WIXLIB

Gale et al. BMC Medical Research Methodology 2013, 7There are a number of approaches to qualitative dataanalysis, including those that pay close attention to languageand how it is being used in social interaction such asdiscourse analysis [15] and ethnomethodology [16]; thosethat are concerned with experience, meaning and languagesuch as phenomenology [17,18] and narrative methods [19];and those that seek to develop theory derived from datathrough a set of procedures and interconnected stages suchas Grounded Theory [20,21]. Many of these approaches areassociated with specific disciplines and are underpinned byphilosophical ideas which shape the process of analysis [22].The Framework Method, however, is not aligned witha particular epistemological, philosophical, or theoreticalapproach. Rather it is a flexible tool that can be adaptedfor use with many qualitative approaches that aim togenerate themes.The development of themes is a common feature ofqualitative data analysis, involving the systematic searchfor patterns to generate full descriptions capable ofshedding light on the phenomenon under investigation. Inparticular, many qualitative approaches use the ‘constantcomparative method’, developed as part of GroundedTheory, which involves making systematic comparisonsacross cases to refine each theme [21,23]. Unlike GroundedTheory, the Framework Method is not necessarily concerned with generating social theory, but can greatlyfacilitate constant comparative techniques through thereview of data across the matrix.Perhaps because the Framework Method is so obviouslysystematic, it has often, as other commentators have noted,been conflated with a deductive approach to qualitativeanalysis [13,14]. However, the tool itself has no allegianceto either inductive or deductive thematic analysis; wherethe research sits along this inductive-deductive continuumdepends on the research question. A question such as, ‘Canpatients give an accurate biomedical account of theonset of their cardiovascular disease?’ is essentially ayes/no question (although it may be nuanced by the extentof their account or by appropriate use of terminology) andso requires a deductive approach to both data collectionand analysis (e.g. structured or semi-structured interviewsand directed qualitative content analysis [24]). Similarly, adeductive approach may be taken if basing analysis on apre-existing theory, such as behaviour change theories, forexample in the case of a research question such as ‘Howdoes the Theory of Planned Behaviour help explain GPprescribing?’ [11]. However, a research question such as,‘How do people construct accounts of the onset of theircardiovascular disease?’ would require a more inductiveapproach that allows for the unexpected, and permits moresocially-located responses [25] from interviewees that mayinclude matters of cultural beliefs, habits of food preparation, concepts of ‘fate’, or links to other important eventsin their lives, such as grief, which cannot be predicted byPage 3 of 8the researcher in advance (e.g. an interviewee-led openended interview and grounded theory [20]). In all thesecases, it may be appropriate to use the Framework Methodto manage the data. The difference would becomeapparent in how themes are selected: in the deductiveapproach, themes and codes are pre-selected based onprevious literature, previous theories or the specificsof the research question; whereas in the inductiveapproach, themes are generated from the data thoughopen (unrestricted) coding, followed by refinement ofthemes. In many cases, a combined approach is appropriatewhen the project has some specific issues to explore, butalso aims to leave space to discover other unexpectedaspects of the participants’ experience or the way theyassign meaning to phenomena. In sum, the FrameworkMethod can be adapted for use with deductive, inductive, orcombined types of qualitative analysis. However, there aresome research questions where analysing data by case andtheme is not appropriate and so the Framework Methodshould be avoided. For instance, depending on the researchquestion, life history data might be better analysed usingnarrative analysis [19]; recorded consultations betweenpatients and their healthcare practitioners using conversation analysis [26]; and documentary data, such as resourcesfor pregnant women, using discourse analysis [27].It is not within the scope of this paper to considerstudy design or data collection in any depth, but beforemoving on to describe the Framework Method analysisprocess, it is worth taking a step back to consider brieflywhat needs to happen before analysis begins. The selectionof analysis method should have been considered at theproposal stage of the research and should fit with theresearch questions and overall aims of the study. Manyqualitative studies, particularly ones using inductive analysis, are emergent in nature; this can be a challenge andthe researchers can only provide an “imaginative rehearsal”of what is to come [28]. In mixed methods studies, the roleof the qualitative component within the wider goals of theproject must also be considered. In the data collectionstage, resources must be allocated for properly trainedresearchers to conduct the qualitative interviewing becauseit is a highly skilled activity. In some cases, a research teammay decide that they would like to use lay people, patientsor peers to do the interviews [29-32] and in this case theymust be properly trained and mentored which requirestime and resources. At this early stage it is also useful toconsider whether the team will use Computer AssistedQualitative Data Analysis Software (CAQDAS), which canassist with data management and analysis.As any form of qualitative or quantitative analysis isnot a purely technical process, but influenced by thecharacteristics of the researchers and their disciplinaryparadigms, critical reflection throughout the researchprocess is paramount, including in the design of the

Gale et al. BMC Medical Research Methodology 2013, 7study, the construction or collection of data, and theanalysis. All members of the team should keep a researchdiary, where they record reflexive notes, impressions of thedata and thoughts about analysis throughout the process.Experienced qualitative researchers become more skilled atsifting through data and analysing it in a rigorous andreflexive way. They cannot be too attached to certainty, butmust remain flexible and adaptive throughout the researchin order to generate rich and nuanced findings thatembrace and explain the complexity of real social life andcan be applied to complex social issues. It is important toremember when using the Framework Method that, unlikequantitative research where data collection and data analysis are strictly sequential and mutually exclusive stages ofthe research process, in qualitative analysis there is, to agreater or lesser extent depending on the project, ongoinginterplay between data collection, analysis, and theorydevelopment. For example, new ideas or insights fromparticipants may suggest potentially fruitful lines of enquiry,or close analysis might reveal subtle inconsistencies in anaccount which require further exploration.Procedure for analysisStage 1: TranscriptionA good quality audio recording and, ideally, a verbatim(word for word) transcription of the interview is needed.For Framework Method analysis, it is not necessarilyimportant to include the conventions of dialogue transcriptions which can be difficult to read (e.g. pauses or twopeople talking simultaneously), because the content is whatis of primary interest. Transcripts should have large marginsand adequate line spacing for later coding and makingnotes. The process of transcription is a good opportunity tobecome immersed in the data and is to be strongly encouraged for new researchers. However, in some projects, thedecision may be made that it is a better use of resources tooutsource this task to a professional transcriber.Stage 2: Familiarisation with the interviewBecoming familiar with the whole interview using the audiorecording and/or transcript and any contextual or reflectivenotes that were recorded by the interviewer is a vital stagein interpretation. It can also be helpful to re-listen to all orparts of the audio recording. In multi-disciplinary or largeresearch projects, those involved in analysing the data maybe different from those who conducted or transcribed theinterviews, which makes this stage particularly important.One margin can be used to record any analytical notes,thoughts or impressions.Stage 3: CodingAfter familiarization, the researcher carefully reads thetranscript line by line, applying a paraphrase or label(a ‘code’) that describes what they have interpreted inPage 4 of 8the passage as important. In more inductive studies,at this stage ‘open coding’ takes place, i.e. coding anythingthat might be relevant from as many different perspectivesas possible. Codes could refer to substantive things(e.g. particular behaviours, incidents or structures), values(e.g. those that inform or underpin certain statements,such as a belief in evidence-based medicine or in patientchoice), emotions (e.g. sorrow, frustration, love) and moreimpressionistic/methodological elements (e.g. intervieweefound something difficult to explain, interviewee becameemotional, interviewer felt uncomfortable) [33]. In purelydeductive studies, the codes may have been pre-defined(e.g. by an existing theory, or specific areas of interestto the project) so this stage may not be strictly necessaryand you could just move straight onto indexing, althoughit is generally helpful even if you are taking a broadlydeductive approach to do some open coding on at least afew of the transcripts to ensure important aspects of thedata are not missed. Coding aims to classify all of the dataso that it can be compared systematically with other partsof the data set. At least two researchers (or at least onefrom each discipline or speciality in a multi-disciplinaryresearch team) should independently code the first fewtranscripts, if feasible. Patients, public involvement representatives or clinicians can also be productively involvedat this stage, because they can offer alternative viewpointsthus ensuring that one particular perspective does notdominate. It is vital in inductive coding to look out for theunexpected and not to just code in a literal, descriptiveway so the involvement of people from different perspectives can aid greatly in this. As well as getting a holisticimpression of what was said, coding line-by-line can oftenalert the researcher to consider that which may ordinarilyremain invisible because it is not clearly expressed or doesnot ‘fit’ with the rest of the account. In this way the developing analysis is challenged; to reconcile and explainanomalies in the data can make the analysis stronger.Coding can also be done digitally using CAQDAS, whichis a useful way to keep track automatically of new codes.However, some researchers prefer to do the early stages ofcoding with a paper and pen, and only start to useCAQDAS once they reach Stage 5 (see below).Stage 4: Developing a working analytical frameworkAfter coding the first few transcripts, all researchersinvolved should meet to compare the labels they haveapplied and agree on a set of codes to apply to all subsequent transcripts. Codes can be grouped together intocategories (using a tree diagram if helpful), which arethen clearly defined. This forms a working analyticalframework. It is likely that several iterations of the analytical framework will be required before no additionalcodes emerge. It is always worth having an ‘other’ codeunder each category to avoid ignoring data that does not

Gale et al. BMC Medical Research Methodology 2013, 7fit; the analytical framework is never ‘final’ until the lasttranscript has been coded.Stage 5: Applying the analytical frameworkThe working analytical framework is then applied byindexing subsequent transcripts using the existing categories and codes. Each code is usually assigned a number orabbreviation for easy identification (and so the full namesof the codes do not have to be written out each time) andwritten directly onto the transcripts. Computer AssistedQualitative Data Analysis Software (CAQDAS) is particularly useful at this stage because it can speed up the processand ensures that, at later stages, data is easily retrievable. Itis worth noting that unlike software for statistical analyses,which actually carries out the calculations with the correctinstruction, putting the data into a qualitative analysis software package does not analyse the data; it is simply an effective way of storing and organising the data so that theyare accessible for the analysis process.Stage 6: Charting data into the framework matrixQualitative data are voluminous (an hour of interviewcan generate 15–30 pages of text) and being able tomanage and summarize (reduce) data is a vital aspect ofthe analysis process. A spreadsheet is used to generate amatrix and the data are ‘charted’ into the matrix.Charting involves summarizing the data by categoryfrom each transcript. Good charting requires an abilityto strike a balance between reducing the data on the onehand and retaining the original meanings and ‘feel’ ofthe interviewees’ words on the other. The chart shouldinclude references to interesting or illustrative quotations.These can be tagged automatically if you are using CAQDASto manage your data (N-Vivo version 9 onwards has thecapability to generate framework matrices), or otherwise acapital ‘Q’, an (anonymized) transcript number, page andline reference will suffice. It is helpful in multi-disciplinaryteams to compare and contrast styles of summarizing inthe early stages of the analysis process to ensure consistency within the team. Any abbreviations used should beagreed by the team. Once members of the team are familiarwith the analytical framework and well practised at codingand charting, on average, it will take about half a day perhour-long transcript to reach this stage. In the early stages,it takes much longer.Page 5 of 8Gradually, characteristics of and differences between thedata are identified, perhaps generating typologies, interrogating theoretical concepts (either prior concepts or onesemerging from the data) or mapping connections betweencategories to explore relationships and/or causality. If thedata are rich enough, the findings generated through thisprocess can go beyond description of particular cases toexplanation of, for example, reasons for the emergence of aphenomena, predicting how an organisation or other socialactor is likely to instigate or respond to a situation, or identifying areas that are not functioning well within an organisation or system. It is worth noting that this stage oftentakes longer than anticipated and that any project planshould ensure that sufficient time is allocated to meetingsand individual researcher time to conduct interpretationand writing up of findings (see Additional file 1, Section 7).DiscussionThe Framework Method has been developed and usedsuccessfully in research for over 25 years, and has recentlybecome a popular analysis method in qualitative healthresearch. The issue of how to assess quality in qualitativeresearch has been highly debated [20,34-40], but ensuringrigour and transparency in analysis is a vital component.There are, of course, many ways to do this but in theFramework Method the following are helpful: Summarizing the data during charting, as well as Stage 7: Interpreting the dataIt is useful throughout the research to have a separatenote book or computer file to note down impressions,ideas and early interpretations of the data. It may beworth breaking off at any stage to explore an interestingidea, concept or potential theme by writing an analyticmemo [20,21] to then discuss with other members ofthe research team, including lay and clinical members. being a practical way to reduce the data, means thatall members of a multi-disciplinary team, includinglay, clinical and (quantitative) academic memberscan engage with the data and offer their perspectivesduring the analysis process without necessarilyneeding to read all the transcripts or be involved inthe more technical parts of analysis.Charting also ensures that researchers pay closeattention to describing the data using eachparticipant’s own subjective frames and expressions inthe first instance, before moving onto interpretation.The summarized data is kept within the widercontext of each case, thereby encouraging thickdescription that pays attention to complex layers ofmeaning and understanding [38].The matrix structure is visually straightforward andcan facilitate recognition of patterns in the data byany member of the research team, including throughdrawing attention to contradictory data, deviantcases or empty cells.The systematic procedure (described in this article)makes it easy to follow, even for multi-disciplinaryteams and/or with large data sets.It is flexible enough that non-interview data (such asfield notes taken during the interview or reflexiveconsiderations) can be included in the matrix.

Gale et al. BMC Medical Research Methodology 2013, 7 It is not aligned with a particular epistemologicalviewpoint or theoretical approach and therefore canbe adapted for use in inductive or deductive analysisor a combination of the two (e.g. using pre-existingtheoretical constructs deductively, then revising thetheory with inductive aspects; or using an inductiveapproach to identify themes in the data, beforereturning to the literature and using theoriesdeductively to help further explain certain themes). It is easy to identify relevant data extracts toillustrate themes and to check whether there issufficient evidence for a proposed theme. Finally, there is a clear audit trail from original rawdata to final themes, including the illustrativequotes.There are also a number of potential pitfalls to thisapproach: The systematic approach and matrix format, as wenoted in the background, is intuitively appealing tothose trained quantitatively but the ‘spreadsheet’look perhaps further increases the temptation forthose without an in-depth understanding ofqualitative research to attempt to quantifyqualitative data (e.g. “13 out of 20 participants saidX). This kind of statement is clearly meaninglessbecause the sampling in qualitative research is notdesigned to be representative of a wider population,but purposive to capture diversity around aphenomenon [41]. Like all qualitative analysis methods, the FrameworkMethod is time consuming and resource-intensive.When involving multiple stakeholders and disciplinesin the analysis and interpretation of the data, as isgood practice in applied health research, the timeneeded is extended. This time needs to be factoredinto the project proposal at the pre-funding stage. There is a high training component to successfullyusing the method in a new multi-disciplinary team.Depending on their role in the analysis, members ofthe research team may have to learn how to code,index, and chart data, to think reflexively about howtheir identities and experience affect the analysisprocess, and/or they may have to learn about themethods of generalisation (i.e. analytic generalisationand transferability, rather than statisticalgeneralisation [41]) to help to interpret legitimatelythe meaning and significance of the data.While the Framework Method is amenable to theparticipation of non-experts in data analysis, it is criticalto the successful use of the method that an experiencedqualitative researcher leads the project (even if the overallPage 6 of 8lead for a large mixed methods study is a different person).The qualitative lead would ideally be joined by other rese

Keywords: Qualitative research, Qualitative content analysis, Multi-disciplinary research The Framework Method for the management and analysis of qualitative data has been used since the 1980s [1]. The method originated in large-scale social policy research but is becoming an increasingly popular approach in medical and health research; however .