Thematic analysis is a method for finding, interpreting and reporting patterns of meaning, called themes, across qualitative data such as interview transcripts, focus groups or open survey answers. This guide follows the approach set out by Virginia Braun and Victoria Clarke, which moves through six phases: getting to know the data, coding it, building themes from the codes, reviewing those themes, defining and naming them, and writing them up.
If you’re about to analyse interviews for a dissertation or thesis, this guide takes you through each phase, then works through a short example from interview extract to finished theme. It also covers the choices your examiners will ask about, such as inductive or deductive coding, and ends with a template for describing your analysis in your methods chapter.
Where thematic analysis comes from
Braun and Clarke’s 2006 paper, ‘Using thematic analysis in psychology’, published in Qualitative Research in Psychology, set out their approach. Their own guide describes thematic analysis as popular for analysing qualitative data ‘in many disciplines and fields’, and as ‘an umbrella term for a set or family of approaches’. They now call their version reflexive thematic analysis.
Two ideas sit at the centre of their approach:
- A theme is not a topic. Braun and Clarke define themes as a ‘pattern of shared meaning underpinned by a central concept or idea’. ‘Supervision’ is a topic; ‘supervision as a negotiation of independence’ could be a theme.
- You generate themes; they don’t ’emerge’. Their FAQs say they ‘do not conceptualise themes as "emerging" from data’, and their guide describes patterns as generated by the researcher through familiarisation, coding and theme development.
The six phases of thematic analysis
Braun and Clarke’s site lists six phases. They’re sequential, but their guide describes analysis as typically recursive, so expect to go back to an earlier phase as your understanding grows.
| Phase | What you do | What you have at the end |
|---|---|---|
| 1. Familiarising yourself with the dataset | Read and re-read transcripts, listen to recordings, make brief notes on first impressions | Familiarity with the whole dataset and a page of early ideas |
| 2. Coding | Work through every item and attach short labels (codes) to anything relevant to your research question; their guide recommends two or more rounds | A set of codes, each with the extracts it applies to |
| 3. Generating initial themes | Cluster codes that seem to share a central idea into candidate themes | A first map of possible themes |
| 4. Developing and reviewing themes | Check each theme against its coded extracts and against the whole dataset; merge, split or drop themes that don’t hold | A smaller set of themes that fit the data |
| 5. Refining, defining and naming themes | Write a short definition of each theme’s central idea and choose a name that says what it is | A definition and name for each theme |
| 6. Writing up | Write the analysis, weaving together extracts and your interpretation, and relate it to the literature | Your findings or results chapter |
How many themes and how much data?
Braun and Clarke’s FAQs suggest that two to six themes (and subthemes) is about right for a single journal article or an undergraduate project, and that five interviews, however long, is probably a bit too small for thematic analysis. A PhD thesis may report more, spread across chapters, but the same test applies: every theme needs a clear central idea and enough data behind it.
A worked thematic analysis example
The extract below was written for this guide, not taken from real research. It imagines a second-year PhD student answering the question ‘How would you describe your relationship with your supervisor?’.
‘At the start I sent her everything, every graph, and waited for her to tell me what to do next. Now I mostly decide and then check. She’s still there if I’m stuck, but she’s started saying "you’re the expert on this now", which is nice but also a bit scary. Meetings used to be weekly; now it’s more like once a month, and I have to be the one who sets the agenda. Sometimes I’m not sure if she’s stepping back because she trusts me or because she’s busy.’
Phase 2: coding the extract
Here is one first round of codes. Each code is a short label that captures something relevant to the research question (how students experience supervision).
| Extract | Code |
|---|---|
| ‘At the start I sent her everything… waited for her to tell me what to do next’ | early dependence on supervisor |
| ‘Now I mostly decide and then check’ | growing decision-making |
| ‘She’s still there if I’m stuck’ | supervisor as safety net |
| ‘"you’re the expert on this now", which is nice but also a bit scary’ | ambivalence about expertise |
| ‘Meetings used to be weekly; now it’s more like once a month’ | less frequent contact |
| ‘I have to be the one who sets the agenda’ | student takes responsibility for meetings |
| ‘not sure if she’s stepping back because she trusts me or because she’s busy’ | uncertainty about supervisor’s reasons |
Notice that the codes are mostly semantic: they stay close to what the student said. ‘Ambivalence about expertise’ is more latent, because it interprets a feeling the student only hints at. Both kinds are useful.
Phases 3 to 5: from codes to themes
Across a full dataset you’d have codes like these from many participants. Suppose similar codes recur in other interviews. You might cluster them like this:
| Candidate theme | Codes it brings together | Central idea |
|---|---|---|
| Earning independence | early dependence on supervisor; growing decision-making; student takes responsibility for meetings | Independence is something students feel they grow into and have to prove |
| Stepping back or stepping away? | less frequent contact; uncertainty about supervisor’s reasons; supervisor as safety net | The same change in contact can read as trust or neglect |
| Becoming the expert | ambivalence about expertise | Being treated as the expert is both affirming and unsettling |
In phase 4 you’d test these against the whole dataset. ‘Becoming the expert’ rests on one code here, so you’d check whether other participants describe the same mix of pride and unease. If they don’t, you might fold it into ‘Earning independence’ as a subtheme. In phase 5 you’d write a two- or three-sentence definition of each surviving theme and settle its name.
Phase 6: writing up a theme
Using the invented data, a short write-up of the second theme might read:
Stepping back or stepping away? Most participants described contact with their supervisor becoming less frequent over the PhD. What varied was how they read it. For some, fewer meetings signalled trust; for others, the same change raised doubts about whether their supervisor was still invested, as one student put it: ‘Sometimes I’m not sure if she’s stepping back because she trusts me or because she’s busy’. This uncertainty matters because…
The write-up states the theme’s central idea, shows the variation within it, uses an extract as evidence and then says why it matters, which is the interpretation an examiner looks for.
Inductive or deductive, semantic or latent?
Two choices shape how you code. Braun and Clarke describe them as orientations to the data:
| Choice | One end | The other end |
|---|---|---|
| Where your codes come from | Inductive: coding and themes are directed by the content of the data | Deductive: coding and themes are directed by existing concepts or ideas, such as a theory you’re testing |
| How deep your codes go | Semantic: codes reflect the explicit content of the data | Latent: codes capture the concepts and assumptions underpinning the explicit content |
You can combine them: for example, an inductive first round of coding, followed by a later look at how the themes relate to a framework from the literature. Whatever you choose, say so in your methods chapter and explain why it fits your research question.
Which type of thematic analysis are you using?
Writing for the Social Research Association, Braun and Clarke group thematic analysis into three types, and warn that not recognising the differences can compromise the quality of research.
| Type | How coding works | Codebook? | Multiple coders and agreement scores? |
|---|---|---|---|
| Coding reliability TA | A fixed codebook is applied to the data | Yes | Yes: independent coders, with statistically tested agreement treated as evidence of reliable coding |
| Codebook TA (for example framework or template analysis) | A structured codebook, often developed early, is used to map or chart the analysis | Yes | Depends on the method |
| Reflexive TA (Braun and Clarke) | Open, organic coding that changes as the analysis develops | No | No: coding by more than one person is for reflexivity, not agreement |
If you follow Braun and Clarke’s six phases, you’re doing reflexive thematic analysis, so don’t report an inter-rater reliability score: their FAQs explain that the approach treats meaning as open to more than one interpretation. If your supervisor or department expects a codebook and coder agreement, use a coding reliability or codebook approach instead and cite it.
Software is a separate choice. Their FAQs say using a program such as NVivo is ‘not incompatible’ with reflexive thematic analysis.
Thematic analysis vs grounded theory and IPA
How does thematic analysis differ from two other qualitative methods? Braun and Clarke’s FAQs draw the line like this:
- Grounded theory aims to build a theory from the data. Thematic analysis can produce conceptually rich interpretations but ‘does not attempt to develop a theory’.
- Interpretative phenomenological analysis (IPA) is their suggestion for small studies that look closely at each individual’s experience; thematic analysis for larger datasets where you’re looking for patterns across participants.
Our guide to types of research explains where qualitative methods sit alongside quantitative and mixed-methods designs.
Common mistakes to avoid
- Themes that are just topics. A theme called ‘Funding’ that lists everything participants said about money is a topic summary. Ask what the central idea is.
- Saying themes ’emerged’. It hides your role. Write that you developed or generated them.
- Mixing approaches. Citing Braun and Clarke’s reflexive approach while reporting a codebook and an inter-rater reliability score suggests confusion about method.
- Too many themes. Ten thinly supported themes are weaker than three well-evidenced ones.
- Extracts without analysis. Quotes show what was said; your job is to say what it means and why it matters.
How to describe thematic analysis in your methods chapter
Examiners want to know what you did and why. A template you can adapt:
‘Interview transcripts were analysed using reflexive thematic analysis (Braun and Clarke, 2006), following the six phases of familiarisation, coding, generating initial themes, developing and reviewing themes, refining, defining and naming themes, and writing up. Coding was [mainly inductive / guided by X framework] and captured both semantic and latent meanings. I coded the full dataset in [two] rounds using [software, if any], and reviewed candidate themes against the coded extracts and the whole dataset. I kept a reflexive journal to record how my own position as [your relationship to the topic] shaped the analysis.’
Then cite the version you used and give a short example of your coding, perhaps a table like the one above, in your appendix. Our guide to thesis appendices explains what else belongs there. Your methods chapter should also describe how you collected the data; our guide to research instruments explains the tools researchers use to collect data, and if part of your project is quantitative, see our guide to the statistical treatment of data.
Next steps
Start with phase 1: read your transcripts through once without coding anything, and note what strikes you. Then code in at least two rounds before you look for themes. If you’re planning a PhD that uses qualitative methods, you can search current PhD opportunities to find projects in your area.
