternative hypothesis A hypothesis making statement of expected difference.
Ambiguity of questions and responses Errors made in the transmission of information in
interpreting written or spoken words or behavior
Analysis The categorizing, the aggregating into constituent parts, and the manipulation of data
to obtain answers to the research question(s) underlying the research project.
Analysis of variance (ANOVA) A test of the differences in k group means for one or more
variables (factors).
Applied research See decisional research.
Area sample A probability sample where the primary sampling is of geographical units.
Arithmetic mean See mean.
ASP model See Online Application Service Provider.
Associative variation A measure of the extent to which occurrences of two variables, or
changes of two variables, are associated. Also known as “concomitant variation.”
Attitude measurement Measurement of the person’s mental state toward a certain stimuli
according to specified sets of dimensions and instructions.
Audimeter An electromechanical devise used to record automatically the times a household’s
television set is turned on and off and the stations which it is tuned to.
Availability of information Availability of information when a decision is being made.
Balanced scale A rating scale with an equal number of response alternatives in opposite
directions from some mid-point, e.g., positive and negative alternatives.
Balancing See matching.
G-2
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Bases for inferring causation The means by which causation can be inferred: (1) associative
variation, (2) sequence of events, and (3) absence of other possible causal factors.
Basic research See fundamental research.
Before-after with one control group design A true experimental design with one treatment
and with measurements of the experimental and control groups made both “before” and “after”
the treatment is administered.
Behaviorally-anchored rating scale (BARS) A rating scale using behavioral incidents to define
each position on the rating scale rather than verbal, graphic, or numeric labels.
Beta error See Type II error.
Bias The difference between the true value of that which is being measured and the average
value derived from a number of independent measures of it.
Bivariate analysis The analysis of relationships between two variables.
Bounded recall An approach for reducing telescoping by asking questions about events of
concern in previous time periods as well as the time period of research interest.
Bulletin Board Qualitative research online technology at a website that allows users to register
and then participate in unmoderated discussions of topics of interest to participants of the
bulletin board.
Canonical correlation analysis A generalization of multiple correlation analysis to two or more
criterion variables.
Canonical correlation coefficient In discriminant analysis it is a measure of the association that
summarizes how related the discriminant function is to the groups, and in canonical correlation
analysis it is a measure of the association between the independent variables and the set of
dependent variables.
CATI A Computer Aided Telephone Interview in which the interviewer reads the questions from
a computer and enters the responses directly.
Causal study Research design which attempts to determine the causes of what is being
predicted—i.e., the “reasons why.”
Census All members of the population are included for study.
Central tendency error Reluctance of respondents to give extreme scores or to use an extreme
position on an individual scale item.
Cheater question A question included in a questionnaire that will disclose a respondent’s
giving fabricated answers.
Chi-square test For a simple tabulation it tests the goodness-of-fit between the observed
distribution and the expected distribution of a variable, and for cross tabulation it tests whether
the observed association or relationship between the variables is statistically significant.
Classification matrix A cross-tabulation-type table showing the results of discriminant analysis’
ability to correctly classify observations into categories of a criterion variable
G-3
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Cluster analysis A class of statistical techniques whose objective is to separate objects into
groups such that the similarity of objects within each group is maximized while maximizing the
difference between groups.
Cluster sampling A probability sample in which a simple random or stratified random sample is
selected of all primary sampling units, and then all elements within the selected primary units
are sampled.
Codebook/coding manual A manual which shows how the data have been coded for analysis.
Coding The process by which responses are assigned to data categories and symbols (usually
numbers) are assigned to identify them with the categories.
Coefficient alpha A measure of internal consistency reliability for a multi-item measure which
is a type of mean reliability coefficient for all possible ways to split the items into two groups.
Coefficient of determination (R2
) A measure of the strength of association between variables in
regression, it specifies how much of the variation in the dependent variable can be explained by
the variation in the independent variable(s).
Coefficient of multiple correlation Correlation coefficient when the number of independent
variables is two or more.
Commercial data Data sold in the form of syndicated services. Collected by commercial
marketing research firms or industry associations. See syndicated services.
Comparative rating scale Objects are rated in comparison with other objects.
Completely randomized design A statistical design where experimental treatments are
assigned to test units on a random basis.
Concept An abstraction formed by generalization about particulars.
Conceptual definition A construct is defined in terms of other constructs. Also known as
constitutive definition.
Concept testing How people, without prompting, interpret deliberately a sketchy idea for a
new product or service.
Concomitant variation See associative variation.
Concurrent validity See criterion validity.
Confidence interval A range of values with a given probability of covering the true population
value (parameter).
Confounding In an experiment, it is the tangling effects of two or more levels of a treatment
variable or two or more treatment variables.
Confusion matrix See classification matrix.
Conjoint analysis A technique of research that measures psychological judgments by
decomposing a set of overall responses to a set of factorially designed stimuli so that the utility
of each stimulus attribute and attribute level can be inferred from the respondent’s overall
evaluations of the stimuli.
G-4
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Constant sum question A question in which the respondent is asked to allocate a fixed sum of
points to options to show the importance (or some other attribute or characteristic) of each
option.
Constitutive definition See conceptual definition.
Construct A concept that is the conscious invention of researchers to be used for a special
research purpose.
Construct validity A form of external validity that assesses the extent to which generalizations
can be made about higher-order constructs from research operations; it is a measurement issue
and is concerned with how and why a measurement works.
Content analysis Coding of free responses to open-end questions.
Content validity The extent to which a scale or measurement instrument represents the
universe of the property or characteristic being measured.
Continuity correction A correction factor in chi-square analysis that adjusts for the use of a
continuous distribution to estimate probability in a discrete distribution. Known as the Yates
continuity correction.
Continuous panel A sample of individuals, households, or firms from whom information is
obtained at successive time periods.
Controlled experiment A research design in which the investigator intervenes by manipulating
at least one assumed causal variable and in which subjects (respondents) are assigned randomly
to experimental and control groups.
Convergent validity A type of construct validity concerned with the correspondence in results
between measuring the same construct by two or more independent methods.
Convenience sample A nonprobability sample chosen by a convenient process and because
elements are easy to obtain.
Conversational interviewing In personal or telephone interviewing the interviewer answers
queries from respondents about question meaning and understanding.
Correlation analysis The analysis of the extent to which changes in one variable are related to
changes in one or more other variables.
Correlation coefficient A measure of the association between two or more variables.
Correspondence analysis A special case of canonical correlation analysis which examines the
relations between the categories of two discrete variables.
Coverage error Occurs when the sample frame or group from which the sample is drawn does
not represent the population as a whole. See also frame error.
Criterion validity The extent to which the measurement instrument works in predicting the
future (predictive validity) or reflects the present (concurrent validity). Also known as pragmatic
validity.
Criterion variable See dependent variable.
Critical path method (CPM) A network approach in which the component activities are
diagrammed in sequence of performance and a time estimate for each activity is presented. See
Program Evaluation and Review Technique (PERT).
G-5
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Cross-over design A statistical experimental design in which different treatments are applied to
the same test units in different time periods.
Cross sectional design A research design where several groups are measured at the same time,
with each group having been exposed to a different level of the treatment variable.
Cross tabulation The simultaneous counting of the number of observations that occur in each
of the data categories of two or more variables.
Cross validation A procedure in regression analysis and discriminant analysis for examining
whether the predictive equations derived hold up beyond the data on which parameters are
based.
Cumulative scale A scale constructed of a set of items with which the respondent indicates
agreement or disagreement, it is unidimensional, and there is a pattern of item responses that is
related to the total score.
Data collection techniques The means by which data are collected by communication or by
observation.
Data matrix A rectangular array of data entries where the rows are a respondent’s responses
and the columns are the variables or data fields for the responses. Also known as the basic data
matrix.
Decision model See problem-situation model.
Decisional research Applied research which attempts to use existing knowledge to aid in the
solution of some given problem (s).
Dependent variable The effect of interest or outcome in an experiment.
Depth interview An often unstructured interview that is used to explore the underlying
predisposition, needs, desires, feelings, and emotions of consumers toward products and
services. May consist of direct and/or indirect questions.
Descriptive study Provides information such as describing market characteristics or functions
on groups and phenomena that already exist; there is prior formulation of specific research
questions. Also known as an “observational study.”
Determinants of a research project The problem, the researcher, the respondent/subject, and
the client.
Deterministic cause. Any event that is necessary and sufficient for the subsequent occurrence
of another event.
Dichotomous question A multiple-choice question with only two alternative responses.
Differential scale A rating scale, assumed to be interval, in which a respondent is asked to
agree with only a subset of items (statements) about an object, with each item having a
predetermined scale value (position on the scale), and the items agreed with correspond to the
respondent’s position on the dimension being measured.
Direct interview An interview in which the purposes of the questions are not purposely
disguised.
Direct-judgment rating method A respondent is asked to give a numerical rating to each
stimulus with respect to some designated attribute.
G-6
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Directory-based sampling A directory or other physical listing is used as a sample frame to
select sample elements to be called in a telephone survey.
Discriminant analysis A multivariate technique of analysis that examines the ability of
predictor variables to discriminate (i.e., to separate) between categories of a criterion
(dependent) variable, or groups.
Discriminant validity A form of construct validity which assesses the extent to which a measure
is unique and not simply a reflection of other variables.
Disk-by-Mail A type of survey where the questionnaire is placed on a personal computer
diskette and mail is used to send it to a respondent and by the respondent to return it to the
research organization.
Disproportionate stratified sampling A stratified sample in which characteristics other than
just relative size, such as relative size of stratum variances, are taken into account.
Distance A characteristic of the real number system where the differences between ordered
numbers are ordered.
Double opt-in mailing list A list of people with e-mail addresses who have indicated agreement
to receive surveys.
Drop-off/pick-up survey A survey where the questionnaire is left with a respondent and a
representative of the research organization returns later to pick up the completed
questionnaire, or the questionnaire can be returned by mail.
Editing The process of reviewing the data to ensure maximum accuracy and clarity.
Electronic surveying See online research.
E-mail submission form used when the researcher builds a HTML online survey, distributes
actively to the respondent, and receives the respondent’s answers as part of an e-mail message
that is directed back to the researcher.
Emic vs. etic issue An issue in cross-national/cultural research in which a researcher must
decide whether constructs and methods are culture-specific (emic) or culture-free (etic).
E-panel Doing panel research using the Internet for data collection.
Equal-appearing intervals See differential scale.
Ethical behavior Behavior conforming to professional standards of conduct; it is what most
people in a given society view as being moral, good, or right.
Ethics Moral principles, quality, or practice.
Ethnography A qualitative approach to research that studies human behavior within a cultural
context.
Executive summary That part of a formal research report which reduces the essentials of the
study—the why, the what, the how, the conclusions, and the recommendations—to one or two
pages.
Experimental error Noncorrespondence of the true (or actual) impact of, and the impact
attributed to, the independent variable (s).
G-7
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Experimentation A research method where there is researcher intervention and control over
the factors affecting the response variable of interest, thus allowing for the establishment of
causal relationships.
Explicit model A model described verbally, graphically or diagrammatically, mathematically
(symbolically), or as a logical sequence of questions (logical flow).
Ex post facto design A quasi-experiment in which the test and control groups are not known
until after the treatment has been administered.
Exploratory study A study whose purposes include the identification of problems, more precise
formulation of problems (including identification of relevant variables), and the formulation of
new alternative courses of action.
External secondary information Secondary information that must be obtained from outside
sources.
External validity The generalizability of a relationship beyond the circumstances under which it
is observed.
Extraneous variable A variable other than the manipulated independent variable that could
influence the dependent variable.
Factor A variable or construct that is not directly observable but is developed as a linear
combination of observed variables.
Factor analysis A class of statistical techniques whose purpose often is data reduction and
summarization which is accomplished by representing a set of observed variables, persons, or
occasions in terms of a smaller number of hypothetical, underlying and unknown dimensions
which are called factors.
Factorial design A statistical experimental design where there is an equal number of
observations made of all combinations involving at least two levels of at least two variables.
False negative error A respondent reports not to have an attitude when he or she really does
have one.
False positive error Statements by respondents that appear to be complimentary, but really
are not, or when respondents appear to have an attitude and they do not.
Field experiment An experiment conducted in a natural environmental setting.
Fixed-size sampling The number of elements to be included in the sample is decided upon in
advance.
Focus group A group of topic knowledgeable people who jointly participate in an interview that
does not use a structured question-and-answer methodology. Usually consists of 8 to 12 people
selected purposively.
Formal research report Consists of a number of components which an be organized in to three
components: prefatory pages, report body, and appended parts.
Four-group, six-study design Combines an after-only with control group design and a beforeafter
with control group design.
Fractionation A rating scale in which the respondent is given two stimuli at a time and is asked
to give some numerical estimate of the ratio between them, with respect to some attribute.
G-8
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Frame error Noncorrespondence of the sought sample to the required sample. Occurs when
the sample frame is incomplete, has multiple entries for elements, or has elements included
that are not in the relevant population.
Framing effects The difference in response to objectively equivalent information depending
upon the manner in which the information is labeled or framed.
Free answer question A question that has no fixed alternatives to which the answer must
conform. Also known as open-ended text.
Frequency distribution See simple tabulation.
Full profile conjoint analysis Conjoint analysis where different stimulus (e.g., a product)
descriptions are developed and presented to the respondent for acceptability or preference
evaluations.
Fundamental research Seeks to extend the boundaries of knowledge in a given area with no
necessary immediate application to existing problems.
Funnel approach An approach to questionnaire design that specifies a sequence of questions
where one proceeds from the general to the specific or from the easier questions to answer to
those that are more difficult to answer.
Goodness-of-fit test An analysis of whether the data obtained in a research study fit or
conform to a model or distribution.
Graphic positioning scale A semantic differential used for multiple object ratings where all
objects are evaluated on each scale item.
Graphic rating scale A rating scale in which a respondent indicates his/her rating of a stimulus
on a graphical response item.
Guided imagery A modified TAT where participants are asked to appraise a product or brand
by concentrating on creating and experiencing an associated image.
Guttman scalogram analysis See cumulative scale.
History Events outside an experimental design that affect the dependent variable.
Hypothesis An assertion about the “state of nature” or the relation between things that often,
from a practical standpoint, implies a possible course of action with a prediction of the outcome
if the course of action is followed.
Implicit model A model that guides a decision but has not been specified in an explicit or
formal manner.
Inaccuracy in response Errors made in the formulation of information. May be concurrent or
predictive (e.g., when reported intentions are not carried out).
Independent variable In an experiment, it is a variable whose effect upon some other variable
the experiment is designed to measure; it is the variable that is manipulated and is also known
as the treatment variable.
Indexes of agreement Measures of the strength of association between two variables in a
cross tabulation, including the phi correlation coefficient, the contingency coefficient, the
lambda-asymmetric coefficient, and the lambda-symmetric coefficient.
G-9
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Indirect interview An interview that is neither fully structured nor unstructured, and in which
the purposes of the questions asked are intentionally disguised.
Information Recorded experience that is useful for decision making; communicated knowledge
which changes the state of knowledge of the person who receives it.
In-store interviewing A type of mall intercept in which the interviews take place in a single
store, usually at the point of purchase.
Instrument effect Changes in the measuring instrument or process that may affect the
measurement obtained in an experiment.
Intentions Presently planned actions to be taken in a specified future time period.
Interaction The situation in an experiment where the response to changes in the levels of one
treatment variable is dependent on the level of some other treatment variable(s).
Interactive interviewing Interviews that are conducted by having a respondent respond on a
personal computer. Some software may customize new questions based on responses to
previously answered questions.
Intercoder reliability The reliability of coding done by multiple persons.
Internal consistency reliability Reliability within single testing occasions in which the variables
are grouped.
Internal secondary information Secondary information that is available from within the
company or the organization.
Internal validity Assesses whether the observed effect is due solely to the experimental
treatments and not due to some extraneous variable(s).
Interpretation The process of taking the results of analysis, making inferences relevant to the
research relationships studied, and drawing managerially useful conclusions about these
relationships.
Interval scale A measurement scale that possesses the characteristics of order and distance,
and the zero point of the scale is arbitrary.
Interview A form of person-to-person (dyadic) communication between two parties that
involves the asking and answering of questions.
Interviewer A person who asks questions in an interview of a respondent.
Judgment sample A nonprobability sample where the elements to be included are selected on
the basis of the researcher’s sound judgment or expertise and an appropriate strategy.
Kolmogorov-Smirnov one-sample test A goodness-of-fit test of the agreement between an
observed distribution of a set of sample values and some specified theoretical distribution
Kolmogorov-Smirnov two-sample test A test of whether two independent samples come from
the same population or from populations with the same distribution.
Kurtosis The shape of a data distribution in terms of height or flatness.
Laboratory experiment An experiment conducted in a controlled laboratory or laboratory-type
setting.
G-10
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Laddering See means-end analysis.
Leniency error occurs when respondents consistently use the extreme positions on a rating
scale with relatively little use of intermediate scale positions.
Likert scale A balanced rating scale in which a respondent is asked to indicate extent of
agreement with a series of statements, using a set of verbal categories from “strongly agree to
“strongly disagree” for response. See also summated scale.
Limited-response category scale A rating scale in which a respondent is limited to choosing
from a predetermined set of response categories.
Logit A type of multiple regression analysis where the categorical dependent variable is
assumed to follow a logistic distribution.
Mail interview A type of survey where the questionnaire is sent to a respondent by mail and
the respondent returns the completed questionnaire by mail.
Make or buy decision The decision by a research client whether the research is to be done inhouse
(make) or by an outside supplier (buy).
Mall intercept Interviews are stationed at selected places in a shopping mall or other
centralized public place and they request interviews from people who pass by.
Management summary See executive summary.
Mann-Whitney U test A test of whether two independent groups providing data are from the
same population and whether there is a relationship between two variables.
Marketing Decision Support System (MDSS) A coordinated collection of data, systems, tools,
and techniques with supporting software and hardware by which an organization gathers and
interprets relevant information from the business and the environment and turns it into a basis
for marketing action.
Marketing information system A “formal” system within an organization for obtaining,
processing, and disseminating decision information. Subsystems are marketing research,
internal records, marketing intelligence, and information analysis.
Marketing intelligence A subsystem of a MIS in which a set of procedures and sources are used
to provide information about relevant developments in the marketing environment.
Marketing research The systematic and objective search for, and analysis of, information
relevant to the identification and solution of any problem in the field of marketing.
Matching A control technique where subjects are equated on the variable(s) to be controlled.
Also known as balancing.
Maturation Changes that occur with the passage of time in the people involved in an
experimental design.
Mean The point on a scale around which the values of a distribution balance; it is the sum of all
the values divided by the number of respondents.
Means-end analysis An in-depth one-on-one interviewing technique that identifies the linkages
people make between product attributes (means), the benefits derived from those attributes
(the consequences), and the values that underlie why the consequences are important (the
ends). Also known as “Laddering” and “Means-End Chain.”
G-11
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Measurement A way of assigning symbols to represent the properties of persons, objects,
events, or states, which symbols have the same relevant relationships to each other as do the
things represented.
Measurement error The difference between the information obtained and the information
wanted by the researcher; it is generated by the measurement process itself.
Median The midpoint of the data in a distribution.
Memory error Inaccuracy in response that occurs when a respondent gives the wrong factual
information because of not remembering an event asked about.
Method of Choices A procedure for indirectly arriving at paired comparison proportions of the
form p(B>A) by asking respondents to choose the one of a set of stimuli that has the “most of,”
is the “best,” or is “preferred,” etc. on the basis of the attribute or characteristic being studied.
Method of inquiry The broad approach to conducting a research project and the philosophy
underlying the approach. Methods include objectivist, subjectivist, Bayesian, and
phenomenologist.
Metric measurement Direct numerical judgments made by a respondent which are assumed to
be either interval- or ratio-scaled.
Metric multidimensional scaling Multidimensional scaling in which the input data are ratio-
scaled.
Mind Track A brainwave-to-computer interface developed by Advanced Neurotechnologies,
Inc. that measures direct emotional response to most any communication medium.
MIS See marketing information system.
MIS activities Discovery, collection, interpretation, analysis, and intra-company dissemination
of information.
Misunderstanding error Inaccuracy in response often due to careless question design.
Mode The typical or most frequently occurring value in a distribution.
Model The linking of propositions together in a way that provides a meaningful explanation for
a system or process.
Moderator A person conducting a focus group whose job is to direct the group’s discussion to
the topics of interest.
Monadic rating scale Each object is rated by itself independently of any other objects being
rated.
Multicollinearity A condition in multiple regression analysis where the predictor variables
show very high correlation among themselves.
Multidimensional scaling A set of techniques that portray psychological relations among
stimuli—either empirically obtained similarities or preferences (or other kinds of orderings)—as
geometric relationships among points in a multidimensional space.
Multi-item scale A scale consisting of a number of closely related individual rating scales whose
responses are combined into a single index or composite score or value. See also summated
scale.
G-12
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Multiple choice question A question that has at least two fixed alternative response categories
and the respondents can select k out of n choices.
Multiple correlation analysis Correlation analysis when the number of independent variables is
two or more.
Multiple regression analysis Regression analysis with two or more independent variables.
Multiplicity sample See snowball sample.
Multistage sampling A multilevel probability sample in which a sample is selected of larger
areas (or groups), and then a sample is selected from each of the areas (groups) selected at the
first level, and so on.
Multitrait Multimethod Matrix A generalized approach for establishing the validity and
reliability of a set of measurements (traits).
Multivariate analysis Statistical procedures that simultaneously analyze measurements of
multiple variables on each individual or object under study.
Natural experiment An experiment in which the investigator intervenes only to the extent
required for measurement, and there is no manipulation of an assumed causal variable. The
variable of interest has occurred in a natural setting, and the investigator looks at what has
happened.
Nominal scale A measurement scale that does not possess the characteristics of order,
distance, and origin.
Nomogram A graphic instrument for specifying sample size relating allowable error, confidence
level, mean or proportion, and standard deviation.
Nomological validity A form of construct validity which attempts to relate measurements to a
theoretical model that leads to further deductions, interpretations, and tests.
Nonmetric multidimensional scaling Multidimensional scaling in which input data are rank
order data (ordinally-scaled), but which output is interval-scaled.
Nonparametric statistical methods Distribution-free methods in which inferences are based on
a test statistic whose sampling distribution does not depend upon the specific distribution of the
population from which the sample is drawn.
Nonprobability sample A sample selected based on the judgment of the investigator,
convenience, or by some other means not involving the use of probabilities.
Nonresponse error Noncorrespondence of the obtained sample to the original sample.
Nonsampling error All errors other than sampling error that are associated with a research
project; typically is a systematic error but can have a random component.
Null hypothesis A hypothesis which states no difference.
Numerical comparative scale A semantic differential used for multiple object ratings where all
objects are evaluated on each scale item using a verbally-anchored numerical scale.
Numerical rating scale A rating scale that uses a series of integers that may, or may not have
verbal descriptions, to represent degrees of some property.
G-13
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Observation technique Information on respondents’ behavior is obtained by observing it
rather than by asking about it.
“One more question” syndrome The tendency to add an additional question to a survey
because the cost is very low to do so.
One-on-one interview See depth interview.
Online Application Service Provider (ASP) Accessed through the Internet, where surveys are
built online, requiring no user software, server, or IT support.
Online research Using the Internet as a mode of data collection. Often used in conjunction
with e-mail.
Operational definition Assigns meaning to a variable by specifying what is to be measured and
how it is to be measured.
Order A characteristic of the real number series in which the numbers are ordered.
Ordered-category sorting A respondent assigns (sorts) a set of stimuli into different categories,
which are ordered on the basis of some property.
Ordinal scale A measurement scale that possesses only the characteristic of order; it is a
ranking scale.
Origin A characteristic of the real number series where there is a unique origin indicated by the
number zero.
Paired comparisons The respondent is asked to choose one of a pair of stimuli on the basis of
some property of interest.
Pantry audit A data collection technique whereby a field worker takes an inventory of brands,
quantities, and package sizes that a consumer has on hand.
Parameter A summary property of a collectivity, such as a population, when that collectivity is
not considered to be a sample.
Partially structured indirect interview An interview using a predevised set of words,
statements, cartoons, pictures, or other representation to which a person is asked to respond,
and the interviewer is allowed considerable freedom in questioning the respondent to ensure a
full response.
Personal interview An interviewer asks questions of respondents in a face-to-face situation.
Pictogram A pictorial chart that depicts data with the help of symbols such as stars, stacks of
coins, trees, facial expressions, caricatures of people, and so forth.
Pilot study A small-scale test of what a survey will be, including all activities that will go into
the final survey.
Planned information Exists when a manager recognizes a need and he or she makes a request
that information be obtained.
Politz-Simmons method A method of estimating both the direction and magnitude of
nonresponse error.
G-14
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Population The totality of all the units or elements (individuals, households, organizations, etc.)
possessing one or more particular relevant features or characteristics in common, to which one
desires to generalize study results.
Population specification error Noncorrespondence of the required population to the
population selected by the researcher.
Popular report A research report that minimizes technical details and emphasizes simplicity.
Postcoding Coding done after the data are collected.
Power of a hypothesis test It is 1 minus the probability of a Type II error (1-β).
Practical significance See substantive significance.
Pragmatic validity See criterion validity.
Precision Refers to sampling error and the size of the confidence limits placed on an estimate.
Precoding Coding done before the data are collected.
Predictive validity See criterion validity.
Predictor variable See independent variable.
Pre-experimental design A research design with total absence of control.
Pretesting The testing of a questionnaire or measurement instrument before use in a survey or
experiment.
Probabilistic cause Any event that is necessary, but not sufficient, for the subsequent
occurrence of another event.
Probability sampling Every element in the population has a known nonzero probability
(chance) of being selected for inclusion in a study.
Probit A type of multiple regression analysis where the categorical dependent variable is
assumed to be normally distributed.
Problem formulation A stage in the research process in which a management problem is
translated into a research problem.
Problem-situation model A conceptual scheme that specifies a measure of the outcome(s) to
be achiever, the relevant variables, and their functional relationship to the outcomes(s).
Program Evaluation and Review technique (PERT) A probabilistic scheduling approach using
three time estimates: optimistic, most likely, and pessimistic. See also critical path method
(CPM).
Projection A research technique whereby a respondent projects his/her personality
characteristics, etc. to a non-personal, ambiguous situation that he/she is asked to describe,
expand, or build a structure around it.
Proportionate stratified sampling A stratified sample in which the sample that is drawn from
each stratum is proportionate in size to the relative size of the stratum in the population.
Proposition A statement of the relationship between variables, including the form of the
relationship.
G-15
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Protocol A record of a respondent’s verbalized thought processes while performing a decision
task or while problem solving (concurrent) or just after the task is completed (retrospective)
Psychogalvanometer A device for measuring the extent of a subject’s response to a stimulus,
such as an advertisement.
Purposive sampling See judgment sample.
Q-sort A scaling technique in which the respondent is asked to sort a number of statements or
other stimuli into a predetermined number categories, formed on the basis of some criterion,
with a specified number having to be placed in each category.
Quasi-experimental design A controlled experiment design where there is manipulation of at
least one assumed causal variable but there is not random assignment of subjects to experiment
and control groups.
Questionnaire An instrument for data collection that requests information from respondents
by asking questions.
Quota sample A nonprobability sample in which population subgroups are classified on the
basis of researcher judgment and the individual elements are selected by interviewer judgment.
Random-digit-dialing A probability sampling procedure used in telephone surveys where the
telephone number to be called is generated by selecting random digits.
Randomized response technique A technique for obtaining information about sensitive
information.
Random sampling error See sampling error.
Rank correlation The correlation between variables that are measured by ranking. Measures
used are Spearman rho and Kendall tau.
Ranking Respondents are asked to order stimuli with respect to some designated property.
Rank order question A question where the answer format requires the respondent to assign a
rank (order) position for the first, second,…, to the nth item to be ordered.
Rating A measurement method where a respondent paces that which is being rated along a
continuum or in one of an ordered set of categories.
Ratio scale A measurement scale possessing all the characteristics of the real number series:
order, distance, and origin.
Reactive effects of experimental situation Effects that may arise from subjects’ reacting to the
situation surrounding the conduct of an experiment rather than the treatment variable.
Reactive effects of testing The learning or conditioning of the persons involved in an
experimental design as a result of knowing that their behavior is being observed and/or that the
results are being measured.
Regression analysis The mathematical relationship between a dependent variable and one or
more independent variables.
Regression coefficient Represented by b, it shows the amount of change that will occur in the
dependent variable for a unit change in the independent variable it represents.
G-16
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Relevancy of information Pertinence and applicability of information to the decision.
Reliability The consistency of test results over groups of individuals or over the same individual
at different times.
Repeated measures design A research design where subjects are measured more than once on
a dependent variable. See also cross-over design.
Repertory grid A partially structured measurement technique that requires the respondent to
compare objects along dimensions that he or she selects.
Representative sample A relatively small piece of the population that mirrors the various
patterns and subclasses of the population.
Research design The specification of methods and procedures for acquiring the information
needed to structure or to solve problems. The operational design stipulates what information is
to be collected, from which sources, and by what procedures.
Research method Experimental or non-experimental; the major difference between the two
lies in the control of extraneous variables and the manipulation of at least one assumed causal
variable by the investigator in an experiment.
Research plan A formal written document that serves as the overall master guide for
conducting a research project.
Research process A series of interrelated steps that define what a research project is all about,
starting with problem formulation and ending with the research report.
Research proposal A shorter and less technical version of a research plan that is used to elicit
the project and gain a commitment of funding.
Research question States the purpose of the research, the variables of interest and the
relationships to be examined.
Research report The major vehicle by which researchers communicate by a written statement
and/or oral presentation research results, recommendations for strategic and tactical action,
and other conclusions to management in the organization or to an outside organization.
Respondent A person who participates in a research project by responding and answering
questions verbally, in writing, or by behavior.
Response bias See response error.
Response error The difference between a reported value and the true value of a variable.
Robust statistical technique A technique of analysis whereby if certain assumptions underlying
the proper use of the technique are violated, the technique performs okay and can handle such
a violation.
Sample A subset of the relevant population selected for inclusion in a research study.
Sample design A statement about a sample that specifies where the sample is to be selected,
the process of selection, and the size of the sample; it is the theoretical basis and the practical
means by which data are collected so that the characteristics of the population can be inferred
with known estimates of error.
G-17
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Sample frame A means of accounting for the elements in a population, usually a physical listing
of the elements, but may be a procedure which produces a result equivalent to a physical listing,
from which the sampled elements are selected.
Sampling distribution The probability distribution of a specified sample statistic (e.g., the
mean) for all possible random samples of a given size n drawn from the specified population.
Sampling error Variable error resulting from the chance specification of population from
elements according to the sampling plan. Often called random sampling error, it is the noncorrespondence
of the sample selected by probability means and the representative sample
sought by the researcher.
Sampling unit A population element which is actually chosen by the sampling process.
Scaling Generation of a continuum on which measured objects are located.
Scanner data Data on products purchased in retail stores that are obtained by electronic
scanning at checkout of the Universal Product Code (UPC); unit and price information are
recorded.
Scree chart In factor analysis, it is a discrete line chart that relates the amount of variance
accounted for by each factor to the factor number (1 … k).
Secondary information Information that has been collected by persons or agencies for
purposes other than the solution of the problem at hand, and which is available for the project
at hand.
Selection error The sampling error for a sample selected by a nonprobability method. It is also
a term used for the effect of the selection procedure for the test (treatment) and control groups
on the results of an experimental study.
Self-hosted server software Survey building software that requires housing on the researcher’s
server.
Semantic differential A rating procedure in which the respondent is asked to describe a
concept or object by means of ratings on a set of bipolar adjectives or phrases, with the
resulting measurements assumed to be interval-scaled.
Sentence completion test A respondent is given a sentence stem (the beginning phrase) and is
asked to complete the sentence with the first thought that occurs to him or her.
Sequential sample An approach to selecting a sample size whereby a previously determined
decision rule is used to indicate when sampling is to be stopped during the process of data
collection.
Simple random sample A probability sample where each sample element has a known and
equal probability of selection, and each possible sample of n elements has a known and equal
probability of being the sample actually selected.
Simple tabulation A count of the number of responses that occur in each of the data categories
that comprise a variable. Also known as marginal tabulation.
Simulation A set of techniques for manipulating a model of some real-world process for the
purpose of finding numerical solutions that are useful in the real process that is being modeled.
Single-source data Obtaining all data from one research supplier on product purchases and
causal factors such as media exposure, promotional influences, and consumer characteristics
from the same household.
G-18
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Skewness A measure of a given data distribution’s asymmetry.
Snowball sampling A nonprobability sample in which initial respondents are selected randomly
but additional respondents are obtained by referrals or by some other information provided by
the initial respondents.
Socioeconomic characteristics The social and economic characteristics of respondents,
including for example, income, occupation, education level, age, gender, marital status and size
of family.
Split-half reliability A measure of internal consistency reliability where the items in a multiitem
measure are divided into two equivalent groups and the item responses are correlated.
Standard deviation A measure of dispersion (variation) around the sample mean, it is the
square root of the variance.
Standard error The standard deviation of the specified sampling distribution of a statistic.
Standard error of the difference The standard deviation of the sampling distribution of the
difference between statistics such as means and proportions.
Standardized interviewing In a survey using personal or telephone interviewing the
interpretation of questions asked is left up to the respondent as the interviewer is not allowed
to answer any query.
Stapel scale An even-numbered balanced nonverbal rating scale that is used in conjunction
with single adjectives or phrases.
State of nature An environmental condition.
Static-group comparison A quasi-experimental design in which a group exposed to a treatment
is compared to a group that was not exposed.
Statistical conclusion validity Involves the specific question whether the presumed
independent and dependent variables are indeed related.
Statistical experimental design After-only designs in which there are at least two treatment
levels. Includes completely randomized, factorial, Latin-Square, randomized block, and
covariance designs.
Statistical power Ability of a sample to protect against the type II error (beta risk).
Statistical regression The tendency with repeated measures for scores to regress to the
population mean of the group.
Stepwise regression Multiple regression analysis in which the independent variable explaining
the most variance is sequentially included one at a time.
Story completion A qualitative research technique where a respondent is presented with the
beginning of a situational narrative and is asked to complete it.
Stratified sampling A probability sample where the population is broken into different strata or
subgroups based on one or more characteristics and then a simple random sample is taken from
each stratum of interest in the population.
Structured interview An interview in which a formal questionnaire has been developed and the
questions asked in a prearranged order.
G-19
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Stub-and-banner table A table that presents one dependent variable cross-tabulated by
multiple independent variables.
Substantive significance An association that is statistically significant and of sufficient strength.
Sufficiency of information Degree of completeness and/or detail of information to allow a
decision to be made.
Summated scale A rating scale constructed by adding scores from responses to a set of Likert
scales with the purpose of placing respondents along an attitude continuum of interest. See
also Likert scale and multi-item scale.
Surrogate information error Noncorrespondence of the information being sought and that
required to solve the problem.
Survey A research method in which the information sought is obtained by asking questions of
respondents.
Survey tracking and address books Online survey technology that uses imbedded codes to
facilitate the identification and tracking of survey respondents and non-respondents.
Syndicated services Information collected and tabulated on a continuing basis by research
organizations for purposes of sale to firms; data are made available to all who wish to subscribe.
See commercial data.
Systematic error See nonsampling error.
Systematic sampling A probability sample where the population elements are ordered in some
way and then after the first element is selected all others are chosen using a fixed interval.
Tabulation The process of sorting data into previously established categories, making initial
counts of responses and using summarizing measures.
Technical report A research report that emphasizes the methods used and underlying
assumptions, and presents the findings in a detailed manner.
Telephone interview Interviews that are conducted by telephone.
Telescoping A response error that occurs when a respondent reports an event happening at a
time when it did not happen. It may be forward (report it happening more recently than it did)
or backward (reporting it happening earlier than it did).
Testing effect The effect of a first measurement on the scores of a second measurement.
Test of independence A test of the significance of observed association involving twop or more
variables.
Test-retest reliability The stability of response over time.
Thematic Apperception Test (TAT) A test consisting of one or more pictures or cartoons that
depict an ambiguous situation relating to the subject being studied, and research subjects are
asked to make up a story about what is happening, or the subject is asked to assume the role of
a person in the situation and then describe what is happening and who the others in the scene
are.
G-20
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Third-person technique A projective qualitative research method in which a respondent is
indirectly interviewed by asking for his or her view of what a neighbor or some other person
would respond to the interview.
Thurstone Case V Scaling Based on the Thurstone’s Law of Comparative Judgment, this
method allows the construction of a unidimensional interval scale using responses from ordinal
measurement methods, such as paired comparisons.
Time series design Data are obtained from the same sample (or population) at successive
points in time.
Total study error Sampling error plus non-sampling error.
Treatment variable See independent variable.
Trend design Data are obtained from statistically matched samples drawn from the same
population over time.
True experiment See controlled experiment.
t – test A test of the difference in mans of two groups of respondents that focuses on sample
means and variances.
Type I error The probability that one will incorrectly reject Ho, the null hypothesis of no
difference, or any hypothesis.
Type II error The probability that one will incorrectly accept a null hypothesis, or any
hypothesis.
Unlimited-response category scale A direct-judgment rating scale where the respondent is
free to choose his/her own number or insert a tick mark along some line to represent his/her
judgment about he magnitude of the stimulus relative to some reference points.
Unobtrusive measures Nonreactive measures of behavior, past and present.
Unsolicited information Information which may, in fact, exist within and be obtainable within
the company, but which potential users do not know is available unless they happen to chance
upon it.
Unstructured interview An interview in which there is no formal questionnaire and the
questions may not be asked in a prearranged order.
Useful information Information which is accurate, current, sufficient, available, and relevant.
Validity of measurement The extent to which one measures what he or she believes is being
measured.
VALS A syndicated segmentation scheme known as Values and Lifestyle segmentation which
combines demographic, attitudinal, and psychographic data, according to pre-defined segments,
Variance A measure of dispersion, it is the mean of the squared deviation of individual
measurements from the arithmetic mean of the distribution.
Variation in measurement Differences in individual scores within a set of measurements that
may be due to the characteristic or property being measured (the true difference) and/or the
measurement process itself.
G-21
Scott M. Smith and Gerald S. Albaum, An Introduction to Marketing Research, © 2010
Verbal measures Include spoken and written responses, including responses provided
interactively with a personal computer.
Verbal rating scale A rating scale using a series of verbal options for rating an object.
Warranty form of interview A type of mail interview where the questions asked are included
on the warranty card to be returned to the manufacturer.
Weighting data Procedures used to adjust the final sample so that the specific respondent
subgroups of the sample are found in identical proportions to those found in the poulation.
Wilcoxon rank sum (T) test A test of the relationship between two sets of measurements from
dependent samples in which the data are collected in matched pairs.
Wilks’ lambda In discriminant analysis it is a multivariate measure of group differences over
discriminating variables.
Word association test A series of stimulus words are presented to a respondent who is asked
to answer with the first word that comes to mind after hearing each stimulus word.