Summary
The episode works through four questions that students commonly find difficult in the opening lesson of the course: whether every measured variable should be adjusted for, what crude associations are for when the causal diagram chooses the adjustment set, how missing-value codes and complete-case decisions change results, and why the shape of an outcome is checked before any model is fitted. The hosts illustrate collider bias with a hypothetical referral clinic and with early COVID-19 reports on smoking, and they relate missing-value codes to the reserved codes used in the Canadian Community Health Survey. Sarah works a hidden-code calculation aloud, corrects her error with a sanity check, and then computes bounds for the true proportion. A harder example shows how pooling two clinics can produce apparent overdispersion, and the hosts connect this result to clustered data and standard errors that are too small. The episode closes with a discussion of cut-points for continuous measures, drawing on the 2017 change to the American hypertension threshold and Canadian estimates under both thresholds, and the hosts agree on a practical rule for their use.