How to analyse focus group data without losing the group

Scope: focus group transcripts with speaker labels, in applied and academic research. All example data are invented.

Short answer. Analyse each group on its own before you compare groups. Decide whether a claim is about an individual speaker or about the group. For claims about how widely a view is held, count speakers or groups, not turns. A short "Mhm. That's exactly it." shows alignment with an earlier turn, but it is not a separate account. Exchanges such as challenges and corrections are data in their own right, and silence is not consent. Which analytic method you use (thematic analysis, template analysis, content analysis, the Framework Method or a rapid approach) depends on your question; these group-level rules apply to all of them.

Most focus group guides stop at "transcribe, code, find themes". That is the part focus groups share with interviews. This guide covers what is different: a focus group transcript contains individual statements and an interaction between people who react to each other (Kitzinger, 1994; Kidd & Parshall, 2000). For a full practical treatment of the analysis process, Krueger and Casey (2015, ch. 6) remain a standard reference.

What makes focus group data different

Sim (1998) draws a practical conclusion: across groups you can infer whether a view or issue was present, but not how strongly it was held. Comparisons should therefore be about presence, variation and conditions, not about intensity.

Before you start: prepare the transcripts

• Label every speaker consistently within each group (N1, N2…), and keep the moderator separate.
• Number the turns (A-13 = group A, turn 13) so that every claim can point to its source.
• Keep groups as separate sources. Do not merge transcripts into one file.
• Record case details: who was in which group and in what role, without identifying people.
• Keep the guide apart from the data. The guide may contain the client's wording or results from an earlier group. Nothing that appears only in the guide should be attributed to participants.

Step by step

1. Read each group as a whole

Before coding, read each transcript from start to finish. Note how the discussion went: which episodes it fell into, where it shifted, who opened which topics, and which passages need reading in the original. Do this for one group at a time. Comparing too early lets the larger or louder group set the frame for the others.

2. Decide the unit of analysis and the level of counting

Decide, and write down, when you are making a claim about individuals and when about a group (Kidd & Parshall, 2000). Then match the counting level to the claim:

Turn-level detail can be analysed in its own right. Onwuegbuzie et al. (2009) describe a "micro-interlocutor" analysis of who responds to each question, in what order and how. What turn counts cannot do is stand in for how many participants hold a view. Define the intended unit and claim level before analysis, document any justified revisions, and do not switch levels opportunistically to strengthen a result. "Most nurses" must mean most nurse speakers, not most turns.

3. Code or summarise within each group

Apply your chosen method inside each group first. Include the turn before the passage, often the moderator's question, as context: without it you cannot tell what a participant is answering.

4. Handle agreement turns

Short agreement turns ("Mhm", "Exactly", "Same here") do not provide a separate substantive account, but they are interactional evidence of expressed alignment. Record each one with the turn it responds to, and do not count it as a new independent statement. In your counts, show agreeing speakers separately from speakers who made the point themselves. Without a rule, agreement either disappears or inflates the frequency of whatever it follows.

5. Record interaction episodes

Some meaning lives in a sequence: a claim, a challenge, a correction, a shift of position. Mark the episode as a range of turns (for example B-31 to B-33) and write a memo on what it shows. Interaction is analytic material (Kitzinger, 1994), and Duggleby (2005) discusses how interaction data can be analysed and reported. You may still need to code the individual turns, but turn-by-turn coding on its own loses the episode.

6. Note dominance, silence and hierarchy

• Report whether a finding is shared or carried by one strong voice.
• Silence is not agreement. If nobody objected, that does not mean everybody agreed (Sim, 1998).
• An empty cell for a participant who barely spoke says nothing about what they think.
• Where the commissioning organisation manages the participants, consider the risk that criticism is muted. Separating groups by hierarchy is a common design choice. Mixed groups can be the right choice when interaction across positions is what you want to study.

7. Distinguish prompted from spontaneous

Mark whether a topic came up in answer to the moderator's question or was raised by participants. Both are data, but they support different statements: "when asked, nurses described…" is not the same as "nurses raised…".

8. Compare groups, then write

Compare the groups only after each one has been read and analysed separately. A difference between groups can be part of a finding, or a finding of its own. Disagreement within a group can matter as much as differences between groups. When you write, say who carries each finding, and keep speakers unidentifiable.

Worked example

Invented example. Rovina Regional Hospital (fictional) pilots a new rota on three wards: twelve-hour shifts and a shared team of nurses who move between wards as needed. A moderator runs two focus groups with the same topic guide: FG-A with eight ward nurses (N1–N8) and FG-B with six ward managers (W1–W6).

[A-15] N1: Depends which ward you're on. On ours the swaps are kept track of;
           on internal medicine they aren't.
[A-16] N3: Kept track of… covered up, more like. Someone from the shared team
           comes in, doesn't know the patients, and I have to show her
           everything anyway. So I'm doing two jobs.
[A-17] N6: Mhm. That's exactly it.

[B-31] W2: The nurses complain, but overtime has gone down. That's a fact.
[B-32] W4: It's gone down on the spreadsheet. When they swap a shift between
           themselves, nobody records it as overtime.
[B-33] W2: But that's not the rota's fault.

Counting the claim "the shared team creates extra work for nurses":

Also look at A-15 to A-16. N1 says swaps are "kept track of" on her ward, and N3 picks up the phrase and recasts it as "covered up". That is a challenge or a reframing, and possibly a disagreement. Check the wider transcript to see whether the two nurses are talking about the same practice on the same ward.

Report wording that matches the evidence: "One nurse described having to show shared-team colleagues everything as 'doing two jobs' (N3, A-16), and another agreed (N6, A-17). Ward managers did not discuss this in the excerpt."

Wording that overstates it: "Nurses widely reported that the shared team doubles their workload."

The episode B-31 to B-33. Coded turn by turn, W2's first turn reads "overtime has gone down". Read as an episode, the three turns show managers disputing what the overtime figures mean and who is responsible for them. That is a different analytic object, and it shows that the managers' group does not hold a single view either. A memo on the range B-31 to B-33 keeps this visible, and the report can quote it.

Checklist before a finding goes into the report

• Does it say which group or groups it comes from?
• If it is a claim about how widely a view is held, is it counted in speakers or groups rather than turns, at the level you defined?
• Are agreeing speakers shown separately from those who made the point?
• Does it mention any disagreement within the group, or a dominant voice?
• Was the topic prompted by the moderator or raised by participants, and does the wording reflect that?
• Does every quotation have a speaker label and turn number, and does it match the transcript word for word?
• Could a reader identify a participant from the wording? If so, change it.

Which analysis method for focus group data?

The steps above sit underneath whatever method you choose. The choice of method depends on your question:

In thematic analysis of group data, decide early whether you are interested in the content of what people say or also in how the group builds meaning together, and record that decision. Themes are not decided by vote. Describe who carries a theme and whether it is shared.

When this guide is not enough

• Interaction is the research object. If you want to study how turns are organised, how agreement is produced or how identities are performed, use conversation or discourse analysis. These need detailed transcription (pauses, overlaps, intonation), which an ordinary transcript does not have. SurveysAI Qualitative Analysis Workflow does not offer them.
• You need to generalise to a population. Theoretical generalisation from focus group data is more feasible than empirical generalisation, which is at best provisional (Sim, 1998).
• There is only one group. You can still analyse it, but you cannot compare groups, and any claim about "nurses" is a claim about this group.

Analysing focus groups in the app

SurveysAI Qualitative Analysis Workflow works with transcripts that have speaker labels. Two of its analyses fit most focus group projects:

• Template analysis for applied briefs, which follows the template-analysis logic, suits a client brief with specific questions where you also want room for the unexpected. It produces a draft analysis question by question, first within each group and then across groups.
• Rapid qualitative analysis suits projects that need findings soon. It summarises each group by domain, with positions, speakers, quotations and contradictions, and compares the groups in a group × domain matrix.

In both, program code checks quotations word for word against the transcripts. An AI critic reads key outputs; it is the same model that wrote them, not a human and not a second model, so this is not independent validation. Check agreement turns and interaction episodes yourself: the report does not replace that reading. The app produces draft analytic outputs, and you remain responsible for reviewing them against your transcripts and for the interpretation.

• Create an account for a template analysis. If the analysis is not preselected, choose Template analysis for applied briefs in the start dialog.
• Create an account for a rapid analysis. If the analysis is not preselected, choose Rapid qualitative analysis in the start dialog.

References

• Duggleby, W. (2005). What about focus group interaction data? Qualitative Health Research, 15(6), 832–840. https://doi.org/10.1177/1049732304273916
• Kidd, P. S., & Parshall, M. B. (2000). Getting the focus and the group: Enhancing analytical rigor in focus group research. Qualitative Health Research, 10(3), 293–308. https://doi.org/10.1177/104973200129118453
• Kitzinger, J. (1994). The methodology of focus groups: The importance of interaction between research participants. Sociology of Health & Illness, 16(1), 103–121. https://doi.org/10.1111/1467-9566.ep11347023
• Krueger, R. A., & Casey, M. A. (2015). Focus groups: A practical guide for applied research (5th ed.). SAGE.
• Morgan, D. L. (1997). Focus groups as qualitative research (2nd ed.). SAGE. https://doi.org/10.4135/9781412984287
• Onwuegbuzie, A. J., Dickinson, W. B., Leech, N. L., & Zoran, A. G. (2009). A qualitative framework for collecting and analyzing data in focus group research. International Journal of Qualitative Methods, 8(3), 1–21. https://doi.org/10.1177/160940690900800301
• Sim, J. (1998). Collecting and analysing qualitative data: Issues raised by the focus group. Journal of Advanced Nursing, 28(2), 345–352. https://doi.org/10.1046/j.1365-2648.1998.00692.x