Before anything else: a theme is not a topic. If your findings chapter has a section called “Benefits of the programme,” you have written a topic summary, and your examiner will say so. Braun and Clarke — whose 2006 paper is the reason you are doing this at all — name that exact construction as the classic error. Fixing it is most of the work below.
This guide follows the six phases in the authors’ own current wording. Note the qualifier “current”: they state on their own site that “the names for these phases have changed since first articulated in 2006,” so a textbook chapter written in 2014 will give you different labels for the same process.
Step 0: Decide which thematic analysis you are doing
This step takes ten minutes and prevents a methods chapter that contradicts itself.
Thematic analysis is not one method. Braun and Clarke describe it as “an umbrella term for a set or family of approaches” and identify three clusters: coding reliability TA, codebook TA, and reflexive TA. They differ “enormously in terms of both underlying philosophy and procedures for theme development.”
Most student projects that cite Braun and Clarke are doing reflexive TA, which is their approach. If you are doing reflexive TA, several things your methods textbook told you to do are things its authors explicitly do not want (see Step 2). Name the version in your methods chapter, cite the corresponding paper, and be consistent thereafter.
Then fix two more orientations, because your examiner will ask which you took:
| Choice | Option A | Option B |
|---|---|---|
| Where codes come from | Inductive — directed by the content of the data | Deductive — directed by existing concepts or ideas |
| How deep you read | Semantic — the explicit content of the data | Latent — the concepts and assumptions underpinning the overt content |
| What you take the data to be | Critical realist — an assumed reality evident from the data | Constructionist — realities produced within the data |
Braun and Clarke describe these as continua rather than “exclusionary oppositions,” and note that inductive, semantic and critical realist orientations often cluster together, as do deductive, latent and constructionist ones. What matters, in their words, is “that the analysis is theoretically congruent and consistent.”
Step 1: Familiarise yourself with the dataset
The published description: “reading and re-reading the data, to become immersed and intimately familiar with its content, and making notes on your initial analytic observations and insights, both in relation to each individual data item (e.g. an interview transcript) and in relation to the entire dataset.”
Two notes are being asked for, not one — per transcript, and across the set. Keep them in a dated familiarisation memo. It costs nothing, and it becomes the paper trail that shows the analysis was a process rather than an opinion.
Transcribe your own interviews if you can bear it. It is slow, and it is the fastest route to the immersion this phase is asking for.
Step 2: Code the entire dataset
The published description: “generating succinct labels (codes!) that capture and evoke important features of the data that might be relevant to addressing the research question. It involves coding the entire dataset, with two or more rounds of coding, and after that, collating all the codes and all relevant data extracts, together for later stages of analysis.”
Three operative words. Succinct — a code is a few words, not a sentence. Entire — all of it, not the interesting parts. Two or more rounds — your first pass will not be your coding.
A worked example. Data extract:
“I’d email my supervisor and then just… wait. Two weeks sometimes. And you can’t chase them, because they’re busy and you’re the one who’s late. So you just work on something else and pretend that’s a plan.”
Weak codes: supervision; email; delay. Those are topics. Better codes: waiting as an unmarked cost; student absorbs responsibility for supervisor’s delay; substituting activity for progress. Those are interpretations, and they can be built into something.
Do I need a second coder and a Cohen’s kappa?
Not if you are doing reflexive TA, and this surprises almost every student who has been told otherwise.
Braun and Clarke describe coding reliability approaches as ones that emphasise “the measurement of the accuracy or reliability of coding through the use of a structured codebook and multiple independent coders, and the measurement of coding reliability… calculated with statistical tests such as Cohen’s Kappa – a Kappa of >.80 indicates a very good level of coding agreement.” Their own position is explicit: coding with another researcher can be helpful for bouncing ideas around, but “this doesn’t necessarily result in ‘better’ coding… just different coding. This is why we don’t advocate the use of a coding frame, or the calculation of inter-coder reliability scores associated with coding reliability approaches.”
So the answer depends entirely on Step 0. If your supervisor requires a kappa, you are doing coding reliability TA, and your methods chapter should say so and cite Boyatzis or Guest rather than Braun and Clarke. Presenting a kappa inside a reflexive TA is the contradiction an examiner notices. Numerical reliability coefficients belong to a different family of measurement altogether — the conventions there are set out in our guide to what counts as an acceptable Cronbach’s alpha, and they do not transfer.
Step 3: Generate initial themes
The published description: “examining the codes and collated data to begin to develop significant broader patterns of meaning (potential themes). It then involves collating data relevant to each candidate theme, so that you can work with the data and review the viability of each candidate theme.”
Physically or digitally, cluster the codes. Sticky notes on a wall still beats software for this phase, because you need to see all of them at once and move them repeatedly.
One phrase to strike from your vocabulary here: themes emerged. Braun and Clarke object to it directly — the language “suggests that meaning is self-evident and somehow ‘within’ the data waiting to be revealed, and that the researcher is a neutral conduit,” whereas themes are “active co-productions on the part of the researcher, the data/participants and context.” Write “I developed” or “I generated,” not “themes emerged.” It is a two-word edit that signals you read past the abstract.
Step 4: Develop and review the themes
The published description: “checking the candidate themes against the coded data and the entire dataset, to determine that they tell a convincing story of the data, and one that addresses the research question. In this phase, themes are further developed, which sometimes involves them being split, combined, or discarded.”
And here is the definition that decides whether your chapter works: “In our TA approach, themes are defined as pattern of shared meaning underpinned by a central concept or idea.”
The test. State the theme’s central concept in one sentence beginning “This theme is about…”. If the sentence names a subject, it is a topic summary. If it makes a claim, it is a theme.
| Topic summary (weak) | Theme (strong) |
|---|---|
| Benefits of the programme | “Permission to be a beginner”: the programme’s value was licence to stop performing competence |
| Experiences of supervision | Waiting as unpaid labour: delay was absorbed by students as their own failure |
| Barriers to participation | Eligible but not invited: formal access without informal signalling produced non-attendance |
Braun and Clarke are precise about why the left column fails: names like these “identify that, for example, ‘benefits of X’ was an important area of the data in relation to the research question(s), but they don’t communicate the essence of this theme; they don’t tell the reader something specific about these benefits.”
How many themes should I have?
They answer this one with a number, which is rare and worth quoting: “In general, 2-6 themes (and subthemes) is about right for a single journal article, an undergraduate project, an Honours or Masters dissertation, and a single analytic chapter in a doctoral thesis.” They add that in a report of about 10,000 words, an overview “is unlikely to be able to sufficiently cover more than six themes (including subthemes) in any depth.”
Nine themes is not a richer analysis. It is usually a coding list that never went through Phase 4.
Step 5: Refine, define and name the themes
The published description: “developing a detailed analysis of each theme, working out the scope and focus of each theme, determining the ‘story’ of each. It also involves deciding on an informative name for each theme.”
Write a theme definition of 100–150 words for each one, for yourself rather than for the thesis. It should say what the theme is about, what falls inside it, what deliberately falls outside it, and which research question it answers. If you cannot write the boundary, the theme is not finished.
Then name it. A two-part name — a short participant phrase, a colon, then the analytic claim — does the most work in the least space, as in the right-hand column above.
Step 6: Write up
The published description: “weaving together the analytic narrative and data extracts, and contextualising the analysis in relation to existing literature.”
Weaving is the instruction. The commonest weak findings chapter presents a theme name, then three long quotations, then a sentence of commentary. Invert that ratio. A workable paragraph shape:
- State the analytic claim in your own words.
- Give a short extract that instantiates it, with a participant identifier.
- Say what the extract shows and why it supports the claim.
- Give a contrasting or complicating extract.
- Say what the tension means.
The contextualising half — connecting each theme back to the literature — is the step that turns a description into a contribution, and it works best when the review chapter was already organised by claim; our step-by-step guide to writing a literature review for a thesis sets out that structure.
Two Canadian specifics worth settling early
Bilingual data. If you interview in French and write in English, code in the language of the interview. Coding a translation analyses the translator’s choices rather than the participant’s, and a latent reading of a translated phrase is not defensible. Present extracts in the original with an English gloss, and say in your methods who translated them and when.
Ethics attaches before any of this. Interviews are research involving living human participants, and approval has to be in place before recruitment rather than before analysis — our guide to REB approval and TCPS 2 in Canada covers what your board will expect to see, including the consent language that governs how you may quote a participant later.
One further consequence to plan for now: your examiner may ask you to justify a specific coding decision on a specific transcript, which is one of the standard question types set out in our guide to what gets asked at a Canadian thesis defence. Dated memos from Phases 1 and 5 are what let you answer it.
Once the themes are settled, the chapter still has to be built and stay consistent while it is rewritten. Draft your findings chapter in Tesify and keep structure and references stable while every interpretation remains 100% written by you.
Frequently asked questions
What are the six phases of thematic analysis?
Familiarising yourself with the dataset; coding; generating initial themes; developing and reviewing themes; refining, defining and naming themes; and writing up. The names have changed since the 2006 paper, so use the current wording.
Is thematic analysis one method?
No. It is an umbrella term for a family of approaches, grouped into coding reliability TA, codebook TA and reflexive TA, which differ in both philosophy and procedure.
How many themes should a master’s thesis have?
Two to six themes including subthemes is described by the method’s authors as about right for an Honours or Masters dissertation, or for a single analytic chapter of a doctoral thesis.
Do I need inter-coder reliability for thematic analysis?
Not for reflexive TA, whose authors state they do not advocate coding frames or inter-coder reliability scores. If your programme requires a kappa, you are doing coding reliability TA and should cite that literature instead.
What is the difference between a theme and a topic summary?
A theme is a pattern of shared meaning underpinned by a central concept. A topic summary collects what participants said about a subject without a shared meaning holding it together. “Benefits of X” is a topic summary.
Should I say my themes “emerged”?
No. The phrase implies meaning was waiting in the data to be revealed. Themes are co-produced by the researcher, the data and the context, so write that you developed or generated them.
Do I have to code every transcript?
Yes — the published description of the coding phase specifies coding the entire dataset, over two or more rounds.
Inductive or deductive: which should I choose?
Whichever matches your research question, and then say so. Inductive coding is directed by the data’s content, deductive coding by existing concepts. They are ends of a continuum rather than a binary.
How many interviews do I need?
No number is published as a rule, and the method’s authors report having had datasets of around 20 interviews rejected by some journals as too small. Justify your number against your research question and your discipline’s norms rather than against a general figure.
Can I do thematic analysis without software?
Yes. Highlighters, printouts and a spreadsheet complete a thesis-scale analysis. Software organises codes; it does not do the interpreting.
What is the difference between thematic analysis and IPA?
IPA is closer to a methodology than a method: it fixes the theoretical framework, the kinds of question you may ask, and a small homogenous sampling strategy. Thematic analysis leaves those choices to you, which is why you must state them.
Where do participant quotations go?
Inside the analytic narrative, short and interleaved with your interpretation, rather than in blocks that the reader is left to interpret alone.