ABSTRACT
Bayesian model averaging has traditionally been used in the context of prediction, however, it has recently been applied to a different context: estimation of the effect of a treatment on an outcome. To estimate treatment effects, prior distributions should prioritize variables that are associated with both the treatment and outcome variables, commonly referred to as confounders. In this chapter, we begin with a brief overview of basic topics in causal inference, and then discuss recent work aimed at integrating traditional Bayesian model averaging techniques within a causal inference framework. We discuss different approaches to prior distributions and we illustrate how, if done properly, informative prior distributions can lead to substantial improvements infinite samples. A brief vignette and all R code used to implement the models described in this section can be found in the Supplementary Materials.
