Thematic analysis, step by step

What is thematic analysis, and which version are you using?

With thematic analysis you look for patterns of meaning across your dataset, put them in order and read what they mean. People like it because it is flexible and not tied to any one theory, and you can use it on interviews, focus groups, open survey answers, diaries or documents. But the same flexibility is why so many times it gets done badly.

First understand that “thematic analysis” is a name for more than one method. Braun and Clarke have written several papers on this. Many studies, they found, cite the 2006 paper and then do something quite different from it. In a 2021 paper in Qualitative Research in Psychology, titled “One size fits all?”, they described three broad families of TA.

Pick the one you are doing and say it clearly in your methodology chapter, and after that don’t mix it with the other two. A common case is a thesis that cites Braun and Clarke and then gives a kappa score to show the themes are “reliable”. An examiner who knows this literature will catch it. Reliability in quantitative studies is a separate matter, and our reliability and validity guide takes it up.

Choices to make before coding

• Inductive or deductive. Do the codes come mainly from the data itself, or from a theory or framework you already have? Mostly it is a mix, and you tell which side leads.
• Semantic or latent. Semantic is coding the words people openly said, and latent is going under them to the assumptions and ideas.
• Philosophical position. Broadly it is realist, where the data tells you about the participants’ reality, or constructionist, where the data shows how meaning gets constructed. Whichever you take, it changes how you word your claims.

What are the six phases of thematic analysis?

Braun and Clarke themselves call these phases a guide only, so you need not go one, two, three in order. You will keep going back and forth, mostly between phases 3, 4 and 5. The names changed a little between the 2006 paper and the 2022 book. That was on purpose, to stress that the analyst builds the themes and does not simply find them.

Phase 1: Familiarisation

If you can manage, transcribe your interviews yourself. It is slow, but nothing brings you closer to your data. When an agency or auto tool has typed the transcript, check every one against the recording. Read each transcript at least twice. Note down ideas and surprises while reading, and how you yourself reacted, because these notes start your reflexive record. For planning and recording the interviews, see our semi-structured interviews guide.

Many scholars are doing their interviews in Tamil, Hindi or another language. Decide this early. One way is to analyse in the original language itself and translate only the extracts you quote. Other way is to translate everything first. Mostly meaning is safer in the original, and whichever way you pick, mention it in your methods.

Phase 2: Coding

Now go through the full dataset and label every segment which is relevant to your research question. That label is your code. Write what is interesting in that bit, say “feeling judged by senior nurses”, not a vague “emotions”. Some relevant bits won’t match what you expected, still code them. Codes will change as you go. Pass through the data at least twice.

The main commercial tools are NVivo, ATLAS.ti and MAXQDA. Taguette and QualCoder cost nothing. For a small dataset even Word comments or a spreadsheet will do. Whichever software you use, it only help you keep the coding organised, and the analysis you have to do yourself.

Phase 3: Generating initial themes

Put together the codes that seem to share one central idea as candidate themes. When Braun and Clarke say theme, they mean a pattern of shared meaning held together by one central concept. It is not a box for everything people said on a topic. Make a thematic map, because it shows how the themes connect.

Phase 4: Developing and reviewing themes

Check each candidate theme at two levels. First against its own coded extracts. Do they really fit together? Then against the full dataset. See if the theme reflects the data, and if something important is left outside all your themes. Merge, split or drop as needed. In many good analyses the themes change a lot here.

Phase 5: Refining, defining and naming themes

Write a few sentences for every theme. Say what it is about, what it is not about, and how it connects with the other themes. If you cannot state its central idea in two sentences, most probably it is not a theme yet. “Proving I belong” as a name tells the reader something, and a flat “Identity” tells almost nothing. You can have sub-themes, but keep only a few.

Phase 6: Writing up

In reflexive TA the writing is part of the analysis itself, not a report attached at the end. Write each theme as an argument. Say what it is, show it with extracts, and give your own reading around each extract. Extracts back up the analysis but cannot replace it. Then connect the themes to your research question and to earlier studies. How to do that linking is there in our discussion guide.

What is the difference between a theme and a topic summary?

In PhD thematic analyses this is the most common problem, and Braun and Clarke have written on it many times. A topic summary gathers everything participants said on one subject, mostly one of the interview questions, like “Challenges faced”, “Support received” or “Suggestions for improvement”. There is summary there but no interpretation.

In reflexive TA, a theme has a central organising idea, and that idea says something about the data. See the two versions below.

• Topic summary: “Challenges faced by first-generation PhD scholars”, and after it a list of financial, academic and family challenges.
• Theme: “Carrying the family’s investment”, which captures how scholars spoke of financial, academic and family pressures as one single feeling of debt to parents who had sacrificed for their education.

The second one tells the reader something they could not guess from the interview schedule. A quick check is to put your theme names next to your interview questions. If the two lists match, you most likely have topic summaries.

What mistakes do examiners look for?

Braun and Clarke went through published researches using TA and listed the common problems in their 2021 paper “One size fits all?”. Below are the ones we see most in Indian theses.

• Themes that “emerged”. A theme does not come up by itself. You are the one who develops it. Writing “emerged” makes it sound like themes came out on their own while you waited, and reflexive TA isn’t like that.
• Citing Braun and Clarke for a coding reliability procedure. Kappa scores, many coders reaching “agreement” and a fixed codebook belong to other approaches.
• Claiming saturation. Braun and Clarke published “To saturate or not to saturate?” in 2021 in Qualitative Research in Sport, Exercise and Health. There they argue that saturation does not suit reflexive TA. The analyst generates meaning, and meaning is not some fixed stock in the data waiting to get used up. Justify your sample size by the aims, the richness of the data and practical limits.
• Too many themes. When a chapter has ten themes and four sub-themes under each, usually it is really a list of codes. Most good analyses end up with two to six main themes.
• Extracts with no analysis. Long quotations joined by “as one participant said” are description, and the analysis is missing.
• No reflexivity. In reflexive TA your own perspective is shaping the analysis all the time. So describe your position, for example a teacher studying teachers or an insider in the community, and explain how you reflected on it.

• The methodology chapter names your version of TA (reflexive, codebook or coding reliability) and the paper or book you followed.
• You have stated your orientation. Inductive or deductive, semantic or latent, and also your philosophical position.
• Each theme has its own central organising idea, and is more than a topic heading like “Challenges” or “Benefits”.
• You write that the themes were developed or constructed by you. You don’t write that they “emerged”.
• Each theme is shown with extracts from several participants, with your analytic comments around each extract.
• The number of themes is manageable, mostly two to six main themes for one study or chapter.
• Your sample size rests on the study’s aims and the richness of data, not only on a saturation claim.
• You have described your own position as a researcher and how you reflected on it.

The 2006 paper also gives a 15-point checklist for a good thematic analysis, and some examiners do use it. For describing each phase in a methods section, Byrne’s 2022 worked example in Quality & Quantity is a helpful model to follow.

How do you write thematic analysis in a PhD thesis?

In the methodology chapter

Name your approach and cite the version you followed. For reflexive TA, that is Braun and Clarke’s 2006 paper and their later work also. Mostly scholars cite the 2019 paper “Reflecting on reflexive thematic analysis” and the 2022 book. After that describe each phase as you actually did it in your study, not a copied textbook summary. Mention your orientation, inductive or deductive and semantic or latent, along with your position as a researcher. Give the details of transcription and translation also, and name the software.

In the findings chapter

Open with an overview of themes and sub-themes, as a table or thematic map. Then give every theme its own section. There define the central idea first, then bring extracts from several participants, each identified by a pseudonym or a code, and interpret every extract. Some examiners like to see the count of participants for each theme. You can give these counts. But explain that in reflexive TA a theme does not become important just because it is frequent.

Using AI tools for coding

Many AI tools are now marketed to generate codes or themes from transcripts, so be careful. Reflexive TA stands on the researcher’s own interpretation, and giving that work to a tool weakens the method. Confidentiality is one more issue, because putting their transcripts on a public AI service can break what your consent form and ethics approval promised. Read our guide to AI tools in a PhD before you try any of them.