Thematic analysis is a qualitative method for identifying, analysing, and reporting patterns of meaning (“themes”) across a dataset such as interview transcripts, focus-group notes, or open-ended survey responses. In short: you read your data closely, attach short labels (codes) to meaningful segments, cluster related codes into candidate themes, then refine and name those themes to tell a coherent, evidence-based story that answers your research question. It is flexible, accessible to first-time qualitative researchers, and compatible with most epistemologies, which is why it is one of the most widely used approaches in dissertations across psychology, education, business, health, and the social sciences.
Use thematic analysis when your data are textual or transcribed, your aim is to understand experiences, perceptions, or meanings rather than to count and measure them, and your research question is exploratory (for example, “How do final-year nursing students experience their first clinical placement?”). The canonical framework is the six-phase approach set out by Virginia Braun and Victoria Clarke in their landmark 2006 paper, which this guide follows in detail below.
What is thematic analysis?
Thematic analysis (TA) is a method, not a methodology tied to a single theory. That distinction matters: because it is theoretically flexible, you can apply TA within a realist framework (treating accounts as a fairly direct window onto experience) or a constructionist one (treating accounts as shaped by language and context). What stays constant is the goal — identifying patterned meaning across a dataset and reporting it rigorously. A theme captures something important about the data in relation to your research question and represents a level of patterned response or meaning. Crucially, a theme is not simply a topic or a domain summary; it is an analytic claim with a central organising concept, supported by coded data extracts.
TA is a form of qualitative analysis applied to the kind of data produced by interviews, focus groups, diaries, and open-ended questionnaire items. If you are still designing the instruments that will generate that data, our guides on methods of data collection and the qualitative research questionnaire will help you produce transcripts rich enough to analyse thematically.
When should you use thematic analysis?
Thematic analysis is a strong choice when:
- Your research question is exploratory and concerns experiences, views, perceptions, or meanings (“What”, “How”, and “Why” questions).
- Your data are textual or transcribed — interviews, focus groups, free-text survey answers, reflective journals, or documents.
- You want a method that is accessible and does not require you to commit to the full theoretical machinery of grounded theory, IPA, or discourse analysis.
- You need flexibility to work inductively (themes from the data) or deductively (themes shaped by existing theory), and at a semantic or a latent level.
It is a weaker choice when you want to test causal hypotheses, quantify the prevalence of categories with statistical rigour, or analyse the fine mechanics of language use — for the last of these, a more linguistically focused approach is appropriate. If you mainly want to count the frequency of predefined categories, you are closer to content analysis than thematic analysis (see the comparison below).
Key choices before you start: inductive vs deductive, semantic vs latent
Before coding a single transcript, make two explicit decisions. First, will your themes be driven by the data (inductive) or by an existing theory or coding frame (deductive)? Second, will you analyse what participants explicitly say (semantic themes) or interpret the underlying assumptions and ideas beneath the surface (latent themes)? These are not strict either/or boxes — most real analyses sit somewhere on a continuum — but stating your position makes your analysis transparent and defensible.
| Dimension | Option | What it means | When to choose it |
|---|---|---|---|
| Coding orientation | Inductive (“bottom-up”) | Codes and themes are derived from the data itself, without trying to fit them into a pre-existing frame. | Under-researched topics; you want participants’ own framing to lead. |
| Coding orientation | Deductive (“top-down”) | Coding is guided by existing theory, prior research, or a predefined codebook. | You are testing or extending an established framework or comparing against theory. |
| Level of meaning | Semantic | Themes describe the explicit, surface meaning of what participants said. | You want to summarise and organise the explicit content. |
| Level of meaning | Latent | Themes interpret underlying ideas, assumptions, and ideologies beneath the surface. | You are working interpretively/constructionist and want analytic depth. |
A psychology dissertation exploring an unfamiliar coping strategy might run inductive + semantic to stay close to participants’ words; a sociology study examining how gendered assumptions shape workplace talk might run deductive + latent, reading beneath the surface through a theoretical lens.
The six phases of thematic analysis (Braun & Clarke, 2006)
Braun and Clarke describe TA as a recursive — not linear — process. You move back and forth between phases rather than marching through them once. The six phases below are the spine of almost every credible TA dissertation, and examiners expect to see them evidenced in your methods chapter.
“Thematic analysis is a method for identifying, analysing, and reporting patterns (themes) within data.” (Source: Braun & Clarke, 2006)
Phase 1 — Familiarising yourself with the data
What you do: Transcribe your recordings (if you have not already), then read and re-read the entire dataset actively, noting initial ideas. Immersion is the point — you cannot identify patterns in data you do not know intimately. Keep a notebook or memo file of first impressions; these are not codes yet, just analytic hunches.
Mini example: While transcribing interviews with new graduates about remote onboarding, you repeatedly notice people mentioning “not knowing who to ask” — you jot it down as an early observation.
Phase 2 — Generating initial codes
What you do: Work systematically through the whole dataset, attaching concise codes to features of the data that are interesting and relevant to your question. A code is a short label (a word or short phrase) summarising the essence of a data segment. Code inclusively, code for as many potential patterns as you can, and keep a little surrounding context with each extract. You can code by hand, in a spreadsheet, or in software such as NVivo or ATLAS.ti.
Mini example: The sentence “I didn’t want to bother my manager with what felt like a stupid question” might be coded as fear of appearing incompetent and reluctance to ask for help.
Phase 3 — Searching for themes
What you do: Shift from codes to themes. Sort your many codes into potential themes and collate the relevant coded extracts under each. A useful technique is a thematic map or even physical cards — grouping codes that seem to share a central idea, and identifying relationships between themes and sub-themes. Some codes will form main themes, some sub-themes, and some may be set aside as a “miscellaneous” pile for now.
Mini example: Codes such as reluctance to ask for help, fear of appearing incompetent, and no informal channels cluster into a candidate theme: “The high social cost of asking questions remotely.”
Phase 4 — Reviewing themes
What you do: Refine your candidate themes against two levels. Level one: re-read the collated extracts for each theme — do they form a coherent pattern? Level two: re-read the entire dataset — do the themes accurately reflect the data as a whole, and have you missed anything? Themes that lack enough supporting data collapse; themes that are too broad split; some merge. By the end, the boundaries of each theme should be clear.
Mini example: You realise “The high social cost of asking questions” overlaps heavily with a separate “isolation” theme, so you merge and rework them into a sharper theme about the erosion of informal learning channels.
Phase 5 — Defining and naming themes
What you do: For each theme, write a short definition (a couple of sentences) capturing its central organising concept — what is unique and central about it, and the story it tells in relation to your question and the other themes. Give each theme a concise, informative, ideally engaging name. If you cannot define a theme in a sentence or two, it probably is not yet a coherent theme.
Mini example: You name a theme “Learning by osmosis — and its quiet collapse online,” defined as the loss of the incidental, overheard learning that physical offices provide and that remote onboarding fails to replace.
Phase 6 — Producing the report
What you do: Write up the analysis. Select vivid, compelling extract examples, relate the analysis back to your research question and the literature, and build an argument — do not just describe. Good TA writing weaves data extracts and your interpretation together so the reader can see why your themes are warranted, not merely that you have them.
Mini example: Under each theme you present two or three quotations, interpret them, and connect the pattern to existing onboarding and organisational-socialisation literature.
A worked example: from codes to a theme
The clearest way to understand TA is to watch a short interview extract turn into coded segments and then into a theme.
“Honestly, every meal feels like a maths exam now. I check the labels, I count the carbs, and even then I’m never sure I’ve got it right — so I just feel guilty most of the time, like I’m failing at something I should be able to do.”
Coding (Phase 2):
• “every meal feels like a maths exam” → cognitive burden of monitoring
• “never sure I’ve got it right” → uncertainty about correct self-management
• “I just feel guilty… like I’m failing” → self-blame / moralised illness
Clustering (Phase 3–5): Across the dataset, cognitive burden, uncertainty, and self-blame recur together. They are clustered, reviewed, and defined into one latent theme: “Managing diabetes as a moral test, not just a medical task” — capturing how participants frame everyday self-care as a question of personal virtue and failure, not only clinical behaviour. Notice the theme is an interpretive claim, not a topic label like “diet.”
Reflexive TA vs codebook and coding-reliability approaches
Braun and Clarke have since clarified that “thematic analysis” is really a family of methods, and they distinguish three broad types. Knowing where your approach sits prevents you from mixing incompatible procedures — a common examiner criticism.
- Reflexive TA — the approach their 2006 paper grew into. Coding is organic and flexible, themes are actively generated by the researcher as analytic outputs (not “found”), and the researcher’s subjectivity is treated as a resource. There is no codebook fixed in advance and no expectation of multiple coders agreeing.
- Codebook TA — uses a structured coding frame (often developed partly in advance) to map the data, blending some structure with qualitative interpretation. Useful in applied or team research with predefined questions.
- Coding-reliability TA — emphasises a fixed codebook and inter-rater reliability (e.g., multiple coders, agreement statistics), reflecting a more (post)positivist stance.
An important practical point: in reflexive TA, seeking inter-coder “reliability” is not a virtue and can even be conceptually inconsistent — themes are interpretations, not objective facts to be agreed upon. So decide your variant first, then justify your quality criteria accordingly rather than importing reliability checks that belong to a different tradition.
Thematic analysis vs content analysis
Students frequently confuse these two. Both work with textual data and both involve coding, but their logic differs. Content analysis is often (though not always) more quantitative: it tends to count the frequency of predefined categories and can report numerical prevalence. Thematic analysis is qualitative throughout: it seeks patterns of meaning and builds interpretive themes, and prevalence (“how many participants”) is not what makes something a theme. As a rule of thumb, if your output is mainly a frequency table, you are doing content analysis; if it is a set of argued, interpreted themes illustrated with extracts, you are doing TA. Our dedicated guide to content analysis sets out that method in full.
Common mistakes (and how to avoid them)
- Themes that are really topic summaries. “Barriers,” “Benefits,” or “Diet” are domain labels, not themes. A genuine theme has a central organising concept and makes a point. Rename “Barriers” into the specific claim your data actually support.
- Too many themes (or too many sub-themes). A typical undergraduate or master’s study reports roughly three to six well-developed themes. Fifteen thin themes signal under-analysis — you have organised the data but not interpreted it.
- Using interview questions as themes. If your themes mirror your topic guide one-to-one, you have summarised the interview structure, not analysed the data.
- Thin evidence. Each theme needs enough coded extracts from across (ideally) multiple participants. One striking quote is an anecdote, not a theme.
- Anecdotalism — cherry-picking. Choosing only quotes that fit your story, while ignoring disconfirming data, undermines trustworthiness. Actively look for and report variation.
- Mismatched quality criteria. Applying inter-rater reliability to a reflexive TA, or claiming themes “emerged” (which implies they were passively found rather than actively constructed).
How to do thematic analysis well: a checklist
- State your position upfront: which TA variant (reflexive / codebook / coding-reliability), inductive or deductive, semantic or latent.
- Transcribe carefully and immerse yourself before coding.
- Code the whole dataset systematically and keep an audit trail of how codes became themes.
- Build, review, and refine a thematic map; check themes against both the extracts and the full dataset.
- Define and name each theme with a one- or two-sentence central organising concept.
- Write an analytic report — interpret extracts, link to literature, and address disconfirming cases.
- Keep reflexive notes on how your perspective shaped the analysis.
Struggling to turn transcripts into credible themes?
Our qualified academics can help you design, analyse, and write up a methodologically sound thematic analysis chapter.
In summary
Thematic analysis is a flexible, rigorous way to make sense of qualitative data by building interpreted themes from systematic coding. Anchor your work in Braun and Clarke’s six recursive phases, decide your variant and your inductive/deductive and semantic/latent positions before you start, support every theme with evidence, and treat themes as analytic claims rather than topic headings. Do that, and your analysis will be both transparent and convincing — exactly what examiners reward.
Related methodology guides
- Qualitative Data Analysis
- Grounded Theory
- Narrative Analysis
