- Thematic analysis (TA) finds and reports patterns of meaning across qualitative data, such as interview transcripts.
- Braun and Clarke’s six phases are: familiarisation, coding, generating initial themes, reviewing themes, defining and naming themes, and writing up.
- A code labels one small idea. A theme is a pattern of shared meaning built from several codes, not just a topic.
- Report each theme with a name, a definition, short quotes and your own analysis.
- In reflexive TA your judgement is part of the method, so a reflexive account and an audit trail matter more than a reliability score.
The problem: ten transcripts and no idea what to do next
You have finished your interviews. Ten final-year students with part-time jobs talked to you for 40 minutes each, and you now have about 60 pages of transcripts. Your supervisor says, “Use thematic analysis.” Where do you start?
This is a common sticking point in qualitative dissertations, interview-based IB Extended Essays and A-level NEA projects. Students either summarise each interview in turn or list topics (“work”, “money”) and call them themes. Neither earns high marks.
This guide follows one interview extract the whole way: into codes, into themes, into a codebook and into a written-up theme with quotes.
What thematic analysis is
Virginia Braun and Victoria Clarke’s 2006 paper made thematic analysis a named, teachable method [1]. They describe TA as a method for identifying, analysing and reporting patterns (themes) within data. It works with interviews, focus groups, open survey answers, diaries or documents.
Two other terms matter:
- Code: a short label for a feature of the data that is relevant to your research question. One sentence can carry several codes.
- Data item and data set: one transcript is a data item. All the transcripts you analyse together are the data set.
Reflexive TA and the other versions
Since 2006, Braun and Clarke have refined their approach and now call it reflexive thematic analysis [2]. The key ideas are simple:
- Themes do not “emerge” from the data. You, the researcher, actively build them.
- Your own position (background, assumptions, theory) is a resource, not a flaw, but you must reflect on it openly.
- Quality comes from deep, thoughtful engagement with the data, not from two coders agreeing.
They contrast this with “coding reliability” approaches, which fix codes in advance and use several coders to check consistency [2]. Between the two sit “codebook” styles, which use a structured codebook but are less concerned with agreement statistics. Whichever you use, name it and stay consistent with it.
| Version | Where codes come from | Typical quality check |
|---|---|---|
| Reflexive TA | Built from the data, revised throughout | Reflexive journal, audit trail, rich quotes |
| Codebook TA | A codebook developed early, then refined | Clear code definitions, team discussion |
| Coding reliability TA | Fixed codebook agreed before coding | Inter-coder agreement (for example Cohen’s kappa) |
The six phases at a glance
Braun and Clarke set out six phases [1]. In their later book they renamed some of them to stress that themes are generated, not found [3]. The phases are not a strict line. You will loop back, especially between phases 3, 4 and 5.
| Phase | 2006 name | Later name | What you actually do |
|---|---|---|---|
| 1 | Familiarising yourself with the data | Familiarising yourself with the dataset | Transcribe, read and re-read, note first ideas |
| 2 | Generating initial codes | Coding | Label every relevant feature across the whole data set |
| 3 | Searching for themes | Generating initial themes | Cluster codes into candidate themes |
| 4 | Reviewing themes | Developing and reviewing themes | Check themes against the coded extracts and the full data set |
| 5 | Defining and naming themes | Refining, defining and naming themes | Write a definition and a sharp name for each theme |
| 6 | Producing the report | Writing up | Weave quotes and analysis into an argument |
Phase 1: Familiarise yourself with the data
Transcribe your own interviews if you can. Then read every transcript at least twice. Jot down anything that strikes you: a repeated phrase, a feeling, a contradiction. These are memos, not codes.
Also start a short reflexive journal. Note what you expected to find and why. If you work part-time yourself, say so; it helps you notice when you read your own story into someone else’s words.
Phase 2: Code the whole data set
Go through every transcript line by line, labelling each meaningful chunk. Code everything relevant to your research question, keeping enough surrounding text for the quote to make sense later.
Codes can be semantic (close to what the person said, such as “can’t plan the week”) or latent (what lies underneath, such as “feeling like a fake student”). Most good analyses use both.
Code this extract from an interview with a final-year student who works in a supermarket. The research question is: “How do final-year students in paid work experience their studies?” (The extract is illustrative.)
Interviewer: How do you fit your shifts around your course?
P4: Honestly, I don’t. The course fits around the shifts. My manager puts the rota up on a Thursday, so I can’t plan my week until then. I’ve missed two seminars this term because I got called in. I feel guilty about it, like I’m not a proper student. But if I don’t take the hours, I can’t pay rent. My friends who don’t work go to the library together after lectures, and I’m just… not there. I’ve started recording lectures and listening on the bus. That helps. And one lecturer lets me hand in on Monday instead of Friday, which honestly saved me.
Result: ten codes from one extract of about 100 words. Across ten transcripts you may have 100 or more codes, many repeated. That is normal.
Phase 3: Generate initial themes
List every code on cards or in a spreadsheet and sort them into clusters that share a core idea. Ask of each cluster: “What is the central organising concept here?” If the answer is just a topic word, such as “work”, it is not yet a theme.
Group the codes from all ten transcripts into candidate themes. Brackets show how many participants had each code. (The counts are illustrative.)
Result: three candidate themes. Counts show spread, but frequency alone does not make a theme important; relevance to the research question does.
Phase 4: Review the themes
Review at two levels. Do the extracts under each theme hang together? Do the themes capture the whole data set? Expect to merge, split or drop themes. Here, “Time is not mine” and “Not a proper student” are linked (missed seminars feed the guilt) but each has its own central concept, so keep both and show the link in a thematic map.
Phase 5: Define and name the themes
For each theme, write a two- or three-sentence definition: what it is about, what it is not about, and how it answers the research question. Then choose a name that does analytic work. A short participant quote often makes a strong name (“Time is not mine”); “Work issues” tells the reader nothing. Now tidy your codebook too. In reflexive TA it is optional, but an appendix codebook makes your decisions visible.
Write codebook entries for three codes from the study, ready for an appendix. (The entries are illustrative.)
| Code | Definition | Include when | Exclude when | Example | Theme |
|---|---|---|---|---|---|
| Rota released late | Shift times known too late to plan study | Short notice of shifts, being “called in” | Long shifts known well ahead (code as “work is not optional”) | “puts the rota up on a Thursday” | 1 |
| Proper student ideal | Comparing self to an imagined full-time student | “Real” or “proper” student talk, feeling like a fake | Plain dissatisfaction with grades | “like I’m not a proper student” | 2 |
| Flexible deadlines | Staff adjusting rules to fit work | Extensions, later hand-in days, recorded sessions offered by staff | Formal extenuating circumstances claims | “lets me hand in on Monday” | 3 |
Result: an examiner can trace exactly how a piece of data ended up in a theme. That transparency is much of what “rigour” means in qualitative work.
Phase 6: Write up
Your findings chapter should read as an argument, not a list of quotes. For each theme:
- Claim. State what the theme is in one or two sentences.
- Evidence. Give one or two short quotes, each with a participant code (P4, P7).
- Analysis. Say what the quote shows and why it matters for your research question.
- Link. Connect the theme to the literature or to the next theme.
Here is how the start of theme 1 might read:
Theme 1, “Time is not mine”, captures how paid work, rather than the course, set the rhythm of participants’ weeks. Eight of the ten participants described fitting study around shifts: “the course fits around the shifts” (P4). Unpredictability mattered as much as hours: “my manager puts the rota up on a Thursday, so I can’t plan my week until then” (P4). Most framed work as non-negotiable because it paid the rent. This loss of control over time also fed the second theme, as missed seminars became evidence, in participants’ eyes, that they were not “proper” students.
The quotes are short, the analysis does the work, and the paragraph points to the next theme.
Reliability debates: do you need inter-coder agreement?
In coding reliability approaches, several people code the same data with a fixed codebook, and agreement is measured, often with Cohen’s kappa [4]. Kappa corrects raw agreement for the agreement you would expect by chance:
where \(p_o\) is observed agreement and \(p_e\) is chance agreement. Suppose two coders each decide, for 40 transcript segments, whether the code “proper student ideal” applies. Both say yes for 14 segments and both say no for 20; they disagree on 6 (3 each way). Then \(p_o = 34/40 = 0.85\). Each coder says yes 17 times, so \(p_e = 0.425^2 + 0.575^2 = 0.511\). That gives \(\kappa = (0.85 – 0.511)/(1 – 0.511) = 0.69\). (The figures are illustrative.)
In reflexive TA, Braun and Clarke argue that this kind of check does not fit. Coding is interpretive, and quality comes from reflexive, thoughtful engagement rather than coder consensus [2].
So what should you do?
- If your department or journal expects inter-coder agreement, use a codebook or coding reliability approach and report kappa properly.
- If you use reflexive TA, do not bolt on kappa. Show quality through a reflexive journal, an audit trail (data, codes, candidate and final themes), a thematic map and quotes from many participants.
For qualitative quality criteria such as credibility and transferability, see Bryman’s Social Research Methods [5] and our guide to reliability and validity in research.
Where to go next
For every step of a qualitative project in one place, the Research Methods Made Simple textbook covers interviews and thematic analysis with a codebook template and 10 companion interview transcripts. The Research Methods Made Simple Practice Questions book adds qualitative tasks on those same transcripts, with worked answers.
Practice questions
EasyQ1. Put Braun and Clarke’s six phases of thematic analysis in order, using the 2006 names. Show worked answer
2. Generating initial codes.
3. Searching for themes.
4. Reviewing themes.
5. Defining and naming themes.
6. Producing the report.
The process is recursive: you may loop back, for example from phase 4 to phase 3.
EasyQ2. What is the difference between a code and a theme? Show worked answer
A theme is a pattern of shared meaning across the data set, built from several codes around one central organising concept.
Example: “rota released late” is a code; “Time is not mine” (paid work controls students’ time) is a theme.
MediumQ3. A student’s findings chapter has three themes: “Family”, “Money” and “University”. Explain the problem and suggest how to fix it. Show worked answer
Fix: return to the coded extracts under each heading and ask what shared meaning they carry.
For example, “Money” might become “Work is not optional: paid hours as a condition of staying at university”.
Each theme should state a central idea that answers the research question.
MediumQ4. Two coders rate 50 segments for one code. Both say “yes” to 18, both say “no” to 24, and they disagree on 8 (coder A says yes to 4 of these; coder B says yes to the other 4). Calculate Cohen’s kappa. Show worked answer
Coder A says yes to 18 + 4 = 22 segments, so 0.44. Coder B also says yes to 22, so 0.44.
Chance agreement: \(p_e = 0.44 \times 0.44 + 0.56 \times 0.56 = 0.1936 + 0.3136 = 0.5072\).
\(\kappa = (0.84 – 0.5072)/(1 – 0.5072) = 0.3328/0.4928 = 0.675\), about 0.68.
Note: report kappa only if you are using a codebook or coding reliability approach, not reflexive TA.
ExamQ5. “Themes emerged from the data.” Evaluate this statement as a description of reflexive thematic analysis. [6 marks] Show worked answer
Mark scheme.
1 mark: “emerged” implies a passive researcher.
1 mark: reflexive TA sees the researcher as actively generating themes.
1 mark: role of research question, theory or subjectivity in shaping themes.
1 mark: counter-point that themes must be grounded in the data.
1 mark: names a check (phase 4 review, audit trail, thematic map).
1 mark: supported judgement with better wording (“generated”, “constructed”).
Frequently asked questions
Is thematic analysis a methodology or a method?
It is a method: a set of procedures for analysing qualitative data. It is not tied to one theory, so state your own position and keep your analysis consistent with it.
Should I code inductively or deductively?
Inductive coding starts from the data; deductive coding starts from a theory or prior framework. Many projects mix the two. Say which you did, and if you used a theory, show how it shaped your codes.
How many interviews and themes do I need?
There is no magic number. Justify your sample from the research question, the depth of each interview and your time, and explain it in your methodology chapter. Two to six well-developed themes is typical for a dissertation; ten usually means some are codes or topics.
Can I use thematic analysis for open survey questions?
Yes. TA works with any text data. Short answers give less depth, so expect more semantic codes. Report how many respondents answered the question.
References
- Braun, V. and Clarke, V. (2006) ‘Using thematic analysis in psychology’, Qualitative Research in Psychology, 3(2), pp. 77–101. https://doi.org/10.1191/1478088706qp063oa
- Braun, V. and Clarke, V. (2019) ‘Reflecting on reflexive thematic analysis’, Qualitative Research in Sport, Exercise and Health, 11(4), pp. 589–597. https://doi.org/10.1080/2159676X.2019.1628806
- Braun, V. and Clarke, V. (2021) Thematic Analysis: A Practical Guide. London: SAGE. https://www.sagepub.com/en-us/cab/thematic-analysis/book248481
- Cohen, J. (1960) ‘A coefficient of agreement for nominal scales’, Educational and Psychological Measurement, 20(1), pp. 37–46. https://doi.org/10.1177/001316446002000104
- Clark, T., Foster, L., Sloan, L. and Bryman, A. (2021) Bryman’s Social Research Methods. 6th edn. Oxford: Oxford University Press.

