Focus groups and group elicitation: interaction as the unit of analysis

A focus group is easy to commission for the wrong reason. Five people in one room look like five interviews for the price of one session. Morgan (1997) rejects that reading: the group, not the individual, is the unit of analysis. If that claim holds, then conformity, polarisation and the losses a group inflicts on its own output are not noise to be averaged away. They are the result.

Kitzinger (1994) names the feature that separates it from a one-to-one interview: interaction between participants, to be explored rather than minimised. This eighth of the ten units asks what changes when the account is produced by a room rather than by one person.

An interview with five people is not a focus group

Morgan (1997) sets out focus groups as a qualitative method with its own logic. He covers when it is appropriate, how to design one, how to run a discussion and how to analyse it. The organising decision is the unit of analysis. A study that wants individual opinions should interview individuals. A study that wants to know how a group arrives at a position has a reason to convene one.

Kitzinger (1995) covers composition, conduct and analysis, and makes an access argument that is easy to overlook. Focus groups do not discriminate against people who cannot read or write, and they can draw in people reluctant to be interviewed alone. That strength is the hinge of the contested question.

Wilkinson (1998) identifies three traditions with different views of what group interaction is for. She isolates three central features: access to participants' own language and concerns, more fully articulated accounts, and the observation of collective sense-making in action. Barbour (2005) finds the same confusion in practice. Focus groups in medical education are often chosen for convenience and then read as though they produced individual opinions.

What the group produces that the individual cannot

Collective sense-making has a structure. Hollander (2004) argues that discussions are shaped by four overlapping contexts: associational, status (especially gender), conversational and relational. He uses data on the effects of violence on everyday life. These produce both problematic silences, where a topic is not disclosed, and problematic speech, where a comment is strategically shaped. A disclosure that appears in one group and not another may reflect the status composition of that room rather than a difference in experience. The processes that limit focus groups for understanding individual thoughts make them an excellent site for analysing social interaction.

Smithson (2000) lists the interactive features an analysis has to face: domination by some participants, the construction of the Other, normative discourse, and conflict. She argues that the analysis of interaction is not optional. A group that falls silent, or agrees too quickly, has told the researcher something an interview would not.

The room is now often a screen. Chai and colleagues (2024) systematically reviewed 45 studies comparing face-to-face with online focus groups. Word counts were higher face to face, and 62.5 per cent of the studies reported more detailed responses there, while the number of ideas expressed did not differ significantly. Where the review asked who spoke, online participation was more evenly distributed and face-to-face groups were often dominated by a few speakers. That levelling assumes each participant has a private, connected place to join from, and the gap in who speaks returns where the connection is shared or metered. The medium is a design variable whose interactional signature differs, not a fallback.

Discussion does not average opinion, and a group is not a multiplier of ideas

The clearest evidence comes from outside the method literature. Moscovici and Zavalloni (1969) compared individual and collective ratings of 140 male secondary school students, aged eighteen and nineteen. The students rated their attitudes towards General de Gaulle and towards Americans on Likert scales. Discussion to consensus produced statistically significant shifts towards the extremes of the scales that persisted in post-consensus individual ratings. That challenges the assumption that group judgements are less extreme than individual ones. The authors suggested normative commitment as the underlying variable.

Isenberg (1986) settled what the finding means. His meta-analysis covers 21 published articles from 1974 to 1982, giving 33 independent effects. He compares social comparison with persuasive argumentation. Both processes act in combination, but persuasive argumentation effects are larger, averaging r = 0.75 against r = 0.44 for mere exposure. A moderator who reports it as one reports a quantity the method does not produce.

Diehl and Stroebe (1987) ran four experiments. They tested free riding, evaluation apprehension and production blocking as explanations of a known shortfall. Real brainstorming groups generate fewer ideas than the same number of individuals working alone. Production blocking, the inability to speak while another holds the floor, accounted for most of the loss. A group session therefore does not safely contain the ideas its members brought. It can suppress them.

Structured alternatives keep the loss but lose the interaction

Delbecq and Van de Ven (1971) set out a group process model for problem identification and program planning. Its meeting formats involve different reference groups in successive phases. The nominal group technique is the procedure associated with it, in which participants generate ideas individually before the group discusses them. That removes the floor-taking Diehl and Stroebe identified. Dalkey and Helmer (1963) report an experimental application of Delphi, in which experts answered questionnaires in depth over successive rounds interspersed with controlled opinion feedback. They found convergence towards a more reliable consensus than one round produced.

• Focus group. Treats interaction between participants as the data, so the group's movement cannot be read back as individual opinion (Kitzinger, 1994; Morgan, 1997).
• Nominal group technique. Generates ideas individually before discussion. That removes the floor-taking Diehl and Stroebe found, but it also loses the unplanned exchange Wilkinson (1998) prizes (Delbecq and Van de Ven, 1971).
• Delphi. Runs anonymous rounds with controlled feedback, which suppresses pressure at the cost of any observation of interaction (Dalkey and Helmer, 1963).

Contested question: does the group buy breadth, depth, or neither?

The literature agrees that a focus group is not a cheap interview and disagrees about what it is instead. The first position is held by Morgan (1997), Kitzinger (1994), Smithson (2000) and Hollander (2004). Group data must be analysed as interaction and never aggregated into individual opinions. Onwuegbuzie and colleagues (2009) give it a procedure. Their framework includes a micro-interlocutor analysis, recording which participant answers each question, in what order and with what signals.

The second position treats the group as an efficient route to more participants and to disclosures a one-to-one setting suppresses. Kitzinger (1995) supports it. The nearest direct test is Guest and colleagues (2017). They randomised 350 African-American men in Durham, North Carolina to a focus group or an individual interview. Both arms received the same open-ended questions on health-care seeking. At the event level the two generated similar numbers of unique items. Per person, interviews were more effective at breadth. Sensitive disclosures were more likely in a focus group. Some sensitive themes occurred only there, while none emerged exclusively in an interview.

The evidence cuts both ways. On breadth per person the interview wins. On sensitive disclosure the group can win. The reasonable reading is that the two elicit different kinds of information rather than different quantities of the same thing. That is what Wilkinson (1998) meant when she said the traditions differ over what interaction is for. Barbour (2005) adds the warning. A study that collects at the group level and analyses at the individual level has confused the level at which its findings hold.

A second dispute sits inside the polarisation finding. Moscovici and Zavalloni (1969) treated the shift to extremes as a challenge to assumptions about group judgement. Isenberg (1986) treats it as a mechanism to be explained. A product team reading a focus group consensus must decide whether it reflects the wider population or only the room.

What would settle it is a design that reports at both levels and pre-specifies which level the claim holds at. Guest and colleagues (2017) come closest, getting opposite answers at the two levels. A randomised comparison on a sensitive topic, with disclosure verified against an independent record, would test the access argument directly.

The two positions draw on different kinds of study. Morgan, Kitzinger, Smithson and Hollander argue from methodology and from close analysis of sessions. Guest and colleagues offer the strongest direct evidence, a randomised comparison of three hundred and fifty participants. Isenberg's meta-analysis settles how the shift happens rather than whether a group is worth convening. While the dispute is unresolved, decide the level before the session. If the claim is about individual opinion, run interviews. The strongest objection is that a structured session can recover breadth and protect its members at once.

Where the argument is heading

The movement is towards analysis at the level the method produces. Onwuegbuzie and colleagues (2009) push furthest, borrowing conversation analysis to code who speaks, in what order and with what signal. Guest and colleagues (2017) argue the choice should be aligned with the kind of information a study needs. Barbour (2005) points the same way from practice, insisting the level of analysis be stated rather than assumed.

The sources this unit rests on

• Chai, C. A., Barrios, M., Gomez-Benito, J., Berrio, A. I. and Guilera, G. (2024). Information retrieval in face-to-face and online focus groups: a systematic review. International Journal of Qualitative Methods 23. doi:10.1177/16094069241286856 Systematically reviews 45 studies that ran both face-to-face and online focus groups. Word counts were higher face to face, and 62.5 per cent reported more detailed responses there, but the number of ideas did not differ significantly between the formats. Online participation was more evenly distributed.
• Morgan, D. (1997). Focus Groups as Qualitative Research. SAGE Publications, Inc. doi:10.4135/9781412984287 Sets out focus groups as a qualitative method with its own logic, covering design, conduct and analysis, and states that the group is the unit of analysis. The unit's central claim.
• Kitzinger, J. (1994). The methodology of Focus Groups: the importance of interaction between research participants. Sociology of Health and Illness. doi:10.1111/1467-9566.ep11347023 Argues that interaction between participants, not efficiency, is what separates focus groups from one-to-one interviews. The unit's founding statement of purpose.
• Kitzinger, J. (1995). Qualitative Research: Introducing focus groups. BMJ. doi:10.1136/bmj.311.7000.299 Introduces focus group methodology and advises on composition, conduct and analysis, noting that the method does not discriminate against people who cannot read or write. It supplies the access argument later work tests.
• Wilkinson, S. (1998). Focus group methodology: a review. International Journal of Social Research Methodology. doi:10.1080/13645579.1998.10846874 Reviews the method across the social sciences and identifies three traditions with different views of what group interaction is for. It isolates three features: access to participants' own language, fuller accounts, and the observation of collective sense-making.
• Smithson, J. (2000). Using and analysing focus groups: Limitations and possibilities. International Journal of Social Research Methodology. doi:10.1080/136455700405172 Identifies the interactive features that complicate interpretation, among them domination, the construction of the Other, normative discourse and conflict, and argues that analysing interaction is not optional, since those dynamics distinguish the method.
• Hollander, J. A. (2004). The Social Contexts of Focus Groups. Journal of Contemporary Ethnography. doi:10.1177/0891241604266988 Uses focus group data on the effects of violence on everyday life to show that discussions are shaped by four overlapping contexts: associational, status, conversational and relational. These produce problematic silences and speech, limiting the method for experience but making it strong for interaction.
• Barbour, R. S. (2005). Making sense of focus groups. Medical Education. doi:10.1111/j.1365-2929.2005.02200.x Reviews focus group use in medical education research and argues that the method is often chosen for convenience and then interpreted as though it produced individual opinions, confusing the level at which findings hold.
• Onwuegbuzie, A. J., Dickinson, W. B., Leech, N. L. and Zoran, A. G. (2009). A Qualitative Framework for Collecting and Analysing Data in Focus Group Research. International Journal of Qualitative Methods. doi:10.1177/160940690900800301 Provides a framework for collecting and analysing focus group data and introduces a micro-interlocutor analysis recording who answers each question, in what order and with what signals. It gives the interactional claim a procedure.
• Moscovici, S. and Zavalloni, M. (1969). The group as a polarizer of attitudes. Journal of Personality and Social Psychology. doi:10.1037/h0027568 Compared individual and collective ratings of 140 male secondary school students on attitudes towards General de Gaulle and towards Americans. Discussion to consensus shifted ratings towards the scale extremes, and the shift persisted in later ratings, so groups do not average out extreme views.
• Isenberg, D. J. (1986). Group polarization: A critical review and meta-analysis. Journal of Personality and Social Psychology. doi:10.1037/0022-3514.50.6.1141 Reviews and meta-analyses 21 group polarisation articles from 1974 to 1982, giving 33 independent effects, and compares social comparison with persuasive argumentation. Both act in combination, but argumentation effects are larger, averaging r = 0.75 against r = 0.44.
• Diehl, M. and Stroebe, W. (1987). Productivity loss in brainstorming groups: Toward the solution of a riddle. Journal of Personality and Social Psychology. doi:10.1037/0022-3514.53.3.497 Reports four experiments testing free riding, evaluation apprehension and production blocking as explanations of the shortfall between real groups and the same individuals alone. Production blocking explained most of the loss.
• Delbecq, A. L. and Van de Ven, A. H. (1971). A Group Process Model for Problem Identification and Program Planning. The Journal of Applied Behavioural Science. doi:10.1177/002188637100700404 Sets out a group process model for problem identification and planning, with meeting formats involving reference groups in successive phases. The origin of the nominal group technique, in which ideas are generated individually before discussion.
• Dalkey, N. and Helmer, O. (1963). An Experimental Application of the DELPHI Method to the Use of Experts. Management Science. doi:10.1287/mnsc.9.3.458 Reports an experiment using Delphi, in which experts answered questionnaires in depth interspersed with controlled opinion feedback. It found convergence towards a more reliable consensus than one round produced.
• Guest, G., Namey, E., Taylor, J., Eley, N. and McKenna, K. (2017). Comparing focus groups and individual interviews: findings from a randomized study. International Journal of Social Research Methodology. doi:10.1080/13645579.2017.1281601 Randomised 350 African-American men to focus groups or individual interviews on health-care seeking. The methods produced similar numbers of unique items per event, but per person interviews were better at breadth, and sensitive disclosures were more likely in a focus group.