Validity in Observational Studies

Summary

Lays out the four pillars of validity as internal, external, construct, and statistical conclusion, treating internal validity as necessary but not sufficient for external validity. Walks through the three classical internal threats of selection bias including Berkson, the healthy-worker effect, differential loss to follow-up, and volunteer bias, then information bias through recall, social desirability, observer, and detection mechanisms. Closes with confounding in its classical, by-indication, and time-varying forms, and introduces restriction, matching, multivariable adjustment, propensity scores, instrumental variables including Mendelian randomization, and marginal structural models with inverse probability of treatment weighting.