Thematic analysis in NVivo without losing the audit trail

The criticism qualitative analysis attracts most often is that it is unsystematic — themes appear in the findings with no visible route from the transcripts.

Software does not fix that by itself. What it does is make the route recordable, so long as you use it deliberately.

Name the approach, and follow it

"Thematic analysis" covers several distinct methods with different assumptions. Cite the one you are using and apply it consistently.

• Braun and Clarke reflexive TA — themes are developed by the researcher, and subjectivity is treated as a resource. Inter-coder reliability is explicitly not the goal.
• Codebook or framework analysis — a structured matrix, well suited to applied and policy research, and to team coding.
• Content analysis — counts and categorises, closer to quantitative logic.
• Grounded theory — open, axial and selective coding building toward theory, with theoretical sampling.

Prepare transcripts before importing

1. Transcribe consistently — decide in advance how you handle pauses, fillers and overlapping speech.
2. Anonymise at this stage. Replace names with consistent pseudonyms and keep the key separately and securely.
3. Use speaker names as headings so NVivo can auto-code by speaker.
4. Apply Heading styles to structure, so auto-coding by section works.
5. Import as .docx rather than PDF; formatting survives and coding is far easier.

Coding: codes first, themes later

The most common structural mistake is creating theme-level nodes at the start and filing quotes into them. That is not analysis — it is sorting into pre-existing boxes.

Code closely to the data first. Themes are constructed afterwards from patterns across codes.

• Make codes descriptive and close to the data — "worries about being judged", not "anxiety".
• Code generously in the first pass. Merging is easy; finding uncoded material later is not.
• Write a definition for every code in its Description field, including what does not belong in it.
• Use memos constantly. Your reasoning at the moment you made a decision is what forms the audit trail.
• Code the same extract to multiple codes where it genuinely speaks to several things.

From codes to themes

1. Review your full code list once initial coding is finished.
2. Group codes that share a meaning into candidate themes.
3. Check each candidate against the coded extracts — does it hold together and is it distinct?
4. Use NVivo's Matrix Coding Query to see how themes distribute across participant groups.
5. Name themes so they say something. "Trust" is a topic; "Trust is earned through visible competence, not seniority" is a theme.
6. Build a thematic map showing how themes and sub-themes relate.

The audit trail you will be asked for

• Code definitions with inclusion and exclusion criteria.
• Dated memos recording analytical decisions as you made them.
• A record of how the codebook changed, and why.
• Sample coded extracts for each theme, attributed to participants by pseudonym.
• Where relevant, an inter-coder agreement statistic — but only if your chosen method treats it as meaningful.

Questions this raises