Summary
The episode works through four questions that students often find difficult when designing observational studies against bias. It uses the birth weight paradox to show that the direction of causal arrows determines whether a variable is a confounder, a mediator or a collider, and that adjusting for a mediator that shares causes with the outcome can create bias. A made-up telephone survey shows that response depending on the outcome alone can bias a prevalence while leaving an odds ratio unbiased, and the hosts then correct an odds ratio for assumed sampling fractions. A three-category example shows how non-differential under-reporting can bias a middle category away from the null and make a threshold resemble a dose-response relationship. The hosts close with a discussion of the strengths and limits of the E-value and agree to report it alongside bias factors for named confounders.