Reflexive Thematic Analysis: the Braun and Clarke approach

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.

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.

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.

• 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.

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.

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.

Semantic and latent coding are the two levels reflexive TA works at, and most analyses use both.

Template analysis and framework analysis are the codebook approaches — structured, revisable frameworks, better suited to teams and to applied deadlines.

Constructivist grounded theory shares the interpretivist commitment and aims at theory rather than at patterned meaning.

IPA is the alternative where individual sense-making, not cross-case pattern, is the question.

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.