Site Logo
All articles
Research MethodsThematic Analysis

Reflexive Thematic Analysis: the Braun and Clarke approach

Evidano8 min read

Thematic analysis is the most used and most misused qualitative method in circulation. Much of what is published as "thematic analysis following Braun and Clarke" is something the authors have spent two decades explicitly disowning: themes as topic summaries, codebooks fixed in advance, inter-rater reliability, and a data-saturation claim. Reflexive TA is a specific approach with specific commitments, and the commitments are what make the six phases work. Reading only the phases is how most people get it wrong.

Three commitments the phases rest on

Themes are generated, not discovered. They do not lie in the data waiting to be found; they are analytic outputs the researcher builds from codes. The language matters, because "themes emerged" implies a passive process that the method rejects and that conceals the analytic work.

Researcher subjectivity is a resource. Reflexive TA does not treat the analyst as a source of bias to be minimised through procedural controls. What the researcher brings — theoretical position, disciplinary training, personal relationship to the topic — shapes the analysis and should be declared and reflected on, not neutralised.

A theme is a pattern of shared meaning organised around a central concept. Not a topic. "Communication" is a topic; "being informed but not consulted" is a theme, because it has a central organising concept and makes a claim. This single distinction accounts for most of the difference between strong and weak thematic analysis.

From these follows what reflexive TA rejects: codebooks fixed before analysis, inter-rater reliability coefficients, and data saturation. Each assumes there is one correct coding to converge on. Braun and Clarke argue that assumption is incoherent inside an approach where coding is interpretive — a position elaborated in Conceptual and design thinking for thematic analysis.

Origins, and why the terminology changed

The founding paper is Braun and Clarke's Using thematic analysis in psychology, written because thematic analysis was widely practised and nowhere properly specified — researchers were doing it without being able to say what they had done.

The name changed later. What is now reflexive thematic analysis was distinguished from coding reliability approaches (Boyatzis, Guest) and codebook approaches (framework analysis, template analysis), because all three were being cited as "thematic analysis" while making incompatible assumptions. Their current statement is Thematic Analysis: A Reflexive Approach.

They have also written directly about how badly the method is reported — see Supporting best practice in reflexive thematic analysis reporting, which is unusually blunt about the specific errors reviewers should be catching.

Three families that all get called thematic analysis

Reflexive TACoding reliability TACodebook TA
Coding frameDevelops throughout; never fixedFixed early, applied by multiple codersStructured, revisable (framework, template)
Multiple codersOptional; not for agreementCentral — agreement is the pointCommon, for consistency
Reliability statisticsRejected as incoherentRequiredSometimes used
Researcher subjectivityA resource, declaredA bias to controlManaged through transparency
ThemesGenerated late, from codesOften specified earlyOften specified early
SaturationRejectedCommonly claimedSometimes claimed

The six phases

1. Familiarising yourself with the data

Read the whole dataset actively, more than once, making notes on what strikes you. Transcribing yourself counts as familiarisation and is worth the time if the schedule allows.

Active reading means reading with questions — what is this person doing, what is assumed here, what is not said — rather than reading to remember.

2. Generating initial codes

Code systematically across the whole dataset. Codes are short analytic labels for something interesting about the data; they can be semantic (close to what was said) or latent (about assumptions and meanings underneath).

Code generously and allow overlap. A first coding pass that produces 30 codes for 20 interviews has almost certainly missed things — 150 to 300 is unremarkable.

3. Generating initial themes

Cluster codes into candidate themes, each organised around a central concept. This is where the analytic step happens and where most published work stops short, producing groups-of-codes-by-topic instead.

The test for a candidate theme is whether you can state its central organising concept in a sentence that makes a claim. If the sentence is a noun phrase, it is a topic.

4. Reviewing themes

Check each candidate theme against its coded extracts, then against the whole dataset. Themes get split, merged, promoted, demoted or abandoned here, and abandoning one that took a week to build is normal.

A theme should have internal coherence and external distinctiveness: the extracts inside it belong together, and it is doing work no other theme is doing.

5. Refining, defining and naming themes

Write a short definition of each theme — what it captures, what it excludes, what it contributes to the overall analysis. Writing the definition is a test: a theme you cannot define in a paragraph is not ready.

Name themes for what they claim. "Barriers" is a container; "the system assumes someone at home during the day" is a theme name that carries the finding.

6. Writing up

The write-up is part of the analysis, not a report of it. Extracts must be embedded in an argument, not listed under headings with a sentence of commentary.

A results section that could be read as "here are six topics people mentioned" has not completed phase five.

Worked example: turning topics into themes

A study of 18 interviews with carers of people with dementia produced an early set of candidate themes: "getting information", "dealing with services", "impact on work", "support from family". Four topics. Everything anyone mentioned fitted into one of them, which is the warning sign — a set of categories that accommodates all the data usually accommodates it by saying nothing.

Phase four sent the analyst back through the extracts. Under "getting information" sat two quite different things: carers who could not obtain information, and carers who were given a great deal of it at a moment they could not use it. Under "dealing with services" sat accounts that were mostly about being positioned as a relative when the carer was functioning as a case manager.

The revised themes made claims. "Information arrives on the system's schedule, not the carer's" brought together the timing accounts from three original topics. "Doing the coordination without the standing to do it" captured the positional problem, and drew extracts from services, work and family alike. A third theme, "the competence you acquire has no name", was built from scattered codes about carers becoming expert in something with no recognised status.

Three themes replaced four topics, drew on the same extracts, and said something. The original set could have been produced from the topic guide without reading the data — which is the practical test worth applying to any thematic analysis.

Common mistakes

  • Themes as topics. The dominant failure. If a theme name is a noun phrase from the interview schedule, phase three has not happened.
  • Themes announced as emerging. Language that hides the analytic work, and the authors object to it specifically.
  • Reporting inter-rater reliability. Coherent in coding-reliability TA, incoherent here. Citing Braun and Clarke alongside a kappa coefficient signals the method was not read.
  • Claiming data saturation. Rejected within this approach. Meaning generation does not stop.
  • A codebook fixed in advance. That is codebook TA — legitimate, and a different method that should be cited as one.
  • No reflexivity statement. Subjectivity as a resource requires saying what the researcher brought.
  • Results as a quote gallery. Extracts need an argument around them.
  • Counting. "Twelve of eighteen participants mentioned…" imports a frequency logic the method does not use.

How quality is judged

The clearest markers: themes that are conceptually organised rather than topical; a stated theoretical position (is the analysis inductive or deductive, semantic or latent, essentialist or constructionist?); a reflexivity statement with content; extracts embedded in an argument; and a coding account that shows development rather than application.

Braun and Clarke's own reporting guidance is worth reading before submission — a great deal of reflexive TA is rejected or heavily revised for errors that are entirely avoidable, chiefly the mismatch between a claimed reflexive approach and reported reliability statistics.

Limitations

Reflexive TA is flexible, and flexibility cuts both ways: without the theoretical position stated, an analysis can be anything, and the method offers less structural protection against superficiality than framework or template analysis.

It does not preserve individual accounts. Cross-case patterning is the point, and if the research question is about how one person made sense of something, IPA or narrative analysis is the better fit.

It also does not build theory in the grounded-theory sense, and its rejection of reliability measures makes it a harder sell in fields where reviewers expect them — a real practical cost, whatever the epistemological merits.

Where software helps

Coding 18 interviews generously produces a few hundred codes that then have to be clustered, split, merged and re-clustered several times. Doing that by hand is possible and is mostly clerical; doing it in a tool leaves time for the part that matters.

The specific things worth having: retrieving all extracts for a candidate theme at once, so phase four can actually be done against the data rather than from memory; and keeping the coding history, so the development of a theme is recoverable when the write-up needs it. Evidano supports reflexive thematic analysis as a named methodology. Generating themes is interpretive work by definition — a tool that returns finished themes has done coding-reliability TA under a reflexive label, and the distinction is the whole point of the method.

Topics

  • reflexive thematic analysis
  • thematic analysis
  • Braun and Clarke
  • coding
  • qualitative analysis
  • reflexivity
  • semantic coding

Other methods in thematic analysis

Written guides are linked directly; the rest have a reference entry in the methodology directory.

Published research using these methods

Studies and evaluations where this family of method was applied with Evidano — the work, not the claim.

Keep reading

Browse all articles
Company
About
Newsletter

Product updates, research, and tips — straight to your inbox.

© Evidano, All Rights Reserved.