8
Sampling
Chapter outline
Introduction to survey research 184
Introduction to sampling 186
Sampling error 188
Types of probability sample 190
Simple random sample 190
Systematic sample 191
Stratified random sampling 192
Multi-stage cluster sampling 193
The qualities of a probability sample 195
Sample size 197
Absolute and relative sample size 197
Time and cost 198
Non-response 199
Heterogeneity of the population 200
Kind of analysis 201
Types of non-probability sampling 201
Convenience sampling 201
Snowball sampling 202
Quota sampling 203
Limits to generalization 205
Error in survey research 205
Key points 206
Questions for review 206
Sampling184
Introduction to survey research
Once the research questions have been formulated, the
planning of the fieldwork can begin. In practice, decisions
relating to sampling and the research instrument will
overlap, but they are presented in Figure 8.1 as part of a
sequence. The figure is meant to illustrate the main phases
of a survey, and these different steps (other than those to
do with sampling, which will be covered in this chapter)
will be followed through in Chapters 9–11 and 15–16.
The survey researcher needs to decide what kind of
population is suited to the investigation of the topic and
also needs to formulate a research instrument and how
it should be administered. By ‘research instrument’ is
meant simply something like a structured interview
schedule or a self-completion questionnaire. Moreover,
there are several different ways of administering such
instruments. Figure 8.2 outlines the main types that are
likely to be encountered. Types 1 through 4 are covered
in Chapter 9. Types 5 and 6 are covered in Chapter 10.
Types 7 through 9 are covered in Chapter 28 in the context
of the use of the Internet generally.
Chapter guide
This chapter and the three that follow it are very much concerned with principles and practices
associated with social survey research. Sampling principles are not exclusively concerned with survey
research; for example, they are relevant to the selection of documents for content analysis (see
Chapter 13). However, in this chapter the emphasis will be on sampling in connection with the selection
of people who would be asked questions by interview or questionnaire. The chapter explores:
• the role of sampling in relation to the overall process of doing survey research;
• the related ideas of generalization (also known as external validity) and of a representative sample;
the latter allows the researcher to generalize findings from a sample to a population;
• the idea of a probability sample—that is, one in which a random selection process has been employed;
• the main types of probability sample: the simple random sample; the systematic sample; the stratified
random sample; and the multi-stage cluster sample;
• the main issues involved in deciding on sample size;
• different types of non-probability sample, including quota sampling, which is widely used in market
research and opinion polls;
• potential sources of error in survey research.
This chapter is concerned with some important aspects
of conducting a survey, but it presents only a partial picture,
because there are many other steps. In this chapter
we are concerned with the issues involved in selecting
individuals for survey research, although the principles
involved apply equally to other approaches to quantitative
research, such as content analysis. Chapters 9, 10,
and 11 deal with the data-collection aspects of conducting
a survey, while Chapters 15 and 16 deal with issues to
do with the analysis of data.
Figure 8.1 aims to outline the main steps involved in
doing survey research. Initially, the survey will begin
with general research issues that need to be investigated.
These are gradually narrowed down so that they become
research questions, which may take the form of hypotheses,
but this need not necessarily be the case. The
movement from research issues to research questions is
likely to be the result of reading the literature relating to
the issues, such as relevant theories and evidence (see
Chapters 1 and 4).
Sampling 185
Figure 8.1Figure 8.1
Steps in conducting a social survey
Issue(s) to be researched
Review literature/theories relating to topic/areg a
Formulate research question(s)
Consider whether a social survey is appropriate (if not, consider an alternative research design)
Consider what kind of population will be appropriate
Consider what kind of sample design will be employed
Explore whether there is a sampling frame that can be emplog yed
Decide on sample size
Decide on mode of administration (face-to-face; telephone; postal; email; Web)
Develop questions (and devise answer alternatives for closed questions)
Review questions and assess face validity
Pilot questions
Revise questions
Finalize questionnaire/schedule
Sample from population
Administer questionnaire/schedule to sample
Follow up non-respondents at least once
Transform completed questionnaires/schedules into computer readable data (coding)
Enter data into statistical analysis program like SPSS
Analyse data
Interpret findings
Consider implications of findings for research questions
Sampling186
Introduction to sampling
resources to conduct a survey of all these students. It is
unlikely that you would be able to send questionnaires
to all 9,000 and even more unlikely that you would be
able to interview all of them, since conducting survey research
by interview is considerably more expensive and
time consuming, all things being equal, than by postal
questionnaire (see Chapter 10). It is almost certain that
you will need to sample students from the total population
of students in your university.
The need to sample is one that is almost invariably
encountered in quantitative research. In this chapter I
will be almost entirely concerned with matters relating
to sampling in relation to social survey research involving
data collection by structured interview or questionnaire.
Other methods of quantitative research involve sampling
considerations, as will be seen in Chapters 12 and 13,
when we will examine structured observation and content
analysis respectively. The principles of sampling
involved are more or less identical in connection with
Many of the readers of this book will be university or
college students. At some point in your stay at your
university (I will use this term from now on to include
colleges) you may have wondered about the attitudes of
your fellow students to various matters, or about their
behaviour in certain areas, or something about their
backgrounds. If you were to decide to examine any or all
of these three areas, you might consider conducting
structured interviews or sending out questionnaires in
order to find out about their behaviour, attitudes, and
backgrounds. You will, of course, have to consider how
best to design your interviews or questionnaires, and the
issues that are involved in the decisions that need to be
made about designing these research instruments and
administering them will be the focus of Chapters 9–11.
However, before getting to that point you are likely to
be confronted with a problem. Let us say that your university
is quite large and has around 9,000 students.
It is extremely unlikely that you will have the time and
gu e 8.Figure 8.2
Main modes of administration of a survey
Survey
Structured interview Self-completion
questionnaire
Face-to-face Telephone Supervised
5
Postal
6
Internet
CAPI
2
CATI
4
Email Web
9
Embedded
7
Attached
8
Notes: CAPI is computer-assisted personal interviewing; CATI is computer-assisted telephone interviewing.
Sampling 187
these other methods, but frequently other considerations
come to the fore as well.
But will any old sample suffice? Would it be sufficient
to locate yourself in a central position on your campus
(if it has one) and then interview the students who
come past you and whom you are in a position to interview?
Alternatively, would it be sufficient to go around
your student union asking people to be interviewed?
Or again to send questionnaires to everyone on your
course?
The answer, of course, depends on whether you want
to be able to generalize your findings to the entire student
body in your university. If you do, it is unlikely that any of
the three sampling strategies proposed in the previous
paragraph would provide you with a representative
sample of all students in your university. In order to be
able to generalize your findings from your sample to the
population from which it was selected, the sample must
be representative. See Key concept 8.1 for an explanation
of key terms concerning sampling.
Key concept 8.1
Basic terms and concepts in sampling
• Population: basically, the universe of units from which the sample is to be selected. The term ‘units’ is
employed because it is not necessarily people who are being sampled—the researcher may want to sample
from a universe of nations, cities, regions, firms, etc. Finch and Hayes (1994), for example, based part of their
research upon a random sample of wills. Their population, therefore, was a population of wills. Thus,
‘population’ has a much broader meaning than the everyday use of the term, whereby it tends to be
associated with a nation’s entire population.
• Sample: the segment of the population that is selected for investigation. It is a subset of the population.
The method of selection may be based on a probability or a non-probability approach (see below).
• Sampling frame: the listing of all units in the population from which the sample will be selected.
• Representative sample: a sample that reflects the population accurately so that it is a microcosm of the population.
• Sampling bias: a distortion in the representativeness of the sample that arises when some members of the
population (or more precisely the sampling frame) stand little or no chance of being selected for inclusion in
the sample.
• Probability sample: a sample that has been selected using random selection so that each unit in the
population has a known chance of being selected. It is generally assumed that a representative sample is
more likely to be the outcome when this method of selection from the population is employed. The aim of
probability sampling is to keep sampling error (see below) to a minimum.
• Non-probability sample: a sample that has not been selected using a random selection method. Essentially,
this implies that some units in the population are more likely to be selected than others.
• Sampling error: error in the findings deriving from research due to the difference between a sample and the
population from which it is selected. This may occur even though probability sampling has been employed.
• Non-sampling error: error in the findings deriving from research due to the differences between the
population and the sample that arise either from deficiencies in the sampling approach, such as an
inadequate sampling frame or non-response (see below), or from such problems as poor question wording,
poor interviewing, or flawed processing of data.
• Non-response: a source of non-sampling error that is particularly likely to happen when individuals are being
sampled. It occurs whenever some members of the sample refuse to cooperate, cannot be contacted, or for
some reason cannot supply the required data (for example, because of mental incapacity).
• Census: the enumeration of an entire population. Thus, if data are collected in relation to all units in a
population, rather than in relation to a sample of units of that population, the data are treated as census data.
The phrase ‘the census’ typically refers to the complete enumeration of all members of the population of a
nation state—that is, a national census. This form of enumeration currently occurs once every ten years in the
UK, although there is some uncertainty at the time of writing about whether another census will take place.
However, in a statistical context, like the term population, the idea of a census has a broader meaning than this.
Sampling188
Why might the strategies for sampling students previously
outlined be unlikely to produce a representative
sample? There are various reasons, of which the following
stand out.
• The first two approaches depend heavily upon the
availability of students during the time or times that
you search them out. Not all students are likely to be
equally available at that time, so the sample will not
reflect these students.
• They also depend on the students going to the locations.
Not all students will necessarily pass the point
where you locate yourself or go to the student union,
or they may vary hugely in the frequency with which
they do so. Their movements are likely to reflect such
things as where their halls of residence or accommodation
are situated, or where their departments are
located, or their social habits. Again, to rely on these
locations would mean missing out on students who do
not frequent them.
• It is possible, not to say likely, that your decisions
about which people to approach will be influenced by
your judgements about how friendly or cooperative
the people concerned are likely to be or by how comfortable
you feel about interviewing students of the
same (or opposite) gender to yourself, as well as by
many other factors.
• The problem with the third strategy is that students
on your course by definition take the same subject as
each other and therefore will not be representative of
all students in the university.
In other words, in the case of all of the three sampling
approaches, your decisions about whom to sample are
influenced too much by personal judgements, by prospective
respondents’ availability, or by your implicit
criteria for inclusion. Such limitations mean that, in the
language of survey sampling, your sample will be biased.
A biased sample is one that does not represent the population
from which the sample was selected. Sampling
bias will occur if some members of the population
In order to appreciate the significance of sampling error
for achieving a representative sample, consider Figures
8.3–8.7. Imagine we have a population of 200 people
have little or no chance of being selected for inclusion in
the sample. As far as possible, bias should be removed
from the selection of your sample. In fact, it is incredibly
difficult to remove bias altogether and to derive a truly
representative sample. What needs to be done is to
ensure that steps are taken to keep bias to an absolute
minimum.
Three sources of sampling bias can be identified (see
Key concept 8.1 for an explanation of key terms).
1. If a non-probability or non-random sampling method is
used. If the method used to select the sample is not
random, there is a possibility that human judgement
will affect the selection process, making some members
of the population more likely to be selected than
others. This source of bias can be eliminated through
the use of probability/random sampling, the procedure
for which is described below.
2. If the sampling frame is inadequate. If the sampling
frame is not comprehensive or is inaccurate or suffers
from some other kind of similar deficiency, the sample
that is derived cannot represent the population, even
if a random/probability sampling method is employed.
3. If some sample members refuse to participate or cannot
be contacted—in other words, if there is non-response.
The problem with non-response is that those who
agree to participate may differ in various ways from
those who do not agree to participate. Some of the
differences may be significant to the research question
or questions. If the data are available, it may be
possible to check how far, when there is non-response,
the resulting sample differs from the population. It is
often possible to do this in terms of characteristics
such as gender or age, or, in the case of something like
a sample of university students, whether the sample’s
characteristics reflect the entire sample in terms of
faculty membership. However, it is usually impossible
to determine whether differences exist between the
population and the sample after non-response in
terms of ‘deeper’ factors, such as attitudes or patterns
of behaviour.
and we want a sample of 50. Imagine as well that one of
the variables of concern to us is whether people watch
soap operas and that the population is equally divided
Sampling error
Sampling 189
between those who do and those who do not. This split
is represented by the vertical line that divides the population
into two halves (Figure 8.3). If the sample is
representative we would expect our sample of 50 to be
equally split in terms of this variable (Figure 8.4). If there
is a small amount of sampling error, so that we have one
person too many who does not watch soap operas and
one too few who does, it will look like Figure 8.5. In
Figure 8.6 we see a rather more serious degree of overrepresentation
of people who do not watch soaps. This
Figure 8.3Figure 8.3
Watching soap operas in a population of 200
Watch soaps Do not watch soaps
Figure 8.4Figure 8.4
A sample with no sampling error
Watch soaps Do not watch soaps
gu e 8.5Figure 8.5
A sample with very little sampling error
Watch soaps Do not watch soaps
gu e 8.6Figure 8.6
A sample with some sampling error
Watch soaps Do not watch soaps
Sampling190
time there are three too many who do not watch them
and three too few who do. In Figure 8.7 we have a very
serious over-representation of people who do not watch
soaps, because there are 35 people in the sample who do
not watch them, which is much larger than the 25 who
should be in the sample.
It is important to appreciate that, as suggested above,
probability sampling does not and cannot eliminate sampling
error. Even with a well-crafted probability sample,
a degree of sampling error is likely to creep in. However,
probability sampling stands a better chance than nonprobability
sampling of keeping sampling error in check
so that it does not end up looking like the outcome
featured in Figure 8.7. Moreover, probability sampling
allows the researcher to employ tests of statistical significance
that permit inferences to be made about the
sample from which the sample was selected. These will
be addressed in Chapter 15.
gu e 8.Figure 8.7
A sample with a lot of sampling error
Watch soaps Do not watch soaps
Imagine that we are interested in levels of alcohol consumption
among university students and the variables
that relate to variation in levels of drinking. We might
decide to conduct our research in a single nearby university.
This means that our population will be all students
in that university, which will in turn mean that we will
be able to generalize our findings only to students of
that university. We simply cannot assume that levels of
alcohol consumption and their correlates will be the same
in other universities. We might decide that we want
our research to be conducted only on full-time students,
so that part-time students are omitted. Imagine too that
there are 9,000 full-time students in the university.
Simple random sample
The simple random sample is the most basic form of
probability sample. With random sampling, each unit of
the population has an equal probability of inclusion in
the sample. Imagine that we decide that we have enough
money to interview 450 students at the university. This
means that the probability of inclusion in the sample is
450
9,000
, i.e. 1 in 20
This is known as the sampling fraction and is expressed
as
n
N
where n is the sample size and N is the population size.
The key steps in devising our simple random sample
can be represented as follows.
1. Define the population. We have decided that this will
be all full-time students at the university. This is our N
and in this case is 9,000.
2. Select or devise a comprehensive sampling frame. It
is likely that the university will have an office that
keeps records of all students and that this will enable
us to exclude those who do not meet our criteria for
inclusion—i.e. part-time students.
3. Decide your sample size (n). We have decided that
this will be 450.
Types of probability sample
Sampling 191
4. List all the students in the population and assign them
consecutive numbers from 1 to N. In our case, this will
be 1 to 9,000.
5. Using a table of random numbers, or a computer
program that can generate random numbers, select
n (450) different random numbers that lie between 1
and N (9,000).
6. The students to which the n (450) random numbers
refer constitute the sample.
Two points are striking about this process. First, there
is almost no opportunity for human bias to manifest
itself. Students would not be selected on such subjective
criteria as whether they looked friendly and approachable.
The selection of whom to interview is entirely
mechanical. Second, the process is not dependent on
the students’ availability. They do not have to be walking
in the interviewer’s proximity to be included in the
sample. The process of selection is done without their
knowledge. It is not until they are contacted by an interviewer
that they know that they are part of a social
survey.
Step 5 mentions the possible use of a table of random
numbers. These can be found in the appendices of many
statistics books. The tables are made up of columns of
five-digit numbers, such as:
09188
90045
73189
75768
54016
08358
28306
53840
91757
89415
The first thing to notice is that, since these are five-digit
numbers and the maximum number that we can sample
from is 9,000, which is a four-digit number, none of the
random numbers seems appropriate, except for 09188
and 08358, although the former is larger than the largest
possible number. The answer is that we should take just
four digits in each number. Let us take the last four digits.
This would yield the following:
9188
0045
3189
5768
4016
8358
8306
3840
1757
9415
However, two of the resulting numbers—9188 and 9415
—exceed 9,000. We cannot have a student with either of
these numbers assigned to him or her. The solution is
simple: we ignore these numbers. This means that the
student who has been assigned the number 45 will be the
first to be included in the sample; the student who has
been assigned the number 3189 will be next; the student
who has been assigned the number 5768 will be next;
and so on.
However, this somewhat tortuous procedure may be
replaced in some circumstances by using a systematic
sampling procedure (see next section) and more generally
can be replaced by enlisting the computer for
assistance (see Tips and skills ‘Generating random
numbers’).
Systematic sample
A variation on the simple random sample is the systematic
sample. With this kind of sample, you select units
directly from the sampling frame—that is, without resorting
to a table of random numbers.
We know that we are to select 1 student in 20. With
a systematic sample, we would make a random start
between 1 and 20 inclusive, possibly by using the last
two digits in a table of random numbers. If we did this
with the ten random numbers above, the first relevant
one would be 54016, since it is the first one where the last
two digits yield a number of 20 or below, in this case 16.
This means that the sixteenth student on our sampling
frame is the first to be in our sample. Thereafter, we
take every twentieth student on the list. So the sequence
will go:
16, 36, 56, 76, 96, 116, etc.
Sampling192
This approach obviates the need to assign numbers to
students’ names and then to look up names of the students
whose numbers have been drawn by the random
selection process. It is important to ensure, however, that
there is no inherent ordering of the sampling frame,
since this may bias the resulting sample. If there is some
ordering to the list, the best solution is to rearrange it.
Stratified random sampling
In our imaginary study of university students, one of the
features that we might want our sample to exhibit is a
proportional representation of the different faculties to
which students are attached. It might be that the kind
of discipline a student is studying is viewed as relevant
to a wide range of attitudinal features that are relevant
to the study of drinking. Generating a simple random
sample or a systematic sample might yield such a representation,
so that the proportion of humanities students
in the sample is the same as that in the student population
and so on. Thus, if there are 1,800 students in the
humanities faculty, using our sampling fraction of 1 in
20, we would expect to have 90 students in our sample
from this faculty. However, because of sampling error,
it is unlikely that this will occur and that there will be
a difference, so that there may be, say, 85 or 93 from
this faculty.
Because it is almost certain that the university will
include in its records the faculty in which students are
based, or indeed may have separate sampling frames for
each faculty, it will be possible to ensure that students are
accurately represented in terms of their faculty membership.
In the language of sampling, this means stratifying
the population by a criterion (in this case, faculty membership)
and selecting either a simple random sample or
a systematic sample from each of the resulting strata. In
the present example, if there are five faculties we would
have five strata, with the numbers in each stratum being
one-twentieth of the total for each faculty, as in Table 8.1,
which also shows a hypothetical outcome of using a
simple random sample, which results in a distribution of
students across faculties that does not mirror the population
all that well.
The advantage of stratified random sampling in a case
like this is clear: it ensures that the resulting sample will
be distributed in the same way as the population in terms
of the stratifying criterion. If you use a simple random or
systematic sampling approach, you may end up with a
distribution like that of the stratified sample, but it is
unlikely. Two points are relevant here. First, you can conduct
stratified sampling sensibly only when it is relatively
easy to identify and allocate units to strata. If it is not
possible or it would be very difficult to do so, stratified
sampling will not be feasible. Second, you can use more
than one stratifying criterion. Thus, it may be that you
would want to stratify by both faculty and gender or
Tips and skills
Generating random numbers
The method for generating random numbers described in the text is what might be thought of as the classic
approach. However, a far neater and quicker way is to generate random numbers on the computer. For example,
the following website provides an online random generator which is very easy to use:
www.psychicscience.org/random.aspx (accessed 9 August 2010).
If we want to select 450 cases from a population of 9,000, specify 450 after Generate, the digit 1 after random
integers between and then 9000 after and. You will also need to specify from a drop-down menu ‘with no repeats’.
This means that no random number will be selected more than once. Then simply click on GO and the 450 random
numbers will appear in a box below OUTPUT. You can then copy and paste the random numbers into a document.
Table 8.1
The advantages of stratified sampling
Faculty Population Stratified
sample
Hypothetical
simple random
or systematic
sample
Humanities 1,800 90 85
Social sciences 1,200 60 70
Pure sciences 2,000 100 120
Applied sciences 1,800 90 84
Engineering 2,200 110 91
TOTAL 9,000 450 450
Sampling 193
faculty and whether students are undergraduates or
postgraduates. If it is feasible to identify students in
terms of these stratifying criteria, it is possible to use
pairs of criteria or several criteria (such as faculty membership
plus gender plus undergraduate/postgraduate).
Stratified sampling is really feasible only when the
relevant information is available. In other words, when
Multi-stage cluster sampling
In the example we have been dealing with, students to
be interviewed are located in a single university. Interviewers
will have to arrange their interviews with the
sampled students, but, because they are all close together
(even in a split-site university), they will not be involved
in a lot of travel. However, imagine that we wanted a
national sample of students. It is likely that interviewers
would have to travel the length and breadth of the UK to
interview the sampled students. This would add a great
deal to the time and cost of doing the research. This kind
of problem occurs whenever the aim is to interview a
sample that is to be drawn from a widely dispersed population,
such as a national population, or a large region, or
even a large city.
One way in which it is possible to deal with this potential
problem is to employ cluster sampling. With cluster
data are available that allow the ready identification of
members of the population in terms of the stratifying criterion
(or criteria), it is sensible to employ this sampling
method. But it is unlikely to be economical if the identification
of population members for stratification purposes
entails a great deal of work because there is no available
listing in terms of strata.
sampling, the primary sampling unit (the first stage of
the sampling procedure) is not the units of the population
to be sampled but groupings of those units. It is the
latter groupings or aggregations of population units that
are known as clusters. Imagine that we want a nationally
representative sample of 5,000 students. Using simple
random or systematic sampling would yield a widely
dispersed sample, which would result in a great deal of
travel for interviewers. One solution might be to sample
universities and then students from each of the sampled
universities. A probability sampling method would need
to be employed at each stage. Thus, we might randomly
sample ten universities from the entire population of universities,
thus yielding ten clusters, and we would then
interview 500 randomly selected students at each of the
ten universities.
Now imagine that the result of sampling ten universities
gives the following list:
Student experience
Probability sampling for a student project
Joe Thompson describes the sampling procedure that he and the other members of his team used for their study
of students living in halls of residence at the University of East Anglia as a stratified random sample. The following
description suggests that they employed a systematic sampling approach for finding students within halls.
Stratified random sampling was used to decide which halls of residence each member of the research team
would go to and obtain questionnaire responses. This sampling method was the obvious choice as it meant
there could be no fixing/bias to which halls the interviewee would go to and also maintained the
representative nature of the research.
The stratified random sampling method known as the ‘random walk process’ was used when conducting the
interviews. Each member of the research group was assigned a number between 4 and 8 as a sampling fraction
gap: I was assigned the number 7 and ‘Coleman house block 1’ as my accommodation block. This meant that,
when conducting my interviews, I would go to Coleman house and knock on the 7th door, and then the
14th door, adding 7 each time, until I had completed five interviews. If I encountered a lack of response from
the 6th door, I would return to the first flat but add one each time to avoid periodicity. This sampling method
was determined by the principles of standardization, reliability, and validity.
To read more about Joe’s research experiences, go to the Online Resource Centre that accompanies this book at:
www.oxfordtextbooks.co.uk/orc/brymansrm4e/
Sampling194
• Glasgow Caledonian
• Edinburgh
• Teesside
• Sheffield
• University College Swansea
• Leeds Metropolitan
• University of Ulster
• University College London
• Southampton
• Loughborough
This list is fine, but interviewers could still be involved
in a great deal of travel, since the ten universities are
quite a long way from each other. North American and
In a sense, cluster sampling is always a multi-stage
approach, because one always samples clusters first, and
then something else—either further clusters or population
units—is sampled.
Many examples of multi-stage cluster sampling entail
stratification. We might, for example, want to stratify
universities in terms of whether they are ‘old’ or ‘new’
universities—that is, those that received their charters
after the 1991 White Paper for Higher Education, Higher
Australian readers who examine this last comment by
looking at a map of the United Kingdom may view the
universities as in fact very close to each other!
One solution is likely to be to group all UK universities
by standard region (see Research in focus 8.1 for an example
of this kind of approach) and randomly to sample
two standard regions. Five universities might then be
sampled from each of the two lists of universities and
then 500 students from each of the ten universities.
Thus, there are separate stages:
• group UK universities by standard region and sample
two regions;
• sample five universities from each of the two regions;
• sample 500 students from each of the ten universities.
Education: A New Framework. In each of the two regions,
we would group universities along the old/new university
criterion and then select two or three universities
from each of the two strata per region.
Research in focus 8.1 provides an example of a
multi-stage cluster sample. It entailed three stages: the
sampling of parliamentary constituencies, the sampling
of polling districts, and the sampling of individuals.
In a way, there are four stages, because addresses are
Research in focus 8.1
An example of a multi-stage cluster sample
For their study of social class in modern Britain, Marshall et al. (1988: 288) designed a sample ‘to achieve 2,000
interviews with a random selection of men aged 16–64 and women aged 16–59 who were not in full-time
education’.
• Sampling parliamentary constituencies
— Parliamentary constituencies were ordered by standard region (there are eleven).
— Constituencies were allocated to one of three population density bands within standard regions.
— These subgroups were then reordered by political party voted to represent the constituency at the
previous general election.
— These subgroups were then listed in ascending order of percentage in owner–occupation.
— 100 parliamentary constituencies were then sampled.
— Thus, parliamentary constituencies were stratified in terms of four variables: standard region; population
density; political party voted for in last election; and percentage of owner–occupation.
• Sampling polling districts
— Two polling districts were chosen from each sampled constituency.
• Sampling individuals
— Nineteen addresses from each sampled polling district were systematically sampled.
— One person at each address was chosen according to a number of pre-defined rules.
Sampling 195
sampled from polling districts and then individuals
are sampled from each address. However, Marshall et al.
(1988) present their sampling strategy as involving just
three stages. Parliamentary constituencies were stratified
by four criteria: standard region, population density,
voting behaviour, and owner–occupation.
The advantage of multi-stage cluster sampling should
be clear by now: it allows interviewers to be far more
geographically concentrated than would be the case
if a simple random or stratified sample were selected.
The advantages of stratification can be capitalized upon
because the clusters can be stratified in terms of strata.
However, even when a very rigorous sampling strategy
is employed, sampling error cannot be avoided, as the
example in Research in focus 8.2 shows.
Research in focus 8.2
The 1992 British Crime Survey
The British Crime Survey (BCS) is a regular survey, funded by the Home Office, of a national sample drawn from
the populations of England and Wales. The survey was conducted on eight occasions between 1982 and 2000
and has been conducted annually since 2001. In each instance, over 10,000 people have been interviewed. The
main object of the survey is to glean information on respondents’ experiences of being victims of crime. There is
also a self-report component in which a selection of the sample are interviewed on their attitudes to crime and
to report on crimes they have committed. Before 1992, the BCS used the electoral register as a sampling frame.
Relying on a register of the electorate as a sampling frame is not without problems in spite of appearing robust:
it omits any persons who are not registered, a problem that was exacerbated by the Community Charge (poll
tax), which resulted in a significant amount of non-registration, as some people sought to avoid detection in
order not to have to pay the tax. In 1992 the Postcode Address File was employed as a sampling frame and has
been used since then. Its main advantage over the electoral register as a sampling frame is that it is updated
more frequently. It is not perfect, because the homeless will not be accessible through it. The BCS sample itself
is a stratified multi-stage cluster sample. The sampling procedure produced 13,117 residential addresses. Like
most surveys, there was some non-response, with 23.3 per cent of the 13,117 addresses not resulting in a ‘valid’
interview. Just under half of these cases were the result of an outright refusal. In spite of the fact that the BCS is
a rigorously selected and very large sample, an examination of the 1992 survey by Elliott and Ellingworth (1997)
shows that there is some sampling error. By comparing the distribution of survey respondents with the 1991
census, they show that certain social groups are somewhat under-represented, most notably: owner–occupiers,
households in which no car is owned, and male unemployed. However, Elliott and Ellingworth show that, as the
level of property crime in postcode address sectors increases, the response rate (see Key concept 8.2) decreases.
In other words, people who live in high-crime areas tend to be less likely to agree to be interviewed. How far
this tendency affects the BCS data is difficult to determine, but the significance of this brief example is that,
even when a sample of this quality is selected, the existence of sampling and non-sampling error cannot be
discounted. The potential for a larger spread of errors when levels of sampling rigour fall short of a sample like
that selected for the BCS is, therefore, considerable.
The reason why probability sampling is such an important
procedure in social survey research is that it is
possible to make inferences from information about a
random sample to the population from which it was
selected. In other words, we can generalize findings
derived from a sample to the population. This is not to
say that we treat the population data and the sample data
as the same. If we take the example of the level of alcohol
consumption in our sample of 450 students, which we
will treat as the number of units of alcohol consumed in
The qualities of a probability sample
Sampling196
the previous seven days, we will know that the mean
number of units consumed by the sample (X) can be used
to estimate the population mean (m) but with known
margins of error. The mean, or more properly the arithmetic
mean, is the simple average.
In order to address this point it is necessary to use
some basic statistical ideas. These are presented in Tips
and skills ‘Generalizing from a random sample to the
population’ and can be skipped if just a broad idea of
sampling procedures is required.
Tips and skills
Generalizing from a random sample to
the population
Let us say that the sample mean is 9.7 units of alcohol consumed (the average amount of alcohol consumed in
the previous seven days in the sample). A crucial consideration here is: how confident can we be that the mean
level of alcohol consumption of 9.7 units is likely to be found in the population, even when probability sampling
has been employed? If we take an infinite number of samples from a population, the sample estimates of the
mean of the variable under consideration will vary in relation to the population mean. This variation will take the
form of a bell-shaped curve known as a normal distribution (see Figure 8.8). The shape of the distribution implies
that there is a clustering of sample means at or around the population mean. Half the sample means will be at
or below the population mean; the other half will be at or above the population mean. As we move to the left
(at or lower than the population mean) or the right (at or higher than the population mean), the curve tails off,
implying fewer and fewer samples generating means that depart considerably from the population mean. The
variation of sample means around the population mean is the sampling error and is measured using a statistic
known as the standard error of the mean. This is an estimate of the amount that a sample mean is likely to
differ from the population mean.
This consideration is important because sampling theory tells us that 68 per cent of all sample means will lie
between + or − 1 standard error from the population mean and that 95 per cent of all sample means will lie
between + or − 1.96 standard errors from the population mean. It is this second calculation that is crucial,
because it is at least implicitly employed by survey researchers when they report their statistical findings.
The distribution of sample means
1.96
SE
1.96
SE
Population
mean
Numberofsamples
Value of the mean
Notes: 95 per cent of sample means will lie within the shaded area. SE = standard error of the mean.
Figure 8.8
Sampling 197
As someone who is known as a teacher of research
methods and a writer of books in this area, I often get
asked questions about methodological issues. One question
that is asked almost more than any other relates
to the size of the sample—‘how large should my sample
be?’ or ‘is my sample large enough?’ The decision about
sample size is not a straightforward one: it depends on a
number of considerations, and there is no one definitive
answer. This is frequently a source of great disappointment
to those who pose such questions. Moreover, most
of the time decisions about sample size are affected by
considerations of time and cost. Therefore, invariably
decisions about sample size represent a compromise
between the constraints of time and cost, the need for
precision, and a variety of further considerations that
will now be addressed.
Absolute and relative sample size
One of the most basic considerations, and one that is possibly
the most surprising, is that, contrary to what you
might have expected, it is the absolute size of a sample
that is important not its relative size. This means that a
national probability sample of 1,000 individuals in the
UK has as much validity as a national probability sample
of 1,000 individuals in the USA, even though the latter
has a much larger population. It also means that increasing
the size of a sample increases the precision of a sample.
They typically employ 1.96 standard errors as the crucial criterion in how confident they can be in their findings.
Essentially, the criterion implies that you can be 95 per cent certain that the population mean lies within + or
− 1.96 sampling errors from the sample mean.
If a sample has been selected according to probability sampling principles, we know that we can be 95 per cent
certain that the population mean will lie between the sample mean + or − 1.96 multiplied by the standard error
of the mean. This is known as the confidence interval. If the mean level of alcohol consumption in the previous
seven days in our sample of 450 students is 9.7 units and the standard error of the mean is 1.3, we can be 95 per
cent certain that the population mean will lie between
9.7 + (1.96 × 1.3)
and
9.7 − (1.96 × 1.3)
i.e. between 12.248 and 7.152.
If the standard error was smaller, the range of possible values of the population mean would be narrower; if the
standard error was larger, the range of possible values of the population mean would be wider.
If a stratified sample is selected, the standard error of the mean will be smaller because the variation between
strata is essentially eliminated because the population will be accurately represented in the sample in terms of
the stratification criterion or criteria employed. This consideration demonstrates the way in which stratification
injects an extra increment of precision into the probability sampling process, since a possible source of sampling
error is eliminated.
By contrast, a cluster sample without stratification exhibits a larger standard error of the mean than a comparable
simple random sample. This occurs because a possible source of variability between students (i.e. membership
of one university rather than another, which may affect levels of alcohol consumption) is disregarded. If, for
example, some universities had a culture of heavy drinking in which a large number of students participated, and
if these universities were not selected because of the procedure for selecting clusters, an important source of
variability would have been omitted. It also implies that the sample mean would be on the low side, but that is
another matter.
Sample size
Sampling198
This means that the 95 per cent confidence interval
referred to in Tips and skills ‘Generalizing from a random
sample to the population’ narrows. However, a large
sample cannot guarantee precision, so that it is probably
better to say that increasing the size of a sample increases
the likely precision of a sample. This means that, as sample
size increases, sampling error decreases. Therefore,
an important component of any decision about sample
size should be how much sampling error one is prepared
to tolerate. The less sampling error one is prepared to
tolerate, the larger a sample will need to be. Fowler (1993)
warns against a simple acceptance of this criterion. He
argues that in practice researchers do not base their
Time and cost
Time and cost considerations become very relevant in
this context. In the previous paragraph it is clearly being
suggested that the larger the sample size the greater the
precision (because the amount of sampling error will
be less). However, by and large, up to a sample size of
around 1,000, the gains in precision are noticeable as the
sample size climbs from low figures of 50, 100, 150, and
so on upwards. After a certain point, often in the region
of 1,000, the sharp increases in precision become less
pronounced, and, although it does not plateau, there is a
decisions about sample size on a single estimate of a
variable. Most survey research is concerned to generate
a host of estimates—that is, of the variables that make
up the research instrument that is administered. He also
observes that it is not normal for survey researchers to be
in a position to specify in advance ‘a desired level of precision’
(Fowler 1993: 34). Moreover, since sampling error
will be only one component of any error entailed in an
estimate, the notion of using a desired level of precision
as a factor in a decision about sample size is not realistic.
Instead, to the extent that this notion does enter into
decisions about sample size, it usually does so in a general
rather than in a calculated way.
slowing-down in the extent to which precision increases
(and hence the extent to which the sample error of
the mean declines). Considerations of sampling size are
likely to be profoundly affected by matters of time and
cost at such a juncture, since striving for smaller and
smaller increments of precision becomes an increasingly
uneconomic proposition. As Hazelrigg (2004: 85) succinctly
puts it: ‘The larger the size of the sample drawn
from a population the more likely (X) converges to m;
but the convergence occurs at a decelerating rate (which
means that very large samples are decreasingly cost
efficient).’
Tips and skills
Sample size and probability sampling
As I have said in the text, the issue of sample size is the matter that most often concerns students and others.
Basically, this is an area where size really does matter—the bigger the sample, the more representative it is likely
to be (provided the sample is randomly selected), regardless of the size of the population from which it is drawn.
However, when doing projects, students clearly need to do their research with very limited resources. You should
try to find out from your department whether there are any guidelines about whether samples of a minimum size
are expected. If there are no such guidelines, you will need to conduct your mini-survey in such a way as to
maximize the number of interviews you can manage or the number of postal questionnaires you can send out,
given the amount of time and resources available to you. Also, in many if not most cases, a truly random
approach to sample selection may not be open to you. The crucial point is to be clear about and to justify what
you have done. Explain the difficulties that you would have encountered in generating a random sample. Explain
why you really could not include any more in your sample of respondents. But, above all, do not make claims
about your sample that are not sustainable. Do not claim that it is representative or that you have a random
sample when it is clearly not the case that either of these is true. In other words, be frank about what you have
done. People will be much more inclined to accept an awareness of the limits of your sample design than claims
about a sample that are patently false. Also, it may be that there are lots of good features about your
sample—the range of people included, the good response rate, the high level of cooperation you received from
the firm. Make sure you play up these positive features at the same time as being honest about its limitations.
Sampling 199
Non-response
However, considerations about sampling error do not
end here. The problem of non-response should be borne
in mind. Most sample surveys attract a certain amount of
non-response. Thus, it is likely that only some members
of our sample will agree to participate in the research.
If it is our aim to ensure as far as possible that 450 students
are interviewed and if we think that there may be
a 20 per cent rate of non-response, it may be advisable
to sample 540–50 individuals, on the grounds that
approximately 90 will be non-respondents.
The issue of non-response, and in particular of refusal
to participate, is of particular significance, because it has
been suggested by some researchers that response rates
to social surveys (see Key concept 8.2) are declining
in many countries. This implies that there is a growing
tendency towards people refusing to participate in
social survey research. In 1973 an article in the American
magazine Business Week carried an article ominously
entitled ‘The Public Clams up on Survey Takers’. The
magazine asked survey companies about their experiences
and found considerable concern about declining
response rates. Similarly, in Britain, a report from a
working party on the Market Research Society’s Research
and Development Committee in 1975 pointed to similar
A further interesting issue in connection with nonresponse
is that of how far researchers should go in order
to boost their response rates. In Chapter 10, a number
of steps that can be taken to improve response rates to
postal questionnaires, which are particularly prone to
concerns among market research companies. However,
an analysis of this issue by T. W. Smith (1995) suggests
that, contrary to popular belief, there is no consistent evidence
of such a decline. Moreover, Smith shows that it is
difficult to disentangle general trends in response rates
from such variables as the subject matter of the research,
the type of respondent, and the level of effort expended
on improving the number of respondents to individual
surveys. However, an overview of non-response trends
in the USA based on non-response rates for various
continuous surveys suggests that there is a decline in the
preparedness of households to participate in surveys
(Groves et al. 2004). Further evidence comes from a
study by Baruch (1999) of questionnaire-based articles
published in 1975, 1985, and 1995 in five academic
journals in the area of management studies. This article
found an average (mean) response rate of 55.6 per cent,
though with quite a large amount of variation around
this average. The average response rate over the three
years was 64.4 per cent in 1975, 55.7 per cent in 1985,
and 48.4/52.2 per cent in 1995. Two percentages were
provided for 1995 because the larger figure includes
a journal that publishes a lot of research based on top
managers, who tend to produce a poorer response rate.
Response rates were found that were as low as 10 per
cent and 15 per cent.
poor response rates, are discussed. However, boosting
response rates to interview-based surveys can prove
expensive. Teitler et al. (2003) present a discussion of
the steps taken to boost the response rate of a US sample
that was hard to reach—namely, both parents of newly
Key concept 8.2
What is a response rate?
The notion of a response rate is a common one in social survey research. When a social survey is conducted,
whether by structured interview or by self-completion questionnaire, it is invariably the case that some people
who are in the sample refuse to participate (referred to as non-response). The response rate is, therefore, the
percentage of a sample that does, in fact, agree to participate. However, the calculation of a response rate is a
little more complicated than this. First, not everyone who replies will be included: if a large number of questions
are not answered by a respondent or if there are clear indications that he or she has not taken the interview or
questionnaire seriously, it is better to employ only the number of usable interviews or questionnaires as the
numerator. Similarly, it also tends to occur that not everyone in a sample turns out to be a suitable or appropriate
respondent or can be contacted. Thus the response rate is calculated as follows:
number of usable questionnaires
total sample – unsuitable or uncontactable members of the sample
× 100
Sampling200
born children, where most of the parents were not
married. They found that, although there was evidence
that increasing the response rate from an initial 68 per
cent to 80 per cent meant that the final sample resembled
more closely the population from which the sample had
been taken, diminishing returns undoubtedly set in. In
other words, the improvements in the characteristics
of the sample necessitated a disproportionate outlay of
resources. However, this is not to say that steps should
not be taken to improve response rates. For example,
following up respondents who do not initially respond to
a postal questionnaire invariably results in an improved
response rate at little additional cost. A study based on
a survey of New Zealand residents by Brennan and
Charbonneau (2009) provides unequivocal evidence of
the improvement in response rate that can be achieved
by at least two follow-up mailings to respondents to
postal questionnaire surveys, which tend to achieve
lower response rates than comparable interview-based
surveys. A chocolate sent with the questionnaire helps
too apparently!
As the previously mentioned study of response rates
by Baruch (1999) suggests, there is wide variation in the
response rates that social scientists achieve when they
conduct surveys. It is difficult to arrive at clear indications
of what is expected from a response rate. Baruch’s
study focused on research in business organizations,
Heterogeneity of the population
Yet another consideration is the homogeneity and
heterogeneity of the population from which the sample
is to be taken. When a population is very heterogeneous,
like a whole country or city, a larger sample will be
and, as he notes, when top managers are the focus of
a survey, the response rate tends to be noticeably lower.
In the survey component of the Cultural Capital and
Social Exclusion (CCSE) project referred to in Research
in focus 2.9, the initial main sample constituted a 53 per
cent response rate (Bennett et al. 2009). The researchers
decided to supplement the initial sample in various
ways, one of which was to have an ethnic boost sample,
in large part because the main sample did not include
sufficient numbers of ethnic-minority members. However,
the response rate from the ethnic boost sample was
substantially below that achieved for the main sample.
The researchers write: ‘In general, ethnic boosts tend to
have lower response rates than cross-sectional surveys’
(Thomson 2004: 10). There is a sense, then, that what
might be anticipated to be a reasonable response rate
varies according to the type of sample and the topics
covered by the interview or questionnaire. While it is
obviously desirable to do one’s best to maximize a response
rate, it is also important to be open about the
limitations of a low response rate in terms of the likelihood
that findings will be biased. In the future, it seems
likely that, given that there are likely to be limits on the
degree to which a survey researcher can boost a response
rate, more and more effort will go into refining ways of
estimating and correcting for anticipated biases in findings
(Groves 2006).
needed to reflect the varied population. When it is relatively
homogeneous, such as a population of students
or of members of an occupation, the amount of variation
is less and therefore the sample can be smaller. The
implication of this is that, the greater the heterogeneity
of a population, the larger a sample will need to be.
Research in focus 8.3
The problem of non-response
In December 2006 an article in The Times reported that a study of the weight of British children had been
hindered because many families declined to participate. The study was commissioned by the Department of
Health and found that, for example, among those aged 10 or 11, 14 per cent were overweight and 17 per cent
were obese. However, The Times writer notes that a report compiled by the Department of Health on the
research suggests that such figures are ‘likely systematically to underestimate the prevalence of overweight and
obesity’ (quoted in Hawkes 2006: 24). The reason for this bias in the statistics is that parents were able to refuse
to let their children participate, and those whose children were heavier were more likely to do so. As a result,
the sample was biased towards those who were less heavy. The authors of the report drew the inference about
sampling bias because they noted that more children were recorded as obese in areas where there was a poorer
response rate.
Sampling 201
Kind of analysis
Finally, researchers should bear in mind the kind of analysis
they intend to undertake. A case in point here is
the contingency table. A contingency table shows the
relationship between two variables in tabular form. It
shows how variation in one variable relates to variation
in another variable. To understand this point, consider
the basic structure of a table in the study by Marshall
et al. (1988) of social class in Britain. This research was
referred to in Research in focus 8.1. The table is based on
the 589 cohabiting couples (1,178 people) of the sample
in which both partners are employed in paid work. The
authors aim to show in the table how far couples are of
thesameoradifferentsocialclassintermsofGoldthorpe’s
seven-category scheme for classifying social class. The
result is a table in which, because each variable comprises
7 categories, there are 49 cells in the table (i.e. 7 × 7). In
order for there to be an adequate number of cases in each
cell, a fairly large sample was required. Imagine that
Marshall et al. had conducted a survey on a much smaller
sample in which they ended up with just 150 couples. If
the same kind of analysis as Marshall et al. carried out
was conducted, it would be found that these 150 couples
would be very dispersed across the 49 cells of the table. It
is likely that many of the cells would be empty or would
have very small numbers in them, which would make it
difficult to make inferences about what the table showed.
In fact, quite a lot of the cells in the actual table in Marshall
et al. have very small numbers in them (8 cells contain 1 or
0). This problem would have been even more pronounced
if they had ended up with a much smaller sample of
couples. Consequently, considerations of sample size
should be sensitive to the kinds of analysis that will be
subsequently required, such as the issue of the number of
cells in a table. In a case such as this, a larger sample will
be necessitated by the nature of the analysis to be conducted
as well as the nature of the variables in question.
The term ‘non-probability sampling’ is essentially an
umbrella term to capture all forms of sampling that are
not conducted according to the canons of probability
sampling outlined above. It is not surprising, therefore,
that the term covers a wide range of different types
of sampling strategy, at least one of which—the quota
sample—is claimed by some practitioners to be almost
as good as a probability sample. In this section we will
cover three main types of non-probability sample: the
convenience sample; the snowball sample; and the
quota sample.
Convenience sampling
A convenience sample is one that is simply available to
the researcher by virtue of its accessibility. Imagine that
a researcher who teaches education at a university is
interested in the kinds of features that teachers look for
in their headmasters. The researcher might administer a
questionnaire to several classes of students, all of whom
are teachers taking a part-time master’s degree in education.
The chances are that the researcher will receive
all or almost all of the questionnaires back, so that there
will be a good response rate. The findings may prove
quite interesting, but the problem with such a sampling
strategy is that it is impossible to generalize the findings,
because we do not know of what population this sample
is representative. They are simply a group of teachers
who are available to the researcher. They are almost
certainly not representative of teachers as a whole—the
very fact they are taking this degree programme marks
them off as different from teachers in general.
This is not to suggest that convenience samples should
never be used. Let us say that our lecturer/researcher
is developing a battery of questions that are designed
to measure the leadership preferences of teachers. It is
highly desirable to pilot such a research instrument
before using it in an investigation, and administering it
to a group that is not a part of the main study may be a
legitimate way of carrying out some preliminary analysis
of such issues as whether respondents tend to answer
in identical ways to a question, or whether one question
is often omitted when teachers respond to it. In other
words, for this kind of purpose, a convenience sample
may be acceptable though not ideal. A second kind of
context in which it may be at least fairly acceptable to use
a convenience sample is when the chance presents itself
to gather data from a convenience sample and it represents
too good an opportunity to miss. The data will not
allow definitive findings to be generated, because of the
Types of non-probability sampling
Sampling202
problem of generalization, but they could provide a
springboard for further research or allow links to be
forged with existing findings in an area.
It also perhaps ought to be recognized that convenience
sampling probably plays a more prominent role than is
sometimes supposed. Certainly, in the field of organization
studies it has been noted that convenience samples
are very common and indeed are more prominent
Snowball sampling
In certain respects, snowball sampling is a form of convenience
sample, but it is worth distinguishing because it
has attracted quite a lot of attention over the years. With
this approach to sampling, the researcher makes initial
contact with a small group of people who are relevant
to the research topic and then uses these to establish
than are samples based on probability sampling (Bryman
1989a: 113–14). Social research is also frequently based
on convenience sampling. Research in focus 8.4 contains
an example of the use of convenience samples in social
research. Probability sampling involves a lot of preparation,
so that it is frequently avoided because of the
difficulty and costs involved.
contacts with others. I used an approach like this to
create a sample of British visitors to Disney theme parks
(Bryman 1999).
Research in focus 8.5 describes the generation of
a snowball sample of marijuana-users for what is often
regarded as a classic study of drug use. Becker’s comment
on this method of creating a snowball sample
is interesting: ‘The sample is, of course, in no sense
Research in focus 8.4
A convenience sample
Miller et al. (1998) were interested in theories concerning the role of shopping in relation to the construction of
identity in modern society. Since many discussions of this issue have been concerned with shopping centres
(malls), they undertook a study that combined quantitative and qualitative research methods in order to explore
the views of shoppers at two London shopping centres: Brent Cross and Wood Green. One phase of the research
entailed structured interviews with shoppers leaving the centres. The interviews were conducted mainly during
weekdays in June and July 1994. Shoppers were chiefly questioned as they left the main exits, though some
questioning at minor exits also took place. The authors tell us: ‘We did not attempt to secure a quota [see below]
or random sample but asked every person who passed by, and who did not obviously look in the other direction
or change their path, to complete a questionnaire’ (Miller et al. 1998: 55). Such a sampling strategy produces a
convenience sample because only people who are visiting the centre and who are therefore self-selected by
virtue of their happening to choose to shop at these times can be interviewed.
Research in focus 8.5
A snowball sample: Becker’s study of
marijuana-users
In an article first published in 1953, Becker (1963) reports on how he generated a sample of marijuana-users.
He writes:
I conducted fifty interviews with marijuana users. I had been a professional dance musician for some years
when I conducted this study and my first interviews were with people I had met in the music business.
I asked them to put me in contact with other users who would be willing to discuss their experiences with
me. . . . Although in the end half of the fifty interviews were conducted with musicians, the other half covered
a wide range of people, including laborers, machinists, and people in the professions. (Becker 1963: 45–6)
Sampling 203
“random”; it would not be possible to draw a random
sample, since no one knows the nature of the universe
from which it would have to be drawn’ (Becker 1963:
46). What Becker is essentially saying here (and the same
point applies to my study of Disney theme park visitors)
is that there is no accessible sampling frame for the population
from which the sample is to be taken and that the
difficulty of creating such a sampling frame means that
a snowball sampling approach is the only feasible one.
Moreover, even if one could create a sampling frame of
marijuana-users or of British visitors to Disney theme
parks, it would almost certainly be inaccurate straight
away, because this is a shifting population. People will
constantly be becoming and ceasing to be marijuanausers,
while new theme park visitors are arriving all the
time.
The problem with snowball sampling is that it is very
unlikely that the sample will be representative of the
population, though, as I have just suggested, the very
notion of a population may be problematic in some
circumstances. However, by and large, snowball sampling
is used not within a quantitative research strategy, but
within a qualitative one: both Becker’s and my study
were carried out within a qualitative research framework.
Concerns about external validity and the ability
to generalize do not loom as large within a qualitative research
strategy as they do in a quantitative research one
(see Chapter 17). In qualitative research, the orientation
to sampling is more likely to be guided by a preference
for theoretical sampling than with the kind of statistical
sampling that has been the focus of this chapter (see Key
concept 18.3). There is a much better ‘fit’ between snowball
sampling and the theoretical sampling strategy of
qualitative research than with the statistical sampling
approach of quantitative research. This is not to suggest
that snowball sampling is entirely irrelevant to
quantitative research: when the researcher needs to
focus upon or to reflect relationships between people,
tracing connections through snowball sampling may be a
better approach than conventional probability sampling
(Coleman 1958).
Quota sampling
Quota sampling is comparatively rarely employed in
academic social research, but is used intensively in commercial
research, such as market research and political
opinion polling. The aim of quota sampling is to produce
a sample that reflects a population in terms of the relative
proportions of people in different categories, such as
gender, ethnicity, age groups, socio-economic groups,
and region of residence, and in combinations of these
categories. However, unlike a stratified sample, the sampling
of individuals is not carried out randomly, since
the final selection of people is left to the interviewer.
Information about the stratification of the UK population
or about certain regions can be obtained from sources
like the census and from surveys based on probability
samples such as the General Household Survey, British
Social Attitudes, and the British Household Panel Survey.
Once the categories and the number of people to be
interviewed within each category (known as quotas)
have been decided upon, it is then the job of interviewers
to select people who fit these categories. The quotas will
typically be interrelated. In a manner similar to stratified
sampling, the population may be divided into strata in
terms of, for example, gender, social class, age, and ethnicity.
Census data might be used to identify the number of
people who should be in each subgroup. The numbers to
be interviewed in each subgroup will reflect the population.
Each interviewer will probably seek out individuals
who fit several subgroup quotas. Accordingly, an interviewer
may know that among the various subgroups
of people he or she must find, and interview, five Asian,
25–34-year-old, lower-middle-class females in the area
in which the interviewer has been asked to work (say, the
Wirral). The interviewer usually asks people who are
available to him or her about their characteristics (though
gender will presumably be self-evident) in order to determine
their suitability for a particular subgroup. Once a
subgroup quota (or a combination of subgroup quotas)
has been achieved, the interviewer will no longer be concerned
to locate individuals for that subgroup.
The choice of respondents is left to the interviewer,
subject to the requirement of all quotas being filled, usually
within a certain time period. Those of you who have
ever been approached on the street by a person toting
a clipboard and interview schedule and have been asked
about your age, occupation, and so on, before being
asked a series of questions about a product or whatever,
have almost certainly encountered an interviewer with a
quota sample to fill. Sometimes, he or she will decide not
to interview you because you do not meet the criteria
required to fill a quota. This may be due to a quota
already having been filled or to the criteria for exclusion
meaning that a person with a certain characteristic you
possess is not required.
A number of criticisms are frequently levelled at quota
samples.
• Because the choice of respondent is left to the
interviewer, the proponents of probability sampling
Sampling204
argue that a quota sample cannot be representative.
It may accurately reflect the population in terms of
superficial characteristics, as defined by the quotas.
However, in their choice of people to approach, interviewers
may be unduly influenced by their perceptions
of how friendly people are or by whether the
people make eye contact with the interviewer (unlike
most of us, who look at the ground and shuffle past
as quickly as possible because we do not want to be
bothered in our leisure time).
• People who are in an interviewer’s vicinity at the times
he or she conducts interviews, and are therefore available
to be approached, may not be typical. There is a
risk, for example, that people in full-time paid work
may be under-represented and that those who are
included in the sample are not typical.
• The interviewer is likely to make judgements about
certain characteristics in deciding whether to approach
a person, in particular, judgements about age.
Those judgements will sometimes be incorrect—for
example, when someone who is eligible to be interviewed,
because a quota that he or she fits is unfilled,
is not approached because the interviewer makes an
incorrect judgement (for example, that the person is
older than he or she looks). In such a case, a possible
element of bias is being introduced.
• It has also been argued that the widespread use of
social class as a quota control can introduce difficulties,
because of the problem of ensuring that interviewees
are properly assigned to class groupings
(Moser and Kalton 1971).
• It is not permissible to calculate a standard error of the
mean from a quota sample, because the non-random
method of selection makes it impossible to calculate
the range of possible values of a population.
All this makes the quota sample look a poor bet, and
there is no doubt that it is not favoured by academic
social researchers. It does have some arguments in its
favour, however.
• It is undoubtedly cheaper and quicker than an interview
survey on a comparable probability sample. For
example, interviewers do not have to spend a lot of
time travelling between interviews.
• Interviewers do not have to keep calling back on
people who were not available at the time they were
first approached.
• Because calling back is not required, a quota sample
is easier to manage. It is not necessary to keep track
of people who need to be recontacted or to keep track
of refusals. Refusals occur, of course, but it is not
necessary (and indeed it is not possible) to keep a
record of which respondents declined to participate.
• When speed is of the essence, a quota sample is
invaluable when compared to the more cumbersome
probability sample. Newspapers frequently need to
know how a national sample of voters feel about a
certain topic or how they intend to vote at that time.
Alternatively, if there is a sudden major news event,
such as a terrorist incident like the London bombs of
July 2005, the news media may seek a more or less
instant picture of the nation’s views about personal
security or people’s responses more generally. Again,
a quota sample will be much faster.
• As with convenience sampling, it is useful for conducting
development work on new measures or on research
instruments. It can also be usefully employed
in relation to exploratory work from which new
theoretical ideas might be generated.
• Although the standard error of the mean should not
be computed for a quota sample, it frequently is. As
Moser and Kalton (1971) observe, some writers argue
that the use of a non-random method in quota sampling
should not act as a barrier to such a computation
because its significance as a source of error is small
when compared to other errors that may arise in
surveys (see Figure 8.9). However, they go on to argue
that at least with random sampling the researcher can
calculate the amount of sampling error and does not
have to be concerned about its potential impact.
There is some evidence to suggest that, when compared
to random samples, quota samples often result in biases.
They under-represent people in lower social strata,
people who work in the private sector and manufacturing,
and people at the extremes of income, and they
over-represent women in households with children and
people from larger households. On the other hand, it has
to be acknowledged that probability samples are often
biased too—for example, it is often suggested that they
under-represent men and those in employment (Marsh
and Scarbrough 1990; Butcher 1994).
Sampling 205
Limits to generalization
Error in survey research
One point that is often not fully appreciated is that, even
when a sample has been selected using probability sampling,
any findings can be generalized only to the population
from which that sample was taken. This is an obvious
point, but it is easy to think that findings from a study
have some kind of broader applicability. If we take our
imaginary study of alcohol consumption among students
at a university, any findings could be generalized only
to that university. In other words, you should be very
cautious about generalizing to students at other universities.
There are many factors that may imply that the
level of alcohol consumption is higher (or lower) than
among university students as a whole. There may be a
higher (or lower) concentration of pubs in the university’s
vicinity, there may be more (or fewer) bars on the
campus, there may be more (or less) of a culture of drinking
at this university, or the university may recruit a
higher (or lower) proportion of students with disposable
income. There may be many other factors too. Similarly,
we should be cautious of overgeneralizing in terms
of locality. Lunt and Livingstone’s (1992: 173) study of
consumption habits was based on a postal questionnaire
sent to ‘241 people living in or around Oxford during
September 1989’. While the authors’ findings represent a
fascinating insight into modern consumption patterns,
we should be cautious about assuming that they can
be generalized beyond the confines of Oxford and its
environs.
We can think of ‘error’, a term that has been employed on
a number of occasions, as being made up of four main
factors (Figure 8.9).
1. Sampling error. See Key concept 8.1 for a definition.
This kind of error arises because it is extremely
unlikely that one will end up with a truly representative
sample, even when probability sampling is
employed.
2. We can distinguish what might be thought of as
sampling-related error. This is error that is subsumed
under the category non-sampling error (see Key conThere
could even be a further limit to generalization
that is implied by the Lunt and Livingstone (1992)
sample. They write that the research was conducted in
September 1989. One issue that is rarely discussed in this
context and that is almost impossible to assess is whether
there is a time limit on the findings that are generated.
Quite aside from the fact that we need to appreciate that
the findings cannot (or at least should not) be generalized
beyond the Oxford area, is there a point at which we
have to say, ‘well, those findings applied to the Oxford
area then but things have changed and we can no longer
assume that they apply to that or any other locality’? We
are, after all, used to thinking that things have changed
when there has been some kind of prominent change.
To take a simple example: no one would be prepared to
assume that the findings of a study in 1980 of university
students’ budgeting and personal finance habits would
apply to students in the early twenty-first century. Quite
aside from changes that might have occurred naturally,
the erosion and virtual dismantling of the student grant
systemhaschangedthewaysstudentsfinancetheireducation,
including perhaps a greater reliance on part-time
work (Lucas 1997), a greater reliance on parents, and the
use of loans. But, even when there is no definable or recognizable
source of relevant change of this kind, there is
none the less the possibility (or even likelihood) that
findings are temporally specific. Such an issue is impossible
to resolve without further research (Bryman 1989b).
cept 8.1) but that arises from activities or events that
are related to the sampling process and that are connected
with the issue of generalizability or external
validity of findings. Examples are an inaccurate
sampling frame and non-response.
3. There is also error that is connected with the
implementation of the research process. We might
call this data-collection error. This source of error
includessuchfactorsas:poorquestionwordinginselfcompletion
questionnaires or structured interviews;
poor interviewing techniques; and flaws in the
administration of research instruments.
Sampling206
4. Finally, there is data-processing error. This arises from
faulty management of data, in particular, errors in the
coding of answers.
The third and fourth sources of error relate to factors
that are not associated with sampling and instead relate
much more closely to concerns about the validity of measurement,
which was addressed in Chapter 7. However,
the kinds of steps that need to be taken to keep these
sources of error to a minimum in the context of social
survey research will be addressed in Chapters 9–11.
Key points
● Probability sampling is a mechanism for reducing bias in the selection of samples.
● Ensure you become familiar with key technical terms in the literature on sampling such as:
representative sample; random sample; non-response; population; sampling error; etc.
● Randomly selected samples are important because they permit generalizations to the population and
because they have certain known qualities.
● Sampling error decreases as sample size increases.
● Quota samples can provide reasonable alternatives to random samples, but they suffer from some
deficiencies.
● Convenience samples may provide interesting data, but it is crucial to be aware of their limitations in
terms of generalizability.
● Sampling and sampling-related error are just two sources of error in social survey research.
Questions for review
● What do each of the following terms mean: population; probability sampling; non-probability
sampling; sampling frame; representative sample; and sampling and non-sampling error?
● What are the goals of sampling?
● What are the main areas of potential bias in sampling?
gu e 8.9Figure 8.9
Four sources of error in social survey research
Data-processing
error
Sampling
error
Sampling-related
error
Data-collection
error
Error
Sampling 207
Sampling error
● What is the significance of sampling error for achieving a representative sample?
Types of probability sample
● What is probability sampling and why is it important?
● What are the main types of probability sample?
● How far does a stratified random sample offer greater precision than a simple random or systematic
sample?
● If you were conducting an interview survey of around 500 people in Manchester, what type of
probability sample would you choose and why?
● A researcher positions herself on a street corner and asks 1 person in 5 who walks by to be
interviewed. She continues doing this until she has a sample of 250. How likely is she to achieve a
representative sample?
The qualities of a probability sample
● A researcher is interested in levels of job satisfaction among manual workers in a firm that is
undergoing change. The firm has 1,200 manual workers. The researcher selects a simple random
sample of 10 per cent of the population. He measures job satisfaction on a Likert scale comprising
ten items. A high level of satisfaction is scored 5 and a low level is scored 1. The mean job satisfaction
score is 34.3. The standard error of the mean is 8.58. What is the 95 per cent confidence interval?
Sample size
● What factors would you take into account in deciding how large your sample should be when
devising a probability sample?
● What is non-response and why is it important to the question of whether you will end up with a
representative sample?
Types of non-probability sample
● Are non-probability samples useless?
● In what circumstances might you employ snowball sampling?
● ‘Quota samples are not true random samples, but in terms of generating a representative sample
there is little difference between them, and this accounts for their widespread use in market research
and opinion polling.’ Discuss.
Limits to generalization
● ‘The problem of generalization to a population is not just to do with the matter of getting a
representative sample.’ Discuss.
Error in survey research
● ‘Non-sampling error, as its name implies, is concerned with sources of error that are not part of the
sampling process.’ Discuss.
Online Resource Centre
www.oxfordtextbooks.co.uk/orc/brymansrm4e/
Visit the Online Resource Centre that accompanies this book to enrich your understanding of
sampling. Consult web links, test yourself using multiple choice questions, and gain further guidance
and inspiration from the Student Researcher’s Toolkit.
9
Structured
interviewing
Chapter outline
Introduction 209
The structured interview 209
Reducing error due to interviewer variability 210
Accuracy and ease of data processing 211
Other types of interview 212
Interview contexts 213
More than one interviewee 213
More than one interviewer 214
In person or by telephone? 214
Computer-assisted interviewing 216
Conducting interviews 217
Know the schedule 217
Introducing the research 217
Rapport 218
Asking questions 219
Recording answers 219
Clear instructions 219
Question order 220
Probing 223
Prompting 224
Leaving the interview 225
Training and supervision 225
Problems with structured interviewing 227
Characteristics of interviewers 227
Response sets 227
The problem of meaning 228
The feminist critique 228
Key points 229
Questions for review 230
Problems with structured interviewing 227
Characteristics of interviewers 227
ResResponponsese setsets 222727
TheThe prprproblobloblemem ofof meameaninning 22ggg 2282828
The femininnnistististi ccrcriititique 2222222288
KeKeKeKeKeyKeKeyKeyKe popointntsss 22922922922
QuQueQueQueQueQuueue tstististististioonsons fofor rrr revieviewew 230230
Structured interviewing 209
Introduction
In the social research interview, the aim is for the
interviewer to elicit from the interviewee or respondent, as
he or she is frequently called in survey research, all manner
of information: interviewees’ own behaviour or that
of others; attitudes; norms; beliefs; and values. There are
many different types or styles of research interview, but
the kind that is primarily employed in survey research is
the structured interview, which is the focus of this chapter.
Other kinds of interview will be briefly mentioned in
this chapter but will be discussed in greater detail in later
chapters.
emphasized in this chapter. The structured interview is
one of the two main ways of administering a survey
research instrument, and its main forms are briefly outlined
in Figure 8.2. This figure should be consulted as a
background to this chapter and Chapter 10.
Chapter guide
The structured interview is one of a variety of forms of research interview, but it is the one that is most
commonly employed in survey research. The goal of the structured interview is for the interviewing of
respondents to be standardized so that differences between interviews in any research project are
minimized. As a result, there are many guidelines about how structured interviewing should be carried
out so that variation in the conduct of interviews is small. The chapter explores:
• the reasons why the structured interview is a prominent research method in survey research; this issue
entails a consideration of the importance of standardization to the process of measurement;
• the different contexts of interviewing, such as the use of more than one interviewer and whether the
administration of the interview is in person or by telephone;
• various prerequisites of structured interviewing, including: establishing rapport with the interviewee;
asking questions as they appear on the interview schedule; recording exactly what is said by interviewees;
ensuring there are clear instructions on the interview schedule concerning question sequencing and
the recording of answers; and keeping to the question order as it appears on the schedule;
• problems with structured interviewing, including: the influence of the interviewer on respondents and
the possibility of systematic bias in answers (known as response sets); the feminist critique of
structured interview, which raises a distinctive cluster of problems with the method, is also examined.
The interview is a common occurrence in social life,
because there are many different forms of interview. There
are job interviews, media interviews, social work interviews,
police interviews, appraisal interviews. And then
there are research interviews, which represent the kind
of interview that will be covered in this and other chapters
(such as Chapters 20 and 21). These different kinds of
interview share some common features, such as the eliciting
of information by the interviewer from the interviewee
and the operation of rules of varying degrees of formality
or explicitness concerning the conduct of the interview.
The research interview is a prominent data-collection
strategy in both quantitative and qualitative research.
The survey is probably the chief context within which
social researchers employ the structured interview (see
Key concept 9.1) in connection with quantitative research,
and it is this form of the interview that will be
The structured interview
Structured interviewing210
The reason why survey researchers typically prefer the
structured interview is that it promotes standardization
of both the asking of questions and the recording
of answers. This feature has two closely related virtues
from the perspective of quantitative research: reducing
error due to variation in the asking of questions, and
greater accuracy in and ease of processing respondents’
answers.
Reducing error due to interviewer
variability
The standardization of both the asking of questions and
the recording of answers means that, if the interview is
properly executed, variation in people’s replies will be
due to ‘true’ or ‘real’ variation and not due to the interview
context. To take a simple illustration, when we ask a
question that is supposed to be an indicator of a concept,
we want to keep error to a minimum, an issue that was
touched on at the end of Chapter 8. We can think of the
answers to a question as constituting the values that a
variable takes. These values, of course, exhibit variation.
This could be the question on alcohol consumption
among students that was a focus of Chapter 8 at certain
points. Students will vary in the number of alcohol units
they consume (as in Figure 9.1). However, some respondents
may be inaccurately classified in terms of the variable.
There are a number of possible reasons for this (see
Thinking deeply 9.1).
Most variables will contain an element of error, so that
it is helpful to think of variation as made up of two components:
true variation and error. In other words:
variation = true variation + variation due to error.
The aim is to keep the error component to a minimum
(see Figure 9.2), since error has an adverse effect on the
Key concept 9.1
What is a structured interview?
A structured interview, sometimes called a standardized interview, entails the administration of an interview
schedule by an interviewer. The aim is for all interviewees to be given exactly the same context of questioning.
This means that each respondent receives exactly the same interview stimulus as any other. The goal of this style
of interviewing is to ensure that interviewees’ replies can be aggregated, and this can be achieved reliably only if
those replies are in response to identical cues. Interviewers are supposed to read out questions exactly and in the
same order as they are printed on the schedule. Questions are usually very specific and very often offer the
interviewee a fixed range of answers (this type of question is often called closed, closed ended, pre-coded, or fixed
choice). The structured interview is the typical form of interview in survey research.
Thinking deeply 9.1
Common sources of error in survey research
There are many sources of error in survey research in addition to those associated with sampling. This is a list of
the principal sources of error:
1. a poorly worded question;
2. the way the question is asked by the interviewer;
3. misunderstanding on the part of the interviewee;
4. memory problems on the part of the interviewee;
5. the way the information is recorded by the interviewer;
6. the way the information is processed, either when answers are coded or when data are entered into the
computer.
Structured interviewing 211
validity of a measure. If the error component is quite
high (see Figure 9.3), validity will be jeopardized. The
significance for error of standardization in the structured
interview is that two sources of variation due to error—
the second and fifth in Thinking deeply 9.1—are likely to
be less pronounced, since the opportunity for variation in
interviewer behaviour in these two areas (asking questions
and recording answers) is reduced.
The significance of standardization and of thereby reducing
interviewer variability is this: assuming that there
is no problem with an interview question due to such
things as confusing terms or ambiguity (an issue that will
be examined in Chapter 11), we want to be able to say as
far as possible that the variation that we find is connected
with true variation between interviewees and not to
variation in the way a question was asked or the answers
recorded in the course of the administration of a survey
by structured interview. Variability can occur in either of
two ways. First, intra-interviewer variability, whereby an
interviewer is not consistent in the way he or she asks
questions and/or records answers. Second, when there is
more than one interviewer, there may be inter-interviewer
variability, whereby interviewers are not consistent with
each other in the ways they ask questions and/or record
answers. Needless to say, these two sources of variability
are not mutually exclusive; they can coexist, compounding
the problem even further. In view of the significance of
standardization, it is hardly surprising that some writers
prefer to call the structured interview a standardized
interview (e.g. Oppenheim 1992) or standardized survey
interview (e.g. Fowler and Mangione 1990).
Accuracy and ease of data processing
Like self-completion questionnaires, most structured
interviews contain mainly questions that are variously
referred to as closed, closed ended, pre-coded, or fixed
choice. This issue will be covered in detail in Chapter 11.
However, this type of question has considerable relevance
to the current discussion. With the closed question,
the respondent is given a limited choice of possible
answers. In other words, the interviewer provides respondents
with two or more possible answers and asks
them to select which one or ones apply. Ideally, this procedure
will simply entail the interviewer placing a tick
in a box by the answer(s) selected by a respondent or
circling the selected answer or using a similar procedure.
The advantage of this practice is that the potential for
interviewer variability is reduced: there is no problem
of whether the interviewer writes down everything that
the respondent says or of misinterpretation of the reply
given. If an open or open-ended question is asked, the
interviewer may not write down everything said, may
embellish what is said, or may misinterpret what is said.
However, the advantages of the closed question in the
context of survey research go further than this, as we
will see in Chapter 11. One advantage that is particularly
significant in the context of the present discussion is that
closed questions greatly facilitate the processing of data.
When an open question is asked, the answers need to
be sifted and coded in order for the data to be analysed
quantitatively. Not only is this a laborious procedure,
particularly if there is a large number of open questions
and/or of respondents; it also introduces the potential
for another source of error, which is the sixth in Thinking
deeply 9.1: it is quite likely that error will be introduced
as a result of variability in the coding of answers. When
open questions are asked, the interviewer is supposed to
write down as much of what is said as possible. Answers
gu e 9.Figure 9.1
A variable
Variation
Figure 9.3Figure 9.3
A variable with considerable error
True variation Variation due
to error
Figure 9.2Figure 9.2
A variable with little error
True variation
Variation due to error
Structured interviewing212
can, therefore, be in the form of several sentences. These
answers have to be examined and then categorized, so
that each person’s answer can be aggregated with other
respondents’ answers to a certain question. A number
will then be allocated to each category of answer, so that
the answers can then be entered into a computer database
and analysed quantitatively. This general process is
known as coding and will be examined in greater detail
in Chapter 11.
Coding introduces yet another source of error. First, if
the rules for assigning answers to categories, collectively
known as the coding frame, are flawed, the variation
that is observed will not reflect the true variation in interviewees’
replies. Second, there may be variability in the
ways in which answers are categorized. As with interviewing,
there can be two sources: intra-coder variability,
whereby the coder varies over time in the way in which
the rules for assigning answers to categories are implemented,
and inter-coder variability, whereby coders differ
from each other in the way in which the rules for
assigning answers to categories are implemented. If
either (or both) source(s) of variability occur, at least part
of the variation in interviewees’ replies will not reflect
true variation and instead will be caused by error.
The closed question sidesteps this problem neatly,
because respondents allocate themselves to categories.
The coding process is then a simple matter of attaching
a different number to each category of answer and of
entering the numbers into a computer database. It is not
surprising, therefore, that this type of question is often
referred to as pre-coded, because decisions about the
coding of answers are typically undertaken as part of the
design of the schedule—that is, before any respondents
have actually been asked questions. There is very little
opportunity for interviewers or coders to vary in the
recording or the coding of answers. Of course, if some
respondents misunderstand any terms in the alternative
answers with which they are presented, or if the answers
do not adequately cover the appropriate range of possibilities,
the question will not provide a valid measure.
However, that is a separate issue and one that will be returned
to in Chapter 11. The chief point to register about
closed questions for the moment is that, when compared
to open questions, they reduce one potential source of
error and are much easier to process for quantitative data
analysis.
Other types of interview
The structured interview is by no means the only type of
interview, but it is certainly the main type that is likely to
be encountered in survey research and in quantitative research
generally. Unfortunately, a host of different terms
have been employed by writers on research methodology
to distinguish the diverse forms of research interview.
Key concept 9.2 represents an attempt to capture some of
the major terms and types.
All the forms of interview outlined in Key concept 9.2,
with the exception of the structured interview and the
standardized interview, are primarily used in connection
with qualitative research, and it is in that context that they
will be encountered again later in this book. They are
rarely used in connection with quantitative research, and
survey research in particular, because the absence of
standardization in the asking of questions and recording
of answers makes respondents’ replies difficult to aggregate
and to process. This is not to say that they have no
role at all. For example, as we will see in Chapter 11, the
unstructured or semi-structured interview can have a useful
role in relation to developing the fixed-choice alternatives
with which respondents are provided in the kind of
closed question that is typical of the structured interview.
Key concept 9.2
Major types of interview
• Structured interview. See Key concept 9.1.
• Standardized interview. See Key concept 9.1.
• Semi-structured interview. This is a term that covers a wide range of instances. It typically refers to a context
in which the interviewer has a series of questions that are in the general form of an interview schedule but is
able to vary the sequence of questions. The questions are frequently somewhat more general in their frame of
reference from that typically found in a structured interview schedule. Also, the interviewer usually has some
latitude to ask further questions in response to what are seen as significant replies.
Structured interviewing 213
In an archetypal interview, an interviewer stands or sits
in front of the respondent asking the latter a series of
questions and writing down the answers. However, there
are several possible departures from it, although this
archetype is the most usual context for an interview.
More than one interviewee
In the case of group interviews or focus groups, there is
more than one, and usually quite a few more than one,
respondent or interviewee. Nor is this the only context in
which more than one person is interviewed. McKee and
Bell (1985), for example, interviewed couples in their
study of the impact of male unemployment, while, in
my research on visitors to Disney theme parks, not just
couples but often their children took part in the interview
as well (Bryman 1999). However, it is very unusual
for structured interviews to be used in connection with
this kind of questioning. In survey research, it is almost
always a specific individual who is the object of questioning.
Indeed, in survey interviews it is very advisable to
discourage as far as possible the presence and intrusion
ofothersduringthecourseoftheinterview.Investigations
in which more than one person is being interviewed tend
to be exercises in qualitative research, though this is not
always the case: Pahl’s (1990) study of patterns of control
of money among couples employed structured interviewing
of couples and of husbands and wives separately.
• Unstructured interview. The interviewer typically has only a list of topics or issues, often called an interview
guide or aide-mémoire, that are to be covered. The style of questioning is usually informal. The phrasing and
sequencing of questions will vary from interview to interview.
• Intensive interview. This term is employed by Lofland and Lofland (1995) as an alternative term to the
unstructured interview. Spradley (1979) uses the term ethnographic interview to describe a form of interview
that is also more or less synonymous with the unstructured interview.
• Qualitative interview. For some writers, this term seems to denote an unstructured interview (e.g. Mason
1996), but more frequently it is a general term that embraces interviews of both the semi-structured and
unstructured kind (e.g. Rubin and Rubin 1995).
• In-depth interview. Like the term ‘qualitative interview’, this one sometimes refers to an unstructured interview
but more often refers to both semi-structured and unstructured interviewing. The use of this term seems to be
increasing.
• Focused interview. This is a term devised by Merton et al. (1956) to refer to an interview using predominantly
open questions to ask interviewees questions about a specific situation or event that is relevant to them and
of interest to the researcher.
• Focus group. This is the same as the focused interview, but interviewees discuss the specific issue in groups.
See Key concept 21.1 for a more detailed definition.
• Group interview. Some writers see this term as synonymous with the focus group, but a distinction may be
made between the latter and a situation in which members of a group discuss a variety of matters that may
be only partially related.
• Oral history interview. This is an unstructured or semi-structured interview in which the respondent is asked
to recall events from his or her past and to reflect on them. There is usually a cluster of fairly specific research
concerns to do with a particular epoch or event, so there is some resemblance to a focused interview (see the
section on ‘Life history and oral history interviewing’ in Chapter 20.).
• Life history interview. This is similar to the oral history interview, but the aim of this type of unstructured
interview is to glean information on the entire biography of each respondent (see the section on ‘Life history
and oral history interviewing’ in Chapter 20.)
Interview contexts
Structured interviewing214
More than one interviewer
This is a very unusual situation in social research, because
of the considerable cost that is involved in dispatching
two (or indeed more than two) people to interview someone.
Bechhofer et al. (1984) describe research in which
two people interviewed individuals in a wide range of
occupations. However, while their approach achieved a
number of benefits for them, their interviewing style was
of the unstructured kind that is typically employed in
qualitative research, and they argue that the presence of
a second interviewer is unlikely to achieve any added
value in the context of structured interviewing.
There are several advantages of telephone over personal
interviews.
• On a like-for-like basis, they are far cheaper and also
quicker to administer. This arises because, for personal
interviews, interviewers have to spend a great
deal of time and money travelling between respondents.
This factor will be even more pronounced when
a sample is geographically dispersed, a problem that
is only partially mitigated for personal interview
surveys by strategies like cluster sampling. Of course,
telephone interviews take time and hired interviewers
have to be paid, but the cost of conducting a telephone
interview will still be lower than a comparable personal
one. Moreover, the general efficiency of telephone
interviewing has been enhanced with the
advent and widespread use in commercial circles of
computer-assisted telephone interviewing (CATI).
• The telephone interview is easier to supervise than
the personal interview. This is a particular advantage
In person or by telephone?
A third way in which the archetype may not be realized
is that interviews may be conducted by telephone rather
than face-to-face. While telephone interviewing is quite
common in commercial fields like market research,
where it usually takes the form of computer-assisted
telephone interviewing (CATI; see below), it is still
far more customary to read reports of studies based on
face-to-face interviews in academic social research, but
see Research in focus 9.1 for an interesting example.
when there are several interviewers, since it becomes
easier to check on interviewers’ transgressions in the
asking of questions, such as rephrasing questions or
the inappropriate use of probes by the interviewer.
Interviews can be tape-recorded so that data quality
can be assessed, but this raises issues that relate to
data protection and confidentiality, so that this procedure
has to be treated cautiously.
• Telephone interviewing has a further advantage,
which is to do with evidence (which is not as clear-cut
asonemightwant)thatsuggeststhat,inpersonalinterviews,
respondents’ replies are sometimes affected
by characteristics of the interviewer (for example,
class, ethnicity) and indeed by his or her mere presence
(implying that the interviewees may reply in
ways they feel will be deemed desirable by interviewers).
The remoteness of the interviewer in telephone
interviewing removes this potential source of bias to a
significant extent. The interviewer’s personal characteristics
cannot be seen, and the fact that he or she is
Research in focus 9.1
A telephone survey of the unemployed in Sweden
Nordenmark and Strandh (1999) report the findings of an interesting study of mental well-being among the
unemployed in Sweden. Early in 1996 a national random sample of 3,500 was drawn from a register of all
unemployed persons that is maintained by the Swedish Labour Market Board. A telephone survey was
conducted with members of the sample. The response rate was 74 per cent. The interview schedule included
questions on such issues as ‘mental well-being, the economy, work involvement, belief in the future, wage
demands and job search behaviour’ (Nordenmark and Strandh 1999: 585). Nearly two years later, those who
had participated were re-interviewed by telephone with very similar questions. This is, therefore, an example of
a panel study. The authors inform us that only part (around 6 per cent) of the 26 per cent who did not respond
was due to a refusal to participate; the remainder was due to problems of contacting respondents.
Structured interviewing 215
not physically present may offset the likelihood of respondents’
answers being affected by the interviewer.
Telephone interviewing suffers from certain limitations
when compared to the personal interview.
• People who do not own or who are not contactable
by telephone obviously cannot be interviewed by
telephone. Since this characteristic is most likely to
be a feature of poorer households, the potential for
sampling bias exists. Also, many people choose to be
ex-directory—that is, they have taken action for their
telephone numbers not to appear in a telephone directory.
Again, these people cannot be interviewed by
telephone. One possible solution to this last difficulty
is random digit dialling. With this technique, the computer
randomly selects telephone numbers within
a predefined geographical area. Not only is this a
random process that conforms to the rules about
probability sampling examined in Chapter 8; it also
stands a chance of getting at ex-directory households.
But it cannot, of course, gain access to those without a
telephone at all.
• Respondents with hearing impairments are likely to
find telephone interviewing much more difficult for
them than personal interviewing.
• The length of a telephone interview is unlikely to be
sustainable beyond 20–25 minutes, whereas personal
interviews can be much longer than this (Frey 2004).
• The question of whether response rates (see Key
concept 8.2) are lower with surveys by telephone
interview than with surveys by personal interview is
unclear, in that there is little consistent evidence on
this question. However, there is a general belief that
telephone interviews achieve slightly lower rates than
personal interviews (Frey and Oishi 1995; Shuy 2002;
Frey 2004).
• There is some evidence to suggest that telephone
interviews fare less well for the asking of questions
about sensitive issues, such as drug and alcohol use,
income, tax returns, and health. However, the evidence
is not entirely consistent on this point, though
it is probably sufficient to suggest that, when many
questions of this kind are to be used, a personal interview
may be superior (Shuy 2002).
• Developments in telephone communications such as
the growing use of answerphones and other forms of
call screening and of mobile phones have almost certainly
had an adverse effect on telephone surveys in
terms of response rates and the general difficulty of
getting access to respondents through conventional
landlines. Households that rely exclusively on mobile
phones represent a particular difficulty.
• Telephone interviewers cannot engage in observation.
This means that they are not in a position to respond
to signs of puzzlement or unease on the faces of
respondents when they are asked a question. In a
personal interview, the interviewer may respond to
such signs by restating the question or attempting to
clarify the meaning of the question, though this has
to be handled in a standardized way as far as possible.
A further issue relating to the inability of the interviewer
to observe is that, sometimes, interviewers may
be asked to collect subsidiary information in connection
with their visits (for example, whether a house
is dilapidated). Such information cannot be collected
when telephone interviews are employed.
• It is frequently the case that specific individuals in
households or firms are the targets of an interview.
In other words, simply anybody will not do. This
requirement is likely to arise from the specifications
of the population to be sampled, which means that
people in a certain role or position or with particular
characteristics are to be interviewed. It is probably
more difficult to ascertain by telephone interview
whether the correct person is replying.
• The telephone interviewer cannot readily employ
visual aids such as show cards (see below) from which
respondents might be asked to select their replies or to
use diagrams or photographs.
• There is some evidence to suggest that the quality
of data derived from telephone interviews is inferior
to that of comparable face-to-face interviews. A series
of experiments reported by Holbrook et al. (2003) on
the mode of survey administration in the USA using
long questionnaires found that respondents interviewed
by telephone were more likely to: express no
opinion or ‘don’t know’ (see Chapter 11 for more on
this issue); to answer in the same way to a series
of linked questions; to express socially desirable
answers; to be apprehensive about the interview; and
to be more likely to be dissatisfied with the time
taken by the interviews (even though they were invariably
shorter than in the face-to-face mode). Also, telephone
interviewees tended to be less engaged with
the interview process. While these results should be
viewed with caution, since studies like these are bound
to be affected by such factors as the use of a large
questionnaire on a national sample, they do provide
interesting food for thought.
Structured interviewing216
Computer-assisted interviewing
In recent years, increasing use has been made of computers
in the interviewing process, especially in commercial
survey research of the kind conducted by market
research and opinion polling organizations. There are
two main formats for computer-assisted interviewing:
computer-assisted personal interviewing (CAPI) and
computer-assisted telephone interviewing (CATI). A very
large percentage of telephone interviews is conducted
with the aid of personal computers. Among commercial
survey organizations, almost all telephone interviewing
is of the CATI kind nowadays, and this kind of interview
has become one of the most popular formats for such
firms. The main reasons for the growing use of CAPI has
been that the increased portability and affordability of
‘laptop’ computers, and the growth in the number and
quality of software packages that provide a platform for
devising interview schedules, provide greater opportunity
for them to be used in connection with face-to-face
interviews. CAPI and CATI have not infiltrated academic
survey research to anything like the same degree that
they have commercial survey research, although that picture
is likely to change considerably because of the many
advantages they possess. Indeed, the survey element of
the mixed methods study Cultural Capital and Social
Exclusion (CCSE), referred to in Research in focus 2.9,
was administered by CAPI. In any case, many of the large
datasets that are used for secondary analysis (see Chapter
14 for examples) derive from computer-assisted interviewing
studies undertaken by commercial or large
social research organizations.
With computer-assisted interviewing, the questions
that comprise an interview schedule appear on the screen.
As interviewers ask each question, they ‘key in’ the appropriate
reply using the keyboard (for open questions)
or using a mouse (for closed questions) and proceed to
the next question. Moreover, this process has the great
advantage that, when filter questions (see Tips and skills
‘Instructions for interviewers in the use of a filter question’)
are asked, so that certain answers may be skipped
as a result of a person’s reply, the computer can be programmed
to ‘jump’ to the next relevant question. This
removes the possibility of interviewers inadvertently asking
inappropriate questions or failing to ask ones that
should be asked. As such, computer-assisted interviewing
enhances the degree of control over the interview
process and can therefore improve standardization of the
asking and recording of questions. However, there is very
little evidence to suggest that the quality of data deriving
from computer-assisted interviews is demonstrably
superior to comparable paper and pencil interviews
(Couper and Hansen 2002). If the interviewer is out in
the field all day, he or she can either take a disk with the
saved data to the research office or send the data down a
telephone line with the aid of a modem. It is possible that
technophobic respondents may be a bit alarmed by their
use, but, by and large, the use of computer-assisted interviewing
seems destined to grow.
For their part, there is evidence that professional interviewers
generally like computer-assisted interviewing,
often feeling that it improves the image of their occupation,
though there are many who are concerned about
the problems that might arise from technical difficulties
and the inconvenience of correcting errors with a computer
as opposed to with a pen. One issue that sometimes
disconcerts interviewers is the fact that they can see only
part of the schedule at any one time (Couper and Hansen
2002). One potential problem with CAPI and CATI is
‘miskeying’, where the interviewer clicks on the wrong
reply. Whether this is more likely to occur than when the
interviewer is using pen and paper is unknown. In the
CCSE study, as noted in Research in focus 2.9, qualitative
interviews were conducted with some of the survey
respondents. In part this was done so that participants
in the semi-structured interview phase could be asked
about some of the answers they had given in the survey
interview. As a result, the researchers found that sometimes
the participant had been recorded as giving a particular
answer that was in fact incorrect. An example is
a respondent who had been recorded as indicating in
the survey interview as preferring to eat out in Italian
restaurants when in fact it should have been Indian ones
(Silva and Wright 2008). As the researchers note, it is
impossible to know how this error occurred, but miskeying
is one possible reason.
The discussion in the previous section of telephone
interviewing and of CATI in this section presumes that
the medium is a landline. However, with the huge growth
in the use of mobile phones (cellular or cell phones)
there is the prospect that these will have a role in future
years. Since lists of mobile-phone users are unlikely to
be available in the way that telephone directories are,
random digit dialing (RDD) is most likely to be employed
by researchers seeking to interview by mobile phone.
Zuwallack (2009) reports the findings of some CATI
projectsconductedbymobilephoneintheUSAonhealthrelated
issues. The researchers found that a lot of people
hung up when contacted but that those respondents who
persisted formed a useful complement to conventional
landline telephone surveys because many of them had
characteristics often under-represented in such surveys,
Structured interviewing 217
such as young adults and minorities. Of particular interest
is that a large percentage of respondents lived in
households without a landline, suggesting that, if the
number of mobile-only households increases, mobilephone
surveys may become increasingly significant.
Zuwallack also reports that the mobile-phone survey
is more expensive than the equivalent landline CATI
survey.
One further point to register in connection with
computer-assisted interviewing is that the section has not
included Internet surveys. The reason for this is that
such surveys are more properly considered as using selfcompletion
questionnaires rather than structured interviewing
(see Figure 8.2). With such surveys, there is no
interviewer in the sense of a person who verbally asks
questions. Internet surveys are covered in Chapter 28.
Issues concerning the conduct of interviews are examined
here in a very general way. In addition to the matters
considered here, there is clearly the important issue of
how to word the interview questions themselves. This
area will be explored in Chapter 11, since many of the
rules of question-asking relate to self-completion questionnaire
techniques like postal questionnaires as well as
to structured interviews. One further general point to
make here is that the advice concerning the conduct of
interviews provided in this chapter relates to structured
interviews. The framework for carrying out the kinds
of interviewing conducted in qualitative research (such
as unstructured and semi-structured interviewing and
focus groups) will be handled in later chapters.
Know the schedule
Before interviewing anybody, an interviewer should be
fully conversant with the schedule. Even if you are the
only person conducting interviews, make sure you know
it inside out. Interviewing can be stressful for interviewers,
and it is possible that under duress standard interview
procedures like filter questions (see Tips and skills
‘Instructions for interviewers in the use of a filter question’)
can cause interviewers to get flustered and miss
questions out or ask the wrong questions. If two or more
interviewers are involved, they need to be fully trained to
know what is required of them and to know their way
around the schedule. Training is especially important in
order to reduce the likelihood of interviewer variability
in the asking of questions, which is a source of error.
Introducing the research
Prospective respondents have to be provided with a credible
rationale for the research in which they are being
asked to participate and for giving up their valuable
time. This aspect of conducting interview research is of
particular significance at a time when response rates
to survey research appear to be declining, though, as
noted in Chapter 8, the evidence on this issue is the focus
of some disagreement. The introductory rationale may
be either spoken by the interviewer or written down. In
many cases, respondents may be presented with both
modes. It comes in spoken form in such situations as
when interviewers make contact with respondents on
the street or when they ‘cold call’ respondents in their
homes in person or by telephone. A written rationale will
be required to alert respondents that someone will be
contacting them in person or on the telephone to request
an interview. Respondents will frequently encounter
both forms—for example, when they are sent a letter and
then when they ask the interviewer who turns up to
interview them what the research is all about. It is
important for the two accounts to be consistent, as this
could be a test!
Introductions to research should typically contain the
bits of information outlined in Tips and skills ‘Topics
and issues to include in an introductory statement’.
Since interviewers represent the interface between the
research and the respondent, they have an important
role in maximizing the response rate for the survey. In
addition the following points should be borne in mind.
• Interviewers should be prepared to keep calling
back if interviewees are out or unavailable. This will
require taking into account people’s likely work and
leisure habits—for example, there is no point in calling
at home on people who work during the day. In
addition, people living alone may be reluctant to
answer the door when it is dark because of fear of crime.
• Be self-assured. You may get a better response if you
presume that people will agree to be interviewed
rather than that they will refuse.
Conducting interviews
Structured interviewing218
• Reassure people that you are not a salesperson.
Because of the tactics of certain organizations whose
representatives say they are doing market or social
research, many people have become very suspicious
of people saying they would just like to ask you a few
questions.
• Dress in a way that will be acceptable to a wide
spectrum of people.
• Make it clear that you will be happy to find a time to
suit the respondent.
Rapport
It is frequently suggested that it is important for the interviewer
to achieve rapport with the respondent. This
means that very quickly a relationship must be established
that encourages the respondent to want (or at
least be prepared) to participate in and persist with
the interview. Unless an element of rapport can be
established, some respondents may initially agree to be
interviewed but then decide to terminate their participation
because of the length of time the interview is
taking or perhaps because of the nature of the questions
being asked. While this injunction essentially invites
the interviewer to be friendly with respondents and to
put them at ease, it is important that this quality is not
stretched too far. Too much rapport may result in the
interview going on too long and the respondent suddenly
deciding that too much time is being spent on the activity.
Also, the mood of friendliness may result in the
respondent answering questions in a way that is designed
to please the interviewer. The achievement of rapport
between interviewer and respondent is therefore a delicate
balancing act. Moreover, it is probably somewhat
easier to achieve in the context of the face-to-face interview
than in the telephone interview, since in the latter
the interviewer is unable to offer obvious visual cues
of friendliness such as smiling or maintaining good eye
contact, which are also frequently regarded as conducive
to gaining and maintaining rapport.
Tips and skills
Topics and issues to include in an
introductory statement
There are several issues to include in an introductory statement to a prospective interviewee. The following list
comprises the principal considerations.
• Make clear the identity of the person who is contacting the respondent.
• Identify the auspices under which the research is being conducted—for example, a university, a market
research agency.
• Mention any research funder, or, if you are a student doing an undergraduate or postgraduate dissertation or
doing research for a thesis, make this clear.
• Indicate what the research is about in broad terms and why it is important, and give an indication of the kind
of information to be collected.
• Indicate why the respondent has been selected—for example, selected by a random process.
• Make it clear that participation is voluntary.
• Reassure the respondent that he or she will not be identified or be identifiable in any way. This can usually be
achieved by pointing out that data are anonymized when they are entered into the computer and that
analysis will be conducted at an aggregate level.
• Provide reassurance about the confidentiality of any information provided.
• Provide the respondent with the opportunity to ask any questions—for example, provide a contact telephone
number if the introduction is in the form of a written statement, or, if in person, simply ask if the respondent
has any questions.
These suggestions are also relevant to the covering letter that accompanies postal questionnaires, except that
researchers using this method need to remember to include a stamped-addressed envelope!
Structured interviewing 219
Asking questions
It was earlier suggested that one of the aims of the structured
interview is to ensure that each respondent is
asked exactly the same questions. Recall that in Thinking
deeply 9.1 it was pointed out that variation in the ways a
question is asked is a potential source of error in survey
research. The structured interview is meant to reduce the
likelihood of this occurring, but it cannot guarantee that
this will not occur, because there is always the possibility
that interviewers will embellish or otherwise change a
question when it is asked. There is considerable evidence
that this occurs, even among centres of social research
that have a solid reputation for being rigorous in following
correct methodological protocol (Bradburn and
Sudman 1979). The problem with such variation in the
asking of questions was outlined above: it is likely to
engender variation in replies that does not reflect ‘true’
variation—in other words, error. Consequently, it is important
for interviewers to appreciate the importance of
keeping exactly to the wording of the questions they are
charged with asking.
You might say: ‘does it really matter?’ In other words,
surely small variations to wording cannot make a significant
difference to people’s replies? While the impact of
variation in wording obviously differs from context to
context and is in any case difficult to quantify exactly,
experiments in question-wording suggest that even small
variations in wording can exert an impact on replies
(Schuman and Presser 1981). Three experiments in
England conducted by Social and Community Planning
Research concluded that a considerable number of interview
questions is affected by interviewer variability. The
researchers estimated that, for about two-thirds of the
questions that were considered, interviewers contributed
to less than 2 per cent of the total variation in each
question (M. Collins 1997). On the face of it, this is a
small amount of error, but the researchers regarded it as
a cause for concern.
The key point to emerge, then, is the importance of
getting across to interviewers the importance of asking
questions as they are written. There are many reasons
why interviewers may vary question-wording, such as
reluctance to ask certain questions, perhaps because
of embarrassment (M. Collins 1997), but the general
admonition to keep to the wording of the question needs
to be constantly reinforced when interviewers are being
trained. It also needs to be borne in mind for your own
research.
Recording answers
An identical warning for identical reasons can be registered
in connection with the recording of answers by
interviewers, who should write down respondents’ replies
as exactly as possible. Not to do so can result in
interviewers distorting respondents’ answers and hence
introducing error. Such errors are less likely to occur
when the interviewer has merely to allocate respondents’
replies to a category, as in a closed question. This process
can require a certain amount of interpretation on the
part of the interviewer, but the error that is introduced is
far less than when answers to open questions are being
written down (Fowler and Mangione 1990).
Clear instructions
In addition to instructions about the asking of questions
and the recording of answers, interviewers need instructions
about their progress through an interview schedule.
An example of the kind of context in which this is
likely to occur is in relation to filter questions. Filter questions
require the interviewer to ask questions of some
respondents but not others. For example, the question:
For which political party did you vote at the last general
election?
presumes that the respondent did in fact vote. This
option can be reflected in the fixed-choice answers that
are provided, so that one of these is a ‘did not vote’ alternative.
However, a better solution is not to presume anything
about voting behaviour but to ask respondents
whether they voted in the last general election and then
to filter out those who did not vote. The foregoing
question about the political party voted for can then be
asked of those who did in fact vote. Similarly, in a study
of meals, there is no point in asking vegetarians lots of
questions about eating meat. It will probably work out
best to filter vegetarians out and then possibly ask them a
separate series of questions. Tips and skills ‘Instructions
for interviewers in the use of a filter question’ provides a
simple example in connection with an imaginary study of
alcohol consumption. The chief point to register about
this example is that it requires clear instructions for the
interviewer. If such instructions are not provided, there
is the risk that either respondents will be asked inappropriate
questions (which can be irritating for them) or
Structured interviewing220
the interviewer will inadvertently fail to ask a question
(which results in missing information).
Question order
In addition to interviewers being warned about the
importance of not varying the asking of questions and
the recording of answers, they should be alerted to the
importance of keeping to the order of asking questions.
Quite a lot of research has been carried out on the
general subject of question order, but few if any consistent
effects on people’s responses that derive from
For one thing, varying the question order can result in
certain questions being accidentally omitted, because the
interviewer may forget to ask those that have been leapfrogged
during the interview. Also, variation in question
order may have an impact on replies: if some respondents
have been previously asked a question that they should
have been asked whereas others have not, a source of
variability in the asking of questions will have been
introduced and therefore a potential source of error.
asking questions at different points in a questionnaire
or interview schedule have been unveiled. Different
effects have been demonstrated on various occasions.
Tips and skills
Instructions for interviewers in the use of
a filter question
Each of the following questions includes an instruction to the interviewer about how to proceed.
1. Have you consumed any alcoholic drinks in the last twelve months?
Yes ____
No ____
(if No proceed to question 4)
2. (To be asked if interviewee replied Yes to question 1)
Which of the following alcoholic drinks do you consume most frequently?
(Ask respondent to choose the category that he or she drinks most frequently and tick one category only.)
Beer ____
Spirits ____
Wine ____
Liquors ____
Other ____ specify ____________________________________________________
3. How frequently do you consume alcoholic drinks?
(Ask interviewee to choose the category that comes closest to his or her current practice.)
Daily ____
Most days ____
Once or twice a week ____
Once or twice a month ____
A few times a year ____
Once or twice a year ____
4. (To be asked if interviewee replied No to question 1)
Have you ever consumed alcoholic drinks?
Yes ____
No ____
Structured interviewing 221
A study in the USA found that people were less likely to
say that their taxes were too high when they had been
previously asked whether government spending ought
to be increased in a number of areas (Schuman and
Presser 1981: 32). Apparently, some people perceived
an inconsistency between wanting more spending and
lower taxes, and adjusted their answers accordingly.
Research on crime victimization in the USA suggests that
earlier questions may affect the salience of later issues
(Schuman and Presser 1981: 45). Respondents were
asked whether they had been victims of crime in the
preceding twelve months. Some respondents had been
previously asked a series of questions about their attitudes
to crime, whereas others had not. Those who had
been asked about their attitudes reported considerably
more crime than those who had not been asked.
Mayhew (2000) provides an interesting anecdote on
question order in relation to the British Crime Survey.
Each wave of the BCS has included the question:
Taking everything into account, would you say the
police in this area do a good job or a poor job?
In 1988 this question appeared twice by mistake for
some respondents! For all respondents it appeared early
on, but for around half it also appeared later on in the
context of questions on contact with the police. Of those
given the question twice, 66 per cent gave the same
rating, but 22 per cent gave a more positive rating to
the police and just 13 per cent gave a less favourable
one. Mayhew suggests that, as the interview wore on,
respondents became more sensitized to crime-related
issues and more sympathetic to the pressures on the
police.
However, it is difficult to draw general lessons from
such research, at least in part because experiments in
question order do not always reveal clear-cut effects of
varying the order in which questions are asked, even
in cases where effects might legitimately have been
expected. There are two general lessons.
1. Within a survey, question order should not be varied
(unless, of course, question order is the subject of the
study!).
2. Researchers should be sensitive to the possible implications
of the effect of early questions on answers to
subsequent questions.
The following rules about question order are sometimes
proposed.
• Early questions should be directly related to the
topic of the research, about which the respondent
has been informed. This removes the possibility that
the respondent will be wondering at an early stage
in the interview why he or she is being asked apparently
irrelevant questions. This injunction means that
personal questions about age, social background, and
so on should not be asked at the beginning of an
interview.
• As far as possible, questions that are more likely to be
salient to respondents should be asked early in the
interview schedule, so that their interest and attention
are more likely to be secured. This suggestion
may conflict with the previous one, in that questions
specifically on the research topic may not be obviously
salient to respondents, but it implies that as far as
possible questions relating to the research topic that
are more likely to grab their attention should be asked
at or close to the start of the interview.
• Potentially embarrassing questions or ones that may
be a source of anxiety should be left till later. In fact,
research should be designed to ensure that, as far as
possible, respondents are not discomfited, but it has to
be acknowledged that with certain topics this effect
may be unavoidable.
• With a long schedule or questionnaire, questions
should be grouped into sections, since this allows a
better flow than skipping from one topic to another.
• Within each group of questions, general questions
should precede specific ones. Tips and skills ‘A
sequence of questions on the topic of identity cards’
provides an illustration of such a sequence, which
follows the recommendations of Gallup (1947, cited
in Foddy 1993: 61–2). The example is concerned to
demonstrate how the approach might operate in
connection with identity cards, which have been an
area of discussion and some controversy in the UK in
recent years. The question order sequence is designed
with a number of features in mind. It is designed
to establish people’s levels of knowledge of identity
cards before asking questions about it and to distinguish
those who feel strongly about it from those
who do not. According to Foddy (1993), the second
question is always open ended, so that respondents’
frames of references can be established with respect
to the topic at hand. However, it seems likely that, if
sufficient pilot research has been carried out, a closed
question could be envisaged, a point that applies
equally to question 4.
Structured interviewing222
• A further aspect of the rule that general questions
should precede specific ones is that it has been argued
that, when a specific question comes before a general
one, the aspect of the general question that is covered
by the specific one is discounted in the minds of
respondents because they feel they have already
covered it. Thus, if a question about how people feel
about the amount they are paid precedes a general
question about job satisfaction, there are grounds for
thinking that respondents will discount the issue of
pay when responding about job satisfaction.
• It is sometimes recommended that questions dealing
with opinions and attitudes should precede questions
to do with behaviour and knowledge. This is because
it is felt that behaviour and knowledge questions are
less affected by question order than questions that tap
opinions and attitudes.
• During the course of an interview, it sometimes
happens that a respondent provides an answer to a
question that is to be asked later in the interview.
Because of the possibility of a question order effect,
when the interviewer arrives at the question that
appears already to have been answered, it should be
repeated.
However, question order effects remain one of the
more frustrating areas of structured interview and questionnaire
design, because of the inconsistent evidence
that is found and because it is difficult to formulate
generalizations or rules from the evidence that does
point to their operation. An interesting discussion about
question order took place some years ago in connection
with the study of social class and is discussed in
Thinking deeply 9.2.
Tips and skills
A sequence of questions on the topic of
identity cards
1. Have you heard of identity cards?
Yes ____ No ____
2. What are your views about identity cards?
3. Do you favour or not favour identity cards?
Favour ____ Not favour ____
4. Why do you favour (not favour) identity cards?
5. How strongly do you feel about this?
Very strongly ____
Fairly strongly ____
Not at all strongly ____
Thinking deeply 9.2
A debate about question order
An interesting case of the issue of question order becoming a focus of controversy is provided by the research
on social class by Marshall et al. (1988), which is referred to in more detail in Research in focus 7.4 and 9.2. In a
critique of the research, Saunders (1989) argues that it reveals what he calls ‘socialist preconceptions’, implying
that values overtly intruded into the research (see Figure 2.3). Saunders argues that one way in which this was
revealed was the sheer weight of questions about social class prior to respondents being asked about the groups
to which they saw themselves as belonging. Saunders (1989: 4) writes:
Structured interviewing 223
Probing
Probing is a highly problematic area for researchers
employing a structured interview method. It frequently
happens in interviews that respondents need help with
their answers. One obvious case is where it is evident that
they do not understand the question—they may either
ask for further information or it is clear from what they
say that they are struggling to understand the question or
to provide an adequate answer. The second kind of situation
the interviewer faces is when the respondent does
not provide a sufficiently complete answer and has to
be probed for more information. The problem in either
situation is obvious: the interviewer’s intervention may
influence the respondent, and the nature of interviewers’
interventions may differ. A potential source of variability
in respondents’ replies that does not reflect ‘true’ variation
is introduced—that is, error.
Some general tactics with regard to probes are as
follows.
• If further information is required, usually in the context
of an open question, standardized probes can be
employed, such as ‘Could you say a little more about
that?’ or ‘Are there any other reasons why you think
that?’ or simply ‘mmmm . . . ?’.
• If the problem is that when presented with a closed
question the respondent replies in a way that does not
allow the interviewer to select one of the pre-designed
answers, the interviewer should repeat the fixedchoice
alternatives and make it apparent that the
answer needs to be chosen from the ones that have
been provided.
• If the interviewer needs to know about something that
requires quantification, such as the number of visits to
building societies in the last four weeks or the number
of building societies in which the respondent has
accounts, but the respondent resists this by answering
in general terms (‘quite often’ or ‘I usually go to the
building society every week’), the interviewer needs
to persist with securing a number from the respondent.
This will usually entail repeating the question.
The interviewer should not try to second guess a
figure on the basis of the respondent’s reply and then
suggest that figure to him or her, since the latter
may be unwilling to demur from the interviewer’s
suggested figure.
A glance at their questionnaire reveals that respondents were bombarded with questions about class right
from the start of the interview. Following no fewer than 28 detailed questions about the class system,
respondents were then asked if they thought they belonged to any social class and whether there was, by any
chance, any other grouping they identified with apart from their class. Not surprisingly, most agreed that they
did belong to one class or another . . . and that they could not think of any other identity. . . . Armed with their
‘findings’, the authors then conclude that we are all class-oriented after all and that other identities are far less
important.
Two of the book’s authors replied with a spirited rebuttal. They replied that
the question about class was preceded by 30 substantive items. Six of these have no obvious relationship to
the issue of social identities; for example, they elicit perceptions of Britain’s economic performance. . . . No less
than 17 . . . were specifically designed to make interviewees see the world in terms other than those of social
class. They invited people to think of themselves as consumers . . . as voters . . . as members of ethnic or
gender groupings; as employees . . . in short, as everything but members of an identifiable social class.
Interviewees were also asked whether there were ‘any important conflicts in Britain today’ before the word
social class was ever mentioned. Then, and only then, were they quizzed about their perception of Britain as
a specifically class society. (Marshall and Rose 1989: 5)
This is an interesting debate because it raises the issue of the role of values and bias in social research and also
because it relates to the issue of question order while demonstrating the difficulty of being definitive about the
issue. For example, while Marshall and Rose’s reply is convincing, it might be that it is the relative number of
questions about social class preceding questions of identity that may have been influential and to which
Saunders alludes. Nonetheless, the debate usefully demonstrates the difficulty of producing conclusive evidence
about question order effects.
Structured interviewing224
However, from the point of view of standardizing the
asking of questions in surveys using structured interviewing,
probing should be kept to a minimum (assuming
it cannot be eliminated) because it introduces error.
This occurs because it is impossible for interviewers to
probe in a consistent manner and because interviewer
effects are more likely to occur, whereby characteristics
of the interviewer have an impact on the respondent’s
replies (Groves et al. 2004: 281–2).
Prompting
Prompting occurs when the interviewer suggests a
possible answer to a question to the respondent. The
key prerequisite here is that all respondents receive the
same prompts. All closed questions entail standardized
prompting, because the respondent is provided with a
list of possible answers from which to choose. An unacceptable
approach to prompting would be to ask an open
question and to suggest possible answers only to some
respondents, such as those who appear to be struggling
to think of an appropriate reply.
During the course of a face-to-face interview, there are
several circumstances in which it will be better for the
interviewer to use ‘show cards’ rather than rely on reading
out a series of fixed-choice alternatives. Show cards
(sometimes called ‘flash cards’) display all the answers
from which the respondent is to choose and are handed
to the respondent at different points of the interview.
Three kinds of context in which it might be preferable to
employ show cards rather than to read out the entire set
of possible answers are as follows.
• There may be a very long list of possible answers.
For example, respondents may be asked which daily
newspaper they each read most frequently. To read
out a list of newspapers would be tedious, and it is
probably better to hand the respondent a list of newspapers
from which to choose.
• Sometimes, during the course of interviews, respondents
are presented with a group of questions to which
the same possible answers are attached. An example
of this strategy is Likert scaling, an approach to
attitude measurement that was discussed in Key concept
7.2. The components of a Likert scale are often
referred to as items rather than as questions, since
strictly speaking respondents are not being asked
questions but are presented with statements to which
they are asked to indicate their levels of agreement.
See Research in focus 7.2 and 7.5 for examples. It
would be excruciatingly dull to read out all five or
seven possible answers ten times. Also, it may be
expecting too much of respondents to read out the
answers once and then require them to keep the possible
answers in their heads for the entire batch of
questions to which they apply. A show card that can
be used for the entire batch and to which respondents
can constantly refer is an obvious solution. As
was mentioned in Key concept 7.2, most Likert scales
of this kind comprise five levels of agreement/disagreement,
and it is this more conventional approach
that is illustrated in the show card in Tips and skills
‘A show card’.
Tips and skills
A show card
Card 6
Strongly agree
Agree
Undecided
Disagree
Strongly disagree
• Some people are not keen to divulge personal details
such as their age or their income. One way
of neutralizing the impact of such questioning is to
present respondents with age or income bands with
a letter or number attached to each band. They can
then be asked to say which letter/number applies to
Structured interviewing 225
them (see Tips and skills ‘Another show card’). This
procedure will obviously not be appropriate if the
research requires exact ages or incomes. It may be
Leaving the interview
Do not forget common courtesies like thanking respondents
for giving up their time. But the period immediately
after the interview is one in which some care is
necessary, in that sometimes respondents try to engage
the interviewer in a discussion about the purpose of the
interview. Interviewers should resist elaboration beyond
extendable to sensitive areas such as number of sexual
partners or sexual practices for the same kinds of
reason.
their standard statement, because respondents may communicate
what they are told to others, which may bias
the findings.
Training and supervision
On several occasions, reference has been made to the
need for interviewers to be trained. The standard texts on
Tips and skills
Another show card
Card 11
1. Below 20
2. 20–29
3. 30–39
4. 40–49
5. 50–59
6. 60–69
7. 70 and over
Student experience
The need for structure in a survey interview
Joe Thompson’s survey research on students and their views of accommodation and facilities at his university
was part of a team project. After he and other members of his team had piloted the interview schedule, they
decided that it was not sufficiently structured. They felt that they needed to impose more structure and decided
to use show cards (he refers to them by their other common name ‘cue cards’).
The group therefore used opportunistic sampling to test if the questionnaire would be successful when applied
in a social setting, having to give the questionnaire to one person over the week. The following week the group
discussed the issues they had encountered when carrying out the pilot questionnaire, raising amongst others
the concern of not having a standard interview procedure, which would mean that certain biases could affect
the results. Therefore the group decided they would use cue cards when giving the options in answer to the
question, so as to avoid leading questions, etc. After these changes were implemented, the final version of the
questionnaire was produced.
To read more about Joe’s research experiences, go to the Online Resource Centre that accompanies this book at:
www.oxfordtextbooks.co.uk/orc/brymansrm4e/
Structured interviewing226
survey research and on interviewing practice tend to be
replete with advice on how best to train interviewers.
Such advice is typically directed at contexts in which
a researcher hires an interviewer to conduct a large
amount or even all the interviews. It also has considerable
importance in research in which several interviewers
(who may be either collaborators or hired
interviewers) are involved in a study, since the risk of
interviewer variability in the asking of questions needs
to be avoided.
For many readers of this book who are planning to
do research, such situations are unlikely to be relevant
because they will be ‘lone’ researchers. You may be doing
an undergraduate dissertation, or an exercise for a research
methods course, or you may be a postgraduate
conducting research for a Master’s dissertation or for
a thesis. Most people in such a situation will not have
the luxury of being able to hire a researcher to do any
interviewing (though you may be able to find someone
to help you a little). When interviewing on your own, you
must in a sense train yourself to follow the procedures
and advice provided above. This is a very different situation
from a large research institute or market research
agency, which relies on an army of hired interviewers
who carry out the interviews. Whenever people other
than the lead researcher are involved in interviewing,
they will need training and supervision in the following
areas:
• contacting prospective respondents and providing an
introduction to the study;
• reading out questions as written and following instructions
in the interview schedule (for example, in
connection with filter questions);
• using appropriate styles of probing;
• recording exactly what is said;
• maintaining an interview style that does not bias
respondents’ answers.
Fowler (1993) cites evidence that suggests that training
of less than one full day rarely creates good
interviewers.
Supervision of interviewers in relation to these issues
can be achieved by:
• checking individual interviewers’ response rates;
• tape-recording at least a sample of interviews;
• examining completed schedules to determine whether
any questions are being left out or if they are being
completed properly;
• call-backs on a sample of respondents (usually around
10 per cent) to determine whether they were interviewed
and to ask about interviewers’ conduct.
Research in focus 9.2 provides an example of some
of the ingredients of research involving multiple
interviewers.
Research in focus 9.2
An example of research involving multiple
interviewers
This example is taken from the study by Marshall et al. (1988), a team of sociologists from the University of
Essex, of social class in modern Britain (see Research in focus 7.4). The interviewing was carried out by a leading
independent social research institute, Social and Community Planning Research (SCPR). The research aimed to
achieve a sample of 2,000 respondents (1,770 was the number actually achieved; see Research in focus 8.1 for
details of the sampling procedure).
One hundred and twenty-three interviewers were employed on the survey. Six full-time briefing sessions were
held, all of which were attended by a member of the Essex team, and interviewers were also given a full set of
written instructions. The first three interviews conducted by each interviewer were subjected to an immediate
thorough checking in order that critical comments, where appropriate, could be conveyed. During the course
of fieldwork the work of interviewers was subject to personal recall. Ten per cent of issued addresses were
re-issued for recall . . . In addition, 36 interviewers were accompanied in the field by supervisors . . .
(Marshall et al. 1988: 291)
Structured interviewing 227
While the structured interview is a commonly used
method of social research, certain problems associated
with it have been identified over the years. These problems
are not necessarily unique to the structured interview,
in that they can sometimes be attributed to kindred
methods, such as the self-completion questionnaire in
survey research or even semi-structured interviewing in
qualitative research. However, it is common for the structured
interview to be seen as a focus for the identification
of certain limitations, which are briefly examined below.
Characteristics of interviewers
There is evidence that interviewers’ attributes can have
an impact on respondents’ replies, but, unfortunately,
the literature on this issue does not lend itself to definitive
generalizations. In large part, this ambiguity in the
broader implications of experiments relating to the
effects of interviewer characteristics is due to several
problems, such as: the problem of disentangling the
effects of interviewers’ different attributes from each
other (race, gender, socio-economic status); the interaction
between the characteristics of interviewers and
the characteristics of respondents; and the interaction
between any effects observed and the topic of the interview.
Nonetheless, there is undoubtedly some evidence
that effects due to characteristics of interviewers can be
discerned.
The ethnicity of interviewers is one area that has
attracted some attention. Schuman and Presser (1981)
cite a study that asked respondents to nominate two or
three of their favourite actors or entertainers. Respondents
were much more likely to mention black actors or
entertainers when interviewed by black interviewers
than when interviewed by white ones. Schuman and
Converse (1971) interviewed 619 black Detroiters shortly
after Martin Luther King’s assassination in 1969. The
researchers found significant differences in the answers
given between black and white interviewers in around
one-quarter of the questions asked.
Although this proportion is quite disturbing, the fact
that the majority of questions appear to have been largely
unaffected does not give rise to a great deal of confidence
that a consistent biasing factor is being uncovered.
Similarly inconclusive findings tend to occur in relation
to experiments with other sets of characteristics of
interviewers. These remarks are not meant to play down
the potential significance of interviewers’ characteristics
for measurement error, but to draw attention to the
limitations of drawing conclusive inferences about the
evidence. All that needs to be registered at this juncture
is that almost certainly the characteristics of interviewers
do have an impact on respondents’ replies but
that the extent and nature of the impact are not clear
and are likely to vary from context to context.
Response sets
Some writers have suggested that the structured interview
is particularly prone to the operation among respondents
of what Webb et al. (1966: 19) call ‘response
sets’, which they define as ‘irrelevant but lawful sources
of variance’. This form of response bias is especially
relevant to multiple-indicator measures (see Chapter 7),
where respondents reply to a battery of related questions
or items, of the kind found in a Likert scale (see Key concept
7.2). The idea of a response set implies that people
respond to the series of questions in a consistent way
but one that is irrelevant to the concept being measured.
Two of the most prominent types of response set are known
as the ‘acquiescence’ (also known as the ‘yeasaying’ and
‘naysaying’ effect) and the social desirability effect.
Acquiescence
Acquiescence refers to a tendency for some people
consistently to agree or disagree with a set of questions
or items. Imagine a respondent who replied with agreement
to all the items in Research in focus 7.2. The problem
is that agreement with some of the items implies a
low level of commitment to work (items 1–4), whereas
agreement with others implies a high level of commitment
to work (items 5–10). One of the reasons why researchers
who employ this kind of multiple-item measure
use wordings that imply opposite stances (that is, some
items implying a high level of commitment and others
implying a low level of commitment to work) is to weed
out those respondents who appear to be replying within
the framework of an acquiescence response set.
Social desirability bias
The social desirability effect refers to evidence that some
respondents’ answers to questions are related to their
Problems with structured interviewing
Structured interviewing228
perception of the social desirability of those answers. An
answer that is perceived to be socially desirable is more
likely to be endorsed than one that is not. This phenomenon
has been demonstrated in studies on mental health
using psychiatric inventories. These inventories are
meant to be concerned not with chronic mental illness
but with minor neuroses and anxieties. Research in New
York by Dohrenwend (1966) noted that Puerto Ricans
tended to score much higher on the inventory that he
administered than other ethnic groups. He found that
this tendency was not due to a higher level of mental
illness in this ethnic group, but to the effect of social
desirability in respondents’ answers. Puerto Ricans were
much less likely than the other ethnic groups to perceive
the items in the inventory as undesirable. This meant
that what the researcher had found was a link not
between ethnicity and mental health, but between ethnicity
and perceptions of the social desirability of mental
health inventory items. Later research suggested that
variation in the social desirability of mental illness symptoms
was related to the perceived prevalence of those
symptoms among the respondent’s friends and acquaintances
(Phillips 1973). The presence of social desirability
effects has been demonstrated in other settings (e.g.
Arnold and Feldman 1981).
In so far as these forms of response error go undetected,
they represent sources of error in the measurement
of concepts. However, while some writers have
proposed outright condemnation of social research on
the basis of evidence of response sets (e.g. Phillips 1973),
it is important not to get carried away with such findings.
We cannot be sure how prevalent these effects are, and to
some extent awareness of them has led to measures to
limit their impact on data (for example, by weeding out
cases obviously affected by them or by instructing interviewers
to limit the possible impact of the social desirability
effect by not becoming overly friendly with
respondents and by not being judgemental about their
replies).
The problem of meaning
A critique of survey interview data and findings gleaned
from similar techniques was developed by social scientists
influenced by phenomenological and other interpretivist
ideas of the kinds touched on in Chapter 2
(Cicourel 1964, 1982; Filmer et al. 1972; Briggs 1986;
Mishler 1986). This critique revolves around what is
often referred to in a shorthand way as the ‘problem
of meaning’. The kernel of the argument is that when
humans communicate they do so in a way that not only
draws on commonly held meanings but also simultaneously
creates meanings. ‘Meaning’ in this sense is something
that is worked at and achieved—it is not simply
pre-given. Allusions to the problem of meaning in structured
interviewing draw attention to the notion that
survey researchers presume that interviewer and respondent
share the same meanings of terms employed
in the interview questions and answers. In fact, the
problem of meaning implies that the possibility that
interviewer and respondent may not be sharing the
same meaning systems and hence imply different things
in their use of words is simply sidestepped in structured
interview research. The problem of meaning is resolved
by ignoring it.
The feminist critique
The feminist critique of structured interviewing is difficult
to disentangle from the critique launched against
quantitative research in general, which was briefly outlined
in Chapter 2. However, for many feminist social
researchers the structured interview symbolizes more
readily than other methods the limitations of quantitative
research, partly because of its prevalence but also
partly because of its nature. By ‘its nature’ is meant the
fact that the structured interview epitomizes the asymmetrical
relationship between researcher and subject
that is seen as an ingredient of quantitative research: the
researcher extracts information from the research subject
and gives nothing in return. For example, standard
textbook advice of the kind provided in this chapter
implies that rapport is useful to the interviewer but he
or she should guard against becoming too familiar.
This means that questions asked by respondents (for
example, about the research or about the topic of the
research) should be politely but firmly rebuffed on the
grounds that too much familiarity should be avoided
and because the respondents’ subsequent answers may
be biased.
This is perfectly valid and appropriate advice from the
vantage point of the canons of structured interviewing
with its quest for standardization and for valid and reliable
data. However, from the perspective of feminism,
when women interview women a wedge is hammered
between them that, in conjunction with the implication
of a hierarchical relationship between the interviewer
and respondent, is incompatible with its values. An
impression of exploitation is created, but exploitation
of women is precisely what feminist social science seeks
to fight against. Oakley (1981) found in her research on
childbirth that she was frequently asked questions by
Structured interviewing 229
her respondents. It was these questions that typified the
problems of being a feminist interviewing women.
The dilemma of a feminist interviewer interviewing
women could be summarised by considering the
practical application of some of the strategies
recommended in the textbooks for meeting
interviewees’ questions. For example, these advise that
such questions as ‘Which hole does the baby come out
of?’ ‘Does an epidural ever paralyse women?’ and ‘Why
is it dangerous to leave a small baby alone in the house?’
should be met with such responses from the interviewer
as ‘I guess I haven’t thought enough about it to give a
good answer right now’, or ‘a head-shaking gesture
which suggests “that’s a hard one”’ (Goode and Hatt
[1952: 198]). (Oakley 1981: 48)
Such advice still appears in textbooks concerned with
survey research. For example, Groves et al. (2004: 283)
supply the following advice:
1. Interviewers should refrain from expressing views
or opinions on the topics covered by the survey
instrument.
2. Interviewers should refrain from presenting any
personal information that might provide a bias for
inferring what their preferences or values might be
that are relevant to the content of the interview.
3. Although a little informal chatting about neutral
topics, such as the weather or pets, may help to free
up communication, for the most part, interviewers
should focus on the task.
This is in fact good advice from the point of view of
reducing error that might arise from the interviewer
influencing or biasing the interviewee’s replies. As such,
it is likely to reduce error arising from the influence of
the interviewer.
Oakley’s point is that to act according to such canons
of textbook practice would be irresponsible for a feminist
in such a situation. It was this kind of critique of structured
interviewing and indeed of quantitative research
in general that ushered in a period in which a great
many feminist social researchers found qualitative research
more compatible with their goals and norms. In
terms of interviewing, this trend resulted in a preference
for forms of interviewing such as unstructured and semistructured
interviewing and focus groups. These will be
the focus of later chapters. However, as noted in Chapter
2, there has been some softening of attitudes towards the
role of quantitative research among feminist researchers.
For example, Walby and Myhill (2001) have shown how
surveys of violence against women that are dedicated to
uncovering such violence (rather than general crime
surveys like the BCS) reveal higher levels than are often
thought to occur. By paying greater attention to issues
like greater privacy in the interview and special training
in sensitive interviewing, dedicated surveys in some
countries have proved highly instructive about the causes
and incidence of violence against women. Such research,
which is based on structured interviewing, would not
seem to be inconsistent with the goals of most feminist
researchers and indeed may be of considerable significance
for many women. Nonetheless, there is still a
tendency for qualitative research to remain the preferred
research strategy for many feminist researchers.
Key points
● The structured interview is a research instrument that is used to standardize the asking and often the
recording of answers in order to keep interviewer-related error to a minimum.
● The structured interview can be administered in person or over the telephone.
● It is important to keep to the wording and order of questions when conducting survey research by
structured interview.
● While there is some evidence that interviewers’ characteristics can influence respondents’ replies,
the findings of experiments on this issue are somewhat equivocal.
● Response sets can be damaging to data derived from structured interviews and steps need to be
taken to identify respondents exhibiting them.
● The structured interview symbolizes the characteristics of quantitative research that feminist
researchers find distasteful: in particular, the lack of reciprocity and the taint of exploitation.
Structured interviewing230
Questions for review
The structured interview
● Why is it important in interviewing for survey research to keep interviewer variability to a minimum?
● How successful is the structured interview in reducing interviewer variability?
● Why might a survey researcher prefer to use a structured rather than an unstructured interview
approach for gathering data?
● Why do structured interview schedules typically include mainly closed questions?
Interview contexts
● Are there any circumstances in which it might be preferable to conduct structured interviews with
more than one interviewer present?
● ‘Given the lower cost of telephone interviewing compared to face-to-face interviews, the former is
generally preferable.’ Discuss.
Conducting interviews
● Prepare an opening statement for a study of manual workers in a firm, to which access has already
been achieved.
● To what extent is rapport an important ingredient of structured interviewing?
● How strong is the evidence that question order can significantly affect answers?
● How strong is the evidence that interviewers’ characteristics can significantly affect answers?
● What is the difference between probing and prompting? How important are they and what dangers
are lurking with their use?
Problems with structured interviewing
● What are response sets and why are they potentially important?
● What are the main issues that lie behind the critique of structured interviewing by feminist
researchers?
Online Resource Centre
www.oxfordtextbooks.co.uk/orc/brymansrm4e/
Visit the Online Resource Centre that accompanies this book to enrich your understanding of
structured interviewing. Consult web links, test yourself using multiple choice questions, and gain
further guidance and inspiration from the Student Researcher’s Toolkit.