Summary
The episode works through five questions on selection-related bias for students who have completed the lesson. It uses hypothetical cohorts, hospitals and surveys to show that an unrepresentative sample distorts descriptive estimates while an association is biased only when selection depends on both exposure and outcome, that Berkson's bias can run in either direction, and that response rates are weak guides to nonresponse bias. The hosts discuss the published debate over whether epidemiological studies should seek representative samples and agree on a practical rule that starts from the purpose of the study. A harder example shows how a constant risk ratio can give a very different absolute benefit when a trial result is transported to a population with a different mix of baseline risk.