Summary
The episode works through five questions that students commonly find difficult in the measurement and causal specification of epidemiological studies. It shows how random measurement error attenuates a slope and how a reliability estimate from a validation substudy corrects it, and it examines construct-irrelevant variance, differential item functioning and the meaning of self-rated health across groups. The hosts discuss whether Likert responses can be analysed as interval data and agree on a sensitivity analysis that compares a linear coding with category indicators. A harder example classifies the covariates in a hypothetical study of night-shift work and diabetes as confounders, mediators and colliders, and links the selection trap to the obesity paradox. The episode closes by considering what lag analyses reveal about reverse causation and how simultaneity differs from it.