Primary Submission Category: Matching
Covariate-adaptive randomization inference in matched designs
Authors: Samuel Pimentel, Ruoqi Yu,
Presenting Author: Samuel Pimentel*
After matching treated units to controls in observational data, it is common to conduct inference by permuting treatment assignments as in a Fisher randomization test (FRT). This approach may fail to control Type I error. Firstly, it does not account for differences in treatment propensities between matched individuals, which typically do not agree exactly in practice. Secondly, it does not consider whether permuted versions of treatment would have led to the selection of the same matched-pair configuration. Recent proposals to update the FRT procedure using estimated propensity scores help address the first problem, but the second has received little attention. We show that treatment permutations incompatible with the matched pair configuration can lead to substantive Type I error violations and present new computationally efficient graph-based inference procedures for optimal pair matching with propensity scores that eliminate incompatible treatment assignments from consideration. We demonstrate their effectiveness via simulations and a re-analysis of an observational study of health outcomes.