Asia Pacific Public Relations Journal, 6(1), 1–34.
1
Media content analysis: Its uses; benefits and best
practice methodology
Jim Macnamara
University of Technology Sydney
The ‘power’ of media
Mass media are believed to cause violence, sexual promiscuity and contribute to
discrimination against women. Media advertising is used to sell products and services. News
in leading media has been shown to significantly affect stock prices; lead to corporate
collapses; cause falls in sales of products; result in the resignation of senior office-holders –
even bring down Presidents. Further information on the effects of mass media is provided in
Macnamara (2003), Mass Media Effects: A Review of 50 Years of Media Effects Research.
Sociologists have been interested in mass media content since the early 20th
century, starting
with Max Weber who saw media content as a means of monitoring the ‘cultural temperature’
of society (Hansen, Cottle, Negrine & Newbold, 1998, p. 92).
Media content analysis – an overview
Media content analysis is a specialized sub-set of content analysis, a well-established
research methodology. Neuendorf (2002) describes content analysis as “the primary messagecentred
methodology” (p. 9) and cites studies such as Riffe and Freitag (1997) and Yale and
Gilly (1988) which “reported that in the field of mass communication research, content
analysis has been the fastest-growing technique over the past 20 years or so” (Neuendorf,
2002, p.1). Riffe and Freitag (1997) found that the number of content analyses published in
Journalism & Mass Communication Quarterly increased from 6.3% of all articles in 1971 to
34.8% in 1995 – nearly a six-fold increase. Fowler (as cited in Neuendorf (2002) reported
that by the mid-1980s over 84% of masters level research methods courses in journalism in
the US included content analysis (p. 27)
Content analysis is used to study a broad range of ‘texts’ from transcripts of interviews and
discussions in clinical and social research to the narrative and form of films, TV programs
and the editorial and advertising content of newspapers and magazines.
Media content analysis was introduced as a systematic method to study mass media by
Harold Lasswell (1927), initially to study propaganda.
Media content analysis became increasingly popular as a research methodology during the
1920s and 1930s for investigating the rapidly expanding communication content of movies.
In the 1950s, media content analysis proliferated as a research methodology in mass
communication studies and social sciences with the arrival of television. Media content
analysis has been a primary research method for studying portrayals of violence, racism and
women in television programming as well as in films.
Lasswell, Lerner and Pool (1952) said: “… content analysis operates on the view that verbal
behaviour is a form of human behaviour, that the flow of symbols is a part of the flow of
Asia Pacific Public Relations Journal, 6(1), 1–34.
2
events, and that the communication process is an aspect of the historical process … content
analysis is a technique which aims at describing, with optimum objectivity, precision, and
generality, what is said on a given subject in a given place at a given time (p. 34).
Lasswell’s better known statement which succinctly encapsulates what media content
analysis is about, published in 1948, (as cited in Shoemaker and Reese, 1996), describes it as:
Who says what
through which channel
to whom
with what effect (p.12).
A widely used definition of content analysis which illustrates the early focus on quantitative
analysis was provided by Berelson (1952) who described it as a “research technique for the
objective, systematic and quantitative description of the manifest content of communication
(p. 18). While it remains oft-quoted, this definition has been found wanting in several
respects. First, the word “objective” is disputed by researchers including Berger and
Luckman (1966) in their classic text, The Social Construction of Reality, in which they point
out that even the most scientific methods of social research cannot produce totally objective
results. Specifically in relation to media content, they point out that media texts are open to
varied interpretations and, as such, analysis of them cannot be objective. Also, some criticize
the definition as restrictive, pointing out that latent as well as manifest content can be
analysed. But, mostly, the early approach to content analysis was criticized because of its
focus on basic quantitative elements and an inherent assumption that quantitative factors
indicated likely social impact.
Other definitions of content analysis include:
 “Content analysis is any research technique for making inferences by systematically and
objectively identifying specified characteristics within text” (Stone, Dunphy, Smith &
Ogilvie, 1996, with credit given to Holsti, p. 5);
 In more contemporary times, Weber (1990) says: “Content analysis is a research method
that uses a set of procedures to make valid inferences from text” (p. 9);
 Berger (1991) says: “Content analysis … is a research technique that is based on
measuring the amount of something (violence, negative portrayals of women, or
whatever) in a representative sampling of some mass-mediated popular form of art” (p.
25);
 Neuman (1997) lists content analysis as a key non-reactive research methodology (i.e.
non-intrusive) and describes it as: “A technique for gathering and analysing the content of
text. The ‘content’ refers to words, meanings, pictures, symbols, ideas, themes, or any
message that can be communicated. The ‘text’ is anything written, visual, or spoken that
serves as a medium for communication” (pp. 272–273);
 Kimberley Neuendorf (2002) is one of the most prominent contemporary researchers
using, teaching (at Cleveland State University) and writing about media content analysis.
She provides this definition: “Content analysis is a summarizing, quantitative analysis of
messages that relies on the scientific method … and is not limited as to the types of
variables that may be measured or the context in which the messages are created or
presented”. Noteworthy about Neuendorf’s definition is that she argues that media
content analysis is quantitative research, not qualitative, and she strongly advocates use of
Asia Pacific Public Relations Journal, 6(1), 1–34.
3
scientific methods “including attention to objectivity-intersubjectivity, a priori design,
reliability, validity, generalisability, replicability, and hypothesis testing” (p. 10).
Neuendorf argues that qualitative analysis of texts is more appropriately described and
categorized as rhetorical analysis, narrative analysis, discourse analysis, structuralist or
semiotic analysis, interpretative analysis or critical analysis (pp. 5-7). However, she
acknowledges that “with only minor adjustment, many are appropriate for use in content
analysis as well”. In The Content Analysis Guidebook, Neuendorf discusses an
“integrative” model of content analysis and notes that a range of methodologies can be
used for text analysis, even though she maintains a narrow definition of content analysis
(p. 41);
 Shoemaker and Reese (1996) are other prominent authors on media content analysis.
They do not fully support Neuendorf’s strict interpretation of content analysis as
quantitative research only. Shoemaker and Reese categorize content analysis into two
traditions – the behaviourist tradition and the humanist tradition. The behaviourist
approach to content analysis is primarily concerned with the effects that content produces
and this approach is the one pursued by social scientists. Whereas the behaviourist
approach looks forwards from media content to try to identify future effects, the humanist
approach looks backwards from media content to try to identify what it says about society
and the culture producing it. Humanist scholars draw on psychoanalysis and cultural
anthropology to analyse how media content such as film and television drama reveal
‘truths’ about a society – what Shoemaker and Reese term “the media’s symbolic
environment” (pp. 31–32). This dual view of the media also helps explain the age-old
debate over whether mass media create public opinion, attitudes and perceptions (effects)
or reflect existing attitudes, perceptions and culture. Most researchers agree that, with
limitations, mass media do both. Shoemaker and Reese say that social scientists taking a
behaviourist approach to content analysis rely mostly on quantitative content analysis,
while humanist approaches to media content tend towards qualitative analysis. They also
note that social scientists may use both types of research as discussed in the following.
Berelson (1952) suggested five main purposes of content analysis as follows:
 To describe substance characteristics of message content;
 To describe form characteristics of message content;
 To make inferences to producers of content;
 To make inferences to audiences of content;
 To predict the effects of content on audiences.
Carney (as cited in Neunendorf, 2002) broadly agreed with this view summarizing the three
main uses of content analysis as (a) descriptive; (b) hypothesis testing and (c) facilitating
inference (p. 52).
Neuendorf (2002) points out that inferences cannot be made as to producers’ intent or
audiences’ interpretation from content analysis alone, arguing that an integrated approach is
required involving use of content analysis with other research such as audience studies.
However, Neuendorf supports Carney’s view of media content analysis as useful for
“facilitating” inference even though it cannot directly prove it and, further, Neuendorf adds
that content analysis has some predictive capabilities as well as other specialist uses.
Neuendorf concludes that there are four main approaches to and roles of content analysis:
 Descriptive;
 Inferential;
Asia Pacific Public Relations Journal, 6(1), 1–34.
4
 Psychometric; and
 Predictive (p. 53).
While psychometric refers to specialized medical and psychoanalytic uses of content analysis
for interpreting the text of patient interviews or statements, the three other approaches are
highly relevant to a range of applications. The first and most basic role, descriptive, provides
insights into the messages and images in discourse and popular culture represented in mass
media. The inferential and predictive roles of content analysis, even though they are
‘facilitating’ rather than conclusive, allow researchers to go further and explore what media
content says about a society and the potential effects mass media representations may have on
audiences.
However, the reliability of media content analysis for description of mediated discourses, and
particularly for drawing inferences or making predictions concerning likely effects of these
mediated discourses, depends on the methodology employed.
Key methodological decisions and considerations in media content analysis are discussed in
the following.
Quantitative v qualitative content analysis
Shoemaker and Reese (1996) note that media content is characterized by a wide range of
phenomena including the medium, production techniques, messages, sources quoted or
referred to, and context, and they say that the task of content analysis is “to impose some sort
of order on these phenomena in order to grasp their meaning.” They continue: “Part of this
ordering process consists of singling out the key features that we think are important and to
which we want to pay attention. Researchers approach content in different ways, using
different conceptual and methodological tools” (p. 31).
Quantitative content analysis collects data about media content such as topics or issues,
volume of mentions, ‘messages’ determined by key words in context (KWIC), circulation of
the media (audience reach) and frequency. Quantitative content analysis also should consider
media form (eg. visual media such as television use more sophisticated semiotic systems than
printed text and, thus, are generally regarded as having greater impact). Neuendorf (2002)
says: “What’s important is that both content and form characteristics ought to be considered
in every content analysis conducted. Form characteristics are often extremely important
mediators of the content elements” (p. 24).
While Neuendorf argues that media content analysis is quantitative only, Shoemaker and
Reese’s categorization of content analysis into humanist and behaviourist traditions indicates
that content analysis can be undertaken using both approaches. They say: “Behavioural
content analysis is not always or necessarily conducted using quantitative or numerical
techniques, but the two tend to go together. Similarly, humanistic content study naturally
gravitates towards qualitative analysis.” Shoemaker and Reese further note: “Reducing large
amounts of text to quantitative data … does not provide a complete picture of meaning and
contextual codes, since texts may contain many other forms of emphasis besides sheer
repetition” (p. 32).
Researchers who advocate analysing latent as well as manifest content as a way of
understanding meanings of texts integrate qualitative and quantitative message analysis.
Media researchers Newbold et al. (2002) note: “The problem [with quantitative content
analysis] is the extent to which the quantitative indicators are interpreted as intensity of
meaning, social impact and the like. There is no simple relationship between media texts and
Asia Pacific Public Relations Journal, 6(1), 1–34.
5
their impact, and it would be too simplistic to base decisions in this regard on mere figures
obtained from a statistical content analysis” (p. 80).
In simple terms, it is not valid to assume that quantitative factors such as size and frequency
of media messages equate to impact. Nor is it valid to assume that these quantitative factors
are the only or even the main determinants of media impact.
Neuman (1997), in a widely used text on social research methodology, comments on the
quantitative-qualitative dichotomy in content analysis: “In content analysis, a researcher uses
objective and systematic counting and recording procedures to produce a quantitative
description of the symbolic content in a text” but he adds “there are qualitative or
interpretative versions of content analysis”. Neuman notes: “Qualitative content analysis is
not highly respected by most positivist researchers. Nonetheless, feminist researchers and
others adopting more critical and interpretative approaches favour it” (p. 273).
Newbold et al. (2002) note that quantitative content analysis “has not been able to capture the
context within which a media text becomes meaningful” (p. 84) and advocate attention to
qualitative approaches as well. Proponents of qualitative text analysis point out factors that
have a major bearing on audience interpretation and likely effects, include:
 Prevailing perceptions of media credibility (e.g. a report in a specialist scientific or
medical journal which will have greater credibility than a report on the same subject in
popular press);
 Context (e.g. a health article published or broadcast during a disease outbreak will be
read differently than at other times);
 Audience characteristics such as age, sex, race, ethnicity, education levels and socioeconomic
position which will all affect ‘readings’ of media content.
Qualitative content analysis examines the relationship between the text and its likely audience
meaning, recognizing that media texts are polysemic – i.e. open to multiple different
meanings to different readers – and tries to determine the likely meaning of texts to
audiences. It pays attention to audience, media and contextual factors – not simply the text.
Accordingly, qualitative content analysis relies heavily on researcher ‘readings’ and
interpretation of media texts. This intensive and time-consuming focus is one of the reasons
that much qualitative content analysis has involved small samples of media content and been
criticized by some researchers as unscientific and unreliable.
In summary, quantitative content analysis can conform to the scientific method and produce
reliable findings. Qualitative content analysis is difficult and maybe impossible to do with
scientific reliability. But qualitative analysis of texts is necessary to understand their deeper
meanings and likely interpretations by audiences – surely the ultimate goal of analysing
media content. So a combination of the two seems to be the ideal approach.
Within mass media and communication studies, most media researchers do not draw the
sharp definitional distinctions that Neuendorf does between text, content and discourse
analysis. Media researchers and academics such as Newbold et al. (2002), Gauntlett (2002)
and Curran (2002) refer to quantitative and qualitative content analysis and most view the
fields as complementary and part of a continuum of analysing texts to try to determine their
likely meanings to and impact on audiences.
Asia Pacific Public Relations Journal, 6(1), 1–34.
6
Hansen et al. (1998) comment: “… rather than emphasizing its alleged incompatibility with
other more qualitative approaches (such as semiotics, structuralist analysis, discourse
analysis) we wish to stress … that content analysis is and should be enriched by the
theoretical framework offered by other more qualitative approaches, while bringing to these a
methodological rigour, prescriptions for use, and systematicity rarely found in many of the
more qualitative approaches” (p. 91).
Shoemaker and Reese’s (1996) categorization of a humanist approach which studies media
content as a reflection of society and culture, and a behaviourist approach which analyses
media content with a view to its likely effects, is also useful in understanding how media
content analysis should be conducted. Any research exploring media content for both what
influence it may have on and for how it might reflect society – i.e. employing both
behaviourist and humanist traditions – should use a combination of quantitative and
qualitative content analysis.
It can be concluded from Hansen et al. (1998), Shoemaker and Reese (1996) and others cited,
that a combination of quantitative and qualitative content analysis offers the best of both
worlds and, further, that a combination of quantitative and qualitative content analysis
methodologies is necessary to fully understand the meanings and possible impacts of media
texts.
It is important to note that some researchers reject altogether the view that the meanings of
texts can be accessed through analysis of the texts (Newbold, et al., 2002, p. 16). Certainly,
researchers using content analysis need to be cautious in making predictions of likely
audience effects, as already noted. However, while audience research remains a primary
approach to gain direct insights into audience perceptions, it too faces methodological
problems. Respondents forget where they received information from (e.g. many respondents
in interviews and group discussions say “someone told me” when, in fact, they received the
information through mass media). Others lie – perhaps not intentionally, but often people do
not want to admit that they read some ‘trash’ magazine or watched daytime television.
Furthermore, respondents talking directly to a researcher sometimes say what they think the
researcher wants to hear, referred to as ‘response generation’ (interviews and ethnographic
research methods are affected by researcher intrusion). Audience studies also have their own
problematic issues with sample, question construction and interpretation of responses.
Media content analysis is a non-intrusive research method that allows examination of a wide
range of data over an extensive period to identify popular discourses and their likely
meanings. Another benefit of content analysis is that it can be conducted frequently (eg.
every month), whereas audience research such as large-scale surveys are, because of their
cost and time taken, restricted to once per year or every few years. Shoemaker and Reese
(1996) propose that “media content and media effects [i.e. audience] research can be
combined to help our understanding of the role that the mass media play in society” and also
to understand societal attitudes (p. 256).
Human v computer coding
Media content analysis increasingly uses computer programs. Computer software is applied
at two levels:
1. For storing, analysing and reporting research data such as coding and notations by
researchers (including constructing tables, charts and graphs); and
Asia Pacific Public Relations Journal, 6(1), 1–34.
7
2. In some cases, for automatic scanning of texts and identification and coding of words and
phrases. This stage can lead to automation of the entire process of coding and analysis, or
provide partial automation with a combination of computer scanning and coding along
with human notations manually entered into the program.
In the first level, texts are read and coded by humans (usually trained researchers) and
computer software programs are used as tools to assist in the analysis in the same way they
are used to analyse the results of surveys and other research. Programs commonly used at this
level are databases for data storage; SPSS for statistical analysis; Excel for tabulation of data
and calculations such as pivot tables; and Excel or graphics programs for generation of
charts. Also, a range of specialist commercial media content analysis systems are used for
storing, analysing and reporting media analysis data such as CARMA®
(Computer Aided
Research and Media Analysis), Delahaye (now part of AB Observer/Bacon’s Information
Services), Echo Research, IMPACT™ and Millward Brown Précis. Most are database
programs with customized data entry screens and fields created for the specialized needs of
media content analysis. Many of these proprietary programs have specialized features such as
inbuilt media databases providing circulation and audience statistics and sometimes
demographic data which enriches and speeds up media content analysis.
At the second level, computer software automatically conducts either all or a large part of
content analysis including scanning texts using Optical Character Recognition (OCR)
technology and matching words and phrases in texts with ‘dictionaries’ of key words and
phrases previously set up in the software program. Some programs do all coding
automatically, while others allow the researcher to enter notations and comments and tag or
link these to relevant articles. Software programs such as General Inquirer developed at
Harvard University in the 1960s; NUD*IST; NVIVO; TextSmart by SPSS; INTEXT;
TextAnalyst; TEXTPACK 7.0, CATPAC, DICTION 5.0, DIMAP and VBPro perform a
variety of content analysis functions. Mayring (2003) also cites experience using two German
software programs for qualitative text analysis, MAXqda (MAX Qualitative Data Analysis
for Windows) and ATLASti.
A number of social researchers claim that computers are not relevant to media content
analysis, suggesting that it must be done manually by detailed human study (Newbold et al.,
2002, p. 84). This claim, per se, is Luddite, or more likely confuses the two levels of
computerization in media content analysis. Few would argue that using a computer database,
spreadsheet, or a specialized program to store and analyse data entered by researchers is
inconsistent with the scientific method. It is most likely that use of computers enhances
accuracy of analysis.
However, Neuendorf (2002) says that “the notion of the completely ‘automatic’ content
analysis via computer is a chimera … The human contribution to content analysis is still
paramount” (p. 40).
Most content analysts agree with this viewpoint based on professional experience. Automated
(fully computerized) content analysis makes mostly arbitrary associations between words and
phrases. While neurolinguistic software programming and Artificial Intelligence (AI) systems
in which computers are purported to ‘learn’ to interpret the way humans do are developing,
such programs remain unreliable for subtle and sophisticated interpretational work and their
analysis is simplistic. Neuman (1997, p. 275) gives the example of the word ‘red’ and how it
can be used with multiple nuances that are not visible to a computer:
“I read a book with a red cover that is real red herring. Unfortunately, its publisher drowned in red
ink because the editor couldn’t deal with the red tape that occurs when a book is red hot. The
Asia Pacific Public Relations Journal, 6(1), 1–34.
8
book has a story about a red fire truck that stops at red lights only after the leaves turn red. There
is also a group of Reds who carry red flags to the little red schoolhouse. They are opposed by redblooded
rednecks who eat red meat and honour the red, white and blue …”
Machine coding of the above text would be very unlikely to identify the range of meanings of
the word ‘red’.
A further disadvantage of automated computer coding is that it results in what Neuendorf
(2002) terms “black box measurement”. Most software programs do not reveal the details of
their measures or how they construct their scales and indexes. The researcher enters text into
“a veritable black box from which output emerges” (p. 129). This is inconsistent with the
scientific method of research which requires that full information is disclosed on how results
were obtained. Also, it limits replicability as other researchers cannot conduct similar studies
unless they use the same software program and, even then, key functions and calculations are
hidden within the ‘black box’.
When content analysis is conducted across multiple languages and cultures, such as for global
or non-western media studies, the problems of machine coding become even more marked, as
most automated coding systems work with English language text only and computer
translations are unreliable except for the most rudimentary applications.
Furthermore, and perhaps most important of all, computers cannot consider the context of
content; they only view the text which can result in narrow incomplete interpretations.
However, computers can clearly support quantitative and qualitative content analysis by
serving as a repository for coding data and provide powerful tools for analyzing and reporting
research.
When human coding is used, the software employed for data storage and analysis is not
materially significant to the research, provided a reliable program is used. Methodology is
more important, as is the training of the coders who need to conduct the analysis in
accordance with strict criteria.
Quantitative content analysis methodology
Quantitative media content analysis should be conducted in accordance with ‘the scientific
method’, Neuendorf (2002) argues, involving the following elements.
 Objectivity/intersubjectivity
A major goal of any scientific investigation must be to provide a description or
explanation of a phenomenon in a way that avoids or minimizes the biases of the
investigator and, while true objectivity may not be possible, it should strive for
consistency and what scholars term intersubjectivity (Babbie, 1986, p. 27; Lindlof, 1995
as cited in Neuendorf, 2002, p. 11). Objectivity, or intersubjectivity, is maximised by
several techniques, most notably selection of a representative sample. (See ‘Media
content sample’)
 A priori design
Media content analyses often fail the test of objectivity/intersubjectivity because
researchers construct the list of issues and messages being studied as they go, adding
issues and messages as they find them in articles, arguing that they need to begin media
content analysis before they can accurately identify the issues and messages contained in
the content.
Asia Pacific Public Relations Journal, 6(1), 1–34.
9
A deductive scientific approach to research design requires that “all decisions on
variables, their measurement, and coding rules must be made before the observation
begins” (Neuendorf, 2002, p. 11). An inductive approach which measures variables after
they have been observed leads to major biases and invalidity in a study. In effect, it
allows issues, topics and messages to be added to the list of those tracked at the whim of
the researcher, and those added during a study may have been present from the outset but
not observed, leading to inaccuracies in data.
Kuhn’s (1970) observation in his seminal work on paradigms that the scientific
requirement for deduction to be based on past research, theories and bodies of evidence is
self-limiting and does not foster innovation is noted. Equally, the view of some media
researchers that it is difficult to identify the variables for study (issues and messages in
media content analysis) before they begin analysis of media content has some basis.
However, this apparent dichotomy can be overcome. Exploratory work can and should be
done before a final coding scheme is established for content analysis to identify the issues
and messages appropriate for study. Neuendorf (2002) says: “Much as a survey
researcher will use focus groups or in-depth interviewing (qualitative techniques) to
inform his or her questionnaire construction, so may the content analyst use in-depth,
often contemplative and incisive observations from the literature of critical scholars.”
Furthermore, Neuendorf suggests that media content analysts can “immerse himself or
herself in the world of the message pool” by conducting “a qualitative scrutiny of a
representative subset of the content to be examined” – i.e. conduct preliminary reading of
texts within the field (pp. 102–103).
Thus, a grounded theory approach, as explained by Glaser and Strauss (1967) and Strauss
and Corbin (1990), can be applied to identify issues and messages appropriate for analysis
through preliminary reading of existing research literature in the field and reading of a
sub-sample of the media content to be studied.
In media content analysis, a priori design is operationalised in a Coding System. A key
component of a Coding System is a comprehensive written Code Book or Coding List.
This contains the list of variables (units of analysis) to be researched and provides
researchers involved in the project with a consistent framework for conducting the
research.
Content analysis should involve examination of multiple variables (i.e. multivariate
analysis) – not be a simplistic rating of a single variable such as positive, negative or
neutral which is univariate and tells us little about the likely meaning and effects of a text.
The primary units of content analysis (variables) are messages expressed as words or
phrases – e.g. ‘violent’, ‘leader’, ‘funding should be increased’, etc. The Coding List
should establish all the messages (both positive and negative) that are relevant. In
addition, the Coding List may establish certain categories of issues or topics, and may
further identify names of certain sources (individuals or organizations) to be analysed in
association with issues or messages.
All positive messages identified for analysis should be equally matched with their
corresponding negative form, and vice versa, to ensure balance. For instance, if ‘boys in
schools are aggressive and violent’ is analysed, the oppositional positive message ‘boys
in schools are not aggressive or violent or are passive and non-violent’ should equally be
analysed in the research. Failure to apply equal vigour to analysing oppositional messages
can seriously distort and invalidate a study.
Asia Pacific Public Relations Journal, 6(1), 1–34.
10
As well as the specific subject-orientated issues and messages to be analysed as part of a
study, a content analysis coding system should also allow coding of other key variables
that determine the likely impact of a text. In specialist content analysis software
programs, these variables are often built in as standard ‘fields’. If not, they should be
established in the Coding List. Typical variables identified by researchers as important
and required for Best Practice content analysis include:
 Media weighting or categorization to allow high circulation, high rating or highly
influential media to be scored higher than small, less important media;
 Prominence to record impact factors such as page number or order in an electronic
media bulletin and use of photos or visuals;
 Positioning such as headline mentions, first paragraph mentions, prominent mentions,
or passing mentions and ‘share of voice’ in articles;
 Size of articles or length of radio and TV segments;
 Sources quoted including the balance of supportive and opposing sources cited in the
texts and their position/credibility (e.g. an official government authority or known
expert is likely to be more credible than a little known unqualified source).
Samples of coding lists and coding forms are published on the Cleveland State
University, Ohio Web site as an adjunct to The Content Analysis Guidebook authored by
Kimberley Neuendorf (2002) and can be downloaded free of charge from
(http://academic/csuohio.edu/kneuendorf/content/hcoding/patcball/html).
 Intercoder reliability
A rigorous ‘scientific’ approach to media content analysis to gain maximum reliability
requires that two or more coders are used – at least for a sample of content (called the
reliability sub-sample). Even when a primary researcher conducts most of the research, a
reliability sub-sample coded by a second or third coder is important to ensure that, in the
words of Tinsley and Weiss (1975), “obtained ratings are not the idiosyncratic results of
one rater’s subjective judgement” (p. 359).
Neuendorf (2002) says: “There is growing acknowledgement in the research literature
that the establishment of intercoder reliability is essential, a necessary criterion for valid
and useful research when human coding is employed.” Neuendorf adds: “This has
followed a period during which many researchers were less than rigorous in their
reliability assessment” (p. 142). Reporting on an analysis of 486 content analysis studies
published in Journalism and Mass Communication Quarterly from 1971 through 1995,
Riffe and Freitag (1997) found that only 56% of these reported intercoder reliability and
that most failed to report reliability variable by variable, which is recommended. Even as
recently as 2001, a study of 200 content analyses by Lombard, Synder-Duch and Bracken
(2003, 2004) found that only 69% discussed intercoder reliability and only 41% reported
reliability for specific variables.
A number of statistical formulae have been developed for measuring intercoder
reliability. Researchers propose that coding between coder pairs and multiple coders
should be compared at two levels: (a) agreement and (b) co-variation (Neuendorf, 2002,
p. 144). Agreement is a simple comparison of the level of agreement between the coders’
scores and ratings. Co-variation assesses whether, when scores do vary as they no doubt
will in human coding, they go up and down together – i.e. whether there is consistency or
a high level of variance. Bartko and Carpenter (1976) note that in clinical and other
psychological research, researchers report co-variation and not simple agreement, while
Asia Pacific Public Relations Journal, 6(1), 1–34.
11
in communication and business research simple agreement only is reported. Neuendorf,
citing Tinsley and Weiss (1975), concludes: “The best situation, of course, would be one
in which coded scores are shown to have both high agreement and high co-variation”
(Neuendorf, 2002, p. 144).
In terms of specific formulae to use, Lombard et al. (2003) report:
[T]here are few standard and accessible guidelines available regarding the appropriate procedures
to use to assess and report intercoder reliability, or software tools to calculate it. As a result, it
seems likely that there is little consistency in how this critical element of content analysis is
assessed and reported in published mass communication studies. Following a review of relevant
concepts, indices, and tools, a content analysis of 200 studies utilizing content analysis published
in the communication literature between 1994 and 1998 is used to characterize practices in the
field. The results demonstrate that mass communication researchers often fail to assess (or at least
report) intercoder reliability and often rely on percent agreement, an overly liberal index. Based
on the review and these results, concrete guidelines are offered regarding procedures for
assessment and reporting of this important aspect of content analysis.
Lombard et al. (2004) note that there are “literally dozens” of different measures or
indices of intercoder reliability. Popping (1988) reported 39 different “agreement
indices”. However, Lombard et al., Neuendorf (2002) and a number of other researchers
agree that the following indices are the most reliable and important:
 Per cent agreement (basic assessment);
 Scott’s pi ();
 Cohen’s kappa ();
 Spearman’s rho;
 Pearson’s correlation coefficient (r);
 Krippendorf’s apha; and
 Lin’s concordance correlation coefficient (rc).
According to professional and academic content analysts, “… the reliability sub-sample
should probably never be smaller than 50 and should rarely need to be larger than about
300” (Neuendorf, 2002, p. 159).
‘Blind coding’ should be conducted by coders of the intercoder reliability sub-sample (i.e.
neither coder should see coding of the others prior to completion of the assessment) to
minimize what researchers term ‘demand characteristic’ – a tendency of participants in a
study to try to provide what the primary researcher wants or to skew results to meet a
desired goal.
Intercoder reliability should ideally be assessed for each of the variables studied – in the
case of content analysis, for all messages and issues analysed. Thus, in analyses with a
wide range of issues and messages, intercoder reliability assessment is a time-consuming
and challenging process.
The relatively complex formulae for calculating these reliability indices are provided in
Neuendorf (2002). Manual calculation requires familiarity with statistics and considerable
time – no doubt the reason that most content analyses do little more than assess percent
agreement, if that, as reported by Riffe and Freitag (1997 and Lombard et al. (2003).
However, a number of software programs help calculate intercoder reliability assessment,
including statistics programs such as SPSS which can assess Cohen’s kappa () and
Asia Pacific Public Relations Journal, 6(1), 1–34.
12
Simstat from Provalis Research which can calculate a number of intercoder reliability
statistics. Also, specialist software programs have been and are being developed for this
purpose including Popping’s AGREE (1984) and Krippendorf’s Alpha 3.12a, although
the latter is a beta (test) program and not available widely (Lombard et al., 2004). A US
company, SkyMeg Software, in consultation with academics from Cleveland State
University, has developed PRAM (Program for Reliability Assessment of Multiple
Coders) which can calculate reliability statistics for each of the most recommended
indices. PRAM is still in development and an academic version alpha release 0.4.4
available as at January 2004 was found to contain some minor ‘bugs’ and ‘clunky’
features. However, release notes on the program state that all coefficients have been
tested and verified by Neuendorf’s students at Cleveland State University. The program,
which analyses coding data exported to Microsoft Excel®
spreadsheets, provides
reliability statistics for each variable assessed on a scale of 0 – 1 where one is 100%
agreement or co-variation (SkyMeg Software, 2003).
Neuendorf (2002) notes in relation to coder reliability that “most basic textbooks on
research methods in the social sciences do not offer a specific criterion or cut-off figure
and those that do report a criterion vary somewhat in their recommendations” (p. 143).
However, Neuendorf cites Ellis (1994) as offering a “widely accepted rule of thumb”.
Ellis states that correlation coefficients exceeding 0.75 to 0.80 indicate high reliability (p.
91). In relation to specific statistics, Frey, Botan and Kreps (2000) declare 70%
agreement (0.70) is considered reliable. Popping (1988) suggests 0.80 or greater is
required for Cohen’s kappa which he cites as the optimal (ie. strictest) measure, while
Banerjee, Capozzoli, McSweeney and Sinha (1999) propose that a 0.75 score for Cohen’s
kappa indicates excellent agreement beyond chance. Riffe, Lacy and Fico (1998), without
specifying the type of reliability coefficient, recommend high standards and report that
content analysis studies typically report reliability in the 0.80 to 0.90 range.
As Neuendorf (2002) notes, it it is clear from a review of work on reliability of content
analysis that reliability coefficients of 0.80 or greater are acceptable to all and 0.75 is
acceptable in most situations. Furthermore, Neuendorf notes that the ‘beyond chance’
statistics such as Scott’s pi and Cohen’s kappa are afforded a more liberal criterion.
A further principle of sound research is that agreement and co-variation rates between
coders, along with details of the intercoder reliability sample, are reported in the research
(Snyder-Duch, et al, 2001; Neuendorf, 2002). Such data should be appended to a content
analysis report, along with the Code Book/Coding list and details of methodology used.
Strategies to maximize agreement and co-variation and, if necessary, address low
agreement or high variation between coders are:
1. Pre-coding training to familiarize all coders with variables such as issues and
messages for analysis and guidelines for classifications and coding;
2. Pilot coding (doing a test first);
3. Review of the Code Book/List and re-briefing to ensure descriptions and instructions
are clear;
4. Retraining if required.
 Validity
Validity of content analysis is achieved through thoroughly understanding the research
objectives, preliminary reading of a sub-set of relevant content (what Neuendorf calls
Asia Pacific Public Relations Journal, 6(1), 1–34.
13
‘immersion in the message pool’), and careful selection of the sample of media content to
be analysed. (See ‘Media content sample’)
 Generalizabilty
Generalizability refers to the extent to which research findings can be applied to and
taken as a measure of the target population generally (in the case of content analysis, the
target population is the total mass media message pool). Generalisability is largely
determined by selection of a representative and sufficiently large sample, as well as the
overall thoroughness of the methodology. (See ‘Media content sample’)
 Replicability
Replicability, the ability and degree of difficulty or otherwise for other researchers to
replicate the research to confirm or challenge the results, is a key criterion for all
scientific research. Replicability is determined by full disclosure of information on
methodology and procedures. In the case of content analysis, this should include the Code
Book/Coding List; coding guidelines and instructions to coders; method of coding used in
the case of human coding; details of any software programs used; and all data supporting
conclusions. As Neuman (1997) notes, a researcher undertaking content analysis
“carefully designs and documents procedures for coding to make replication possible” (p.
274).
Media content sample
Sampling for media content analysis comprises three steps, Newbold et al. (2002) propose:
1. Selection of media forms (i.e. newspapers, magazines, radio, TV, film) and genre (news,
current affairs, drama, soap opera, documentary, and so on);
2. Selection of issues or dates (the period);
3. Sampling of relevant content from within those media (pp. 80–81).
The simplest form of selecting content for analysis is a census – i.e. selection of all units in
the sampling frame. This provides the greatest possible representation. However, a census
may not be possible in some cases – e.g. where a large volume of media coverage has to be
analysed such as a study over many months or years. In such cases, a sample of media
content may be selected. Sampling needs to be conducted in an objective way, ensuring
reliability is maintained.
Typical methods of sampling for media content analysis include:
 Systematic random (selecting every nth
unit from the total population of articles or
advertisements/commercials for study);
 Purposive such as selecting all articles from key media (and not from less important
media. This is valid provided there is some basis for the criteria applied);
 Quota such as selecting a proportion of articles from each of several regions or areas
(either geographic, demographic, psychographic, or subject category);
 Stratified composite samples constructed by randomly selecting units for analysis (articles
or ads) from certain days or weeks over a period.
Asia Pacific Public Relations Journal, 6(1), 1–34.
14
Riffe, Lacy and Fico (1998), Riffe, Lacy and Drager (1996) and Riffe, Lacy, Nagovan and
Burkum (1996) have identified the most accurate sampling methods for analysing weekday
TV news and media publications over a period and report that stratification by month or week
provides the optimum result. However, often a purposive method focusing on the most
relevant media is appropriate.
Editorial or advertising media content can be collected in a number of ways including:
 Reading and manually ‘clipping’ relevant items from newspapers and magazines and
taping electronic media broadcasts;
 Subscribing to a media monitoring service;
 Downloading items from online media sites. However, it should be noted that online
editions often do not contain all printed and broadcast content – e.g. special supplements
and sections may not be available online;
 Online news services such as Factiva, Lexis-Nexis and Dow Jones. It should be noted
similarly that these services often provide a narrow sample of media content, usually from
major newspapers only.
Two methods are used for recording coding: (a) electronic into a computer system and (b)
‘paper coding’. In modern computerized content analysis systems, the Coding List is usually
contained in software menus or screens and coding data may be entered directly into a
computer system. However, many coders still prefer ‘paper coding’ (i.e. writing coding on to
the articles or transcripts or recording coding on a coding form attached to the text.) ‘Paper
coding’ data is later entered into a computer system for analysis.
During coding, issues and messages are identified by either, or a combination of (a) wordmatching
(i.e. an exact match), and (b) presence of acceptable synonyms or similar phrases.
Acceptable synonyms or similar phrases should be identified in guidelines provided to coders
attached to or as part of the Coding List. For example, if ‘participatory’ is a message for
analysis, acceptable synonyms could be ‘joins in activities’, ‘works with others’, ‘takes part’
and ‘engages’. The more comprehensive the Coding List and guidelines to coders, the more
reliable the analysis will be. Coding guidelines should be strictly followed.
Reading and coding for content analysis is a time-intensive process and produces a veritable
‘data mountain’. However, ‘coding’ allows key data about media articles and programs rather
than the full text to be entered into a computer database, providing data reduction and, when a
scientific method has been employed, quantitative analysis can be carried out using
computer-aided statistical and reporting tools.
Qualitative content analysis
Qualitative content analysis can, to some extent, be incorporated within or conducted
simultaneously with quantitative content analysis. For instance, positive and negative words
and phrases can be analysed to identify the tone of text. Also, analysts can record notations
during coding in relation to contextual factors.
However, in many cases, in-depth analysis of selected content using qualitative research
methods is required to fully understand the potential meanings (manifest and latent) for
audiences and likely effects of texts.
Asia Pacific Public Relations Journal, 6(1), 1–34.
15
The precise methodology best used for qualitative message or text analysis is poorly defined.
McKee (2004) notes that “we have a very odd lacuna at the heart of cultural studies of the
media. Textual analysis is the central methodology, and yet we do not have a straightforward
published guide as to what it is and how we do it”. He explains this as partly:
… the ambivalence of cultural studies practitioners towards disciplinarity and institutionalization
[which] lead (sic) to an odd interpretation of our axioms that knowledge is power, that discourses
define reality and that there is no such thing as ‘objective’ knowledge. We know that every
methodology is partial, producing particular kinds of information. Linked with an anti-displinarian
trend, this seems to have led us to refuse to think seriously about our own methodologies. Instead,
we tend towards a kind of ‘transgressive’ methodological approach, where we do whatever takes
our fancy.
McKee adds: “we insist that the specificity of any methodology must be investigated to
reveal the limits to the kinds of knowledge it can produce, and yet our own central
methodology is woefully under investigated, and still largely intuitive”.
Despite this lack of specific guidelines for qualitative text analysis, research procedures for
qualitative text and message analysis are informed by the work of Denzin and Lincoln
(1994); Hijams (1996); Mayring (2000; 2003); Patton (1990; 2002); Robson (1993); and
Silverman (1993) and these can be drawn on to frame a study with reasonable levels of
reliability and validity.
Qualitative message analysis methods applicable to analysis of media content include text
analysis, narrative analysis, rhetorical analysis, discourse analysis, interpretative analysis and
semiotic analysis, as well as some of the techniques used in literary studies such as critical
analysis, according to Hijams (1996).
Within the broad hermeneutic tradition concerned with text analysis, there are two main
strands particularly relevant to qualitative content analysis. The first, narratology, focuses on
the narrative or story-telling within a text with emphasis on meaning that may be produced by
its structure and choice of words. The second draws on semiotics and focuses attention on
signs and sign systems in texts and how readers might interpret (decode) those signs
(Newbold et al., 2002, p. 84).
Semiotics utilizes a number of different approaches, description of which is outside the scope
of this paper other than a broad summary of their essential elements. Two main streams of
semiotics, sometimes referred to as semiology and semiotics, have evolved from the work of
Swiss linguist Ferdinand de Saussure and American Charles Sanders Peirce respectively.
While quantitative content analysis has its complexities and requires considerable statistical
rigor to comply with the requirements of scientific research, as outlined earlier in this chapter,
the coding task in quantitative analysis is predominantly “one of clerical recording”,
according to Potter and Levine-Donnerstein (1999, p. 265). In comparison, they note
“objectivity is a much tougher criterion to achieve with latent than with manifest variables”
as studied in qualitative content analysis. Newbold et al. (2002) warn:
The logic of deconstructing latent meanings, and privileging them over the more obvious
‘manifest’ ones, is questionable, for the audience may not see this latest dimension; the analysis
may be longer than the text. The task is time-consuming, and often tells us what we already know
in a language we don’t understand (p. 249).
Asia Pacific Public Relations Journal, 6(1), 1–34.
16
Newbold et al. go further in warning of the inherent challenges in semiology, the tradition of
semiotics based on theories developed by de Saussure, in the following terms:
The scientific validity of semiology is questionable – in comparison with traditional positivistic
science, at least – for it is not replicable (it is impossible to repeat with exactly the same results). It
is not easy to show that semiology examines the subject it sets out to study … (p. 249).
However, like others, Newbold et al. acknowledge that there are advantages of using
semiology as a tool. “It exposes the ideological, latent meaning behind the surface of texts,
allowing us to grasp the power relations within them” (p. 249).
The essential concepts of semiotics and semiology are that words and images are signs that
‘stand for’ or ‘signify’ something else beyond their obvious manifest meaning and relate to
one another to form codes or code systems – collectives of signs that produce certain
meanings (Newbold et al., 2002, p. 87; Selby & Cowdery, 1995, p. 47).
Early semiotics took a structuralist approach, seeing the meaning of signs as largely fixed and
interpreted according to a system, whereas later post-structuralist influenced semiotics theory
saw signs as interpreted by audiences – often differently to the intentions of the author and
differently between audiences.
Jensen (1995) brought together what he terms an integrated social semiotics theory of mass
communication which draws on structuralist semiotic research as well as more modern poststructuralist
theories of active audience participation in interpretation of mediated meanings.
In other words, elements of both de Saussure influenced semiology and Peirce influenced
semiotics can be applied and each has something to offer to a comprehensive study of mass
media representations.
Newbold et al. (2002) observe: “So in studying media texts … we can use these ideas as they
can provide a way of assessing the meaning production in a text” (p. 87). They cite Van
Zoonen (1994) who explains that semiotic analysis of a media text can begin by identifying
the signs in the text and their dominant characteristics. Then, citing Selby and Cowdery
(1995), they say “these signs can be analyzed as a result of selection and combination”
(Newbold et al., 2002, p. 87).
Images such as photographs and icons are key signs in media texts. For instance, a
photograph of a man holding a baby suggests fatherhood, family commitment and, depending
on how it is composed, gentleness and caring. For instance, a photograph may contains
several signs such as the man cradling the baby’s head in his hand and or gazing at the baby
with a kind and caring expression (signifying love and protection), or holding the baby with
outstretched arms away from his body and peering quizzically at the infant (signifying
confusion and aversion). Road signs and international symbols such as $ representing dollar
or money, © for copyright and  for ‘No’ (as in No Entry or No Smoking) are examples of
icons and symbols that signify meanings beyond themselves. Similarly, audiences routinely
interpret the sign + as denoting the mathematical function of addition and  as
multiplication, while the slightly different  is symbolic of Christianity or the Christian
Church.
In terms of language, Campbell and Pennebaker (2003) and others identify pronouns as key
signifiers of meaning in texts and a focus of qualitative text analysis. Campbell and
Pennebaker investigated the relationship between linguistic style and physical health using
latent semantic analysis to analyse writing samples provided by students and prison inmates.
Campbell and Pennebaker reported that change in the frequency with which participants used
Asia Pacific Public Relations Journal, 6(1), 1–34.
17
pronouns (e.g. I, me, he, she) is the linguistic feature that best predicts improvement in
physical health. Their data showed that flexibility in pronoun use is associated with improved
physical health (pp. 60–65). Over-use of personal pronouns such as I, me and my can also
indicate self-centredness and egotism.
Other key text elements commonly studied in qualitative content analysis are:
 Adjectives used in descriptions (positive and negative) which give strong indications of a
speaker’s and writer’s attitude (e.g. it was ‘disgusting’);
 Metaphors and similes used (e.g. labelling a car a ‘lemon’ or a person a ‘rat’);
 Whether verbs are active or passive voice;
 Viewpoint of the narrator (i.e. first person, second person, third person);
 Tonal qualities such as aggressiveness, sarcasm, flippancy, emotional language;
 Binaries established in texts and how these are positioned and used;
 Visual imagery in text; and
 Context factors such as the position and credibility of spokespersons or sources quoted
which affects meaning taken from the text (e.g. if one message is presented by a high
profile expert it will generally outweigh a non-expert opinion).
Mayring (2000) developed a number of procedures for qualitative text analysis, among which
he says two are central: inductive category development and deductive category application.
Inductive analysis involves working from specific observations of categories and patterns (eg.
issues or messages) to a broad theory or conclusion. Deductive analysis involves working
from a broad theory or general position to specific observations to confirm or disprove the
former (Trochim, 2002). After inductively determining categories, Mayring (2003) says “the
qualitative step of analysis consists in a methodological controlled assignment of the category
to the passage of the text”.
Mayring’s procedures bring some systematic approach to qualitative text analysis. In essence,
his method involves a priori design of the categories – they should not be created as the
analyst goes along – and, importantly, this method requires matching of a category to a
passage of text; not matching of the text to a category. By starting with pre-determined
categories, which by their nature are specific, this increases the systematicity of qualitative
analysis.
Intercoder reliability assessment also should be used with qualitative analysis to assist
reliability and validity, Mayring (2003) recommends, although he notes that more flexible
measures need to be applied. His studies maximized reliability and validity by using “only
trained members of the project team” and he reduced the standard of coder agreement stating
that Cohen’s kappa () of 0.7 would be sufficient.
Mayring (2003) also notes that several computer programs have been developed for
qualitative analysis, but he stresses that these are to “support (not to replace) steps of text
interpretation”. He reported experience using MAXqda (MAX Qualitative Data Analysis for
Windows) (dressing&pehl GbR & Verbi GmbH, 2004).
Sample for qualitative analysis
Sampling for qualitative analysis is not required to meet the statistically valid formulae of
quantitative analysis. Nevertheless, sampling for in-depth qualitative study should not be
simply drawn at the researcher’s whim, and even random methods may not yield useful data
as the purpose of qualitative research is to investigate certain issues or themes in detail.
Random or even representative methods of sampling may not capture the issues or themes
Asia Pacific Public Relations Journal, 6(1), 1–34.
18
which are the subject of qualitative analysis. Miles and Huberman (1994) argue that sampling
strategies for qualitative research should be driven by a conceptual question, not by concern
for “representativeness” (p. 29). They suggest instead, three techniques which can be used
together to yield rich results in qualitative analysis:
1. Selecting apparently typical/representative examples;
2. Selecting negative/disconfirming examples; and
3. Selecting exceptional or discrepant examples (p. 34).
By choosing a combination of typical, disconfirming and exceptional examples for study,
qualitative analysis can explore the boundaries of the data field and identify the range of
views including discordant ones and extremes in various directions, as well as the typical.
While quantitative research has the benefit of yielding empirical data that is generalisable and
representative with a high probability, it reduces research findings to the average or median
position on key questions. Qualitative analysis using the sampling approach identified by
Miles and Huberman allows exploration of discourse at various points within the range.
An overview of the processes of content analysis by Kimberley Neuendorf (2002) is provided
in Figure 1.
Commercial Media Analysis
Commercially, media content analysis has a number of uses and offers significant benefits to
companies, organizations, government agencies and political parties – particularly those that
receive wide media coverage.
In practical terms, organizations receiving a small amount of publicity can review media
coverage using personal observation. But, when multinational companies and large
organizations receive hundreds or even thousands of mentions in mass media, often in a
number of countries and in multiple languages, simple observation cannot provide reliable
understanding of likely outcomes and effects.
Media content analysis is increasingly used commercially because of the two key roles of the
mass media.
Mass media – the world’s most powerful communication channel
While media effects theory is a complex and ongoing field of research, many research studies
show that mass media have significant impact and effects on public awareness, perceptions
and sometimes behaviour such as buying decisions and voting. CEOs, marketers, advertisers
and PR professionals know that mass media are important influences affecting brands,
reputation, corporate image and the success of marketing and communication campaigns. It is
because of this influence that mass media are used for advertising products and services.
Editorial media content also influences readers, viewers and listeners – sometimes even more
than advertising. But unlike advertising, editorial is highly variable in content and format. It
may be critical, promote competitors, or raise issues impacting an organization. And mass
media are also increasingly global. Reports from far corners of the world can impact a share
price, a brand or reputation. So understanding the content of editorial mass media is
increasingly important for organizations involved in public communication.
Asia Pacific Public Relations Journal, 6(1), 1–34.
19
Media Content Analysis Flowchart
Continued over:
Asia Pacific Public Relations Journal, 6(1), 1–34.
20
Figure 1. A flowchart for the typical process of content analysis research (Neuendorf, 2002).
Asia Pacific Public Relations Journal, 6(1), 1–34.
21
Mass media – one of the world’s largest databases
As well as influencing public opinion, mass media reflect opinion and perceptions through
reporting what other people, companies and organizations are saying and doing. Furthermore,
the media report issues and trends, often ‘breaking’ news and setting or framing the agenda of
public debate. (See Macnamara, 2003 for more information on effects of mass media.)
Media analysis therefore provides two types of research:
1. Evaluation to measure effectiveness of an organization’s communication (PR) to and
through the media including audience reach, messages communicated, ‘share of voice’
and benchmarking its profile against competitors or in its sector;
2. Strategic insights and intelligence through issues tracking (environmental scanning),
competitor analysis and trend identification.
Figure 2 provides an overview of the four roles and uses of media content analysis within
these two areas – i.e. for formative (strategic planning) research and for evaluation.
Figure 2. The four roles of media content analysis
Client
Issues
Sources
PublicMEDIAClient
Issues
Sources
PublicMEDIA
1. Evaluate effectiveness of your
media relations & PR (what
you are getting into the media)
3. Gain strategic insight &
intelligence into issues &
trends reported in the media
2. Gain strategic insight &
intelligence on other sources
including competitors reported in
the media
4. Evaluate messages reaching
target audiences, ‘share of voice’,
etc to determine likely impact on
public opinion
MEDIA
Media Content Analysis
© Jim R. Macnamara, 1999
Client
Issues
Sources
PublicMEDIAClient
Issues
Sources
PublicMEDIA
1. Evaluate effectiveness of your
media relations & PR (what
you are getting into the media)
3. Gain strategic insight &
intelligence into issues &
trends reported in the media
2. Gain strategic insight &
intelligence on other sources
including competitors reported in
the media
4. Evaluate messages reaching
target audiences, ‘share of voice’,
etc to determine likely impact on
public opinion
MEDIA
Media Content Analysis
© Jim R. Macnamara, 1999
Asia Pacific Public Relations Journal, 6(1), 1–34.
22
References
Babbie, E. (1986). Observing ourselves: Essays in social research. Belmont, CA: Wadsworth.
Banerjee, M., Capozzoli, M., McSweeney, L., & Sinha, D. (1999). Beyond kappa: A review of
interrater agreement measures. Canadian Journal of Statistics, 27(1), 3–23.
Bartko, J., & Carpenter, W., Jr. (1976). On the methods and theory of reliability. Journal of Nervous
and Mental Disease, 163, 307–317.
Berelson, B. (1952). Content analysis in communication research. New York: Hafner.
Berger, A. (1991). Media research techniques. Newbury Park, CA: Sage.
Berger, P., & Luckmann, T. (1967). The social construction of reality. New York: Doubleday Anchor.
Campbell, R., & Pennebaker, J. (2003). The secret life of pronouns: flexibility in writing style and
physical health. Psychological Science, 14(1), 60–65. Retrieved July 19, 2004, from
http://jerz.setonhill.edu/weblog/permalink.jsp?id=885
Curran, J. (2002). Media and power. London: Routledge.
Denzin, N., & Lincoln, Y. (Eds.). (1994). Handbook of qualitative research. Thousand Oaks,
California: Sage Publications.
Dressing&pehl GbR & Verbi GmbH. (2004). MAXqda. [Computer Software]. Retrieved July 19,
2004, from http://www.winmax.de
Ellis, L. (1994). Research methods in the social sciences. Madison, WI: WCB Brown & Benchmark.
Frey, L., Botan, C., & Kreps, G. (2000). Investigating communication: an introduction to research
methods (2nd
ed.). Boston: Allyn & Bacon.
Gauntlett, D. (2002). Media, gender & identity. London: Routledge.
Glaser, B., & Strauss, A. (1967). The discovery of grounded theory. Chicago: Aldane.
Hansen, A., Cottle, S., Negrine, R., & Newbold, C. (1998). Mass communication research methods.
London: Macmillan.
Hijams, E. (1996). The logic of qualitative media content analysis: a typology. Communications, 21,
93–109.
Jensen, K. (1995). The Social Semiotics of Mass Communication. London: Sage Publications.
Kuhn, T. (1970). The Structure of Scientific Revolutions (2nd
ed.). University of Chicago Press.
Lasswell, H. (1927). Propaganda techniques in the world war. New York: Knopf.
Lasswell, H., Lerner, D., & Pool, I. de Sola. (1952). The comparative study of symbol: an
introduction. Stanford University Press, Hoover Institute and Library on War, Revolution and Peace.
Lombard, M., Synder-Duch, J., & Bracken, C. C. (2003). Content analysis in mass communication:
assessment and reporting of intercoder reliability. Human Communication Research, 29, 469–472.
Lombard, M., Snyder-Duch, J., & Bracken, C. C. (2004). Practical resources for assessing and
reporting intercoder reliability in content analysis research projects. Retrieved April 28, 2004, from
www.temple.edu/mmc/reliability
Macnamara, J. (2003). Mass media effects: a review of 50 years of media effects research. Retrieved
July 30, 2004, from http://www.archipelagopress.com/jimmacnamara
Mayring, P. (2000). Qualitative inhaltsanalyse. Grundlagen und Techniken (7th
ed.). Weinheim:
Psychologie Verlags Union. (Original work published 1983)
Mayring, P. (2003). Qualitative content analysis. Forum Qualitative Sozialforschung/Forum Social
Research. Online Journal, 1(2). Retrieved December 18, 2003, from http://qualitative-
research.net/fqs/fqs-e/2-00inhalt-e.htm
McKee, A. (2004). A beginner’s guide to textual analysis. Retrieved April 13, 2004, from
http://www/enhancetv.com.au/articles/article3_1.htm
Miles, M., & Huberman, M. (1994). Qualitative data analysis. California: Sage.
Neuendorf, K. (2002). The Content Analysis Guidebook, Thousand Oaks, CA: Sage Publications.
Neuman, W. (1997). Social research methods: qualitative and quantitative approaches. Needham
Heights, MA: Allyn & Bacon.
Newbold, C., Boyd-Barrett, O., & Van Den Bulck, H. (2002). The media book. London: Arnold
(Hodder Headline).
Patton, M. (1990). Qualitative evaluation and research methods. (2nd
ed.). Newbury Park: Sage
Publications.
Patton, M. (2002). Qualitative evaluation and research methods. (3rd
ed.). Newbury Park: Sage
Publications.
Asia Pacific Public Relations Journal, 6(1), 1–34.
23
Popping, R. (1984). AGREE, a package for computing nominal scale agreement. Computational
Statistics and Data Analysis, 2, 182–185.
Popping, R. (1988). On agreement indices for nominal data. In W. E. Saris & I. N. Gallhofer (Eds.),
Sociometric research: Volume 1, data collection and scaling (pp. 90-105). New York: St Martin’s.
Potter, W., & Levine-Donnerstein, D. (1999). Rethinking validity and reliability in content analysis.
Journal of Applied Communication Research, 27.
Riffe, D., & Freitag, A. (1997). A content analysis of content analyses: twenty-five years of
Journalism Quarterly. Journalism and Mass communication Quarterly, 74, 873–882.
Riffe, D., Lacy, S., & Fico, F. (1998). Analysing media messages: Using quantitative content analysis
in research. Mahwah, NJ: Erlbaum.
Riffe, D., Lacy, S., & Drager, M. (1996). Sample size in content analysis of weekly news magazines.
Journalism and Mass Communication Quarterly, 73, 635–644.
Riffe, D., Lacy, S., Nagovan, J., & Burkum, L. (1996). The effectiveness of simple and stratified
random sampling in broadcast news content analysis. Journalism and Mass Communication
Quarterly, 73, 159–168.
Robson, C. (1993). Real world research: a resource for social scientists and practitioner-researchers.
Oxford: Blackwell.
Selby, K., & Cowdery, R. (1995). How to study television. Basingstoke: Macmillan.
Shoemaker, P. & Reese, S. (1996). Mediating the message: theories of influences on mass media
content. White Plains, NY: Longman.
Silverman, D. (1993). Interpreting qualitative data: methods for analysing talk, text and interaction.
London: Sage.
SkyMeg Software. (2004). Program for Reliability Assessment of Multiple Coders’. Alpha release
0.4.4. [Computer Software]. Retrieved January 7, 2004, from
http://www.geocities.com/skymegsoftware/pram/html
Strauss, A., & Corbin, J. (1990). Basics of qualitative research: Grounded theory procedures and
techniques. Newbury Park, CA: Sage.
Synder-Duch, J., Bracken, C., & Lombard, M. (2001). Content analysis in communication:
assessment and reporting of intercoder reliability. Presentation to the International Communication
Association, Washington, DC, May.
Stone, P., Dunphy, D., Smith, M., & Ogilvie, D. (with credit given to Dr Ole Holsti. (1966). The
general inquirer: a computer approach to content analysis. Cambridge: MIT Press.
Tinsley, H., & Weiss, D. J. (1975). Interrater reliability and agreement of subject judgements. Journal
of Counseling Psychology, 22, 358–376.
Trochim, W. (2002). ‘Deductive and inductive thinking. In Research methods knowledge base.
Retrieved June 5, 2006, from http://www.socialresearchmethods.net/kb.dedind.htm
Van Zoonen, L. (1994). Feminist media studies. London: Sage.
Weber, R. (1990). Basic content analysis (2nd
ed.). Newbury Park, CA: Sage.
Yale, L., & Gilly, M. (1988). Trends in advertising research: a look at the content of marketingorientated
journals from 1976 to 1985. Journal of Advertising, 17(1), 12–22.
Published Reference:
Macnamara, J. (2005). Media content analysis: Its uses, benefits and Best Practice
Methodology. Asia Pacific Public Relations Journal, 6(1), 1– 34.
* Jim Macnamara PhD, MA, FPRIA, FAMI, CPM, FAMEC is Professor of Public Communication at the
University of Technology Sydney and Director of the Australian Centre of Public Communication,
positions he took up in 2007 after a 30-year professional career spanning journalism, public
relations, advertising, and media research. He is the author of 12 books including ‘The 21st
Century Media (R)evolution: Emergent Communication Practices’ published by Peter Lang, New
York in 2010.