Summary
Defines confounding through its three conditions and works the hormone-replacement-therapy cardiovascular story in which observational studies suggested 30 to 50 percent reductions while the Women's Health Initiative randomized trial found a 29 percent increase. Walks through confounding by indication, confounding by contraindication, and time-varying confounding in HIV care that required Hernan, Brumback, and Robins' marginal structural models with inverse probability of treatment weighting. Reframes race as a proxy for racism-as-exposure using Krieger, Williams, and Bauer's intersectionality, then turns to model misspecification with J-shaped alcohol-mortality curves, multicollinearity, Simpson's paradox, and missing-data treatment.