Randomization-based inference within principal strata

Abstract

In randomized studies, treatment comparisons conditional on intermediate postrandomization outcomes using standard analytic methods do not have a causal interpretation. An alternative approach entails treatment comparisons within principal strata defined by the potential outcomes for the intermediate outcome that would be observed under each treatment assignment. In this article we develop methods for randomization-based inference within principal strata. We compare our proposed methods with existing large-sample methods as well as traditional intent-to-treat approaches. This research is motivated by HIV prevention studies, where few infections are expected and inference is desired within the always-infected principal stratum, that is, all individuals who would become infected regardless of randomization assignment.