Focus Groups: using group dynamics as the instrument

Interaction as evidence. Participants question each other, demand justifications, and offer counter-examples with a bluntness moderators cannot ethically match. The classic methodological statement — Kitzinger’s Qualitative research: introducing focus groups — puts interaction at the definition’s centre and catalogues what it yields: shared knowledge, group norms, and the reasoning behind positions as they are challenged.

Naturalistic vocabulary. Groups drift toward the words people actually use with peers, which interviews’ formality suppresses — invaluable for message testing, instrument design, and understanding how a topic is talked about in the wild.

Norm visibility. What gets laughed at, corrected, hedged, or silenced in a group maps the social boundaries around the topic — data that exist only in the presence of an audience.

The corollary: sensitive-and-individual topics, decision reconstructions, and anything requiring depth per person belong in interviews. The group format censors as well as reveals, and the design choice is about which effect serves the question.

• Homogeneity within, comparison across. Groups talk most freely among perceived peers; segment by the lines that would silence people if mixed (hierarchy, stake, experience), and run separate groups per segment for comparison.
• Strangers versus existing groups: strangers reduce imported dynamics; existing groups (a ward team, a users’ committee) reveal real shared norms. Choose deliberately — both are legitimate, differently.
• Six to eight participants. Below five, the group loses interactive energy; above nine, it fragments into performance and silence. Over-recruit by two for no-shows.
• Three to five groups per segment is the working norm: patterns stable across groups are findings; anything from a single group is a hypothesis, because each group has its own emergent character.
• The standard design references are Krueger and Casey’s Focus Groups (SAGE) and Morgan’s Focus Groups as Qualitative Research (SAGE).

Focus group analysis has a unit-of-analysis discipline interviews lack: the group, not the individual, is the primary unit. Counting how many participants “said X” across groups misreads the data — utterances are produced in and shaped by a specific group’s dynamic.

Practically: code transcripts with the interaction preserved (who said it, in response to what, with what uptake); analyse within-group first (what did this group build, contest, silence?); then compare across groups and segments for patterns that survive the idiosyncrasy of any one room. Report interactional findings as such — “in all four customer groups, price justifications collapsed under peer challenge” is a different claim, and a stronger one, than “participants disliked the price”.

A local authority tested a school-meals reform with six focus groups: three of parents (segmented by school cluster), two of catering staff, one of secondary students. Mixed parent–staff groups were rejected at design — pilot conversations showed staff went silent around parents.

The parent groups produced what surveys had not. Individually polled (a pre-task), parents rated “healthier menus” top priority; in discussion, the group vocabulary told another story — “healthy” talk was quickly policed as performance (“we all tick that box, come on”), and the negotiated priority became “meals my kid will actually eat without the drama”. A ranking task did the analytic work: watching three groups independently demote the flagship menu option while building consensus around portion flexibility was replication in real time.

The interactional data also caught a norm the authority needed to know: cost concerns were nearly unsayable — raised twice, both times deflected with jokes, both times followed by private post-group disclosures to the moderator. The report flagged means-tested elements as needing individual-channel consultation, quoted the groups’ own vocabulary for the messaging work, and recommended the portion-flexibility redesign that the reform eventually led with.

• Serial interviewing. Moderator-to-participant Q&A around the table, interaction never invited — the format’s value discarded.
• Mixed-power rooms. Managers with staff, providers with patients: the silenced segment’s absence reads as agreement.
• Counting heads. Treating utterance frequency across groups as prevalence; groups are not samples of individuals.
• One group per segment. No way to distinguish the segment’s pattern from that room’s chemistry.
• Consensus as success. Moderating toward agreement; the disagreements were the data.
• Transcripts without speakers. Un-diarized transcripts erase interaction — the analysis can no longer see who moved whom.

Conformity pressure is constitutive, not incidental: groups over-produce the sayable and under-produce the stigmatised, and no moderation fully removes it. Pair with individual methods when the private view matters.

A few voices can carry a room, and group-level findings inherit that skew; the mitigation is moderation craft plus honest debrief notes about whose group it became.

And logistics constrain sampling: those who attend scheduled group sessions differ from those who cannot. Online groups widen reach and change the dynamics — flatter, easier exit, thinner interaction — a trade to make knowingly rather than discover afterwards.

In-depth interviews provide the individual depth and privacy groups structurally cannot.

Most significant change uses structured group deliberation for evaluation, where the selection debate is the data.

Voice-of-customer interviews capture individual experience for product teams before group testing of concepts.

Photovoice runs facilitated group analysis over participant-produced images rather than a discussion guide.

Group data’s analytic tax is the transcript: overlapping speech, eight voices, and findings that depend on knowing who said what to whom. Evidano’s transcription handles multi-speaker diarization, and its coding keeps every theme tied to speaker-attributed quotes — so within-group analysis (who moved, after which argument) and cross-group comparison become retrievable rather than reconstructed, and the groups run in other languages analyse in the same workspace.

Moderation and the group-level interpretive judgement — whose room it was, what the silences meant — remain the researcher’s; the tool ensures the interaction the method exists for survives into the analysis.