Analyzing Talk and Text II: Thematic Analysis

Thematic analysis (TA) is a popular and foundational method of analyzing qualitative policy data. It is concerned with the identification and analysis of patterns of meaning (themes) and constitutes a widely applicable, cost-effective and flexible tool for exploratory research. More generally, it constitutes a cornerstone of qualitative data analysis. Drawing principally on Braun and Clarke’s (2013; 2006) work, the chapter outlines when the use of this method is suitable and makes practical suggestions about how to plan and conduct TA research. Few policy studies employing TA contain a transparent discussion of research methods. This chapter stresses the importance of research transparency and methodological reflexivity: researchers should not only document what they do; they should also explicitly argue how and why they opted for specific methods and discuss implications for future empirical research.


Introduction
This chapter is concerned with the thematic analysis (TA) of data from interviews, (policy) documents or other verbal expressions in relation to media and creative industries policy research. It introduces TA as one of the most straightforward ways of deducing patterns of meaning -referred to as themes -from qualitative data. In its essence, TA consists of the analytical construction of: (a) codes, (b) themes in qualitative verbal expressions; and (c) patterns of recurrence, evaluation or associations within these themes.
TA is of great relevance for media and creative industries policy research using qualitative data for two reasons: first, it is a widely applicable and cost-effective means of exploratory research; second, one understanding of TA, as advanced in this chapter, is that it constitutes the essential starting point of virtually all qualitative data analysis.
TA is particularly suitable for analysing experiences, perceptions and understandings. It can be used to analyse a large variety of qualitative data and is a flexible method which can be applied within various theoretical frameworks. TA is also applicable independently of any initial theory and can be used for purely inductive research. Furthermore, it is suitable for the analysis of small, medium-sized and even large data sets.
TA has been referred to as 'possibly the most widely used qualitative method of data analysis' (Braun and Clarke, 2013: 175). Numerous studies on communications-, media-, creative-and cultural industries policy have employed it as a method (see e.g. Blomkamp, 2014;Darchen & Tremblay, 2015;Herzog & Dias Osório, 2018;Herzog & Karppinen, 2014;McNally et al., 2017;2018;MacLean, 2011;Potschka, 2012;Ruhode, 2016;Stover, 2010). However, when compiling the literature review, which served as a basis for this chapter, it became clear that in the majority of these studies there is little explicit discussion of the empirical methods employed. It was challenging to find ideal-type applications that could illustrate best practice, at least in relation to media and creative industries policy scholarship. A similar finding was reported by Hollifield and Coffey (2006: 579) in their overview of qualitative research in media management and economics: Perhaps the most striking finding of the analysis was that in the qualitative articles examined, there was a tendency toward methodological fogginess rather than transparency. It was rare to find a qualitative study in which the research design and the methods used to select cases, interview subjects, or sources of information were clearly spelled out so as to invite understanding, critique, or replication. 2 These observations inspired one of the arguments we advance in this chapter: the call for research transparency and methodological reflexivity. We aim to make a modest contribution to correct this shortcoming by discussing the conduct of TA. In our endeavour, we draw principally on Braun and Clarke's 'organic' or 'Big Q' approach to TA, which has to be distinguished from other TA approaches that retain a base in quantitative research: 'small q' TA approaches in Braun and Clarke's terminology (see Boyatzis, 1998;Fugard & Potts, 2015;Guest, MacQueen & Namey, 2012;Joffe, 2012). 3 The chapter outlines when the use of TA is suitable, and makes practical suggestions about how to apply TA in research on media and creative industries policy. We describe  six steps of TA and discuss them in the context of this research area. To start with, however, we take a step back and situate TA in qualitative data analysis more broadly.

TA as a Cornerstone of Qualitative Data Analysis and Boundaries to Other Methods
TA is a rather basic, flexible tool. At the heart of it lies a process called coding: the gradual development of labels and their application to segments of potentially relevant data. This categorization can be done inductively, aiming to generate new theory emerging from the data analysed (bottom-up), or deductively, testing theory (top-down) (Clarke & Braun, 2014: 1948. Compared with other long-established methods of qualitative data analysis -such as discourse analysis, grounded theory and interpretative phenomenological analysis -TA has only recently gained widespread prominence as a method on its own (Clarke & Braun, 2013: 120). To some extent, the flexibility of TA makes it hard to establish clear distinctions from other methods of analyzing qualitative data. 4 In fact, a variety of common methods of qualitative data analysis can be seen as special cases or extensions of TA. For instance, analytic induction employs TA but with the aim of refining an initial, hypothetical association between phenomena until a reformulated hypothesis fits all cases under investigation. Grounded theory typically combines TA with the ambition of adopting a pure inductive approach that does not start from pre-existing theory and related hypotheses but develops its results -concepts, categories and descriptions of their essential characteristics and theory about associations between concepts -from qualitative data only. Narrative analysis applies TA to specific types of interviews, in which participants recount continuous stories about their lives and events within. In all these methods of analysing qualitative data (and when developing ideas on how to collect further qualitative data), researchers will implicitly or explicitly develop a matrix of cases and recurrent concepts or themes, which is the essence of TA.
TA can also be conceived as a foundational technique in quantitative content analysis, which is a rapidly growing research area associated with such key words as 'digital humanities', 'text and data mining' or 'big data'. Here, researchers develop a coding scheme that defines specific (combinations of) verbal expressions and counts how often these occur in qualitative data, thus enabling automated processing of great volumes of qualitative data and generating Electronic copy available at: https://ssrn.com/abstract=3068081 quantitative data that can then be analysed using statistical methods. In many instances, at least implicitly, TA guides the researchers who engage in quantitative content analysis in their development of coding schemes.
Furthermore, TA has similarities with qualitative content analysis, which also follows a series of clearly defined steps (Schreier, 2014). Qualitative content analysis, however, emerged from quantitative content analysis and is solely focused on description. As such, it is suited for the analysis of large data-sets. 5 TA, by contrast, involves interpretation and can (also) be used to analyse small and medium-sized data sets (Braun & Clarke, 2013: 50). 6 Since TA operates within a qualitative paradigm, it is at the discretion of the experienced researcher to judge which sample size works best to answer a research question (Braun & Clarke, 2016).
TA is distinct from critical discourse analysis (CDA) in that the latter is concerned with the examination of power structures, inequalities and dominance. CDA ultimately aims for social change. By contrast, TA is an empirical research method which is open-ended regarding any outcome. Furthermore, there is no commonly accepted step-by-step guide for the conduct of CDA. Any researcher interested in the analysis of qualitative data should thus have the tools of TA at their disposal. According to Braun and Clarke (2006: 78), TA 'is the first qualitative method of analysis that researchers should learn, as it provides core skills that will be useful for conducting many other forms of qualitative research'. Without mastering this method, it is improbable that more ambitious data analyses will generate high quality results.
The Role of the Researcher TA helps researchers to establish what research participants (e.g. authors of texts or interviewees) consider important, how they categorize experiences and perceptions, what related attitudes they hold and how various categories are associated with each other. TA can be used to analyse a large variety of qualitative data derived, for example, from interviews, focus groups, Facebook comments and messenger interactions, YouTube videos, policy documents, press articles or any other type of text (see e.g. Ditchfield & Meredith, 2018;Kvale, 2007;Massey, 2011).
TA goes far beyond identifying whether an issue has been covered in a positive, negative or neutral tone. Classifying text into such broad categories can, for instance, be done using text analytics software such as WordStat 8. TA, by contrast, proceeds by first identifying codes, then themes and eventually patterns in a qualitative data set. As meanings are not openly available, this involves interpretation. Coding and theme development as steps in TA are driven by the goal of retaining considerable detail in the data items. Researchers develop and apply modes of classification and interpretation that help to infer non-obvious and credible meaning from this complex data. In TA, eschewing quantification, verbal expressions are reported and analysed by human readers, who deduce meaning in an informal process that tries to capture nuance in long sequences of text. The verbal expressions analysed will not necessarily correspond in a straightforward manner to a pre-existing repertoire of theoretical concepts and associations between them; indeed, researchers may even try not to rely too much on preconceived ideas. This is a mixed blessing. On the one hand, TA can be a useful tool for going beyond pre-conceived ideas. On the other hand, due to the complexity of data and the absence of pre-determined and clear-cut classification criteria, it requires a relatively large degree of individual judgement. We are not aware of any generally applicable way to entirely overcome this challenge in applied qualitative research. However, instances of individual judgement are inevitable in virtually any research process, even the most formulaic quantitative research − if to a lesser degree.
Another issue should be of greater concern. In TA, it can be hard to document the process that leads to the identification of themes, the selection of related statements in the data (e.g. interview extracts), and their interpretation in a transparent and clearly understandable manner . This can make it hard to convince others that the data set has been treated with due diligence and that the research results are trustworthy. As a corrective, we argue in this chapter that researchers applying TA should invest in carefully documenting the various stages of their research process, thereby adding transparency and reflexive practice.
There are several benefits in striving for transparency. First, it will ensure a better understanding of the trustworthiness and limitations of results and increase the impact of high quality empirical work, including its chance to play a role in policy-making processes. 'If decisions or actions are to be based on qualitative research', Ritchie and Spencer (1994: 175) note, 'then policy-makers and practitioners need to know how the findings of the research have been obtained.' Second, transparency is a prerequisite for an informed debate among researchers and readers on possible complements, extensions and improvements of the methods employed. Third, the ambition to document the research process in a transparent manner has a disciplining effect on researchers, who, like most people, face the temptation of favouring confirmation of their pre-conceived ideas or to minimize effort in dealing with empirical data.

Methodological Reflexivity
We have already stressed the importance of being transparent and clearly explaining the methodological and analytical decisions made. Some of the points we mention in this chapter go beyond research transparency and concern forms of research and methodological reflexivity in the sense that researchers scrutinize not only what they do but also how and why they do it (see May & Perry, 2013;Moravcsik, 2014). This involves critically reflecting on the methods employed, discussing the implications regarding the trustworthiness of results and seeking to derive lessons for further research. To illustrate this notion we distinguish three categories of transparency and reflexivity: basic information, detailed information and research reflexivity (see Fig. 22.1). 7 These three categories form a hierarchy. Any publication of TA results needs to document key aspects of the research, preferably in sufficient detail to help readers to assess the trustworthiness of results. In the most ambitious TA applications, researchers explicitly scrutinize and justify their methods themselves.
Basic and detailed information in the context of TA research refer to several types of documentation regarding data: the primary and secondary data collected for a piece of research; the process of data collection; and the distinction and separate documentation of the data corpus and the data set. 8 Where TA involves primary data, data collection and data analysis complement each other in an iterative process. Codes are gradually developed. Preliminary codes and themes accordingly inform the priorities set in further data collection. Furthermore, general decision-rules and criteria as well as changes to the analytic approach employed by the researchers during data analysis should be documented in as far as that is feasible. However, qualitative-interpretivist data analysis cannot be entirely reduced to general rules. The most effective means to document informal aspects of the data analysis (that take greater detail into account) are illustrative examples including extracts from the qualitative data. For best-practice examples see Braun and Clarke (2013: Chapter 13).
The most ambitious TA applications will also aim for research reflexivity. This concerns ethical considerations regarding the research and critical reflections on the choices made with regards to data analysis, the way in which these may have affected the results and whether there are alternatives and lessons for further research on related topics. Explicit discussion of these issues is rarely found in applied TA research regarding media policy or creative industries. Ethical concerns are often not addressed either. These may, for example, emerge in studies that are based on elite interview data (see Empson, 2018;Ganter, 2017;Herzog & Ali, 2015). In TA, the integration of data extracts which best represent themes is an integral part of producing a report. However, if informants do not give their consent for being identified and named, this may cause difficulties, particularly for studies drawing on small data sets, such as a limited number of interviews with the key actors who were in charge of designing and implementing a certain policy (see Herzog, 2016). Furthermore, far too few studies feature critical reflections on the choices made about methods of data collection and analysis. This is a missed opportunity to increase trust in the findings of high quality research and to advance scholarship. A well-performed TA stands up to these requirements. It should clearly outline the underlying assumptions, the options available and explain why researchers set priorities. Ideally, when reporting the research, scholars also include a critical reflection on data analysis (e.g. with regard to coding and theme development). In the following, we present some guidelines on how to fulfil these requirements.

Planning and Conducting a TA Project
As with any empirical project, researchers applying TA should first define a topic or research question and develop an idea of what sources and type of data would ideally be available to answer this question. Under real-world restrictions, researchers then need to decide what kind of data, from which sources, can actually be made available for the specific project at hand, and then set priorities. Sources of data may be documents (e.g. policy documents, mission statements, press articles, blog entries), interview-or focus group transcripts or transcripts from council or business meetings or parliamentary assemblies. Driven by the economics of research, the cost-efficient analysis of secondary data can be tempting but participation in primary data collection gives those conducting a TA a better understanding of the context. It also provides opportunities to ensure data-generation on topics deemed particularly relevant by the researcher. In the following paragraphs we introduce the six phases involved in carrying out a TA, as specified in  accessible and concise step-bystep guide, which we fit to the requirements of media and creative industries policy scholars.
Phase 1 is called 'familiarizing yourself with your data'. During this phase the researcher should break down a data set from the data corpus, read through the entire data set and actively engage with it by searching for patterns of meaning. In fact, the process of analysis Electronic copy available at: https://ssrn.com/abstract=3068081 begins when the researcher starts to search for these patterns. This may already happen when, during the process of an interview, the researcher makes notes linking statements of the interviewee to other data. The production of transcripts of verbal data (e.g. interviews, speeches, television programmes) falls into this phase. A TA can only be applied in relation to a data set in written form. Due to TA's flexibility, transcripts do not need to be prepared in one specific format. The transcripts subjected to a TA are also not required to contain as much detail as, for example, a transcript for a conversation analysis that indicates overlapping utterances, intervals, characteristics of speech delivery and so on (see Braun & Clarke, 2012: 60). In the report emerging from the research, scholars should document their underlying epistemological and other assumptions that influenced how they searched for patterns of meaning. The tool of memoing may be useful in this respect (see box 1).
Phase 2 is concerned with 'generating initial codes'. Codes are labels applied to segments of data which are likely to be relevant in the context of the research question(s). They are 'the building blocks of analysis: If your analysis is a brick-built house with a tile roof, your themes are the walls and roof and your codes are the individual bricks and tiles' (Braun & Clarke, 2012: 61). During this phase the entire data set is organized into meaningful groups. First, all data extracts are coded (for an example see Terry et al., 2017: 27). Then all data extracts with the same code are assembled (Braun & Clarke 2006: 89). 9 As phenomena differ from study to study, it is difficult to give advice on the number of appropriate codes to use. Fereday and Muir-Cochrane (2006: 84) mention an example in which they used six broad code categories, though many researchers may use more. Coding can be carried out in either a 'data-driven' or 'theory-driven' way. It can be done manually or by means of software (e.g. by using NVivo, MAXQDA or ATLAS.ti) (see Gibbs, 2013;Silver & Lewins, 2014). Figure  2 show a coding framework in NVivo 12. The report should document how many codes were generated, whether coding was data-or theory driven and, if applicable, which software was used.
Phase 3, 'searching for themes', can start when the entire data set has been coded and the individual codes have been collated. At the beginning of Phase 3 the researcher should address the crucial question of what does and does not count as a theme (see box 2). Based on these specifications when re-reading through the collated data extracts, the researcher starts constructing themes and sub-themes. 10 Themes do not emerge passively and the researcher needs to take an active role in interpreting and reporting them. In this, TA offers a certain degree of flexibility. It is, however, important to be consistent and transparent about the choices made. A useful way to conceptualize patterns and their relationships within the data is by visualizing them in one or more thematic maps (see Attride-Stirling, 2001). Mapping helps to develop an understanding of the significance of the themes. For this phase the researcher should document the characteristics a theme and sub-theme must fulfil. Phase 4, 'reviewing themes', is concerned with the step from developing provisional or candidate themes into final themes. Braun and Clarke (2006: 91) advise that '[d]ata within themes should cohere together meaningfully, while there should be clear and identifiable distinctions between themes'. This step may, for example, involve a reorganization of some coded data extracts, a grouping together of two provisional themes, renaming a theme and abandoning another. The preliminary thematic map will be refined during Phase 4. The overall aim is that the themes and sub-themes accurately represent the data set. At the end of Phase 4 themes may already have provisional names. Researchers should outline their role in theme interpretation. Given the interpretative nature of qualitative analysis, it is a common analytical shortcoming if themes are given the same name as question categories in interview schedules (Connelly & Peltzer 2016: 52).
During Phase 5, 'defining and naming themes', the names of the themes are reviewed. In this phase the researcher specifies the essence of each theme. What does the theme tell us that is relevant for the research question? How does it fit into the 'overall story' the researcher wants to tell about the data? Theme names need to be evocative, concise, catchy, punchy, and informative in that they 'immediately give the reader a sense what the theme is about' (Braun & Clarke 2006: 93;2013: 258). The themes are finalized when renaming has led to satisfactory results. If the analysis is carried out by a single researcher it may be wise to ask an external expert whether the themes are fully development, clear and capture all relevant data (King, 2004). Overall the finalization of themes is a time-consuming process and the researcher needs to invest sufficient care (see box 2). The researcher should also make clear the efforts invested in theme development. Explanation of the methodology used and research transparency are likely to increase trust in the findings. Phase 6, 'producing the report', is dedicated to the writing up. In fact, writing starts during Phase 1 and continues through Phase 6. Throughout the process of analysis the researcher simultaneously deals with the whole data set, the data extract focused on at any one point in time and the analytical report which is produced. This requires the researcher to move constantly back and forward. The final report should contain such data extracts as interview quotes which best represent a particular theme that emerged from the analysis. According to Braun & Clarke (2006: 93), '[e]xtracts need to be embedded within an analytic narrative that compellingly illustrates the story you are telling about your data, and your analytic narrative needs to go beyond description of the data, and make an argument in relation to your research questions.' In other words: 'The qualitative researcher has to provide some coherence and structure to this cumbersome data set while retaining a hold of the original accounts and observations from which it is derived' (Ritchie & Spencer, 1994: 176). There is detailed guidance available how research should be structured and presented in order to get it published. 11 There is no rule set in stone about following the six phases precisely as outlined. Braun and Clarke (2012: 60) acknowledge that scholars more experienced in the conduct of TA will likely get more insights into their data in the familiarization phase, are quicker in accomplishing the coding process and able to code at a level more conceptual than semantic, and more swiftly to develop themes confidently with less reviewing.

Conclusion
TA has a lot to offer the media and creative industries policy scholar. Its strengths include that it is comparatively easy to learn and flexible to apply. We have argued that TA is a foundational method of qualitative data analysis. Mastering this method equips the researcher with core skills that are valuable for many other forms of qualitative research.
TA is an interpretative method in qualitative data analysis: classifications and interpretations occur through complex and largely informal considerations that cannot be fully expressed in replicable terms such as a complete algorithm. The six phases, as specified by , offer clear guidance on how this method can be applied, though they should be applied flexibly in alignment with the research design, research question(s) and the data analysed. During the phases of TA, writing and analysis go hand in hand and the process of analysis is recursive and not linear in character.
We have started with the observation that in the majority of studies which applied TA in relation to media-and creative industries policy studies there is little explicit discussion of the methods used, and this may hold for qualitative research on many other topics. We suggest guidelines for more rigorous methods and documentation of research processes. A solid TA requires basic and detailed information on data collection, data analysis and the underlying reasoning. Research transparency and reflexivity are ingredients that differentiate a reasonable application from a good one. Qualitative research more generally is often subject to the 'anything goes' critique. More comprehensive documentation and explicit argumentation of why the methods employed were regarded as the best choices available will help counter superficial criticism. It should help to promote the impact of well-crafted qualitative research, to inform others about how to conduct high quality research, and perhaps even foster innovation building on the TA repertoire.

Box 22.1: Memoing
Memo writing refers to systematic and continuous, though flexible, note-making during the entire research process. As a technique it is principally associated with grounded theory research. However, memoing can be useful for a variety of qualitative research projects, including those employing TA. The writing of memos serves as a means for quality control in qualitative research. It supports an in-depth engagement of the researcher with the data. Memoing can be inspiring, foster self-reflexivity and help to transparently debate choices made during the research process and their implications. For instance, memoing will help keep available detailed information on why a particular data-collection method is employed or how perspectives guiding analytical decisions emerged. Memos can be recorded in any medium, from pen to word-processing software or as audio files. Memos should always be titled, dated and cross-referenced. Birks, Chapman and Francis (2008: 71) give a useful description of the benefits of memoing: Memos make it possible to record what you saw and why you saw it in the data. When your perspective changes, you must be able to justify how this will impact on decisions in regard to coding and analysis. This limits the impact of arbitrary, subjective viewing of the data by forcing you to explain any changes made to earlier decisions. These earlier decisions may have been made with a fresh outlook, or an exhausted set of eyes and so changing perspective is not a bad thing, just something that needs to be justified. 12

Box 22.2: What Counts as a Theme?
A theme is the core concept in TA. It can be defined as 'an abstract entity that brings meaning and identity to a recurrent experience and its variant manifestations. As such, a theme captures and unifies the nature or basis of the experience into a meaningful whole' (DeSantis & Ugarriza, 2000: 362). Themes have to be carefully and thoughtfully extracted by the researcher. In Braun and Clarke's (2016: 740) words, they are 'actively crafted', reflecting the interpretative choices the researcher has made. Themes are usually implicit, implied and embedded in repetitive expressions, even though these may be variant. A good theme contains more than one or two words. It can 'stand alone as a meaning statement' (Connelly & Peltzer, 2016: 55). In the final report it has to be supported by good examples from the data that closely match the theme. Underdeveloped themes are a common mistake that can hinder the publication of research findings. Examples of underdeveloped themes are exact words or phrases used, for example, in an interview transcript. In these cases no interpretative efforts have been undertaken. The identification of basic social functions (e.g. social conflict) or structural components (e.g. economics, politics) as themes is also often seen as an indication that themes have not been thoroughly developed. Furthermore, in the latter case the close link between the theme and the actual data set is missing. Researchers should outline a rationale for designating a theme and acknowledge their own perspective and bias in theme development (see Sandelowski & Leeman, 2012).