Qualitative work seeks analytic generalization, not statistical representativeness

What it is

The legitimate way to take qualitative material beyond the cases at hand is analytic generalization: attaching mechanisms, conditions, and patterns of variation that hold in the cases to a theory or a portable proposition—not using the sample to estimate how common a phenomenon is in a population. Statistical representativeness needs known or modelable inclusion probabilities and an estimate aimed at a population parameter. Interviews, observation, and case studies usually do not have that sampling apparatus. Asking whether “twelve people represent users” evaluates the wrong object.

Why it happens

Probabilistic generalization runs from instances to a population parameter: a sample mean or proportion approximates the population under a sampling model. Analytic generalization runs from instances to a proposition: if mechanism M produces result R under condition C, a new setting that still has C can be expected to show R, whether or not it belongs to the original population. The burden therefore moves from “is n large enough” to “is the mechanism well specified by the cases, are conditions written down, and were contrary cases taken seriously.” Purposive sampling exists to cover variation that matters to the proposition, not to mimic a random sample’s demographic profile.

Studying it

Split each claim into two layers: relations already observed inside the cases, and the proposition offered for travel. For the proposition, list supporting cases, undermining cases, and the conditions it depends on. Organize multiple cases by replication logic rather than counting logic: literal replication asks whether similar conditions reproduce the result; theoretical replication asks whether changing a condition changes the result as the proposition predicts. If a report gives percentages, check whether the denominator is a convenience sample; if so, the percentage describes composition inside the corpus, not a population estimate. Review asks whether the next, different, clearly conditioned case could challenge the proposition, not whether n hit a customary number.

Where it stops holding

Analytic generalization cannot passport a theme label that has no mechanism; “many people mentioned waiting” is not a proposition. It also does not forbid a quantitative estimate—that is a different question and needs a different sampling design. Saying “so the sample does not matter” tears out the rationale for purposive sampling as well. Readers may still judge case-to-case fit; that is transferability, which depends on thick description, and is not the same act as promoting a case into theory.

Related

• Same group: Q4.10.1 A conclusion’s scope is bounded by the sample · Q4.10.2 A change of context can void a conclusion · Q4.10.3 Reports must state boundary conditions · Q4.10.5 A single-case finding is not a general law until replicated · Q4.10.6 Recheck premises before transferring across cultures or platforms · Q4.10.7 Packaging small-sample findings as universal truths is a common overreach
• Adjacent: Q1.03 Choosing qualitative and quantitative methods · Q1.04 Sampling and representativeness
• Search terms: analytic generalization · theoretical generalization · transferability