pm o Rabiee: Focus-group interview and data analysis (2004l' o Whittemore and Knafl: The integrative review: updated methodology (2005) o Catherine Pope, Sue Zlebland, Nicholas Mays, Qualitative research in health care . Analysing qualitative data (2000) BMJ 2000;320:LL4-6.
18
Embed
o Rabiee: Focus-group interview and data analysis (2004l - IQ ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
pmo Rabiee: Focus-group interview and data analysis (2004l'
o Whittemore and Knafl: The integrative review: updated
methodology (2005)
o Catherine Pope, Sue Zlebland, Nicholas Mays, Qualitative research
in health care
. Analysing qualitative data (2000) BMJ 2000;320:LL4-6.
Proceedings of the Nutrition Society (2t04),63, 655-660
@ The Author 2004
Focus-group interviews are becoming increasingly popularin health research for exploring what individuals believe orfeel as well as why they behave in the way they do. Theyoffer a useful vehicle for involving users in care manage-ment and sFategy developmen! needs assessment, partici-patory planning and evaluation of health promotion andnubition intervention progra¡nmes (Basch; 1987; Gregory,1991; Duke et aL 1994; Kitzinger, 1995; Higingbottom,1998; Richa¡dson & Rabiee, 2001; Van Dillelr et al.2O03).The main aim is to understand, and explain, the meanings,belieft and culfirres ehat influence the feelings, attitudesand behaviours of individuals. It is ideally suited forexploring the complexity surrounding food choice anddietary and other lifestyle behaviou¡s within the context oflived experience, and in ways encourage the participants toengage positively with the process of the resea¡ch. Draw-ing on relevant literature as well as personal experience,the fi¡st section of the present paper describes some of thecommonly-asked questions about the use of focus-groupinterviews in health resea¡ch and the issues that need to beconsidered before and during this metbod of data collec-tion. The second section provides a detailed and practical
DOI: lCJ079/PNS2004399
account of Kmeger's (1994) framework of data analysis.Kmeger's (1994) framework- analysis has been chosen
because person'al experience has indicated that theapproaches are easily accessible to both researchers and
students, a¡d that it is one of the most useful startingpoints for analysing fccus-group interviews.
What-is a focus-group interview?
A focus group is, according io i.ederman (see Thomas er al.1995), 'a technique involving the use of in-depth groupinterviews in which participants are selected because theyare a purposive, although not necessarily representative,sampling of a specific population, this group being'focused' on a given topic'. Participants in this type ofresearch are, therefore, selected on the criteria that tbeywould have something to say on the topic, are within theage-range, have simila¡ socio-characteristics and would becomfortable talking to the interviewer and each otber(Richardson & Rabiee, 2001). This approach to selectionrelates to the concept of 'Applicability', in which subjects
Focus-group interview ar¡d data analysis
Fatemeh RabieeSchool of Health and Polícy Studies, University of Central England-Birmingham,B42 2SU, UK
In recent years focus-group intewiews, as a me¿rns of-qualiøtive data collection, have gained
popularity amongst professionals within the health and social ca¡e arena. Despite thispopularity, analysing qualitative data, particularly focus-group interviews, poses a challenge to
most practilioner ¡esearchers. The present paper responds to the ¡eeds-¡xpressed by public
health nutritionists, community dietitians and health development specialisc-foliowing-twotraining sessions organised collaboratively by the Health Development Agency, the Nut¡itionSociety and the British Dietetic Association in 2003. The focus of the present paper is on the
concepts and application of framework analysis, especially the use of Kmeger's framework. Itprovides some practical steps for the analysis of individual data, as well as focus-group data
using examples from the author's own research, in such a way as to assist thÈnewcomer toqualitative ¡esea¡ch to engage with the methodology-'Thus, it comPlemen',s the papers byDraper (2004) and Fade (2004) that discuss in detåil the comPlementary roie of qualit'atíve daøin researching human behaviours, feelings and attitudes. Draper (20M) has provided theoreticaland philosophical bases for qualiøtive daø analysis. Fade (2004) has described interpretativephenomenology analysis as a nethod of analysing individual interview data. The Present PaPer,using framework analysis concentrating on focus-group interviews, provides another approach
to qualiøtive data analysis.
Focus-group intervíews: Qualitative data anal¡sis
Corresponding sutùor: Professor Fatemeb Rabiee, fu +44 12133I 5498, email [email protected]
6s6 F. Rabiee
are selected because of their knowledge of the stucly area(Burrows & Kendall, 1997). One of the distinct featuresof focus-group interviews is its group dynamics, hence thetype and range of data generated through the socialinteraction of the group a¡e often deeper and richerthanthose obtained from one-to-one interviews (see Thomaset aL 1995).
Focus gfoups could provide information about a rangeof ideas and feelings that individuals have about certainissues, as well as illuminating the differences in perspec-tive between groups of individuals. For example, usingfocus-group interviews diverse views about health and
issues affecting trealth amongst professionals and themembers of the public working and living in an outer-city deprived a¡ea of Birmingham have been generated' Inthis research the data derived from the community (young
people, single-parent young women and professional womenliving in the area) are distinct and yet cover a range ofissues affecting the health of the participants. The nar¡a-tives generated from professionals working in this area,
although very different, complement the range of issues
raised by tbe public (Rabiee, 1999). Focus grouPs can
geneftic large amounts of daø in a relatively short timespan, and the findings may be used to precede quantitativeprocedures. Like one-to-one interviews, the results offocus-group interviews can be presented in uncomplicatedways using lay terminology supported by quotations fromthe participants. Krueger & Casey (2000), while describingin detail the advantages and disadva¡rtages of focus grouPs,
point out wben to use focus-group interviews and when notto use them.
The uniqueness of a focus group is its abiliry to generate
data based on the synergy of the group interaction (Green
et al.2O03). The members of tbe group should, therefore,feel comfortable with each otber and engage in discussion.Krueger & Casey (2000) point out that for some in-dividuals self-disclosu¡e is natural and comfortable, whilefor others it requires trust and effof. It is fo¡ this reason
that they recommend investing time and effort in selectingmembers of the group. Krueger (1994) believes rich datacan only be generafed if individuals in the group are
prepared to engage fully in the discussion and, for thisreason, advocates the use of a homogenous group. Based
on the topic under investigation Krueger (1994) suggessthat participants should sha¡e similar characteristics:gender group, age-range, ethnic and social class back-ground. Most resea¡chers, although they would notdisagree with the concept of homogeneiry, recommendthat participants sbould not know each other, thus
encouraging mo¡e honest and spontaneous expression ofviews and a wider range of responses It also prevents set
behaviours relating to pre-existing relationships and
patterns of leadership in the group (see Thomas ef ø1.
1995). Kitzinger (1994), on tt¡e other hand, advocates theuse of pre-existing groups, as acquaintances could relate to
each other's comments and may be more able to challengeone another. Personal experience indicates that whenexploring very sensitive and personal issues the use ofpre-existing groups might be advantageous, as there isalready an extent of trust amongst the members of thegroup, which will encourage the expression of views. This
factor is particularly important when very little informationis available on the topic under investigation and tlle data
from erploratory focus-group interviews is to be used toformulate and design a large-scale study. For example, in a
study widening paficipation and increasing access tohigher education amongst Muslim women (Rabiee &Thompson, 2000) found there was clear benefit from pre-
existing acquaintance. The women who all knew each other,felt comfonable talking about a number of very personal
issues that affected their participation, and were able to
express thei¡ views on how to widen participation.Regardless of whether a pre-existing or newly-formed
group is used, the important role of the group facilitator ormoderator should not be underestimated-(Krueger, 1994;
Burrows & Kendall, 1997). A skilful moderator, as well as
being able to manage the existing relationship, could c¡eate
an environment in which the paficipants who do not linoweach otler feel relaxed and encouraged to engage and
exchange feelings, views and ideas about an issue. Apartfrom the facilitator or moderator a note taker should bcpresent to observe non-verbal interactions, indicate theimpact of the group dynamic, document exchanges ofviews and the general content of discussion and note whichstatement is made by which particular individual, therebysupplementing the oral text and enabling a fuller analysisof tbe data (Kitzinger, 1994, 1995).
Another frequently-asked question is about the numberof focus groups. Kmeger (1994) suggests continuing withrunning focus groups until a clea¡ pattern emerges and
subsequent groups produce only repetitious information(theoretical saturation). Howeve¡, several authors, includ-ing Kmeger (1994), suggest that for a simple research
question the number of focus grouPs necessary may onlybe th¡ee or four (for a full discussion of this issue, see
Burrows & Kendall, 1997).The optimum number of participants for a focus grorrg
may vary. Krueger & Casey (2000) suggest between six-and eight participants, as smalle¡ g:oups show greater
potential. However, the number generaily suggested as
being manageable is between six and ten participants;large enough to gain a variety of perspectives and smallenough not to become disorderly or fragmented. Recruitingparticipants for a focus-group interview is a big challenge,particularly if the informants belong to low-income crminority ethnic groups. Experieáce of researching these
groups suggess that lack of confidence and low self-esteem often prevent these individuals participating il a
group discussion. Focus-group interviews could, therefo¡e,be used as a vehicie to emPower the participants fromthese communities. Another potential problem in usingfocus groups is the number of non-attende¡s. Therecommendation is, therefore, to over-recruit by 10-25 Vo,
based on the topic and groups of participants. In order tomaximise participation it is important to obtain an agreed
date from the informants well in advance of the interviewsand to ¡emind them a few days before they start.
Each group interview usually lasts approximately l-2h,based on the complexity of the topic under investigation,number of questions and the number of participants. It is,therefore, ethical and good practice to wa¡n the partici-pants about their time commitment.
Developing qualiøtive research methodskills 657
Data anadysis
Qualitative research and, in particular, focus-group inter-views generate large amounts of data, which tend tooverwhelm novice as well as experienced researchers Alh interview could easily take 5-6h to hanscribe in full,leading to thirty to forty pages of tanscripts. Thus, a
central aim of data analysis, according to Robson (1993),is to reduce data. Yin (1989) points out that data analysisconsists of a number of stages, i.e. examining, categorisingand tabulating or otherwise recombining the evidence, inorder to add¡ess the initial goal of a study. Krueger &Casey (2000) build on this concept and suggest that thepurpose should drive the analysis; they believe that'analysis begins by going back to the intention cf thestudy and survival requires a clea¡ fix on the purpose of thestudy'. Following this concept, although ha¡d at times, isexEemely helpful for managing the data, making sense ofwhat is going on, getting rid of extra and i¡relevantinformation and travelling safely through the maze of largeand complicated paths of information.
The process of qualitative onalysis aims to bringmeaning to a situation rather than the sea¡ch fo¡ tn¡thfocused on by quantitative research. Strauss & Corbin(1998) describe analysis as '... the interplay between¡esearchers and data', acknowledging that there is an
extent of subjective selection and interpretation of thegenerated data. It is important to acknowled.ge thatregardless of the type of research (qualitative or quantita-tive) an extent of subjectivity exits. The distinction shouldbe seen more in relation to the stage of tÏe process ratherthan just the type of subjectivity. For example, the issue ofsubjectivity in surveys is often at the stage of designing thequestionnaire, the pre-set answers at this stage couldprevent the expression of other potential answers; hence,indicating an extent of selective answers and interpreta-tions of the issue under investigation.
Having made this point, in orde¡ to minimise thepotential bias introduced in analysing and interpretingfocus group data Krueger & Casey (2000) point out thatthe analysis should be systematic, sequential, veriûable,and continuous. Following this path provides a trail ofevidence, as well as increasing the extent of dependabiliry,consistency and conformability (Lincoln & Guba, 1989) ofthe data, important issues'for assessing the quality ofqualitative data (Secker et aI.1995).
The first step in establishing a trail of evidence is a clea¡procedure of data analysis, so that the process is clearlydocumented and understood. This step would allowanother resea¡cher to verify the findings; it safeguardsagainst selective perception and increases the rigour of thestudy. In order to achieve this objective, there must besufficient data to constitute a trail of evidence.
Although the main source of data analysis is therecorded spoken language derived f¡om the interview;nevertheless, reflection about the interview, the settingsand capturing the non-verbal communication expressed bythe member of the groups would add a valuable dimensionto the constn¡ction and analysis of data. This record couldbe in the form of aD audiotape or a videotape. It isrecommended that a reflective diary should be kept by the
-facilitator or moderator and that observational notes shouldbe wrinen immediately after each focus-group interview.
,A.pproaches to data analysis
There a¡e a number-of approaches to the analysis ofqualitative data. In practice, as Green & Thorogood (2004)
identified, most -researchers use a combination ofapproaches. The presenL paper describes Krueger's ( I 994)framework analysis, but also incorporates some key stages
of 'fr¿mework analysi+' described by Ritchie & Spencer(1994). The advantage of the Krueger (1994) approach isthat it provides a clear. series of steps, which couldfrrst-time researchers to manage thecomplex nature of qualitative daø
largemuch
amountheþand
more easilyTramework analysi's is used for botb individual and
ularly focus-group analysis, occurs concurrently with datacollection. Krueger (1994) suggests that a heþful way ofthinking about this role is to consider a continuum ofanalysis ranging fronthe mere accumulation of raw datato the interpretation of data:
the analysis continuum:raw data; descriptive statements; interpretation.
It isìmportant to point out that analysis does not take placein a linearform and that one part of the process overlapsanother. 'Frameworlcanalysis' as described by Ritchie &Spencer (1994), is 'an analytical process which involves a
number of distinct though highly interconnected stages'.
The five key stages outlined are: familiarization; identify-ing a themacie fre.rnework; indexing; charting; mappingand interpretation. The other distinctive aspect of frame-work analysis-is that although it uses a thematic approach,
it fllows themes to develop both f,¡om the resea¡chquestions and from the narratives of resea¡ch participants.
The process of data ¡n¡lysis begins during the data
coilection, öy sliilfully facilitating the discussion andgenerating rich data from the interview, complementingthem with the observational notes and typing the ¡ecordedinformation. This stage is followed by familiarisation withthe ûata, which- can be achieved by listening to tapes,
reading-'.he transcripts in their entirety several times and
reading the observational notes taken during interview and
summary notes written immediately afte¡ the interview.The aim is to immerse in the details and get a sense of theinterview as a whole before breaking it into parts. Duringthis process tùe major themes begin to emerge.
The next stage involves identifying a thematic frame-work, by writing memos in the margin of the text in theform of shof phrases, ideas or concepts arising from thetexts and beginning to develop categories At this stage
descriptive statements are formed and an analysis is
carried out on the data under the questioning route.The thi¡d stage, indexing, comprises sifting the data,
highlighting and sorting out quotes and making compari-sons both within and between cases. The fourth stage,
charting, involves lifting the quotes from their originalcontext and re-arranging them under the newly-developed
658 F. Rabiee
Table l. Text transcribed during the process of developing and writing up a policy for mainstream Children's Sêrvjces: indexing and
charting slage
Line no. Transcription Code'
1M165T66
167168169
170171
172173174175176
177'178
179180
ln that they thought it was positive and it actually movedforward, yes I th¡nk so, w¡th some of the feedback yes, yes, I
think so, because that's I think links into the next about theplanning mainstream service, because all the time in terms ofdeveloping you-know, a policy and a sirategy, what was alsofirmly in my mind was well how do we make this hapPên,
"êsÈÞ?ri¡ ,sjärt
ffi because the having a
-ë"!trçElÐoÞffi.änÈËg q
üJ_gäólæ So at the core I'm very sort of practical being, I like toknow how th¡ngs are going to work because that fits into itôé,T,ÉfÏ
jflffiÈ"ffiffiffi lthink öS,il
F"A*iF.nþXrmi
'| .1
1.2
32
-Codes relale to highl¡ghted text.
âppropriate thematic content. Table I grves an examPle ofindexing anC charting from a transcribed text.
Indexing and charting could also be viewed as managingthe data. One of the most important aspects of this task isdata reduction, which is achieved by comparing aadcontrasting data and cutting and pasting similar quotestogether. In order to manage this stage successfullyKrueger (1994) suggests the following practical steps.
Practical steps for managing and sorting out data
Krueger & Casey (2000) advocate the use of either a longtable or a computer-based approach for cutting, pasting,sorting, arranging and rearranging data through comparingand contrasting the relevant information. Although thereis specialised software such as QSR NUT*IST (Richards,1998), it is possible to analyse the transcripts usingMicrosoft Word, or indeed 'by hand'.
The procedure for the 'iong hble' approacb requireshaving_access to either a long table or a room with lots offloor or wall space. Before cutting the transcripts apa¡t, itis important to:
a) number each line of each transcript;b) make two hard copies of each transcript; one to cut uP
and one that stays intact;c) print transcripts on different coloured paper, e.g,
d) ¿urange the working transcript in a reasonable order,i.e. sequence in which the interview took place,categories of participants: age, young people; socialgroup, low-income families or professionals; gender,male or female;
e) have enough pages of flipchart or newsprint. Place thepages on the long table, on the floor or on the wall.'Write on each page one of the focus group questionsto be analysed. It may also be necessary to divide thepages of newsprint or flipcbart into sections to
represent different groups of paficipants, i.e. youngpeople, single parents, professionals.
Having prepared for this stage, Krueger & Casey (2000)suggest that the researcher should read each quote andanswer these four questions:
l. did the participant answer the question that was
asked? If yes, go to question 3; if no, go to question2; ifdon't know, set it aside and review it later;
2. does the comment answer a different question in the
focus group? If yes, move it to the appropriatequestion; if no, go to question 3;
3. does the comment say something of importance aboutthe topic ? If yes, put it under the appropriate question;if no, set it aside;
4. is it something that has been said ea¡lier? If yes, startgrouping like quotes together; if no, sta¡t a separatepile.
Having gone through this systematic process, soon thenewsprint pages are fiIled up with relevant quotes. For the
time being, le¿ve those 'not so relevant' quotes that havebeen set aside, it may be necessary to go back to thosequotes at a later stage.
Interpretation of data
The data are now ready for the ñnal stage of analysis, i.e.mapping and inte¡preting. One of the tasks here is not onlyto make sense of the individual quotes, but also to beimaginative and analytical enough to see the relationshipbetween the quotes, and the links between the data as awhole. Krueger (1994) provides seven established criteria,wbich suggest the following headings as a framework forinterpreting coded data; words; context; intemal consistency;frequency and extensiveness of comments; speciflcity ofcomments; intensity of comments; big ideas.
The following extracts taken from a piece of resea¡ch onexperience of heart attack provides an example of how to
)
Developing qualitative research method skills 659
apply these headings to the data:
JF: :I didn't think I was having a heart attack because itwas my arm.'DS: 'But certainly before I had a heart attacþ I thoughtOh my God a heart attack that would be hor¡endous,you know but having one well I don't know, I reallydidn't feel anything. I wasn't in a great deal ofpain.'EM: 'And they kept saying have you got pain and I saidno and they said you must have pain. I said but Ihaven't. You haven't got it in your arm or your chest. Ïsaid no.'JG: 'Well with tbe frauma of my father dying a weekafter I came out of hospital and the problems we havehad has made it a difficult time. I have found it morestressful.'KZ: 'I don't believe I've had a heart attack, I still find itvery difficult to understand, but the tests showed I didso that's it.'KN: 'Well, what is difficult about it because it's what alot of people say.'KZ:'Well, perhaps it's because my visions or ideas of a
hea¡t attack were that I would be suffering morelainand I didn't. Was it possible I had a lea¡t atrâck a yearago and not know it?'
Consider the actual words used and their meaning'When the participants talk about the term 'heart attack' itbecomes evident that their actual experience shows [ittlerelationship with thei¡ understanding of the term before theevent. There seems to be a belief that a heart atiackinvolves a lot more suffering than they had expeÉenced, socoming to terms with what had happeneci to them involvesre-defining this key term.
Consider lhe context
The wording of the moderators' questions and subsequentcomments made by others in the group infllences thecontext within which the comments a¡e made.
The respondents are never asked directly to talk aboutthe actual experience of having a hea¡t attacþ althoughwhen the conversation gets round to talking about howthey feel now and what they think about the future, itseems that they need to recount their experience. Tlrey alsotalk about previous occasions when simila¡ things hap-pened to them.
Consi.der the frequency and extensiveness of commenls
Frequency relates to consideration ofhow often a commentor view is macie, while having enough insight to 'spot a
gem' when it comes along. The term extensive refers to thenumber of participants who express a particular view.
All participants talk at length about their experienceapart from JG, who had lost his father since coming out ofhospital so had things on his mind other tban his own stateof bealth. They are simply telling the story again, and itseems that this process is about coming to terms with whathad happened to them. Along with the actual hea¡t attackexperiences, the patients talk about previous times when
they had felt unwell and they attempt to make sense ofthem in light of thei¡ recent diagnosis.
KZ: 'Was it possible I had a-hea¡t attack a year ago andnot-krow it?'
Intensíty of the comments
Consider ''.Ie. depth of feeling in which comments orfeelings are expressed; the following are some examples of-'hor'+people feel now'. Women used more negative termsto desciibe their current state of health. There is somevariation-i¡ the ex-patients comments about how they arefeeling now. All the men describe their current state ofhealth.quite positively, whe¡eas several of the women usemore negative terms.
MG: 'It has been a very depressing couple of weeks,'.trinking when am I going to be all right, you know.WhaEean I do, how far can I go.'JF: 'I get very depressed. Sometimes I sit and cry. Icouldn'treally tell you what I am crying for. I mean itsjust I feel a bit low in myself towards tbe end of the daysort of thing. It only lasts a little while.'MH: T.easonable, but scared. It's really lonely and thenI get depressed. I've got no-one to talk to.'
Internal consistency
Ccnside¡ any changes in opinion or position by theprticipants. For example, the following quotes aboutchildren in primary school clearly indicate some changesin participant's views and an extent of consensus:
Teacher A: 'Children in my class are always tired, can'tfocus on anything more than five minutes.'Playgroup leader: 'You are right, most of them come toschool without having a breakfast, what do you expect?'Teacher B: 'A clea¡ lack of parenting skills, they do notsend the children to bed early enough, do they?'Pl4ygroup assistant: 'Haven't we been a bit judgemen-tal? There are some people including myself who don'tlike eating anything frrst thing in the morning.'Teacher B: 'Good point, my husband never eats
brealdast during weekdays, and he can't stop eating onSaturclay, as he gets up around 10.'
Specifc ity of responses
Greater-aflention is placed on responses referring topersonal experience as opposed to hypothetical situations.Forexample, JG's reference to his father's death gives a
spocifrc and related answer.
Big ideas
Consider larger trends or concepts that emerge from an
accumulation of evidence and cut across the variousdiscussions. Krueger (1994) suggests taking a break for a
few days at this stage in order to ¡efocus on the big picture.
Development of the framework
These criteria are reduced to the following five headings in alater publication (Krueger & Casey, 2000): frequency;
)
660
Tabie 2. Haadings to help in the interpretation of focus group data Bu¡rows D & Kendall S (1997) Focus groups: Wha+-are they and
how can they be used in nursing and health care resea¡ch ?
Social Sciences in Health 3,244-253.Draper AK (2004) The principles and application of qualita-
tive resea¡ch. Proceedings of the Nutrition Sociery 63,64i-646.
Duke SS, Gordon-Sosby K, Reynolds KD & Gram IT (1994) Astudy of b{east cancer detection practices and beliefs in blackwomen attending public bealth clinics. Health EducationResearch 9,331-342.
Fade S (20M) Using interpretative phenomenological analysis forpublic health nutrition and dieætic ¡esearch: a practical guide.
Proceedings of the Nutrition Society 63,64'l-653.Green I & Thorogood N (2004) Quclitative Methods in Health
Research. London: Sage Publications.Green JM, Draper AK & Dowler EA (2003) Short cuts to safety:
risk and 'rules of thumb' in accounts of food choice. I]¿al¡å,Risk ond Society 5,33-52.
Gregory S & McKie L (1991) The smea¡ test: listening towomen's views. Nursing Standard 5,32-36.
Higingbottom G (1998) Focus groups: thei¡ use in HealthPromotion research. Community P ractitioner 7 2, 360-363.
Kitzinger J (1994) The methodology of focus groups: the
importance of interactions between research pa¡ticiPants.
Sociology of Health and lllness 16, 103-121.Kitzinger J (1995) Qualitative research: introducing focus groups.
Brítish M edical J oumal 3ll, 299-302.Krueger RA (7994) Focus Groups: A Practical Guide for Applied
Research. Thousand Oaks, CA: Sage Publications.Krueger RA & Casey MA (2000) Focus Groups: A Practical
Guide for Applied Research,3¡d ed. Thousand Oaks, CA: Sage
Publications.Lincoln Y & Guba E (1989) Fourth Generation Evaluation.
Newbury Park, CA: Sage Publications.Rabiee F (1999) Evidence based practice: its relevance to
nutritional intervention programmes. P roceedin gs of NutritionSociery 58, 504.
Rabiee F & Thompson D (2000) Widening Participatíon -Increasing Access to Higher Education for Muslim Women.Birmingham: University of Cent¡al England and BirminghamUniversity.
Richa¡ds L (1998) Qvalitatíve Solution and Research
Robson C (1993) The Real World Research-A Resource forSocial Scientists and P ractitioner-researchers, Oxford: Black-well Publications.
Secker J, Wimbush E, Watson J & Milbum K (1995) Qualitativemethods in health promotion resea¡ch: some criteria forqù^ßry. Health Education Journal 54,'74-87.
Strauss A & Corbin J (1998) Basícs of Qualitative Research,
Techniques and Procedures for Developing Grounded Theory,
2nd ed. Thousand Oaks, CA: Sage Publications.Thomas L, MacMillan J, McColl E, Hale C & Bond S
(1995) Comparison of focus group and individual interviewmethodology in examining patient satisfaction with nursingcate. Social Sciences in Health l,20Ç2L9.
Yin RK (1989) Case Study Research: Design and Methods,2nded. London: Sage Publications.
F. Rabiee
Krueger (1 994)Krueger & Casey
(2000)F Rabiee
recommendalion
1. Words2. Conlext3. lnlernal
consistency4. Frequency and
extensiveness5. lntensity
of comments6. Specificity ol
responses
7. Big ideas
1. Frequency
2. Motion
3. Speciflcity ofresponses
4. Extensiveness5. Big p¡cture
1. Words2. Context3. ¡nternal
cons¡stency4. Frequency
5. lntensityof comments
6. Specificity ofresponses
7. Extensiveness8. Big picture
specificity; emotions; extensiveness; big picture. The maindifference is that words, context and intemal consistencyhave been excluded from the interpretation, frequency and
extensiveness have been separated into two separate
categories, intensity of comments and big ideas have been
reframed as emotions, and the big picture has been in-t¡oduced. Although the developmeDt ofthe new categories iswelcome, as it is now crisp and concise, experience suggests
that students and the first-time practitioner researcher tendto find that incorporating the th¡ee excluded criteria is easierto follow, it can be applied with more rigour and P¡oduce a
richer interpretation. A modifrcation to the latest criteria isrecommended that includes the concePts of word, contextand internal consistency; therefore, making eight criteriarather than five (see Table 2).
Conclusion
The present paper discusses the role of focus-group inter-views in health and nutrition research. An attempt has
also åeen made to answet some of the questions mostfrequently asked by students and practitioner tesearchers.
Some practical guidance is -provided for the analysisof focus-group interviews, using Krueger (1994) and Ritchie& Speneer (1994) framework analyses. The æralysis ofqualitative data requires the development of new skills, butalso imagination, patience, time and practice. Developing
'.his skill is a good investmerit andthe rewa¡ds are nume¡ous!
Acknowledgements
I would like to thank my colleagues Dr Jackie Landman,Mrs Stephanie Fade and Dr Liza Draper for their usefulcomments and reading this paper. Also, special thanks goto my ex-students, Ch¡istine Richardson, Karen Noy and
Anne-Marie Morris, who, with thei¡ dedication and
enthusiasm for learning, enabled me to further myexperience of applying Krueger's framework and add tothe cu¡rent knowledge on qualitative data analysis.
References
Basch C (1987) Focus g¡oup interview: an under-utilised researchtechnique for improving theory and p¡actice in health educ-
^tion. Health Education Quørterly 14, 411-418.
METHODOLOGICAL ISSUES IN NURSING RESEARCH
The integrative review: updated methodology
Robin rùØhittemore PhD APRN
Associate Research Scientist and Lecturer, School of Nursing, Yale Uniuersity, Connecticut, USA
Kathleen Knafl pt'o
Elizabeth N. Gray Distinguisbed Professor and Associate Dean for Research and Facuhy Affairs, School of Nursing, Oregon
An integrative review is a specific review method that
summarizes past empirical or theoreticâl literature to provide
a more comprehensive understanding of a particular phe-
nomenon or healthcare problem (Broome 1993). Integrative
reviews, thus, have the potential to build nursing science,
informing research, practice, and policy initiatives. \ùüell-done
integrative reviews present the state of the science, contribute
'wHITTEMoRE R. ð¿ KNAFL K. (2005) Journal of Aduanced Nursing 52(5), 546-553
The integrative review: updated methodology
Aim. The aim of this paper is to distinguish the integrative review method from
other review methods and to propose methodological strategies specific to the
integrative review method to enhance the rigour of the process.
Background. Recent evidence-based practice initiatives have increased the need for
and the production of all types of reviews of the literature (integrative reviews,
systematic reviews, meta-analyses, and qualitative reviews). The integrative review
method is the only approach that allows for the combinâtion of diverse method-
ologies (for example, experimental and non-experimental research), and has the
potential to play â greater role in evidence-based practice for nursing. With respect
to the integrative review method, strategies to enhance data collection and extrac-
tion have been developed; however, methods of analysis, synthesis, and conclusion
drawing remain poorly formulated.
Discussion. A modiÊed framework for research reviews is presented to address
issues specific to the integrative review method. Issues related to specifying the
review purpose, searching the literature, evaluating data from primary sources,
analysing data, and presenting the results are discussed. Data analysis methods of
qualitative research are proposed as strategies that enhance the rigour of combining
diverse methodologies as well as empirical and theoretical sources in an integrative
fevlew.
Conclusion. An updated integrative review method has the potential to allow fordiverse primary research methods to become a gfe^ten part of evidence-based
ses, and qualitative reviews). The proliferation of all types ofresearch reviews during the past decade has contributed to
more systematic and rigorous methods. Much has been
learned about the methodology associated with combining
546 O 2005 Blackwell Publishing Ltd
Methodological issues in nursing research
disparate studies into integrated results ând conclusions,
particularly with respect to systematic review and meta-
anâlyses approaches (Cooper 1998, Greenh algh 1,9971. Y et,
concern has been raised that these review methods, while
important for evidence-based practice, do not include the
depth and breadth of nursing research âs they overemphasize
the randomized clinical trial and hierarchies of evidence
(Kirkevold 7997,Evans & Pearson 2001). To date, evidence-
based practice initiatives have viewed different types ofevidence (i.e. quantitative and qualitative) as mutually
exclusive (Evans & Pearson 2001).
The integrative review method is an approach that allows
for the inclusion of diverse methodologies (i.e. experimental
and non-experimental research) and has the potential to play
a gÍeater role in evidence-based practice for nursing. The
integrative review contributes to the presentation of varied
perspectives on a phenomenon of concern and has been
advocated as important to nursing science and nursing
practice (Kirkevold !997,Estabrooks 1.998, Evans & Pearson
2001). However, the complexiry inherent in combining
diverse methodologies can contribute to lack of rigour,
inaccuracy, and bias (Beck 1999, O'Mathuna 2000). Meth-
ods to enhance data collection (i.e. literature search) and data
extraction have been developed (Garrard 2004, Conn et al.
2003); however, methods of analysis, synthesis, and conclu-
sion-drawing remain poorly formulated. This is a consider-
able issue, as the data extracted from primary articles ofdiverse methodologies generally consist of a large repertoire
of varied data. Explicit and systematic methods for data
analysis specific to the integrative review method are needed
to protect against bias and improve the accuracy of conclu-
sions. In addition, little attention has been paid to issues
related to combining empirical and theoretical reports. The
purpose of this paper, therefore, is to distinguish the
integrative review method from other review methods and
to propose methodological strategies specific to the integra-
tive review method to enhance its rigour. An updated
integrative review method has the potential to allow fordiverse primary research methods to become a gfeater part ofevidence-based practice initiatives.
Background
Methods of conducting reviews of the health care literature
have been used since the 1970s in an effort to synthesize
findings from discrete primary studies and to increase the
generalizability of data about a phenomenon (Jackson L980).
Methods to improve rigour continue to evolve because of the
complexiry of conducting a thorough review (Greenhalgh
79971. rü/hi\e the¡e are commonalities to all current review
issues of a particular topic (Broome t9931. The varied
sampling frame of integrative reviews in conjunction with the
multiplicity of purposes has the potential to result in a
comprehensive portrayal of complex concepts, theories, or
health care problems of importance to nursing.
Yet, without explicit and systematic methods specific to
undertaking an integrative review, the risk of error increases
exponentially. Systematic bias and error can occur at any
stage of the review (Oxman 1994, Dunkin 1996). For
example, the literature search stage may be incomplete
without consideration of important primary sources. Data
from primary sources can be incorrectly extracted and
interpreted. Most important, data analysis may be incom-
plete or may not be an accurate synthesis of all of the data
from primary sources. Analysing and synthesizing varied
primary sources is a maior challenge in undertaking an
integrative review. Thus, developing data analysis strategies
is an important prioriry in updating the methodology of the
integrative review. These issues will be discussed in the
following section.
Strategies to enhance rigour in integrative reviews
It is well-documented that research reviews are considered
research of research and therefore should meet the same
standards as primary research in methodological rigour(Ganong 1987, Cooper 1998). Cooper (1998) has delineated
the process of conducting a research review as encompassing
a problem formulation stage, a literature search stage, a data
evaluation stage, â data analysis stage, and a presentation
stage. This framework, and the strategies proposed by this
author, are appropriate to all review methods and anyone
conducting an integrative review would benefit f¡om review-
ing this source. However, Cooper's (L998) framework is
primarily aligned with the systematic review or meta-analysis
method. The issues specific to the integrative review method
and the challenges of combining diverse data sources are not
included. Therefore, this framework will be modified toaddress issues specific to the integrative review method. Arecent integrative review on the concept of integration willprovide an example of decisions and issues associated withthe process ('tù(/hittemore 2005b) (Table 1).
Problem identification stage
The initial stage of any review method is a clear identification
of the problem that the review is addressing and the review
purpose. Subsequently, the variables of inte¡est (that is,
concepts, target population, health care problem) and the
appropriate sampling frame are determined (that is, type of
empirical studies, inclusion of theoretical literature). Having
a well-specified review purpose and variables of interest willfacilitate all other stages of the review, particularly the ability
to differentiate between pertinent and extraneous informa-
tion in the data extraction stage. Data extrâction from
primary research reports can be exceedingly complex because
a wide range of variables will have been studied across
multiple reports. Any integrative review can encompass an
infinite number of variables, issues, or populations; therefore,
clarity of the review purpose is important. A well-specified
research purpose in an integrative review will facilitate the
abiliry to accurately operationâlize variables and thus extrâct
appropriate data from primary sources.
Kirkevold (1,9971 advocated that more integrative reviews
should be carried out from an explicit philosophical or
theoretical perspective, focusing a review within a broad and
diverse sampling frame, in contrast to integrative reviews that
are solely descriptive of existing research. For example, in an
integrative review on the concept of integration, empirical
and theoretical sources were included to advance the under-
standing of the process of integration related specifically to
health and illness (Table 1). In any câse, a clear problem
identiÊcation and review purpose are essential to provide
focus and boundaries for the integrative review process.
Literature search stage
Vell-defined literature search strategies are c¡itical for
enhancing the rigour of any type of review because incom-
plete and biased searches result in an inadequate database
and the potential for inaccurate results (Cooper 1998, Conn
et al. 2003a). Ideally, all of the relevant literature on the
problem or topic of interest is included in the review; yet
obtaining this literature can be challenging and costly (Jadad
et a\.1.998). Computerized databases are efficient and
effective; however, limitations associated with inconsistent
search terminology and indexing problems may yield only
about 50% of eligible studies. Thus, other recommended
approaches to searching the literature include ancestry
searching, journal hand searching, nerworking, and searching
research registries (Conn ef al. 2003b1. Depending on the
purpose and type of literature included in an integrative
review, addressing the issue of publication bias may also be
relevant to the literature search stage (Conn et al. 2003b,
Soeken & Sripusanapan 2003).
In general, a comprehensive search for an integrative
review identifies the maximum number of eligible primary
sources, using at least two to three strategies (ladad et al.
1998, Conn et al. 2003b). Purposive sampling can be
combined with a comprehensive search if appropriate to the
for evaluating and interpreting quâlity in research revre\À/s
exists, how quality is evaluated in an integrative review willvary depending on the sampling frame. For example, in an
integrative review where primary sources are of a similar
research design, calculating quality scores and incorporating
these scores into the design (that is, inclusion or exclusion
criteria) or the analysis may be optimal. In an integrative
review with diverse empirical sources, it may only be
reasonable to evaluate quality in sources that represent
outliers (that is, is methodological quality a viable reason for
the discrepant finding?). In an integrative review with a
diverse sampling frame inclusive of empirical and theoretical
sources, an approach to evaluating qualiry similar to histor-
ical research may be appropriate. In this case, the authenti-
city, methodological quality, informational value, and
representativeness of available primary sources is considered
and discussed in the final report (Kirkevold t997). Theoret-
ical reports may also be evaluated using techniques of theory
analysis and critique (lØalker & Avant 1995, Chinn &Kramer 2004). k can be seen that evaluating quality ofprimary sources in an integrative review is complex. Ideally,
consideration of the quality of primary sources in an
integrative review is addressed in a meaningful way. For
example, in a review that encompasses theoretical and
empirical sources, rwo quality criteria instruments could be
developed for each type of source and scores could be used as
criteria for inclusion/exclusion or as â variable in the data
analysis stage as identified in the integrative review of the
concept of integration (Table 1 ). Further practical application
and discussion of these proposed strategies are indicated.
Data analysis stage
Data analysis in research reviews requires that the data from
primary sources are ordered, coded, categorized, and sum-
marized into a unified and integrated conclusion about the
research problem (Cooper 79981. L thorough and unbiased
interpretation of primary sources, along with an innovative
synthesis of the evidence, are the goals of the data analysis
stâge.
Strategies for data analysis with integrative reviews are one
of the least developed aspects of the process, yet are one ofthe most difÊcult aspects and potentially fraught with error.
Therefore, a systematic analytic method should be explicitly
identified before undertaking the review. Primary resea¡ch
methods of analysis developed for mixed-method and qual-
itative designs are particularly applicable to the integrative
review method allowing for iterative comparisons across
primary data sources (Miles & Huberman 1994, Tashakkori
Ec Teddlie 7998, Patton 2002).
A constant comparison method is one overarching
approach used in a broad array of qualitative designs that
converts extracted data into systematic categories, facilitating
rhe distinction of patterns, themes, variations, and relation-
ships (Glaser 1978, Miles & Huberman 7994,Patcon2002l.
Initially, extracted data are compared item by item so that
similar data are categorized and grouped together. Subse-
quently, these coded categories are compared which further
the analysis and synthesis process. In the integrative review
method, this approach to data analysis is compatible with the
use of varied data f¡om diverse methodologies. The method
consists of data reduction, data display, data comparison,
conclusion drawing, and verification (Miles & Huberman
1994). These processes will be explained in more detail.
Data rcduction
The first phase of data reduction involves the determination
of an overall classification system for managing the data from
diverse methodologies. The primary sources included in the
integrative review need to be divided into subgroups
according to some logical system to facilitate analysis. In an
integrative review, this initial subgroup classification can be
based on rype of evidence and analysed sequentially (that is,
examining all qualitative or descriptive studies on topic, then
correlational or compârâtive designs, and lastly any inter-
vention or experimental designs). This initial subgroup clas-
sification can also be based on chronology, settings (that is,
rural or urban), sample characteristics (that is, gender, SES)
or by a predetermined conceptual classification (that is,
experience of participants, attitudes, and behaviours) (Brown
1999,Patton2002l, and analysed by topic. For example, in
considering an integrative review on lifestyle change in type 2
diabetes, the initial categorization may include the perspec-
tive of individuals attempting lifesryle change, the barriers
and facilitators to lifestyle change, and the behaviours or
interventions that promote lifestyle change.
Next, data reduction involves techniques of extracting and
coding data from primary sources to simplify, abstract, focus,
and organize data into a manageable framework. Reliable
and valid coding procedures are essential to ensure meth-
odological rigour (Broome 1993, Brown ¿/ al. 20031.
Predetermined and relevant data of each subgroup classifica-
tion are extracted from all primary data sources and
compiled into a matrix or spreadsheet (Miles Ec Huberman
7994, Garrard 2004). Thus, each primary source is reduced
to a single page with similar data extracted from individual
sources (of each subgroup classification). This approach
provides succinct organization of the literature which facili-
tates the ability to systematically compare primary sources on
specific issues, variables, or sample characteristics.
Brown S.4., Upchurch S.L. & Acton G.J. (2003) A framework fordeveloping a coding scheme for meta-analysis. Westeln Journal ofNursing Research 25, 20 5 -222.
Chinn P. Ea Kramer M. (2004) Tbeory and Nursing: Integrated
Knowledge Deuelopment,6th edn. Mosby, St Louis, MO.Conn V.S. & Rantz M.j. (2003) Managing primaty study quality in
meta-analyses. Research in Nwsing and Health 26,322-333.Conn V.S., Isaramalai S., Rath S., Jantarakupr P., Wadhawan R. &
Dash Y. (2003a) Beyond MEDLINE for literature searches. /ozr-nal of Nursing Scholarship 35, 1'77-1'82.
Evans D. & Pearson A. (2001) Systematic reviews: gatekeepers ofnursing knowledge. Journal of Clinical Nursing 10, 593-599.
Ganong L.H, 11987) Integrative reviews of nursing research.
Research in Nursing and Heøbh 10, 1.-1.1..
Garrard J. Q0041 Heahh Sciences Literuture Reuieø Møde Easy: Tbe
Matrix Method, 2nd edn. Aspen Publication, Caithersburg, MD.Glaser B.G. (1,978) Theoretical Sensitiuity: Aduances in the Meth-
odology of Grounded Theory.'|he Sociology Press, Mill Valley,CA.
Glass G.V. (19761Prímary, secondary, and meta-analysis of research.
Educational Research er 5, 3-8,Greenhalgh f . 09971 Papers that summarize other pâpers (system-
atic r€views and meta-analysesl. Brìtish Medical Journal 315,672-67 5.
Jackson G. (1980) Methods for integrative reviews. Reuiett' of Edu-
cdtional Research 50, 438460,Jadad 4.R., Moher D. E¿ Klassen T.P. (1998) Guides for reading and
interpreting systematic reviews: Il. How did the authors find the
studies and âssess th€ir quality? Archiues of Pediatric and Ado-lescent Medicìne 152, 872-817.
Jensen L.A. & Allen M.N. (1996) Meta-synthesis of qualitativefindings. Qualitatiue Health Researcb 6, 553-560.
Kearney M,H. (1998) Ready-to-wear. Discovering grounded formaltheory. Research in Nursing and Health 27, 179-'1.86.
Kirkevold M. (1,9971 Integrative nursing research - an importantstrategy to further the development of nursing science and nursing
practice. Journøl of Aduanced Nursing 25,977-984.Knafl K.A. Ec Webster D.C. (1988) Managing and analyzing quali-
tative data: a description of tasks, techniques, and materials.'Westefti
tournal of Nursing Research 10, 195-21.0.
Miles M.B. & Huberman A.M. (1994) Qualitatiue Data Anølysis.
Sage Publications, Thousand Oaks, CA.
O'Mathuna D.P. (2000) Evidence-based practice and reviews oftherapeutic touch. Journal of Nursing Scholarship 32,279¿85.
lntegratiue reuiew
Oxman A.D. (1994) Checklists for review articles. British Medical
Journøl 309, 648-651.Oxman A.D. Ec Guyatt C.H. (1988) Guidelines for reading literature
reviews. Canadian Medica I Association J ournal 138, 697-7 03.
Paterson 8.L., Thorne 8.L., Canam C. Ec Jillings C. (2O01\ Meta-
Study of Qualitatiue Heahb Researcú. Sage Publications, Thou-sand Oaks, CA.
Patton M.Q. 120021 Qualitatìue Researcb and Eualuatìon Methods,
3rd edn. Sage Publicarions, Thousand Oaks, CA.
Rodgers B.L. & Cowles K.V. (1993) The qualirative research audittrail: a complex collection of documentation, Research in Nursingand Healtb 16, 21,9-226.
Sandelowski M. (1995) Qualitative analysis: what it is and how to
begin. Research in Nursing and Health 78,371,-375.Sandelowski M. & Barroso J. (2003) Crearing metasummaries of
qualitative findings. Nur sing Research 52, 226-233.Sandelowski M., Docherty S. & Emden C. (19971 Qualitative meta-
synthesis: issues and techniques. Researcb in Nwsing and Heølth
20,365-371..Schilling L.S., Grey M. & Knafl K.A. (2002) The concept of self-
management of type 1 diabetes in children and adolescents: an
evolutionary concept analysis. Journal of Aduanced Nursing 37,
87-99.Soeken K.L. & Sripusanapan A. (2003) Assessing publication bias in
meta-analysis. Nursìng Researcb 52, 57-60.Tashakkori A. & Teddlie C. (1998) Mixed Methodology: Combining
Qualitatiue and Quantitatiue Approaches. Sage Publications,
Thousand Oaks, CA.
\íalker L.O. 6¿ Avant K.C. (1995) Strategies for Theory Construc-
tion in Ntrsing,3rd edn. Appleton & Lange, Norwalk, CT.rùlhittemore R. (2005a) Combining the evidence in nursing research:
methods and implications. Nursìng Research 54, 56-62.ìØhittemore R. (2005b) Analysis of integration in nursing science and
practice. J ournal of Nursing Sch olarship 37, 261-267.
Social PolicyBmnch, TheTieæury, PO Box3?24, Wellington,New Zealand
Nicholæ MaysheaLlh aduisø
Corespondence to:C Popec pope@bristolacuk
Series editors:Catherine Pope andNicholæ Mays
BMJ 2000;3201114-6
Qualitatiue research in health care
Analysing qualitative dataCatherine Pope, Sue Ziebland, Nicholas Mays
Contrary to popular perception, qualitative researchcan produce vast amounts of data. These may includeverbadm notes or transcribed recordings ofinterviewsor focus groups,jotted notes and more detailed "field-notes" ofobservational research, a diary or chronologi-cal accoun! and the researcher's reflecúve notes madeduring the research. These data are not necessarilysmall scale: tra¡rscribing a typical single interview takesseveral hours and can generate 20-40 pages of singlespaced text. Transcripts a¡d notes are the raw data ofthe research. They provide a descriptive record of theresearch, but they cannot provide explanations. Theresearcher has to make sense of the data by sifting andinterpreting them,
Relation between analysis and qualitativedata
In much qualitative research the analytical processbegins during data collection as the data already gath-ered are analysed and shape the ongoing datacollection. This sequentia-l analysis' or interirn analysisthas the advantage ofaJlowing the researcher to go backald refine questions, develop hypotheses, and pursueemerging avenues of inquiry in further depth.CruciaÌly, it also enables the researcher to look fordeviant or negaúve cases; that is, examples of ta-lk orevents úrat run counter to the emerging propositionsor hypotheses and can be used to refine them. Suchcontinuous analysis is almost inevitable in qualitativeresearch: because ùre researcher is "in the field" collect-ing the data, it is impossible not to start thìnking aboutwhat is being heard a¡rd seen.
The analysis
None the less there is still much anaJytical work to doonce the researcher has left the fleld. Textual data (inthe form of fieldnotes or transcrþts) are exploredusing some variant of content analysis. In general,qualitative research does not seek to quandry data.
Qualitative sampling strategies do not aim to identi$r a
staúsúcally representative set of respondents, so
expressing results in relative ffequencies may bemisleading. Simple counts are sometimes used andmay provide a useful summary of some aspects of theanalysis. In most qualitative analyses the data are
preserved in their textual form and "indexed" togenerate or develop analytical categories and theoreti-cal explanations.
Qualitative research uses arìalytical categories todescribe and explain social phenomena. These catego-ries may be derived inducúvely-that is, obtainedgradually fiom the data-or used deductivel¡ either atthe beginning or part way through the anaJysis as a wayof approaching the data. Deductive analysis is less
common in qualitative research but is increasingÌybeing used, for example in the "fiamework approach"
Summary points
Qualitative research produces large amounts oftextual data in the form of uanscripts andobservational fi eldnotes
The systematic arìd rigorous preparation andanalysis of these data is time consuming andlabour intensive
Data analysis often takes place alongside datacollection to allow questions to be refined andnew avenues ofinquiry to develop
Textual data are typically explored inductivelyusing content a¡alysis to generate cate8ories a¡dexplanations; software packages can help withanalysis but should not be viewed as short cuts torigorous and systematic analysis
High quality analysis of qualitative data dependson the skill, vision, and integrity ofthe researcher;it should not be left to the novice
described below. The term grounded dreory is used todescribe the inductive process of identiling analyticalcategories as they emerge from the data (developinghypotheses from the ground or research freld upwardsrather deflning them a priori).' Initially the data areread a¡rd reread to identi$ and index themes and cat-egories: these may centre on particular phrases,incidents, or types of behaviour. Sometimes interestingor unfamiliar terms used by the group studied canform the basis of analytical categories. Becker andGeer's classic study of medical training uncovered thespecialist use of the term "crock" to denote patientswho were seen as less worthwhile to treat by medicalstaffand students.a
All dre data reìevant to each category are identifiedand examined using a process called const¿rnt
comparison, in which each item is checked orcompared with the rest of the data to establish anal¡i-cal categories. This requires a coherent and systematicapproach. The key point about this process is that it isinclusive; categories are added to reflecl as many of thenuances in the data as possible, rather than reducingthe data to a few numerica.l codes. Secdons of thedata-such as discrete incidents-wilì typically includemultiple themes, so it is irnportant to have some systemof cross indexing to deal with this. A number ofcomputer software packages have been developed toassist with this process (see below).
Indexing the data creates a large number of"fuzzycategories" or units.o Informed by ùre anal¡icaì andtheoretical ideas developed during the research, these
categories are further refined a¡rd reduced in number
r74 BMI VOLUME32O 8.|ANUARY2000 M.bmjcom
Education and debate
by grouping them together. It is dren possible to selectkey themes or categories for further investigaúon-typically by "cutting and pasting"-that is, selecting sec-
tions of data on lile or related themes and puttingthem togedrer. Paper systems for this (using multiplephotocopies, cardex systems, matrices, or spread-sheets), although considered somewhat old fashionedand laborious, can help the researcher to develop aninrimate knowledge of the data. Word processors canalso facilitate data searching, and split screen functionsmake this a particularly appealing method for sortingand copying daca into separate frles.
Software packages designed to handlequalitative data
Several software packages designed for qualitative dataanalysis enable complex organisation and retrieval ofdata. Among the most widely used are esR Nuo*Istand err-es.ti.u 7 This evolu[ion has been welcomed as a¡r
important development with the potential to improvethe rigour of analysis.t Such software can allow basic"code and retrieval" of data, and more sophisúcatedanalysis using algorithms to identi$r co-occurringcodes in a range of logically overlapping or nestingpossibilities, annotation of the text, or the creation a¡rdamalgamation of codes. Some packages can be used tomake theoretica.l links or search for "disconirming evi-dence" (for examplg by using boolean operators suchas "or," "a¡rd," "not"). The Hypersoft package uses
"hyperlinks" to capture the conceptual links which areobserved between sections of the data; this can protectthe narrative sfucture of the data to avoid the problemof decontextualisation or data Íiagmentation.e
Using software to help with the more laborious sideof analysis has many potential benefits,but some cautionis advisable. The prospect of computer assisted analysis
may persuade researchers (or those who fund them)that they can manage much larger amounts of data andincrease the apparent "powei' of their study. However,qualitative studies are not desigrred to be representativein terms of statistical generalisabiliry and they may gainlittle Íiom an expanded sample size except a more
cumbersome dataset, The sample size should be
directedby the research question and anal¡ical require-ments, such as dat,a saturation, rather than try theavailable softwa¡e. In some circumstances, a single case
study design may be the most successful way of generat-ing theory. Furthermorg using a computer package may
not make the analysis less time consuming,'o although itmay show that the process is systematic.
Täking the analysis forward-the role ofthe researcher
A computer package may be a useful aid when gather-ing, organising, and reorganising data and helping to
find exceptions, but no package is capable ofperceiving a link between theory and data or definingan appropriate structure for the analysis. To take theanalysis beyond the most basic descriptive andcounting exercise requires the researcher's anaJyticalskills in moving towards hlpotheses or propositionsabout the data.
One way of performing this next stage is caìled
anal¡ic induction.This involves an iterative testing andretesting of theoretical ideas using the data. Bloordescribed his use of this procedure in some detaiÌ
þox)." In essence, the researcher examines a set ofcases, develops hlpoùreses or consmrcts, and examines
further cases to test these propositions.
Inter-rater reliabilitySome researchers have found that the use of morethan one analyst can irnprove the consistency orreliability of anaìyses.o'2'3 lloweve¡ the appropriate-ness of the concept of inter-rater reliability in qualita-tive research is contested.ta None the less there may bemerit in involving more than one analyst in situationswhere researcher bias is especially likely to beperceived to be a probìem-for example, where socialscientists are investigating the work of clinicians. In astudy ofdiagnosis in cardiology, Daly et al deveÌoped amodified form of qualitative anaJysis involvingexternal researchers and the cardiologists who hadmanaged the patients. The researchers identiîed the
Analysis
Stages in the analysis ofûeldnotes in a qualitative snrdy of ear, nose, andthroat surgeons' disposal decisions for child¡en refened for possibletonsillectomy and adenoidectomy (with examples)" :
(1) Provisional classification-for each surgeon all cases categorisedaccording to disposal category used (tonsillectomy and adenoidectomy oradenoidectomy alone)
(2) Identifcation of feanrres of provisional cases-common feahrres of cases
in each disposal category identifred (most tonsillectomy and adenoidectomycases found to have three main dinical signs)
(3) Smrtiny of deviant cases-include in (2) or modi$ (1) to accommodatedeviant cases (tonsillectomy and adenoidectomy performed when only twoof three signs present)
(4) Identiûcation of shared features of cases-featu¡es common to otherdisposal categories (history of several episodes of tonsillitis)(5) Derivation of swgeons' decision rules-from the feacures common tocases (case history more irnportant than physical examination)
(6) Derivation ofsurgeons'search procedures (for each decision ruleþtheparticular clinical signs looded for by each surgeon
Repeat steps (l) to (6) for each disposa.l category
È
BIW VOLUME32O 8JANUARY2000 tw.bmj.com ll5
Education and debate
Five stages of data analysis in the framework approach. Familiarisatim-immersion in the raw data (or typically a pragmaticselection from the data) by listening to tâpes, reading transcripts, studyingnotes md so on, in order to list key ideas and ¡eorrent themes. Idmtilying a thmaticJrammork-ìdentiSiug all the key issues, concepts,and themes by which the data can be examined md referenced. This iscarried out by drawing on a priori issues and questions derived from theaims md objectives of the shrdy as well as issues raised by the respondentsthemselves md views or experiences that recur in the data. The end productof this stage is a deøiled index of the data, which labels the data intomanageable chunks for subsequent retrieval ærd exploration. Ind¿xing-applytng the thematic fiamework or index systematically to allthe data in textuaL fom by annotating the trmsripts with numerical codesfiom the index, usually supported by short text descriptors to elaborate theindex heading. Single passages of text ca¡l often encompass a large numberof different themes, each of which has to be recorded usually in the marginof ùe nanscripto Charting-reznmging ùre data according to the appropriate part of thethematic framework to which they relate, and forming chars For example,there is likely to be a chart fo¡ each key subject area or theme with enlriesfor several respondents. Unlike simp.le ot and paste met]rods that groupverbatim text, the charts contain distilled summaries of views andexperiences. Thus the chæting process involves a considerable amowt ofabstraction md slnthesis. Maþþing a,nì, interþretatim-úsing the charts to define concepts, map therange and nahlre ofphenomen4 create typologies and find associationsbetween rhemes with a view to providing explmations for the findings Theprocess of mapping and interpretation is influenced by the originalresearch objectives as well as by the themes that have emerged from thedata themselves
main aspects of the consultations that seemed to berelated to the use of echocardiograph¡ and theydeveloped criteria which other analysts could use toassess the raw data. The cardioÌogists then independ-ently assessed each case using the raw data in order toproduce an account ofhow and why a test was or was
not ordered and with what consequences. The assess-
ments of the cardioìogists and researchers were com-pared statistically and the level of agreement was
shown to be good. Where there was disagreementbetween the original researchers'analysis and that ofthe cardioìogist, a further researcher repeated thearìalysis and any remaining discrepalcies wereresolved by discussion between the researchers andthe cardiologists. Although there was an element ofcircularity in part of this lengthy process (in that theformal criteria used by the cardiologists were derivedfiom the initial researchers'analysis) and it involvedthe derivation of quantitative gradings and statisticalanalysis of inter-rater agreement that are unusual in a
qualitative study, ùis process meant that clinical criticscould not argue that the findings were simply basedon the subjective judgments of an individualresearcher.
This uticlc is takenfiom the secondedition ofQtaLilaliue Research
in Health Car¿,
cdited by CadrerinePope and NicholroMays, published byBMJ Books
Applied qualitative research
The fiamework approach has been developed in Brit-ain specifrcally for applied or policy relevant qualitativeresearch in which the objectives ofthe investigation aretypically set in advance and shaped by the i¡formationrequirements of the funding body (for example, aheaJth authority).', The timescales of applied researchtend to be short and drere is often a need to ìink theana-lysis wid-r quantitative findings. For these reasons,although the fiamework approach reflects the original
accounts and observations of dre people studied (thatis, "grounded" and inductive), it starts deductively frompre-set a.ims and objectives. The data collection tendsto be more structured dran would be the norm formuch other qualitative research and the analyticalprocess tends to be more expìicit and more stronglyinformed by a priori reasoning þox).6 The analysis is
desigrred so that it can be viewed and assessed bypeople other than the primary analyst.
Conclusions
Anaìysing qualitative data is not a simple or quick task.
Done properl¡ it is systematic and rigorous, and there-fore labour-intensive and time-consuming. Fieldingcontends that "good qualitative analysis is able todocument its claim to reflect some of the truÙr of a
phenomenon by reference to systematically gathereddat4" in contrast, "poor qualitative analysis is
anecdotal, unreflective, descriptive without beingfocused on a coherent line of inquiry."'6 At its heart,good qualitarive analysis relies on the skill, vision a¡dintegrity of the researcher doing that analysis, and as
Dingwall et al have pointed out, this requires trained,and, cruciall¡ experienced researchers.'i
Further reading
Bryman A, Burgess R, eds Analysing qualitatiue dntaLondon: Routledge, 1993
Miles M, Hubermzn A Qualitatiue data analysß.London: Sage, 1984
'l'he vielvs exprcsscd in this paper arc those of ¡he autìrors anddo not neccssarily rcllect the views of thc New Zealand lÌeasur¡in tìrc casc ofNM 'flre'Iieasury takcs no responsibiJity for anycrrors or omissions in, or for thc correctncss of the inlormationcontained in this arLiclc.
Beckerì IS Sociologítaluøk London: ^llen
LÐe, I07lMiles M, Ilul¡erman A ØnLítatiw data anaqs¿r .t.ondonr Sage, 1984
Glmcr.BG, Smuss Â-L The dirøø¡ of grow¿¿ /å¿ø) Cl)icago: ^ldine,1967
Becker I IS, Gecr ts Pdticipant obscrvatioo: tlìe an¡lysis o[ qualintiveûeld datâ Inr Burgcss RG,erl Fieldresearch:a sourtebooh andjeklnøntøLLondon: Allcn and Unwio, l982Peny S Líting øilh nulliþl¿ s(lcrosis. Al¿elshotr
^vcbury, 1 994
Richuds I Richards L QSn NUD*lST,vtsion i úr. London: Sage, 1994MDhî'L ATl,A,S.ti Jt Wirulotts. Berlio: Scienúûc Soltrvare Developlnent1997
8 Kelle U, ed Cotuî^tlù-aided qtn\laliile d¿Ia ana\sis: lheu1, ilelhods andprdrli.a London: Sage, I 995
9 Dey I Øtalilali@ data aml¡sis: a tun lriendl¡ grLùle lor social scíentísts
London: Roudcdge, 1993l0 Lce R, liclding N User's exper icnccs of quzrlintìve clan æalysis softrvarc
ln: Kelle Ç ed Coiltlrtltr eid¿d qualílaliile dalo anabsís: lheü), ùulho¿s and
fi a.l ice. LoxJon: S^ge, 1 995I I Bloor M On tlrc anirlysis ofobservational drn: a discussion o[the worth
and uses ofinduclive techniqucs and respo¡rdcnt validation Sociologr
1978:12t545-52l2 D^ly J, McDonâld I, Willis E Why don't you æk them? A (luaL¡htive
reseru ch fi¿meivork for invesrigaring the diagnosis oI cmdìac nonnalìt¡Inr D¡lyJ, Mc¡onald I, Willis l, eds Researthing heallh mre: desigu, rlîløn-ùe$ ¿istiþlinet Lon(lonr Routledge, I 992: I 89-206
13 WiiEkin B The þolitiß of ile({iøl entuúnløt Nerv I lirven: Y¡Le Un iversityP¡ess, I 991
14 ^r
mstrong D, Gosling Â, WcinmmJ, Mar te¡u T The phce of in(e¡-ñlerreliability in qualitîúve rcscarcb::m cmpirical study. Sodìlog)1997;31:597-606
15 RitchieJ, Spencer L Qu^L¡htivc dah mîlysìs lor îpplied policy reseüchIn Bryrnnn A, Burgcss R, cds .4¿dbriag qilalilali@ dola LondoniRoutìedge, Ì993: l?3-94