Small nonrandom groups cannot estimate population proportions
Aliases: prevalence inference · nonprobability sample · focus-group counting
What it is
Focus groups usually combine small purposive samples with participant interaction, so a count of people who agreed cannot estimate a target-population proportion. The limitation is not sample size alone. It also follows from nonprobability recruitment, dependent observations within a group, and unequal opportunities for a view to be voiced.
Why it happens
Prevalence estimation requires a defined target population, interpretable inclusion probabilities, and quantifiable sampling uncertainty. Focus groups instead recruit for relevant variation and generate material through interaction: one account can elicit, suppress, or transform another. Speaking turns are neither independent observations nor a stable denominator, so percentages calculated from mentions create spurious precision.
Studying it
Use groups to identify types of account, vocabulary, mechanisms, and conditions, and report the recruitment and discussion context in which each arose. Repetition across groups can test whether an interpretation depends on one composition. Adequacy can be judged against the question and new explanatory value, but thematic saturation is not stable population prevalence. When a decision needs a proportion or segment difference, design a probability sample or an explicit quantitative inference model.
Where it stops holding
Failure to estimate prevalence does not make group evidence worthless: a rare account may reveal a severe hazard or new mechanism. A census discussion of a small, bounded team can describe who publicly expressed what, while still qualifying interaction and unspoken views. Adding more purposively recruited groups does not by itself turn the sample into a probability sample.
Applying it
- Report which accounts appeared and under what conditions, not percentages of users.
- Retain counterexamples, eliciting context, and group provenance; do not rank importance by quote count.
- Convert high-consequence, disputed, or scale-dependent themes into subsequent measurement items.
- Label existence evidence separately from frequency evidence in decision materials.