On this page
Ask how the units were chosen
A design can be audited by its frame and randomization rules. A photograph or table showing a pleasing mix of groups does not reveal the selection probabilities. Hand-picked observations can appear balanced on one recorded feature while being selected on other unrecorded features.
Inspect the observed composition
A simple random sample does not force every subgroup to appear in every draw. If a study requires positive samples in defined subgroups, a stratified allocation can enforce that coverage. The rule then belongs in the design, with appropriate estimation weights.
Keep claims about features separate
| Evidence | What it establishes |
|---|---|
| Recorded random selection rule | How selection can occur |
| Matched group percentages | Balance on those measured groups |
| Every subgroup sampled | Coverage of those defined subgroups |
| Large stored row count | A record count, not automatic representativeness |
Report both design and limitations
Document unexpected imbalances, response patterns and missing frame units. The design’s uncertainty calculation and the sample’s substantive limitations answer different questions. Neither should be hidden behind a generic claim that the sample is representative.