This workshop will review the instrumental variable (IV) framework for accounting for unobserved confounding and other sources of selection bias. We will emphasize the assumptions needed for the IV to produce unbiased causal effect estimates in a non-experimental study, and will also introduce analogous IV methods for selection bias. important considerations for practical implementation in epidemiologic practice, with the goal of taming whenever suspected to be present, bias due to unobserved confounding, and selection bias due to nonignorable missing. Drawing from contemporary applications ranging from analysis of randomized controlled trials, Mendelian randomization studies to the study of the infectiousness of obesity will be discussed for illustration. Finally, methods to evaluate empirically and with the use of sensitivity analysis, the credibility of assumptions made in the IV approach will also be given careful consideration. Coding examples will be drawn from SAS, Stata and R.
Workshop Instructors: Maria Glymour and Eric Tchetgen Tchetgen