What is Judgment Sampling?

Judgment sampling (purposive or expert sampling) is a non-probability method in which the researcher selects participants deliberately, based on their knowledge of who holds the characteristics, experience, or perspective the study requires. Where probability sampling hands selection to a random mechanism to earn statistical inference, judgment sampling hands it to expertise to earn relevance: the participants are chosen because they are informative, not because they are typical. This is the native sampling logic of qualitative research, and it is legitimate exactly to the extent that the study's claims are about mechanisms, experiences, and possibilities rather than about how common anything is.

The Forms

Expert sampling. People with specialised knowledge: the power users who have found every workaround, the administrators who configure the product for hundreds of others, the practitioners a Delphi panel convenes.

Criterion sampling. Everyone who meets a defined condition: switched from a competitor in the last quarter, abandoned checkout twice, completed onboarding without support. The screener is the instrument, and behavioural criteria beat self-described ones.

Maximum-variation sampling. Deliberately diverse cases (novice and expert, small and enterprise, three markets) so that any pattern found across them is more likely to be a mechanism than a local quirk, the sampling strategy that best supports analytic transfer.

Extreme and deviant case sampling. The outliers on purpose: the heaviest users, the fastest churners, the people who succeeded where everyone else failed. Extreme cases reveal boundaries and mechanisms the middle conceals.

Typical case sampling. The deliberately ordinary, chosen to illustrate the common experience for stakeholders, without any claim that "typical" was established statistically.

What It Can and Cannot Claim

Judgment samples support claims of existence and mechanism ("this problem occurs, and here is how it unfolds"), of range ("these are the kinds of workaround people build"), and of design implication ("the flow fails at this step for users in this situation"). They cannot support prevalence ("40% of users experience this"), magnitude, or population-level comparison, because nothing about the selection connects the sample proportionally to any population. The most frequent abuse is exactly that leap: eight purposively chosen interviews reported with percentages. The representativeness a prevalence claim requires is a different design with a different logic, and the honest readout of a judgment sample counts participants, describes who they were, and lets the reader judge transfer.

Doing It Well

1. Write the rationale before recruiting.
Which characteristics matter, why, and what mix of cases the question needs. Judgment without a documented rationale is convenience sampling with a confident face.

2. Screen on evidence, not identity.
Recent specific behaviour and, where feasible, observable proof: recruiting a panel through behavioural screeners and then watching participants perform real tasks (as a Ballpark study does) verifies the judgment that put them in the sample.

3. Sample for disconfirmation.
Include the cases most likely to contradict the emerging story; a judgment sample that only contains believers has been assembled to agree, the confirmation failure in recruitment form.

4. Report selection transparently.
Criteria, sources, who was approached, who declined: the reader's only basis for judging what the sample can teach.

The Takeaway

Judgment sampling chooses participants for what they know, not for how common they are: experts, criterion cases, maximum variation, extremes, and typicals, each selected by documented rationale. It buys depth and relevance, and it pays by forfeiting prevalence claims entirely. Sample deliberately, screen on evidence, seek the disconfirming case, and report the selection; then let the claims be about mechanism, which is what the design was built to find.