Summary
The episode works through four questions that students often find difficult in linear and logistic regression. It converts odds ratios into predicted probabilities, uses a constructed population to show why an odds ratio can change after adjustment for a variable that is not a confounder, and explains how to read the coefficients of a model that contains an age-by-smoking product term. It then distinguishes discrimination from calibration, drawing on the validation of the Framingham equations in other cohorts and on a count of events per coefficient. The hosts close by debating whether systolic blood pressure should be turned into a yes-or-no outcome, and they agree on a rule for choosing and reporting cut-points.