Health Care for Women International

ISSN: 0739-9332 (Print) 1096-4665 (Online) Journal homepage: http://www.tandfonline.com/loi/uhcw20

Content analysis: Method, applications, and issues

Barbara Downe‐Wamboldt RN, PhD

To cite this article: Barbara Downe‐Wamboldt RN, PhD (1992) Content analysis: Method,
applications, and issues, Health Care for Women International, 13:3, 313-321, DOI:
10.1080/07399339209516006

To link to this article: http://dx.doi.org/10.1080/07399339209516006

Published online: 14 Aug 2009.

Submit your article to this journal

Article views: 3534

View related articles

Citing articles: 538 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=uhcw20

Download by: [The UC San Diego Library] Date: 04 May 2017, At: 02:14
 CONTENT ANALYSIS: METHOD, APPLICATIONS,
AND ISSUES

Barbara Downe-Wamboldt, RN, PhD

School of Nursing, Dalhousie University, Halifax, Nova Scotia, Canada

Content analysis research methodology is detailed, its procedures

are described, some examples of its application are provided, and
the controversial issues surrounding its use are discussed. Unlike
strictly qualitative designs, content analysis has external validity as
a goal. Because of its focus on human communication, content anal-
ysis offers practical applicability, promise, and relevance for re-
search involving the practice and education of nurses and other
helping professionals.

In recent years there has been a growing recognition that both qualita-
tive and quantitative approaches are needed for advancing nursing sci-
ence. Morse (1991) has suggested, "researchers who purport to sub-
scribe to the philosophical underpinnings of only one research approach
have lost sight of the fact that research methodologies are merely tools,
instruments to be used to facilitate understanding" (p. 122). The most
recent controversial issue is not whether one method is intrinsically bet-
ter than another, but which combination of methods is best to meet the
aims of a particular study. An integral part of contemporary nursing
research, content analysis methodology offers the opportunity to com-
bine what are often thought to be antagonistic approaches to data analy-
sis.
The intellectual basis of content analysis can be traced to the begin-
ning of conscious use of symbols and language. Before World War n ,
the method was primarily restricted to critique of journalistic endeavors,
to document their religious, scientific, or literary content (Krippendorff,
1980; Speed, 1893). The simplistic reliance on counting words or
phrases mat was characteristic of the method in its early development is
attributed to these journalistic roots. According to Weber (1985), the
first large-scale application of the method was during World War n by
the U.S. Office of Special Services to Nazi war propaganda. Although

Health Care for Women International, 13:313-321, 1992

Copyright © 1992 by Hemisphere Publishing Corporation 313
314 B. Downe-mmboldt

used widely in other disciplines as diverse as psychology, sociology, and

political science, I shall discuss the use of content analysis in nursing
science research and provide examples from that discipline.
Content analysis is a research method that provides a systematic and
objective means to make valid inferences from verbal, visual, or written
data in order to describe and quantify specific phenomena. Unfortu-
nately, for some researchers scientific validity is equated with quantifi-
cation. Content analysis is more than a counting game; it is concerned
with meanings, intentions, consequences, and context. To describe the
occurrences of words, phrases, or sentences without consideration of the
contextual environment of the data is inappropriate and inadequate. The
analyst must be cognizant of the context and must justify the findings in
terms of the context or environment that produced the data. The goal of
content analysis is to enhance the inferential quality of the results by
relating the categories to the context or environment that produced the
data. Inferences are not an integral part of the data, but rather come
from the researcher and the research reader. An unobtrusive method,
content analysis can be applied to existing data from one point in time or
to documents that have existed over longer time frames; it can be used
either alone or in conjunction with other methods.
The method can be used for several purposes, such as revealing the
focus of individual, group, institutional, or societal attention; determin-
ing psychological states of persons or groups; reflecting cultural patterns
and beliefs; describing themes, trends, goals, or other characteristics in
communication content; analyzing open-ended survey data; and describ-
ing attributes of the sender of a communication and attitudinal and be-
havioral responses of the recipient(s) to the communication (Berelson,
1952; Krippendorff, 1980; Polit & Hungler, 1991; Weber, 1985). As
with any research strategy, the objective of content analysis is to provide
knowledge and understanding of the phenomena under study.

METHOD AND APPLICATIONS

Krippendorff (1980) disagreed with the contention that content analy-
sis is what each of us may do in reading the morning paper, and he
clearly identified it as a scientific process that differs across disciplines
in terms of its focus. This set of analytic techniques facilitates the de-
scription of the manifest and/or latent content of communication by
measuring the frequency, order, or intensity of occurrences of words,
phrases, or sentences (Krippendorff, 1980; McLaughlin & Marascuilo,
1990; Weber, 1985). According to Weber (1985), "The rules of this
inferential process vary with the theoretical and substantive interest of
the investigator" (p. 9). Although there is no single set of rules and
Content Analysis 315

procedures, the content analysis research method generally encompasses

the following steps:
1. Selecting the unit of analysis,
2. Creating and defining the categories,
3. Pretesting the category definitions and rules,
4. Assessing reliability and validity,
5. Revising the coding rules if necessary,
6. Pretesting the revised category scheme,
7. Coding all the data, and
8. Reassessing reliability and validity.
Designing the Sampling Method and Selecting the Unit of Analysis
External validity is a goal of content analysis, so the use of probabil-
ity sampling techniques is an issue. Content analysis is predictive in
intent. To select a sample, the population of all sources of data must first
be identified, such as all records of hospice volunteers' interactions with
terminally ill patients. The researcher then chooses a plan (random,
stratified, systematic, cluster, proportional, or multistage) to yield sam-
ples that are representative of the phenomena of interest (Krippendorff,
1980). Data collection is influenced by the knowledge level, experi-
ences, biases, and perspectives of the researcher, as well as by what
information the participants are willing or able to provide and what
sources of data are available to the researcher. A fundamental assump-
tion of any research endeavor is that the investigator always considers
potential sources of error in the data. Brink's (1991) account of the
anthropologist who spent a summer collecting data only to be told at his
going-away party that the tribe had consistently lied to him throughout
the summer clearly emphasizes this issue.
One of the most basic and important decisions for the researcher is
the selection of the unit of analysis. As with any study, the investigator
is guided by the research question to be answered. Most frequently, the
unit of analysis includes words, sentences, phrases, paragraphs, or
whole text such as interviews, diaries, or books; themes (entire ideas or
thoughts); and the amount of space and time given to a topic. For in-
stance, in our study involving hospice volunteers we used self-reported
records of the volunteers' interactions with patients (Downe-Wamboldt
& Ellerton, 1986). Our aim was to describe the role of hospice volun-
teers in supporting hospitalized terminally ill patients and their families.
When Beaton (1990) studied the experiences of women admitted to the
hospital in early labor, she used the verbal communication between the
nurse and patient as the unit of analysis. Flaskerud and Rush (1990)
used individual and group interview data as the unit of analysis in their
316 B. Downe-Wamboldt

exploratory study of the health beliefs and practices of Black women.

Creating and Defining the Category System

Whose reality is the "real" one? The intent of content analysis is not
necessarily to document the shared meaning between the researcher and
the researched, but to describe the phenomena of interest for a particular
purpose. Often, people are unable to interpret their own behavior; re-
searchers may attempt to do so by bringing a wider range of knowledge
to the context of the study when they analyze the data. I do not mean to
suggest that category schemes should be the "immaculate conceptions"
of the researcher. Category schemes are invented by the researcher to
generate knowledge and to increase understanding of a particular phe-
nomenon. The schemes are based on the research question, the selected
unit of analysis, the relevant theories, a review of the previous research,
literature, and the data. In content analysis it is acknowledged that the
researcher uses a particular framework or perspective to analyze the
data; what you see in the dark depends on where you choose to focus the
light. This does not imply that the researcher can anticipate all of the
categories before the material is obtained and analyzed. In fact, impos-
ing such constraints on the data could impede the validity of the results.
There is no single meaning to be discovered in the data; rather, multiple
meanings can be identified depending on the purpose of the study. If a
coding scheme can be used reliably and produce relevant results for the
research question, it is appropriate (Fox, 1982).
We developed our category system for the roles of hospice volunteers
(listening and responding, giving and receiving information, socializing,
providing spiritual comfort, providing physical comfort, and referring)
from a review of the literature and a preliminary analysis of the data
(Downe-Wamboldt & Ellerton, 1986). When Beaton (1990) analyzed
her data, she used a previously established category system (that of
Stiles, 1978, the Taxonomy of Verbal Response Modes) that was appro-
priate for the purpose of her investigation.
The coding system should define the critical attributes of specific
categories and distinguish the similarities and differences between cate-
gories. Problems will arise with ambiguous definitions of categories and
of the rules of coding. When there is high agreement on the definitions,
there are fewer problems with coding the data. In defining the categories
it is essential to include attributes that are exhaustive; having a particular
defining attribute does not exclude the possibility of another. Further-
more, different units may have the same attributes but to a different
degree. This variation in intensity of an attribute can provide meaningful
insights that deepen one's understanding of the content under investiga-
Content Analysis 317

tion. Different levels of measurement can be used in the analysis to

distinguish these differences, for example peaceful or afraid (nominal)
and fearful or terrified (ordinal). How narrow or broad the categories
should be depends on the purpose of the investigator. Flaskerud and
Rush (1990), for example, used the broad categories of natural and
supernatural sources of illness and remedies in their study of traditional
health beliefs and practices of Black Americans.
To facilitate the application of statistical analysis and for conceptual
clarity, it is most frequently recommended that the categories created be
mutually exclusive (Babbie, 1986; Polit & Hungler, 1991; Weber, 1985;
Wilson, 1989). If a unit of analysis can be classified in two or more
categories, then most statistical procedures that require that the variables
be mutually exclusive will be useless to the researcher. The total cate-
gory or one or more subcategories, but not both total category and
subcategories, should be used to maintain the statistical independence of
the content variable (Weber, 1985). As long as the categories are mutu-
ally exclusive, computer-assisted content analysis programs are avail-
able to systematically categorize and reduce linguistic data. Ultimately
the decision as to whether or not to use mutually exclusive categories is
made by the researcher on the basis of what is most logical in the con-
text of the question under investigation.

Pretesting the Category System

In the next stage, a small sample of text is carefully pretested to
determine if the rules for classification are clear or ambiguous. When
coding a sample of data, the researcher may discover that some of the
content falls outside preestablished categories. He or she then needs to
create a set of categories based on the themes that appear in the data; to
some degree, all research methods involve the interaction of theoretical
concerns and empirical data. Samples of text that are not easily classi-
fied will provide insights for revisions to the scheme. A benefit of con-
tent analysis is that the researcher has the opportunity to devise the most
appropriate definitions of the categories based on his or her interactions
with the data. Moving back and forth between text and the output of
content analysis provides for progressive refining and validating of the
category scheme. Some how-to suggestions for the process of analysis
include clarifying ideas with other researchers in the area, constantly
comparing items in the data, rephrasing questions to overcome barriers
to perception, thinking of like and unlike cases, listening to others'
ideas, drawing back from the data to facilitate seeing the whole, and
withholding final judgments as long as possible to be open to new per-
spectives.
318 B. Downe-Wamboldt

Assessing Reliability
When is a study science and when is it one person's philosophy?
Controversy exists over whether reliability should be sacrificed for
meaning or vice versa. Researchers frequently face a difficult choice
between depth or level of understanding and repeatability or reliability.
The researcher wants the most empirically meaningful information with-
out too much loss of reliability. Often compromises are essential. When-
ever possible, the best solution to the dilemma between level of under-
standing and specifity is to use both latent and manifest content analysis
approaches to data analysis. In latent content analysis, the researcher is
concerned with the underlying meaning in each passage of the text.
Coding the underlying meaning or latent content focuses on the tone or
implied feeling, whereas coding the manifest content describes only the
visible, surface, or obvious components of communication. Field and
Morse (1985) described latent content analysis as reviewing data within
the context of the entire data set for each participant and manifest con-
tent as a check for specific instances of the categories. For example,
when the researcher's goal is to detail the psychological status of termi-
nally ill persons, using both latent and manifest content analysis to de-
scribe the tone, meaning, and number of references to dying will pro-
vide more insightful and meaningful results than would using either
approach alone.
Stability and agreement reliability are the most pertinent types of
reliability for narrative data. The extent to which the results of content
classification are consistent over time when the data are coded more
than once by the same coder (intrarater reliability) can be ascertained by
using Cohen's kappa for nominal data (Weber, 1985). Interrater reliabil-
ity, or agreement between two different raters coding the same data, can
be assessed using Cohen's kappa or percentage agreement (Polit &
Hungler, 1991). Detailed descriptions of methods to assess interrater
reliability have been provided by Krippendorff (1980) and Berelson
(1952). Test-retest reliability may be used to determine if the narrative
data are consistent from one time to another. Test-retest reliability may
be appropriate in coding interview data when consistency is expected;
conversely, it is inappropriate when change in the data are anticipated.
The larger the unit of analysis, the more difficult it is to achieve satisfac-
tory levels of reliability in coding data.
If the researcher plans to use more than one coder for the study, the
reliability of the coding process must be assessed before all of the data
are analyzed and disputes among coders must be resolved (Weber,
1985). The coding rules will need to be revised if the reliability is low (a
kappa of 0.8 to 0.9 is the desired level, with 0.7 being the minimum
Content Analysis 319

standard). After the revisions, more pretests of the coding scheme need
to be completed until an acceptable level of reliability is achieved and
before all of the data are analyzed. Human errors are always possible in
coding data and may be related to fatigue, personal bias, and perception.
Consequently, systematic checks of accuracy of coding are necessary
throughout the process. Interpretations of definitions that seemed clear
at the beginning of the project may change in subtle ways as familiarity
with the data increases. One of the advantages of computer programs for
coding data is the opportunity to achieve perfect reliability. Once the
content-coding schemes have been defined, an endeavor that requires
imagination, logic, creativity, and theoretical work, a computer will
consistently apply the rules to text.

Assessing Validity
Whose reality is the most accurate relative to the research question?
Few people share the same social, cultural, political, or historical per-
spective. Thus, it is absurd to think that one person can understand the
subjective experience of another person exactly as he or she has lived it.
What the researcher is able to know about the lives and thoughts of other
people is directly influenced by his or her personal history, areas of
interest, and focus. Consider the example of an elderly woman with
arthritis describing for her rheumatologist how she manages her activi-
ties of daily living under the adversity of functional losses associated
with arthritis. In this interaction the elderly woman relates her prob-
lems, as she experienced them, to a medical expert. Their perspectives
are assumed to be different. The rheumatologist may be interested in
detailing the disease process; the elderly woman, in describing her
unique difficulties. Thus meanings are always relative to the perspec-
tives of the sender and the receiver of the message.
The strategy of taking the results to the participants to have them
validate one's interpretations can be a useful approach to authenticate
the experiences of the participants. Achieving intersubjective agreement
would simplify the analysis; however, it is not a presupposition for con-
tent analysis (Krippendorff, 1980). The researcher, who has a broader
understanding of the historical context of social structures that have
influenced the actions of the players, may develop a broader understand-
ing of what is going on, in addition to the understanding that he or she
may share with the participants. Multiple meanings are always present
in data—there is no right meaning, only the most accurate meaning from
a particular perspective.
Validity has to do with what is being measured and how well. Validity
is confirmed or denied by returning to the original text to find examples
320 B. Downe-Wamboldt

of categories and by relating relevant theory to text. Content analysis

relies heavily on face or content validity that can only be determined by
the judgments of experts in the area. The validity of our category
scheme was supported by previous relevant research and literature
(Downe-Wamboldt & Ellerton, 1986). The conceptual congruence of
Beaton's (1990) data to a preexisting category scheme provided support
for its validity. Flaskerud and Rush (1990) used the relevant literature,
the original data, and another nurse researcher to assist with the verifi-
cation of their categories. Submitting a sample of text to an independent
panel of experts for coding is another method to assess the validity of the
category system. An in-depth discussion of the relative merits and limi-
tations of the use of a panel of experts in assessing validity has been
provided by Brink (1991) and Stern (1991).
Factor analysis is a method of identifying clusters of related variables
that has been used in content analysis to identify themes (Weber, 1985).
This statistical approach can be used to identify convergent and discrim-
inant validity when a large set of measures is available. To help establish
external validity or generalizability of the categories, the researcher can
code separate samples of data to determine if the categories are valid
from one sample to another. Finally, if a researcher samples texts within
documents, such as one part of a taped interview, then the sample
should consist of whole paragraphs rather than individual sentences to
maintain the context or semantic coherence of the document.

CONCLUSION
Content analysis provides a mechanism to yield interesting and theo-
retically useful generalizations with minimal loss of information from
the original data. "It may be applied to virtually any form of linguistic
communication to answer the classic questions of who says what to
whom, why, how, and with what effect" (Babbie, 1986, p. 268). Finan-
cial costs are usually minimal because content analysis essentially is a
coding operation involving logical and conceptualizing work that can be
completed effectively by only one researcher. Disadvantages associated
with this method include its being limited to recorded communications
(verbal, visual, or written data), the amount of time required to code
data, and the type of statistical procedures that can be applied to data.
By paying careful attention to the assessment of reliability and validity,
the researcher can minimize problems associated with these issues. Be-
cause of its focus on human communication, content analysis is particu-
larly well suited to research involving the practice and education of
nurses and other helping professionals.
Content Analysis 321

REFERENCES
Babbie, E. (1986). The practice of social research (4th ed.). Belmont, CA: Wadsworth.
Beaton, J. I. (1990). Dimensions of nurse and patient roles in labor. Health Care for
Women International, 11, 393-408.
Berelson, B. (1952). Content analysis in communication research. Glencoe, IL: Free
Press.
Brink, P. J. (1991). Issues in reliability and validity. In J. M. Morse (Ed.), Qualitative
nursing research: A contemporary dialogue (pp. 151-168). Newbury Park, CA:
Sage.
Downe-Wamboldt, B. L., & Ellerton, M. L. (1986). A study of the role of hospice
volunteers. Hospice Journal, 1, 17-31.
Field, P. A., & Morse, J. M. (1985). Nursing research: The application of qualitative
approaches. Rockville, MD: Aspen.
Flaskerud, J. H., & Rush, C. E. (1990). AIDS and traditional health beliefs and prac-
tices of Black women. Nursing Research, 38, 210-215.
Fox, D. J. (1982). Fundamentals of research in nursing. Norwalk, CT: Appleton-
Century-Crofts.
Krippendorff, K. (1980). Content analysis: An introduction to its methodology (4th ed.).
Newbury Park, CA: Sage.
McLaughlin, F. E., & Marascuilo, L. A. (1990). Advanced nursing and health care
quantification approaches. Philadelphia: W. B. Saunders.
Morse, J. M. (1991). Approaches to qualitative-quantitative methodological triangula-
tion. Nursing Research, 40, 120-123.
Polit, D., & Hungler, B. (1991). Nursing research: Principles and methods (4th ed.).
Philadelphia: J. B. Lippincott.
Speed, C. J. (1893). Do news now give news? Forum, 15, 705-711.
Stern, P. N. (1991). Are counting and coding a capella appropriate in qualitative re-
search? In J. M. Morse (Ed.), Qualitative nursing research: A contemporary dia-
logue (pp. 135-147). Newbury Park, CA: Sage.
Stiles, W. B. (1978). Manual for a taxonomy of verbal response modes. Chapel Hill,
NC: Institute for Research in Social Science.
Weber, R. P. (1985). Basic content analysis. Newbury Park, CA: Sage.
Wilson, H. S. (1989). Research in nursing (2nd ed.). Redwood City, CA: Addison-
Wesley.