“Finite-Sample Optimal Estimation and Inference on Average Treatment Effects Under Unconfoundedness,” Tim Armstrong, Yale