10.1177/1049732305276687QHUsieAhL, ISThAaTnInVoEn H /E PARLOTBHL REMESSE WARITCHH I/N NTEoRvember 2005 VIEWS Three Approaches to Qualitative Content Analysis Hsiu-Fang Hsieh Sarah E. Shannon Contentanalysisisawidelyusedqualitativeresearchtechnique.Ratherthanbeingasingle method, current applications of content analysis show three distinct approaches: conventional,directed,orsummative.Allthreeapproachesareusedtointerpretmeaningfromthe content of text data and, hence, adhere to the naturalistic paradigm. The major differences amongtheapproachesarecodingschemes,originsofcodes,andthreatstotrustworthiness. In conventional content analysis,coding categories are derived directly fromthe text data. Withadirectedapproach,analysisstartswithatheoryorrelevantresearchfindingsasguidance for initial codes. A summative content analysis involves counting and comparisons, usuallyofkeywordsorcontent,followedbytheinterpretationoftheunderlyingcontext.The authors delineate analytic procedures specific to each approach and techniques addressing trustworthiness with hypothetical examples drawn from the area of end-of-life care. Keywords: contentanalysis;qualitativeresearch;researchmethodology;end-of-lifecare Content analysis is a research method that has come into wide use in health studiesinrecentyears.Asearchofcontentanalysisasasubjectheadingtermin theCumulativeIndextoNursingandAlliedHealthLiteratureproducedmorethan 4,000 articles published between 1991 and 2002. The number of studies reporting theuseofcontentanalysisgrewfromonly97in1991to332in1997and601in2002. Researchersregardcontentanalysisasaflexiblemethodforanalyzingtextdata (Cavanagh,1997).Contentanalysisdescribesafamilyofanalyticapproachesrangingfromimpressionistic,intuitive,interpretiveanalysestosystematic,stricttextual analyses (Rosengren, 1981). The specific type of contentanalysis approach chosen byaresearchervarieswiththetheoreticalandsubstantiveinterestsoftheresearcher and the problem being studied (Weber, 1990). Although this flexibility has made contentanalysisusefulforavarietyofresearchers,thelackofafirmdefinitionand procedureshaspotentiallylimitedtheapplicationofcontentanalysis(Tesch,1990). Thedifferentiationofcontentanalysisisusuallylimitedtoclassifyingitasprimarilyaqualitativeversusquantitativeresearchmethod.Amorethoroughanalysis ofthewaysinwhichqualitativecontentanalysiscanbeusedwouldpotentiallyilluminatekeyissuesforresearcherstoconsiderinthedesignofstudiespurportingto 1277 AUTHORS’ NOTE: We wish to express our gratitude to Drs. Pamela L. Jordan, Carol J. Leppa, and J. Randall Curtis for their feedback and support in writing this article. QUALITATIVE HEALTH RESEARCH, Vol. 15 No. 9, November 2005 1277-1288 DOI: 10.1177/1049732305276687 © 2005 Sage Publications use content analysis and the analytic procedures employed in such studies, thus avoiding a muddling of methods (Morse, 1991). Ourpurposeinthisarticleistopresentthebreadthofapproachescategorized as qualitative content analysis. We have identified three distinct approaches: conventional,directed,andsummative.Allthreeapproachesareusedtointerprettext data from a predominately naturalistic paradigm. We begin with a brief review of thehistoryanddefinitionsofcontentanalysis.Wethenillustratethethreedifferent approachestoqualitativecontentanalysiswithhypotheticalstudiestoexplicatethe issues of study design and analytical procedures for each approach. BACKGROUND ON THE DEVELOPMENT OF CONTENT ANALYSIS Content analysis has a long history in research, dating back to the 18th century in Scandinavia(Rosengren,1981).IntheUnitedStates,contentanalysiswasfirstused asananalytictechniqueatthebeginningofthe20thcentury(Barcus,1959).Initially, researchers used contentanalysis as either a qualitative or quantitative methodin theirstudies(Berelson,1952).Later,contentanalysiswasusedprimarilyasaquantitative research method, with text data coded into explicit categories and then described using statistics. This approach is sometimes referred to as quantitative analysisofqualitativedata(Morgan,1993)andisnotourprimaryfocusinthisarticle.Morerecently,thepotentialofcontentanalysisasamethodofqualitativeanalysis for health researchers has been recognized, leading to its increased application and popularity (Nandy & Sarvela, 1997). Qualitativecontentanalysisisoneofnumerousresearchmethodsusedtoanalyzetextdata.Othermethodsincludeethnography,groundedtheory,phenomenology,andhistoricalresearch.Researchusingqualitativecontentanalysisfocuseson the characteristics of language as communication with attention to the content or contextual meaning of the text (Budd, Thorp, & Donohew, 1967; Lindkvist, 1981; McTavish & Pirro, 1990; Tesch, 1990). Text data might be in verbal, print, or electronic form and might have been obtained from narrative responses, open-ended survey questions, interviews, focus groups, observations, or print media such as articles,books,ormanuals(Kondracki&Wellman,2002).Qualitativecontentanalysis goes beyond merely counting words to examining language intensely for the purpose of classifying large amounts of text into an efficient number of categories thatrepresentsimilarmeanings(Weber,1990).Thesecategoriescanrepresenteither explicit communication or inferred communication. The goal of content analysis is “to provide knowledge and understanding of the phenomenon under study” (Downe-Wamboldt, 1992, p. 314). In this article, qualitative content analysis is definedasaresearchmethodforthesubjectiveinterpretationofthecontentoftext datathroughthesystematicclassificationprocessofcodingandidentifyingthemes or patterns. Toillustratethepossibleapplicationsofcontentanalysis,weconstructedhypotheticalstudiesdrawnfromtheareaofend-of-life(EOL)research.Contentanalysis has been a popular analytic method in studies related to EOL care, an area of increasing emphasis as demonstrated by its inclusion as one of the five research themessupportedbytheNational InstitutesofHealth,National InstituteofNurs 1278 QUALITATIVE HEALTH RESEARCH / November 2005 ingResearch(NINR)for2003(“EnhancingtheEnd-of-LifeExperienceforPatients and Their Families,” NINR, 2003). CONVENTIONALCONTENT ANALYSIS Researcher X used a conventional approach to content analysis in her study (Table 1). Conventional content analysis is generally used with a study design whoseaimistodescribeaphenomenon,inthiscasetheemotionalreactionsofhospice patients. This type of design is usually appropriate when existing theory or research literature on a phenomenon is limited. Researchers avoid using preconceived categories (Kondracki & Wellman, 2002), instead allowing the categories andnamesforcategoriestoflowfromthedata.Researchersimmersethemselvesin the data to allow new insights to emerge (Kondracki & Wellman, 2002), also described as inductive category development (Mayring, 2000). Many qualitative methods share this initial approach to study design and analysis. If data are collected primarily through interviews, open-ended questions will beused.Probesalsotendtobeopen-endedorspecifictotheparticipant’scomments ratherthantoapreexistingtheory,suchas“Canyoutellmemoreaboutthat?”Data analysis startswithreading all data repeatedlytoachieveimmersionand obtaina senseofthewhole(Tesch,1990)asonewouldreadanovel.Then,dataarereadword by word to derive codes (Miles & Huberman, 1994; Morgan, 1993; Morse & Field, 1995)byfirsthighlightingtheexactwordsfromthetextthatappeartocapturekey thoughtsor concepts.Next, theresearcherapproaches thetextbymaking notesof hisorherfirstimpressions,thoughts,andinitialanalysis.Asthisprocesscontinues, labelsforcodesemergethatarereflectiveofmorethanonekeythought.Theseoften come directly from the text and are then become the initial coding scheme. Codes thenaresortedintocategoriesbasedonhowdifferentcodesarerelatedandlinked. These emergent categories are used to organize and group codes into meaningful clusters(Coffey&Atkinson,1996;Patton,2002).Ideally,thenumbersofclustersare between 10 and 15 to keep clusters broad enough to sort a large number of codes (Morse & Field, 1995). Depending on the relationships between subcategories, researchers can combineororganizethislargernumberofsubcategoriesintoasmallernumberofcategories.Atreediagramcanbedevelopedtohelpinorganizingthesecategoriesintoa hierarchical structure (Morse & Field, 1995). Next, definitions for each category, subcategory, and code are developed. To prepare for reporting the findings, exemplarsforeachcodeandcategoryareidentifiedfromthedata.Dependingonthe purposeofthestudy,researchersmightdecidetoidentifytherelationshipbetween categories and subcategories further based on their concurrence, antecedents, or consequences (Morse & Field, 1995). With a conventional approach to content analysis, relevant theories or other researchfindingsareaddressedinthediscussionsectionofthestudy.InResearcher X’sstudy,shemightcompareandcontrastherfindingstoKübler-Ross’s(1969)theory.Thediscussion wouldinclude asummaryofhowthefindings fromherstudy contributetoknowledgeintheareaofinterestandsuggestionsforpractice,teaching, and future research. The advantage of the conventional approach to content analysis is gaining directinformationfromstudyparticipantswithoutimposingpreconceivedcatego Hsieh, Shannon / PROBLEMS WITH INTERVIEWS 1279 ries or theoretical perspectives. Researcher X’s study depicts a research question appropriate for this approach. Knowledge generated from her content analysis is based on participants’ unique perspectives and grounded in the actual data. Her sampling technique was designed to maximize diversity of emotional reactions, and the analysis techniques were structured to capture that complexity. One challenge of this type of analysis is failing to develop a complete understanding of the context, thus failing to identify key categories. This can result in findings that do not accurately represent the data. Lincoln and Guba (1985) described this as credibility within the naturalistic paradigm of trustworthiness or internal validity within a paradigm of reliability and validity. Credibility can be establishedthroughactivitiessuchaspeerdebriefing,prolongedengagement,persistentobservation,triangulation,negativecaseanalysis,referentialadequacy,and member checks (Lincoln & Guba, 1985; Manning, 1997). Another challenge of the conventional approach to content analysis is that it can easily be confused with other qualitative methods such as grounded theory 1280 QUALITATIVE HEALTH RESEARCH / November 2005 TABLE 1: Hypothetical Research Study Using a Conventional Approach to Content Analysis— Researcher X’s Study Little is known about the emotional reactions of terminally ill patients who are receiving hospice care, possibly because of their reluctance to discuss death issues (Wilson & Fletcher, 2002). Some patients might feel relieved to have active therapy end, whereas others might feel afraid or even abandoned. Researcher X wanted to learn more about the emotional experiences of hospice patients to be able to address their needs more effectively. Because there was no existing theory to serve as a framework for her study, her research question was “What are the emotional reactions of terminally ill patients who are receiving hospice care?” Based on her clinical experience, Researcher X suspected that the emotional reactions of patients who were new to hospice care differed from those who had been in hospice care for a longer period. She also suspected that those receiving home hospice care had different experiences from those receiving in-patient hospice care. Researcher X therefore decided to use a stratified sampling technique to ensure heterogeneity of the sample. The target sample size was 10 home hospice patients and 10 inpatient hospice patients, with 5 from each group being recruited within 48 hours of enrollment into hospice and 5 recruited 7 to 10 days following enrollment. In addition, the sample would include both men and women and both older and middle-aged people. Prior to recruitment and data collection, the research procedures were approved for use with human subjects. Informed consent was obtained from all participants. Researcher X collected data through individual interviews using open-ended questions such as “What has it been like to be in hospice care?” followed by specific probes. All interviews were audiotape-recorded and transcribed verbatim. Researcher X used content analysis to analyze the data. She began by reading each transcript from beginning to end, as one would read a novel. Then, she read each transcript carefully, highlighting text that appeared to describe an emotional reaction and writing in the margin of the text a keyword or phrase that seemed to capture the emotional reaction, using the participant’s words. As she worked through the transcript, she attempted to limit these developing codes as much as possible. After open coding of three to four transcripts, Researcher X decided on preliminary codes. She then coded the remaining transcripts (and recoded the original ones) using these codes and adding new codes when she encountered data that did not fit into an existing code. Once all transcripts had been coded, Researcher X examined all data within a particular code. Some codes were combined during this process, whereas others were split into subcategories. Finally, she examined the final codes to organize them into a hierarchical structure if possible. In the findings, the emotional responses of hospice patients were described using the identified codes and hierarchical structure. In discussion of the findings, the results from this content analysis were compared and contrasted with Kübler-Ross’s (1969) model to highlight similarities and differences. method(GTM)orphenomenology.Thesemethodsshareasimilarinitialanalytical approach but go beyond content analysis to develop theory or a nuanced understandingofthelived experience.Theconventionalapproachtocontentanalysis is limited in both theory development and description of the lived experience, because both sampling and analysis procedures make the theoretical relationship between concepts difficult to infer from findings. At most, the result of a conventionalcontentanalysisisconceptdevelopmentormodelbuilding(Lindkvist,1981). For example, Researcher X might find that patients who are new to hospice care express worry about how their social obligations will be met(such as finding care for a pet), whereas patients who have been in hospice for long periods might expressmoreanticipatorygrief.ResearcherXmightcompareherfindingstothose ofKübler-Ross(1969) andconcludethatanadditionalemotionalreactiontoenteringhospicecareistheprocessof“tyinguplooseends,”whichshemightdefineas making both financial and social arrangements. DIRECTED CONTENT ANALYSIS Sometimes, existing theory or prior research exists about a phenomenon that is incomplete or would benefit from further description. The qualitative researcher might choose to use a directed approach to content analysis, as Researcher Y did (Table2).PotterandLevine-Donnerstein(1999)mightcategorizethisasadeductive useoftheorybasedontheirdistinctionsontheroleoftheory.Howeverthekeytenets of the naturalistic paradigm form the foundation of Researcher Y’s general approachtothestudydesignandanalysis.Thegoalofadirectedapproachtocontentanalysisistovalidateorextendconceptuallyatheoreticalframeworkortheory. Existingtheoryorresearchcanhelpfocustheresearchquestion.Itcanprovidepredictionsaboutthevariablesofinterestorabouttherelationshipsamongvariables, thushelpingtodeterminetheinitialcodingschemeorrelationshipsbetweencodes. This has been referred to as deductive category application (Mayring, 2000). Contentanalysisusingadirectedapproachisguidedbyamorestructuredprocessthaninaconventionalapproach(Hickey&Kipping,1996).Usingexistingtheoryorpriorresearch,researchersbeginbyidentifyingkeyconceptsorvariablesas initial coding categories (Potter & Levine-Donnerstein, 1999). Next, operational definitions for each category are determined using the theory. In Researcher Y’s study, Kübler-Ross’s (1969) five stages of grief served as an initial framework to identify emotional stages of terminally ill patients. If data are collected primarily through interviews, an open-ended question mightbeused,followedbytargetedquestionsaboutthepredeterminedcategories. Afteranopen-endedquestion,ResearcherYusedprobesspecificallytoexploreparticipants’ experiences of denial, anger, bargaining, depression, and acceptance. Codingcanbeginwithoneoftwostrategies,dependingontheresearchquestion.If thegoaloftheresearchistoidentifyandcategorizeallinstancesofaparticularphenomenon,suchasemotionalreactions,thenitmightbehelpfultoreadthetranscript and highlight all text that on first impression appears to represent an emotional reaction.Thenextstepinanalysiswouldbetocodeallhighlightedpassagesusing the predetermined codes. Any text that could not be categorized with the initial coding scheme would be given a new code. Hsieh, Shannon / PROBLEMS WITH INTERVIEWS 1281 Thesecondstrategythatcanbeusedindirectedcontentanalysisistobegincodingimmediatelywiththepredeterminedcodes.Datathatcannotbecodedareidentifiedandanalyzedlatertodetermineiftheyrepresentanewcategoryorasubcategoryofanexistingcode.Thechoiceofwhichoftheseapproachestousedependson thedataandtheresearcher’sgoals.Iftheresearcherwantstobesuretocaptureall possibleoccurrencesofaphenomenon,suchasanemotionalreaction,highlighting identified text without coding might increase trustworthiness. If the researcher feels confident that initial coding will not bias the identification of relevant text, then coding can begin immediately. Depending on the type and breadth of a category, researchers might need to identify subcategories with subsequent analysis. For example, Researcher Y might decide to separate anger into subcategories depending on whom the anger was directed toward. The findings from a directed content analysis offer supporting and nonsupporting evidence for a theory. This evidence can be presented by showing codes withexemplarsandbyofferingdescriptiveevidence.Becausethestudydesignand analysis are unlikely to result in coded data that can be compared meaningfully usingstatisticaltestsofdifference,theuseofrankordercomparisonsoffrequency 1282 QUALITATIVE HEALTH RESEARCH / November 2005 TABLE 2: Hypothetical Research Study Using a Directed Approach to Content Analysis— Researcher Y’s Study Despite their wide acceptance and popularity, Kübler-Ross’s (1969) five stages of grief (denial, anger, bargaining, depression, and acceptance) have not been confirmed through other research. In taking care of terminally ill patients, Researcher Y wondered how well Kübler-Ross’s theory described his patients’ experiences with imminent death. His research question was “How well does Kübler-Ross’s model describe the emotional passages or journeys of patients who have been diagnosed with a terminal illness?” Researcher Y designed a sampling plan to maximize the chance of recruiting participants at different stages. All participants were diagnosed with a terminal illness, but one third were recruited while receiving “last chance” forms of curative therapy, one third after they refused further curative therapy but were not enrolled in hospice care, and one third who were contemplating (or had recently made) the decision to enter hospice care. In addition, the sample was recruited for gender balance and diagnostic diversity, specifically both oncology and non-oncology diagnoses. The target sample size was 18 to 21 participants. Interviews were conducted with individuals using open-ended questions, such as “What has your emotional journey been since being diagnosed with this illness?” Specific probes were developed based on Kübler-Ross’s model, such as Have you felt angry since your diagnosis? After institutional review board approval, informed consent from all participants was obtained. All interviews were audiotape-recorded and transcribed verbatim. Researcher Y developed operational definitions of the five emotional responses (anger, bargaining, etc.) identified in Kübler-Ross’s model. He then reviewed all transcripts carefully, highlighting all text that appeared to describe an emotional response. All highlighted text was coded using the predetermined categories wherever possible. Text that could not be coded into one of these categories was coded with another label that captured the essence of the emotion. After coding, Researcher Y examined the data for each category to determine whether subcategories were needed for a category (e.g., anger toward self, anger toward doctors, anger toward spiritual being). Data that could not be coded into one of the five categories derived from the theory were reexamined to describe different emotional reactions. Finally, Researcher Y compared the extent to which the data were supportive of Kübler-Ross’s theory versus how much represented different emotional responses. The report of study findings described the incidence of codes representing the emotional stages suggested by Kübler-Ross with those that represented different emotional responses by comparing the rank order of all codes. In the discussion section, Researcher Y summarized how the study validated KüblerRoss’s model and what new perspectives were added. ofcodescanbeused(Curtisetal.,2001).ResearcherYmightchoosetodescribehis study findings by reporting the incidence of codes that represented the five main categories derived from Kübler-Ross (1969) and the incidence of newly identified emotional reactions. He also could descriptively report the percent of supporting versus nonsupporting codes for each participant and for the total sample. Thetheoryorpriorresearchusedwillguidethediscussionoffindings.Newly identifiedcategorieseitherofferacontradictoryviewofthephenomenonormight furtherrefine,extend,andenrichthetheory.InResearcherY’sstudy,thediscussion might focus on the extent to which participants’ emotional journeys paralleled Kübler-Ross’s(1969)modelandthenewlyidentifiedemotionalreactionsorstages that were experienced by participants in the study. The main strength of a directed approach to content analysis is that existing theory can be supported and extended. In addition, as research in an area grows, a directed approach makes explicit the reality that researchers are unlikely to be workingfromthenaiveperspectivethatisoftenviewedasthehallmarkofnaturalistic designs. The directed approach does present challenges to the naturalistic paradigm. Using theory has some inherent limitations in that researchers approach the data with an informed but, nonetheless, strong bias. Hence, researchers might be more likelytofindevidencethatissupportiveratherthannonsupportiveofatheory.Second,inansweringtheprobequestions,someparticipantsmightgetcuestoanswer inacertainwayoragreewiththequestionstopleaseresearchers.InResearcherY’s study,somepatientsmightagreewiththesuggestedemotionalstageseventhough they did not experience the emotion. Third, an overemphasis on the theory can blindresearcherstocontextualaspectsofthephenomenon.InResearcherY’sstudy, the emphasis on Kübler-Ross’s (1969) stages of emotional response to loss might have clouded his ability to recognize contextual features that influence emotions. For example, the cross-sectional design of the study might have overemphasized currentemotionalreactions. Theselimitations arerelated toneutrality orconfirmability of trustworthiness as the parallel concept to objectivity (Lincoln & Guba, 1985).Toachieveneutralorunbiasedresults,anaudittrailandauditprocesscanbe used. In Researcher Y’s study, the vague terminology used in Kübler-Ross’s description of the model would be a challenge for the researcher in creating useful operational definitions. Having an auditor review and examine these definitions before the study could greatly increase the accuracy of predetermined categories. SUMMATIVE CONTENT ANALYSIS Typically,astudyusingasummativeapproachtoqualitativecontentanalysisstarts withidentifyingandquantifyingcertainwordsorcontentintextwiththepurpose ofunderstandingthecontextualuseofthewordsorcontent(Table3).Thisquantificationisanattemptnottoinfermeaningbut,rather,toexploreusage.Analyzingfor the appearance of a particular word or content in textual material is referred to as manifest content analysis (Potter & Levine-Donnerstein, 1999). If the analysis stoppedatthispoint,theanalysiswouldbequantitative,focusingoncountingthe frequencyofspecificwordsorcontent(Kondracki&Wellman,2002).Asummative approachtoqualitativecontentanalysisgoesbeyondmerewordcountstoinclude latentcontentanalysis.Latentcontentanalysisreferstotheprocessofinterpretation Hsieh, Shannon / PROBLEMS WITH INTERVIEWS 1283 of content (Holsti, 1969). In this analysis, the focus is on discovering underlying meaningsofthewordsorthecontent(Babbie,1992;Catanzaro,1988;Morse&Field, 1995).InResearcherZ’sstudy,theinitialpartoftheanalysistechnique,tocountthe frequency of death, die, and dying is more accurately viewed as a quantitative approach. However, Researcher Z went on to identify alternative terms for death and to examine the contexts within which direct versus euphemistic terms were used. Hence, Researcher Z used a summative approach to qualitative content analysis. Researchers report using content analysis from this approach in studies that analyze manuscript types in a particular journal or specific content in textbooks. Examples include studies examining content related to EOL care in medical textbooks (Rabow, Hardie, Fair, & McPhee, 2000), EOL care in critical care nursing textbooks (Kirchhoff, Beckstrand, & Anumandla, 2003), palliative care in nursing textbooks (Ferrell, Virani, Grant, & Juarez, 2000), death and bereavement in nursing textbooks (Ferrell, Virani, Grant, & Borneman, 1999), and spirituality in 1284 QUALITATIVE HEALTH RESEARCH / November 2005 TABLE 3: Hypothetical Research Study Using a Summative Approach to Content Analysis— Researcher Z’s Study Talking about death has virtually been banished from our language (Callahan, 1995). Use of the terms die, dying, anddeathremain taboo in U.S. society in favor of euphemisms such aspassing, going to a better place, and so on. A failure to use explicit terms might hinder the effectiveness of communication between physicians and patients (Levy, 2001; Vincent, 1997). Recognizing this problem, Researcher Z wanted to know how often health care providers, patients, or family members used explicit terms versus euphemisms. Under what circumstances are these explicit terms used? Her research question was How are the termsdie, dying, anddeathused in clinician-patient communication when discussing hospice care, and what alternative terms are used? Researcher Z designed a sampling plan to maximize the diversity of the sample around demographic characteristics of both the clinician and the patient/family. Patient characteristics included gender, age, diagnosis, and ethnic background. Clinician characteristics included gender, discipline, and area of specialization. Two types of communication events with patients who had received a terminal diagnosis were sampled. One was discharge teaching for hospitalized patients who were being transferred to home hospice, inpatient hospice, or skilled nursing facilities for endof-life (EOL) care. The other communication event was clinician-patient/family conferences in out- or inpatient settings to plan EOL care. Fifty separate communication events were sampled for 50 different clinicians and patient/family pairs. The research proposal was approved by institutional review boards before data collection. Informed consent was obtained from each participant. All clinicianpatient conversations were audiotape-recorded and transcribed verbatim. Data analysis started with computer-assisted searches for occurrences of the termsdie, death, and dying in the transcripts. Word frequency counts for each of the three death-related terms in a transcript were calculated and compared to the total length of the communication event. Researcher Z also coded the identity of the speaker, such as physician, nurse, patient, or family member. Frequency counts by type of speaker were calculated and compared to the total number of terms coded. Next, Researcher Z tried to identify alternative terms or expressions used instead ofdeath, die, or dying. Occurrences of these terms were counted both as a total number and for each alternative term. Frequencies of euphemisms versus direct terms were compared for type of speaker, demographic characteristics of clinician, and demographic characteristics of patient within each communication event and across the total sample. The major study findings described the occurrences of the three explicit terms used in clinicianpatient communication as compared to euphemistic terms. Comparisons across type of speaker and characteristics of clinicians and patients were made. The discussion of this study focused on exploring possible explanations for differences in the use of explicit versus euphemistic terms when discussing EOL care for different groups and in different situations. nursing textbooks (McEwen, 2004). These researchers started with counting the pages that covered specific topics followed by descriptions and interpretations of thecontent,includingevaluatingthequalityofthecontent.Othershavecompared theresultsofacontentanalysis withotherdatacollectedwithinthesameresearch project,suchascomparingpreferencesforvarioustypesoftelevisionprogramming with socioeconomic indicators of participants (Krippendorff, 1980). In a summative approach to qualitative content analysis, data analysis begins withsearchesforoccurrencesoftheidentifiedwordsbyhandorbycomputer.Word frequencycountsforeachidentifiedtermarecalculated,withsourceorspeakeralso identified.ResearcherZwantedtoknowthefrequencyofwordsthatwereusedto refertodeathbutalsotounderstandtheunderlyingcontextsfortheuseofexplicit versuseuphemisticterms.Heorsheilluminatedthecontextofeuphemisticversus explicittermsbyreportinghowtheirusagedifferedbyvariablessuchasthespeaker (patient versus clinician), the clinician’s specialization, and the age of the patient. Counting is used to identify patterns in the data and to contextualize the codes (Morgan,1993).Itallowsforinterpretationofthecontextassociatedwiththeuseof thewordorphrase.Researcherstrytoexplorewordusageordiscovertherangeof meanings that a word can have in normal use. Asummativeapproachtoqualitativecontentanalysis hascertainadvantages. It is an unobtrusive and nonreactive way to study the phenomenon of interest (Babbie,1992).Itcanprovidebasicinsightsintohowwordsareactuallyused.However,thefindingsfromthisapproacharelimitedbytheirinattentiontothebroader meanings present in the data. As evidence of trustworthiness, this type of study reliesoncredibility.Amechanismtodemonstratecredibilityorinternalconsistency is to show that the textual evidence is consistent with the interpretation (Weber, 1990). For Researcher Z’s study, validation by content experts on what terms are used to replace the death terms would be essential. Alternatively, researchers can check with their participants as to their intended meaning through the process of member check (Lincoln & Guba, 1985). SUMMARY OF KEY ASPECTS Allapproachestoqualitativecontentanalysisrequireasimilaranalyticalprocessof seven classic steps, including formulating the research questions to be answered, selectingthesampletobeanalyzed,definingthecategoriestobeapplied,outlining thecodingprocessandthecodertraining,implementingthecodingprocess,determining trustworthiness, and analyzing the results of the coding process (Kaid, 1989).Wehaveoutlinedhowthisprocessdiffersdependingonthespecificcontent analysis approach used. The success of a content analysis depends greatly on the coding process. The basic coding process in content analysis is to organize large quantities oftextintomuchfewercontentcategories(Weber,1990). Categoriesare patternsorthemesthataredirectlyexpressed in thetextorarederivedfromthem throughanalysis.Then,relationshipsamongcategoriesareidentified.Inthecoding process, researchers using content analysis create or develop a coding scheme to guide coders to make decisions in the analysis of content. A coding scheme is a translationdevicethatorganizesdataintocategories(Poole&Folger,1981).Acodingschemeincludestheprocessandrulesofdataanalysisthataresystematic,logi Hsieh, Shannon / PROBLEMS WITH INTERVIEWS 1285 cal, and scientific. The development of a good coding scheme is central to trustworthiness in research using content analysis (Folger, Hewes, & Poole, 1984). Key differences among conventional, directed, and summative approaches to contentanalysis centeron how initial codes are developed. In a conventional contentanalysis,categoriesarederivedfromdataduringdataanalysis.Theresearcher isusuallyabletogainaricherunderstandingofaphenomenonwiththisapproach. With a directed content analysis, the researcher uses existing theory or prior research to develop the initial coding scheme prior to beginning to analyze the data (Kyngas&Vanhanen,1999).Asanalysisproceeds,additionalcodesaredeveloped, and the initial coding scheme is revised and refined. Researchers employing a directed approach can efficiently extend or refine existing theory. The summative approach to content analysis is fundamentally different from the prior two approaches.Ratherthananalyzingthedataasawhole,thetextisoftenapproachedas singlewordsorinrelationtoparticularcontent.Ananalysisofthepatternsleadsto an interpretation of the contextual meaning of specific terms or content (Table 4). CONCLUSIONS Different research purposes require different research designs and analysis techniques(Knafl&Howard,1984).Thequestionofwhetherastudyneedstouseaconventional,directed,orsummativeapproachtocontentanalysiscanbeansweredby matchingthespecificresearchpurposeandthestateofscienceintheareaofinterest with the appropriate analysis technique. Itisimportantforhealthresearcherstodelineatethespecific approachtocontent analysis they are going to use in their studies before beginning data analysis. Creating and adhering to an analytic procedure or a coding scheme will increase trustworthinessorvalidityofthestudy.Carefuldescriptionofthetypeofapproach to content analysis used can provide a universal language for health researchers and strengthen the method’s scientific base. Examples used in this article were drawnfromtheareaofresearchonendoflife, butthecontentanalysis techniques described could be used in a broad range of studies. Content analysis offers researchersaflexible,pragmaticmethodfordevelopingandextendingknowledgeof the human experience of health and illness. 1286 QUALITATIVE HEALTH RESEARCH / November 2005 TABLE 4: Major Coding Differences Among Three Approaches to Content Analysis Timing of Defining Source of Type of Content Analysis Study Starts With Codes or Keywords Codes or Keywords Conventional content analysis Observation Codes are defined during data analysis Codes are derived from data Directed content analysis Theory Codes are defined before and during data analysis Codes are derived from theory or relevant research findings Summative content analysis Keywords Keywords are identified before and during data analysis Keywords are derived from interest of researchers or review of literature REFERENCES Babbie, E. (1992).The practice of social research. New York: Macmillan. Barcus,F.E.(1959).Communicationscontent:Analysisoftheresearch1900-1958(Acontentanalysisofcontent analysis). Unpublished doctoral dissertation, University of Illinois, Urbana-Champaign. Berelson, B. (1952).Content analysis in communication research. Glencoe, IL: Free Press. Budd, R. W., Thorp, R. K., & Donohew, L. (1967). Content analysis of communications. New York: Macmillan. Catanzaro, M. (1988). Using qualitative analytical techniques. In N. F. Woods & M. Catanzaro (Eds.), Nursing research: Theory and practice(pp. 437-456). St. Louis, MO: C. V. Mosby. Cavanagh, S. (1997). Content analysis: concepts, methods and applications.Nurse Researcher, 4(3), 5-16. Coffey,A.,&Atkinson,P.(1996).Makingsenseofqualitativedata:Complementaryresearchstrategies.Thousand Oaks: Sage. Curtis,J.R.,Wenrich,M.D.,Carline,J.D.,Shannon,S.E.,Ambrozy,D.M.,&Ramsey,P.G.(2001).Understanding physicians’ skills at providing end-of-life care: Perspectives of patients, families, and health care workers. Journal of General Internal Medicine, 16, 41-49. Downe-Wamboldt,B.(1992).Contentanalysis:Method,applications,andissues.HealthCareforWomen International, 13, 313-321. Ferrell,B.,Virani,R.,Grant,M.,&Borneman,T.(1999).Analysisofcontentregardingdeathandbereavement in nursing texts.Psychooncology, 8, 500-510. Ferrell, B., Virani, R., Grant, M., & Juarez, G. (2000).Analysis of palliative care content in nursing textbooks. Journal of Palliative Care, 16, 39-47. Folger,J.P.,Hewes,D.E.,&Poole,M.S.(1984).Codingsocialinteraction.InB.Dervin&M.J.Voigt(Eds.), Progress in communication sciences (pp. 115-161). Norwood, NJ: Ablex. Hickey, G., & Kipping, C. (1996).Issues in research. Amulti-stage approach to the coding of data from open-ended questions.Nurse Researcher, 4, 81-91. Holsti,O.R.(1969).Contentanalysisforthesocialsciencesandhumanities.Reading,MA:Addison-Wesley. Kaid, L. L. (1989). Content analysis. In P. Emmert & L. L. Barker (Eds.), Measurement of communication behavior (pp. 197-217). New York: Longman. Kirchhoff, K. T., Beckstrand,R. L., & Anumandla,P. R. (2003).Analysisof end-of-life content in critical care nursing textbooks. Journal of Professional Nursing, 19, 372-381. Knafl,K.A.,&Howard,M.J.(1984).Interpretingandreportingqualitativeresearch.ResearchinNursing and Health, 7, 17-24. Kondracki,N.L.,&Wellman,N.S.(2002).Contentanalysis:Reviewofmethodsandtheirapplicationsin nutrition education.Journal of Nutrition Education and Behavior, 34, 224-230. Krippendorf, K. (1980).Content analysis: An introduction to its methodology. Beverly Hills, CA: Sage. Kübler-Ross, E. (1969).On death and dying. New York: Macmillan. Kyngas,H.,&Vanhanen,L.(1999).Contentanalysisasaresearchmethod[Finnish].Hoitotiede,11,3-12. Levy,M.M.(2001).End-of-lifecareintheintensivecareunit:Canwedobetter?CriticalCareMedicine,29, N56-N61. Lincoln, Y. S., & Guba, E. G. (1985).Naturalistic inquiry. Beverly Hills, CA: Sage. Lindkvist,K.(1981).Approachestotextualanalysis.InK.E.Rosengren(Ed.),Advancesincontentanalysis (pp. 23-41). Beverly Hills, CA: Sage. Manning,K.(1997).Authenticityinconstructivistinquiry:Methodologicalconsiderationswithoutprescription. Qualitative Inquiry, 3, 93-115. Mayring,P.(2000).Qualitativecontentanalysis.Forum:QualitativeSocialResearch,1(2).RetrievedMarch 10, 2005, from http://www.qualitative-research.net/fqs-texte/2-00/02-00mayring-e.htm McEwen,M.(2004).Analysisofspiritualitycontentinnursingtextbooks.JournalofNursingEducation,43, 20-30. McTavish, D.-G., & Pirro, E.-B. (1990). Contextual content analysis.Quality and Quantity, 24, 245-265. Miles,M.B.,&Huberman,A.M.(1994).Qualitativedataanalysis:Anexpandedsourcebook.ThousandOaks, CA: Sage. Morgan,D.L.(1993).Qualitativecontentanalysis:Aguidetopathsnottaken.QualitativeHealthResearch, 3, 112-121. Morse, J. M. (1991).Qualitative nursing research. Newbury Park, CA: Sage. Morse,J.M.,&Field,P.A.(1995).Qualitativeresearchmethodsforhealthprofessionals (2nded.).Thousand Oaks, CA: Sage. Hsieh, Shannon / PROBLEMS WITH INTERVIEWS 1287 Nandy,B.R.,&Sarvela,P.D.(1997).Contentanalysisreexamined:Arelevantresearchmethodforhealth education.American Journal of Health Behavior, 21, 222-234. NationalInstituteofNursingResearch.(2003).Researchthemesforthefuture.RetrievedOctober10,2003, from http://www.nih.gov/ninr/research/themes.doc Patton, M. Q. (2002).Qualitative research and evaluation methods. Thousand Oaks, CA: Sage. Poole, M. S., & Folger, J. P. (1981). Modes of observation and the validation of interaction analysis schemes.Small Group Behavior, 12, 477-493. Potter,W.J.,&Levine-Donnerstein,D.(1999).Rethinkingvalidityandreliabilityincontentanalysis.Journal of Applied Communication Research, 27, 258-284. Rabow, M. W., Hardie, G. E., Fair, J. M., & McPhee, S. J. (2000).End-of-life care content in 50 textbooks from multiple specialties. Journal of the American Medical Association, 283, 771-778. Rosengren,K.E.(1981).AdvancesinScandinaviacontentanalysis:Anintroduction.InK.E.Rosengren (Ed.),Advances in content analysis(pp. 9-19). Beverly Hills, CA: Sage. Tesch, R. (1990).Qualitative research: Analysis types and software tools. Bristol, PA: Falmer. Vincent, J.-L. (1997). Communication in the ICU.Intensive Care Medicine, 23, 1093-1098. Weber, R. P. (1990).Basic content analysis. Beverly Hills, CA: Sage. Wilson,C.T.,&Fletcher,P.C.(2002).Dealingwithcoloncancer:Onewoman’semotionaljourney.Clinical Nurse Specialist, 16, 298-305. Hsiu-Fang Hsieh, Ph.D., is an assistant professor at Fooyin University, Kaohsiung Hsien, Taiwan. Sarah E. Shannon, Ph.D., is an associate professor at the University of Washington, Seattle. 1288 QUALITATIVE HEALTH RESEARCH / November 2005