A Unified Theory of Covariate Adjustment for Design-Based Inference

• In Event: Causal Inference: Theory and Methods

Abstract

From within a class of covariate-adjusted estimators of the average treatment effect (ATE), I define two that are asymptotically optimal under arbitrary experimental designs. Simulation illustrates the potential for reductions in MSE (relative to OLS or WLS), in particular, for cluster-randomized experiments.

Author

• Joel A. Middleton, UC Berkeley