Primary Submission Category: Randomized Studies
Design-Based Causal Inference with Missing Outcomes: Missingness Mechanisms, Imputation-Assisted Randomization Tests, and Covariate Adjustment
Authors: Siyu Heng, Jiawei Zhang, Yang Feng,
Presenting Author: Siyu Heng*
Design-based causal inference is immune to outcome model misspecification as its statistical validity only comes from the study design (e.g., randomization or matching design) and does not require assuming any outcome-generating distributions or models. However, design-based causal inference may still suffer from other data challenges from outcome variables, among which missingness in outcomes is a significant one. We systematically study the outcome missingness problem in design-based causal inference. First, we use the potential outcomes framework to clarify the minimal assumption (concerning the outcome missingness mechanism) needed for conducting finite-population-exact randomization tests for the null effect (i.e., Fisher’s sharp null) and that needed for constructing finite-population-exact confidence sets with missing outcomes. Second, we propose a general framework called “imputation and re-imputation” for conducting finite-population-exact randomization tests in design-based causal studies with missing outcomes. Our framework can incorporate any existing outcome imputation algorithms and meanwhile guarantee finite-population-exact type-I error rate control. Third, we extend our framework to conduct covariate adjustment in an exact randomization test with missing outcomes and to construct finite-population-exact confidence sets with missing outcomes.