Samples are not representative of the population, skewing conclusions.
Wald's wartime work on returning bombers is the canonical case: armour belongs where the survivors were not hit. The sample was defined by having survived, which reversed the obvious reading.
In practice
- A satisfaction survey sent to current customers cannot reach the ones who left, which is the question it was commissioned to answer.
- A study of successful founders drawn from a list of successful founders can say nothing about what causes success.
- Further reading
- Wikipedia
- Subject
- Science & Epistemology
- Essay
- The ways evidence misleads