KO H10

 10 Identifying patterns across data OVERVIEW Searching for patterns: fr...

 10 Identifying patterns across data OVERVIEW Searching for patterns: from codes to candidate themes Reviewing and revising candidate themes Other ways of identifying patterns across data Can and should I go beyond looking for patterns? After coding the data, it’s time to shift to looking for larger patterns across the dataset. We focus again on thematic analysis (TA) in this chapter, and describe and illustrate key differences in other pattern-based approaches (i.e. interpretative phenomeno- logical analysis [IPA], grounded theory [GT], and pattern-based discourse analysis [DA]). At its most basic, a pattern-based analysis allows you systematically to identify and report the salient features of the data. But much pattern-based analysis goes beyond this, and interrogates and interprets the patterns identified (Chapter 11 covers anal- ysis and interpretation of patterns). Pattern-based analysis rests on the presumption that ideas which recur across a dataset capture something psychologically or socially meaningful. In working out which patterns are relevant and important in relation to your research question, it’s not just a question of which are the most frequent. While frequency is an important factor, it’s also about capturing the different elements that are most meaningful for answering your research question. So it’s about meanings, rather than numbers (Chapter 11 discusses this further). New Zealand health researcher Stephen Buetow (2010) has developed an approach based on, but extending, TA, called 10-Braun & Clarke_Ch-10.indd 223 28/02/2013 7:44:22 PM  224 Successfully analysing qualitative data saliency analysis, which captures the point that something in data can be important without appearing very frequently. SEARCHING FOR PATTERNS: FROM CODES TO CANDIDATE THEMES As you work from codes and coded data to identify broader patterns – themes – you get deeper into the analysis. A theme ‘captures something important about the data in rela- tion to the research question, and represents some level of patterned response or mean- ing within the data set’ (Braun & Clarke, 2006, p. 82). It’s typically broader than a code in that it contains many facets. Remember our brick and tile house example: a theme is like the wall or roof panel of a house, made up of many individual bricks or tiles (codes). A good code will capture one idea; a theme has a central organising concept, but will contain lots of different ideas or aspects related to the central organising concept (each of those might be a code). For example, one of themes we developed from our coding of the weight and obesity focus group (FG) is called modern life is rubbish. The cen- tral organising concept of this theme is that contemporary lifestyles encourage obesity. Particular facets of the theme include ideas around the availability of convenience food, not having time to cook, ubiquitous advertising of ‘unhealthy’ foods and safety concerns curtailing children’s activity. Although a rich and complex code may become a theme without much expansion, the distinction between codes and themes is generally a good one to work with: codes combine to form themes. It’s also useful to distinguish whether you’re identifying a theme or what we would call a feature of the data. Both capture something that’s recurring in the data, but a theme has a central organising concept, which tells us something about the content of the data that’s meaningful, something about how, and in what way, that concept appears in the data: it tells us something meaningful in relation to our research question. For example, when analysing the story completion task data (explained in Chapter 6; for sample data, see the companion website), many of Victoria’s students identified gender as a key theme. However, gender should more accurately be described as a feature of the data, rather than a theme. Why? Because it doesn’t have a central organising concept and sim- ply clusters together a whole lot of different ways gender is evident in the data. This isn’t to say that gender was irrelevant to the analysis. It wasn’t – it was really important. But a theme around gender had to be built around a central organising concept and tell us something meaningful how gender appeared in the data. Good gender themes included ‘stereotyped gender roles’ (the stories typically contained very traditional ideas about appropriate gender roles) and ‘the gendering of emotion’ (male respondents often wrote angry stories; female participants typically wrote nurturing stories). In determining if you have identified a theme or a feature, it’s also worth making sure that it’s not just that you have given your theme a terrible title, which fails to demonstrate the central organising concept (see Box 10.1). 10-Braun & Clarke_Ch-10.indd 224 28/02/2013 7:44:23 PM  Identifying patterns across data 225 IDENTIFYING THEMES AS AN ACTIVE PROCESS It’s quite common to read about ‘themes emerging from the data’. We get quite – maybe even very – grumpy when we read this! Why? Because it falsely suggests analysis is a passive process where you identify something that already exists (Ely, Vinz, Downing, & Anzul, 1997). Developing themes from coded data is an active pro- cess: the researcher examines the codes and coded data, and starts to create potential patterns; they do not ‘discover’ them (Taylor & Ussher, 2001). Searching for patterns is not akin to an archaeologist digging to find hidden treasures buried within the data, pre-existing the process of searching for them. It’s more akin to the process of sculp- ture. Analysts, like sculptors, actively make choices about how they shape and craft their ‘raw data’ (e.g. their piece of marble) into an analysis (like a work of art, such as Michelangelo’s David). Like the sculptor’s block of marble, the dataset provides a material basis for the analysis; it provides some limits or boundaries on what it is pos- sible to produce. However, it does not completely determine the shape of the analysis; it’s possible to create many different analyses from qualitative data, just as it’s pos- sible to create many different sculptures from one piece of marble. Different research- ers, with different tools, can produce different analyses from the same data (e.g. see Coyle & Lyons, 2007; Forrester, 2010) and, like sculpture, the resulting analysis can vary in quality (see Chapter 12). So how do you identify themes? To identify patterns in the data, you need to review the codes and the collated data relating to each code, with the aim of identifying simi- larity and overlap between codes. It might help to look for concepts, topics or issues which several codes relate to, and which could be used as a central organising concept for a theme; some codes, if they are large, rich and complex enough, may themselves be ‘promoted’ to a theme (Charmaz, 2006), a process referred to as subsumption in IPA (Smith et al., 2009). Basically, you want to identify a number of themes (with central organising concepts) that capture the most salient patterns in the data relevant to answering your research question. This makes it sound straightforward, but it’s not always easy; sometimes meaningful, interesting or important, or indeed any, patterns seem elusive. Box 10.1 provides useful questions to ask yourself as you shift from codes to themes. If you’re still having trouble, don’t panic! It would be very unlikely that there are no patterns in the data. It may be that you haven’t given it enough time, and aren’t ‘immersed’ in the data enough and close enough to it (do you feel you know it intimately?). At the same time, you do need some distance from the data. This all takes time! You need to leave lots of time for coding and analysis – they always take far longer than you ever anticipate (practically every student we have ever supervised has made this comment, even though we warn each and every one about how long it takes). So don’t do this in a rush; it won’t help in pattern identification. Sometimes identifying patterns is made harder by the data you have. For instance, we have found that qualitative survey data, where the data are typically quite short responses to particular questions, pose unique and particular challenges for identifying patterns across the dataset (see Box 10.2). 10-Braun & Clarke_Ch-10.indd 225 28/02/2013 7:44:23 PM  226 Successfully analysing qualitative data BOX 10.1 GOOD QUESTIONS TO ASK YOURSELF IN DEVELOPING THEMES Is this a theme (is it just a code or a subtheme)? Is there a central organising concept that unifies the data extracts? What is the quality of this theme? Does the central organising concept tell me something meaningful about a pattern in the data, in relation to my research question? Can I identify the boundaries of this theme? What does it include and exclude? Are there enough (meaningful) data to support this theme? Is the theme too ‘thin’? Is there too much going on in the theme, so that it lacks coherence? Are the data too diverse and wide-ranging? Would using subthemes resolve this problem? Or should it be better split into two or more themes, each with their own central organising concept? How does this (potential) theme relate to other (potential) themes? Is the relationship between (potential) themes hierarchical or linear? What’s the overall story of my analysis? How does this theme contribute to that overall story? Is the central organising concept reflected in the title I have given to the theme (see Chapter 11)? CANDIDATE THEMES IN THE WEIGHT AND OBESITY FG DATA To provide an example of pattern identification, when reviewing our coded FG data, we identified a number of clusterings of codes. For a start, there were some very obvious, semantic-level clusterings: lots of codes related to negative descriptions of contemporary life; lots of codes provided rosy views of life in the past, compared to the present, in relation to obesity and exercise; a large number also clustered around the idea that children’s current socialisation and education is inadequate. Each of these clusterings has a clear and distinct central organising concept, which is captured by these descriptions. Looking a bit more deeply at the data, around the under- lying assumptions or latent ideas informing the things people said, we identified codes clustering around another two central organising concepts: the idea that human beings are naturally gluttonous; the idea that human beings are naturally lazy and exercise is an inherently nega- tive activity. Table 10.1 illustrates the codes we clustered to produce these five candidate themes (the companion website has an extended version of Table 10.1). The five themes are organised underneath two overarching themes – human nature and modern life. There are also two subthemes, which we discuss later. In Table 10.1, the number for each overarching theme, theme and subtheme indicate the hierarchical structure of relationships between them (discussed further later). 10-Braun & Clarke_Ch-10.indd 226 28/02/2013 7:44:23 PM  Identifying patterns across data 227 BOX 10.2 IDENTIFYING PATTERNS IN QUALITATIVE SURVEY DATA Qualitative survey data can present some particular challenges for a pattern-based analysis. We think this stems from certain features: the questions themselves often give a very dominant structure to the data. This impact is potentially exacerbated if you collate data by question for coding and analysis, rather than collating by data item; response to any one question is often quite short; the data are potentially quite ‘bitty’– rather than a longer flowing narrative. You get more discrete responses to particular questions, and these can cover a wide range of issues, but not in much depth. One of the key challenges in identifying patterns is moving away from the organising structure provided by your questions. Remember, it’s important not to confuse questions with patterns or themes. Although sometimes we may want to know the sorts of answers given to particular questions (e.g. ‘What is a feminist’?), in general a pattern-based analysis will not just look within a question, but across the whole dataset to determine themes. Some of the key themes we identified in our critical constructionist thematic analysis of New Zealand responses to the pubic hair survey (discussed in Chapter 6) included: hair removal as personal choice (evident semantically and latently), constructions of pubic hair as private, as interfering with sex, and of removed or reduced pubic hair as cleaner and more attractive (see Braun, Tricklebank & Clarke, under submission). These constructions were not just found around one question, or in relation to one issue; they cut across responses. In more essentialist (or semantic or inductive) TA, it can be harder to move beyond the questions/topics explored to identify themes across responses to different questions. Keeping data collated by participant, rather than by question, which we typically do, should help in seeing patterns across the dataset rather than around questions. In some ways, doing a more constructionist (or latent or theoretical) form of TA or other analysis may be easier with qualitative survey data, because you don’t focus entirely on the explicit content of the data. By trying to get to the meanings and logic underpinning data responses, or in approaching data particular theoretical issue or question in mind (e.g. gendering pubic hair), you remove yourself from the structure imposed by the questions and responses. IMPORTANT POINTS TO REMEMBER AT THIS STAGE There are some important points to note at this stage. The first is that themes identi- fied at this point in the process are provisional; they are candidate themes, and will be revised or refined through the developing analysis. You have to be prepared to let them go. Sometimes, it will be a case that your supervisor or co-researcher does not think they work; sometimes it may be that you realise the analysis either doesn’t fit the data well, or that it doesn’t provide the best or most interesting answer to the research question. You 10-Braun & Clarke_Ch-10.indd 227 28/02/2013 7:44:23 PM  Table 10.1 Candidate themes showing selected associated codes (see the companion website for an extended version) 10-Braun & Clarke_Ch-10.indd 228 1. Human Nature 2. Modern Life 1.1. Sins and sinners 1.1.1. 1.2. Exercise is evil 2.1. Those 2.2. Modern life is 2.2.1. Technology 2.3. They don’t get no Deserving/ halcyon days of rubbish trumps all education undeserving yore obesity ‘Liking food’ as negative; ‘Deserving’ and Choice and exercise ‘Dadadada’ – Cost as a bottom line Children engage Adequate socialisation: associated with ‘undeserving’ (none in past; now common story that determines what in sedentary cooking needs to be overeating obesity: if in we have it, but won’t – a past we all you eat ‘play’ learned (taught in home Convenience (pre- control, can do it) recognise Time poor (money Modern or school) prepared food); judge them; if Constraints and Different rich) technology Children’s socialisation convenience of modern not, can’t supports for regular lifestyles ‘Bad foods’ associated encourages/ is important (but lifestyles is hard to resist Blaming (eat exercise Past – no with positive things in facilitates inadequate) Emotional what he likes) Exercise as negative: such thing ads/marketing obesity/lack of Irresponsible parenting: eating/‘overeating’ and not blaming boring (common as ‘exercise’; exercise adults pander to (he’s fed bad Advertising/ has no validity – not an story); chore and physical activity marketing of junk Negative impacts children; don’t regulate ‘eating disorder’; it’s just food) burden; inherently an integral part food (to children) of technology children’s eating towards gluttony! Doesn’t take unpleasant; inherently of life problematic She’s not healthy foods; feed them Home cooking as a lot to cause lacking fun Times have responsible for unhealthy food obesity Children engage in onerous (time, effort); Exercise as changed sedentary ‘play’ her children’s Kids have an inherent cooking is a hassle External self-indulgent Times have behaviour: tries desire to be able to Humans are naturally factors/life Exercise can be a changed: Commodification of to promote good cook, but education exercise behaviour but system denies them this gluttonous: unless events: obesity luxury/pleasure junk food not controlled will eat too impinges upon Prepared food as powerless in face Socialisation (school PE much/wrong foods; BUT you (you have Exercise easier if part everyday food cheap and therefore of technology teaching) as inadequate of a regular routine – in the past appealing... and ‘modern life’ we should have restraint little control) becomes something you just do 28/02/2013 7:44:24 PM  1. Human Nature 2. Modern Life 10-Braun & Clarke_Ch-10.indd 229 1.1. Sins and sinners 1.1.1. 1.2. Exercise is evil 2.1. Those 2.2. Modern life is 2.2.1. Technology 2.3. They don’t get no Deserving/ halcyon days of rubbish trumps all education undeserving yore obesity Labradoring – pure Emotional Need a motivation to Times have Irresponsible to Technological Socialisation failure; gluttony eating is still exercise (getting away changed (the cook if you can buy inherently across generations: Obesity: caused by potentially from kids); motivation halcyon past – a pre-prepared meal addictive (more parents don’t necessarily laziness under control trumps all obstacles to freedom and cheaply appealing than know how to cook; (some restraint exercise (motivation an active Junk food used to be exercise as young parents not should be also rare) childhood; the a treat ‘leisure’ activity) equipped to socialise applied; Not exercising is easy; hellish present) their children; current completely No ‘modern pantry’: Technology is cooking teaching exercise requires The home no longer unhealthy unrestrained effort (bother) inadequate; government eating is bad) contains the basics for intervention needed (to cooking re-educate) Humans have a natural Society is no longer Socialisation is key: early propensity safe: children as learning sets up later for obesity: a perceived to be attitudes and practices constant threat vulnerable – limits you have to outside play as ‘child actively work safety’ paramount against (if Exercise isn’t part of you become everyday life obese, you’re to Forced into unhealthy blame) eating by modern lifestyles 28/02/2013 7:44:24 PM  230 Successfully analysing qualitative data have to make sure you give yourself enough time to get the analysis ‘wrong’, revise it, or start again; as noted above, time pressure does not generally produce an excellent qualita- tive analysis (see Chapter 12). As a rule of thumb, analysis is likely to take at least twice as long as you imagine it will. The second important point is that, as noted above, themes are not determined in some quantitative fashion (see also Chapter 11), and there is no magical equation or cut- off point to determine what counts as a theme across a dataset, and what doesn’t. As this form of analysis is about identifying patterns across data, themes need to be identified across a proportion of the data (we cannot specify exactly what proportion, as qualita- tive analysis does not work like that), but it does not need to be present in every data item, or even most data items (Braun & Clarke, 2006; Buetow, 2010). Similarly, within each individual data item, some themes will be present, and others not. Some themes may be very prominent in certain data items, but less prominent in others. Determining the importance of a theme is not about counting (e.g. frequency overall, frequency within each data item); it’s about determining whether this pattern tells us something meaningful and important for answering our research question. This means the themes you discuss in any research report will not necessarily be the most common ones (see Braun & Clarke, 2006). The third important, related, point is that your themes don’t have to cover everything in the data – they should be about addressing the research question, and since you are report- ing patterned meaning, some less patterned or irrelevant codes will be excluded. In our data, a large number of codes didn’t obviously address the research question, or didn’t sit within these initial candidate themes. We collated those into a ‘miscellaneous’ category. A category like that can be important to keep at this stage, as the analysis is still very provisional, mean- ings themes may be radically altered, and new themes developed (for an example, see Frith & Gleeson, 2004). Codes which don’t appear to fit anywhere, or cluster with any others, may start to fit as the analysis progresses. Or they may not. Being able to let go of coded material (and indeed, candidate themes) that does not fit within your overall analysis is an important part of qualitative research. Your task in analysing the data is a selective one. It’s about telling a particular story about the data, a story that answers your research question. It isn’t to repre- sent everything that was said in the data. Finally, if you’re doing your analysis with anyone else involved (supervisor, co- researcher), it’s important to realise that some coding and analytic differences are likely when doing qualitative analysis, as we all read data from different perspectives and experiences (see Chapter 12). The key is to work out whether the differences are problematic (e.g. contradictory themes) and, if so, work out where they’re coming from (different theoretical takes, perhaps?), and how to resolve them. In our example, as we previously noted, Virginia, who’s quite ‘sporty’, noted the negative tone of talk around exercise, something that Victoria, who’s not so ‘sporty’, did not. After discussing this point, Victoria understood Virginia’s analysis of exercise constructions, and this eventu- ally became a theme in our analysis (exercise is evil). Qualitative research is not about finding the right answer; what you’re always looking for is the best ‘fit’ of analysis to answer the research question. 10-Braun & Clarke_Ch-10.indd 230 28/02/2013 7:44:24 PM  Identifying patterns across data 239 themes we developed from Sally’s experience also applied to Carla’s, and so in Box 10.3 we just present one summary of the superordinate (master) themes, with a selection of emergent themes related to the superordinate themes, and selected brief illustrative data extracts. We also provide brief summaries, as we are not including the actual analysis. (NB: Both our emergent and superordinate themes sit at the more ‘socially oriented’ end of the IPA spectrum; see Chapter 8.) A compiled table of superordinate (and emergent) themes in IPA, like that shown in Box 10.3, should both capture what you want to say about the data, and the order you want to say it in. It’s effectively a shorthand road-map for your analysis and we recommend presenting it at the start of your results (for an example, see Eatough et al., 2008; see also Chapter 13). After this stage, the IPA analysis moves into a deeper interpretative level (see Chapter 11). WHAT’S DIFFERENT IN PATTERN IDENTIFICATION IN GT? The procedures for GT are in some ways similar to TA, but can seem very different, because GT uses quite different language. Furthermore, unlike IPA, there seem to be as many different processes and terms for coding and analysis in GT as there are authors on the subject (see Birks & Mills, 2011). We will briefly outline some of the unique and key elements specific to GT (drawing on Birks & Mills, 2011; Charmaz, 2006; Pidgeon & Henwood, 1996, 2004). In GT, the shift from the initial to the intermediate stage (Birks & Mills, 2011) of coding signals a different level of engagement with the data, but there’s no pre- determined point at which it’s time to shift from ‘coding’ to ‘theme development’ (unlike TA and IPA); the shift from initial to more ‘focused’ (Charmaz, 2006), ‘core’ (Pidgeon & Henwood, 2004) or ‘intermediate’ (Birks & Mills, 2011) coding in GT does not occur when all data items have been coded. Instead, it’s a judgement on the part of the researcher that they have reached a point where they are starting to see categories, rather than codes, in the data and think that close (line-by-line) coding is no longer productive (Birks & Mills, 2011). This is partly about reaching code satura- tion, when the initial coding phase stops generating anything new in relation to the concepts explored (this might mean not all data are fully coded or that you have made a number of sweeps through the data). However, the shift between coding ‘levels’ is not a direct linear, progressive process; it can be recursive (Charmaz, 2006; Pidgeon & Henwood, 2004). A number of different coding practices can apply in this stage, such as focused and axial coding. Focused coding (to use Charmaz’s, 2006, term) involves refining an index- ing system of initial codes to determine ones that are analytically useful, and using these as a basis for re-engaging with the data at a less fine-grained (i.e. not line-by-line) level. The analytic process is essentially about shifting from narrower codes to broader cat- egories, and building a more theoretical or conceptual analytic take on the data. This may involve simply ‘raising’ codes to categories, or developing categories by divid- ing codes, or by clustering codes (Pidgeon & Henwood, 2004), similar to the process of theme development in TA. This phase also requires writing clear definitions of each category (again, similar to writing theme definitions in TA; see Chapter 11). Exploring 10-Braun & Clarke_Ch-10.indd 239 28/02/2013 7:44:25 PM  240 Successfully analysing qualitative data BOX 10.3 SUPERORDINATE THEMES IN OUR FG DATA The fat self is a restricted self – captures the ways the experience of being fat was described as one of physical restriction, but also importantly one of psychological restriction. Being fat equals physical restriction Sally: I was having back ache leg ache god knows what else Carla: That weight you know really feeling like it was slowing me down and struggling to sort of run after [my children] Being fat equals psychological restriction Sally: I would have been too ashamed and felt I can’t do [a degree] cos I can’t go out in public The slim self is a free self – captures the way slimness was seen as desirable, and associated with being able to live a freer life, and thus be a different person; and the way losing weight was experienced as liberating Slimness is liberatory Sally: Because I’d lost the weight it gave me so much more confidence it was amazing Weight loss is transformative for the self Sally: It’s all had a bigger impact on how I feel about myself A spoiled identity: the stigma and shame of fatness – captures the experience of shame relating to social stigma around fatness, highlighting the emotional and relational experience of fatness, such as the emotional aspects of a diagnosis of obesity. Fatness is shameful Sally: I don’t think I’d have started the university course had I been the weight I was cos I’d have been too ashamed Carla: You look at yourself and think ‘oh I’m disgusting’ Diagnosis as obese as a negative experience Sally: Clinically I was classed as obese I used to find it really degrading and insulting Carla: ‘Oh my god I might as well just go and shoot myself now’ you know My fatness is not my fault – captures the way obesity was characterised as happening to them, rather than something they did themselves. 10-Braun & Clarke_Ch-10.indd 240 28/02/2013 7:44:25 PM  Identifying patterns across data 241 Response to depression Sally: Suddenly you feel a little bit depressed as well so (well) you tend to have the knock-on effect of eating Carla: Depression kicks in and it becomes this vicious cycle Obesity results from metabolic factors beyond the person’s control Sally: I have a very slow metabolism [...] there’s nothing I can do to change it Carla: I have to have huge doses of steroids which why I’m overweight at the moment There’s no easy way out of obesity – captures the way obesity was experienced as an ever- present threat in their lives. Weight is hard to lose Carla: It becomes such a huge mountain to climb Carla: I fought to get it off Weight is easy to gain Sally: If I look at a cream cake I put pounds on Carla: Half of its gone back on again relationships between categories, and mapping or modelling these relationships (Pidgeon & Henwood, 2004), rather than just identifying and defining categories, is important (just as it is in TA). Axial coding (Corbin & Strauss, 1990) is a formalised technique for conceptually mapping the relationships between categories (and sub-categories) through an ‘exhaus- tive coding’ (Pidgeon & Henwood, 2004: 640) process; it is an extra layer of analysis that although potentially useful is not critical (Charmaz, 2006). Charmaz (2006: 61) sug- gests that axial coding can ‘extend or limit your vision’ depending on the focus of your research and your ‘ability to tolerate ambiguity’. Researchers who prefer more structure will benefit from axial coding whereas those who value a more flexible approach may feel limited by it (see also Pidgeon & Henwood, 2004). ‘Advanced’ memos, which are more conceptual and relate to the developing concep- tual analysis (see Box 10.4), can also be useful for developing the relationships between categories. Through different coding phases and practices, the GT researcher produces 10-Braun & Clarke_Ch-10.indd 241 28/02/2013 7:44:25 PM  242 Successfully analysing qualitative data BOX 10.4 AN ADVANCED MEMO Different levels of explanation (18 August 2010)   We’re noticing there are three seemingly quite distinct levels of explanation within the data about the factors that seem to cause individuals to become obese, and also to explain the so-called obesity epidemic. The first level we might call structural – this is factors that relate to the very nature and organisation of society (e.g. Sally: ‘our lifestyle has changed’ [L154]). An example of this would be the idea that we now lead very sedentary lives, compared to what we did in the past, and that this results from the ways our jobs and working lives are organised. NB this explanation seems quite classed; to reflect a middle-class sensibility or conception of what a job entails (not physical labour) and this is applied universally as ‘the way things are’. The second level seems to relate to the way we are brought up to be as people – we might term these socialisation – so education and what we are taught or not in schools would relate here (e.g. Carla: ‘teaching them to cook at school you know’ [L871]). Mostly this is negative. Then there are explanations which are about what individuals do or don’t do. Some of these are things that happen to the person, or that the person has no control over (like their biology; e.g. Sally: ‘I mean metabolism does seem to have a lot to do with it” [L288]); others are things they themselves ‘choose’ (e.g. whether they ‘overeat’; Rebecca: ‘I have a twelve year old brother and he just eats what the hell he likes’ [L497]). The levels seem to interrelate (e.g. structural factors enable personal tendencies; Judy: ‘modern technology, like allows you to be lazy as well’ [L162]). This may form the basis for a model of explanation about why a person becomes obese. lots of information to help them develop their analysis – either a GT-lite version (see Chapter 11 for the model we developed from the FG data), or to continue to move to a full GT. This information includes: definitions of categories; the indexing system; memos; models and maps of the categories and relationships between them; and the coded dataset itself. There are two important points to remember: 1) although the analysis becomes more theorised, and shifts to the more conceptual domain, it needs to remain grounded in the data; 2) the research question can be refined to more clearly fit the developing analy- sis (Pidgeon & Henwood, 2004). WHAT ABOUT PATTERN-BASED DA? Methods of pattern-based DA, in contrast, tend not to offer structured guidance about the steps and processes for identifying ‘different’ patterns in the data, but instead, fol- lowing selective coding (see Chapter 9), involve reading and re-reading the data to iden- tify and interpret patterns and features of the selected data (Coyle, 2007; Parker, 1992; 10-Braun & Clarke_Ch-10.indd 242 28/02/2013 7:44:25 PM