Blocking estimators and inference under the Neyman-Rubin model

TL;DR

This paper derives variance formulas for estimators of average treatment effects within the Neyman-Rubin model, applicable to any blocking scheme and number of treatments, enhancing understanding of treatment effect inference.

Contribution

It provides a general derivation of variances for treatment effect estimators under complex blocking and multiple treatments, extending prior work.

Findings

Variance formulas applicable to arbitrary blocking schemes

Extension to multiple treatments scenarios

Improved understanding of estimator variability

Abstract

We derive the variances of estimators for sample average treatment effects under the Neyman-Rubin potential outcomes model for arbitrary blocking assignments and an arbitrary number of treatments.