Mon 7 Mar 11am NY time:

Bayesian Methods in Causal Inference and Decision Making

Consider the problem of A/B testing (that is, an experiment or observational study designed to estimate the effect of some exposure or treatment). The basic data analysis workflow is to start by comparing the average outcomes under the two groups, and then to estimate varying treatment effects and adjust for pre-treatment imbalances. The basic decision making workflow is to start by looking at statistical significance and then to consider cost-benefit tradeoffs. We consider several places where Bayesian inference enters this workflow: priors on the treatment effect and its variation, priors on adjustment factors, partial pooling across experiments, poststratification, and decision analysis. Many theoretical and practical research questions arise even in the apparently simple case of randomized experiments.