Bayesian Approach to Exclusion Restriction

Presenter: Rick McAlexander

Plan

Many works in political science use an instrumental variables (IV) regression to fix endogeneity problems and to identify causal effects. Standard approaches assume that an instrumental variable is correlated with the endogenous dependent variable, but that the instrument has no direct effect on the dependent variable of interest. Often, we have reason to belief that this assumption-the exclusion restriction-is violated, or at the least, we are uncertain if the assumption holds exactly. In this workshop, I show how to easily relax the exclusion restriction using a Bayesian model coded in Stan. I discuss what the implications are for our inferences from studies using IV models, and how the parameters from these models behave as the type of uncertainty changes. Please have Stan and rstan updated and installed!

Link to Workshop Materials

• Slides for the presentation
• Example R Script