Thematic analysis in practice

The gap between “I read my interviews and these topics came up” and a defensible thematic analysis is method. An examiner needs to see how you got from data to theme, and to be able to trace any claim back to what a participant actually said.

Decide your approach before you code

Inductive analysis derives themes from the data. Deductive analysis codes against a framework established in advance. Many projects are reasonably hybrid — an initial framework from the literature, extended as the data demands — but you have to state which you are doing, because it changes what the analysis can claim.

The same applies to your position on the data. A realist reading treats accounts as reports of experience; a constructionist reading treats them as constructed in the interview itself. Both are legitimate; drifting between them without noticing is not.

The phases, and what actually happens in each

1. Familiarisation — read the whole dataset before coding anything, taking notes rather than codes
2. Initial coding — label segments systematically across every transcript, not just the interesting ones
3. Searching for themes — group codes into candidate themes that say something, not just name a topic
4. Reviewing themes — test each against the coded extracts and against the full dataset
5. Defining and naming — write a definition per theme, including what it excludes
6. Writing up — build an argument, using extracts as evidence rather than illustration

Coding across the whole set

The most common shortcut is coding the first three transcripts thoroughly and skimming the rest for confirmation. It produces themes that reflect your earliest impressions and it is visible in the write-up, because extracts cluster around a few participants. Code the whole set at the same level of attention, and revisit the early transcripts once the coding framework has stabilised.

Using extracts properly

Each extract should earn its place by evidencing an analytic claim. A findings chapter that runs quote, brief paraphrase, quote, brief paraphrase is a transcript with formatting. The pattern that works is: state the analytic claim, present the extract that supports it, then explain what in the extract does the supporting work.

Include disconfirming cases. A theme that holds for eleven participants and not for the twelfth is more credible when you say so and discuss why, than when the twelfth is quietly omitted.

Demonstrating trustworthiness

• An audit trail: coding framework, coded data, and the decisions taken between them
• Reflexivity: how your position shaped access, questions and interpretation
• Where the design supports it, a second coder on a subset with disagreements discussed
• Enough contextual detail for a reader to judge transferability to their setting

None of this requires software, though NVivo, ATLAS.ti and MAXQDA make the audit trail easier to produce. A documented manual process is perfectly defensible; an undocumented one is not, whichever tool produced it.