Scholion.ai

A Bayesian Criterion for Rerandomization

Zhaoyang Liu, Tingxuan Han, Donald B. Rubin, Ke Deng

Journal of the American Statistical Association · 2025 · Volume 120 · Issue 552

DOI: 10.1080/01621459.2025.2507432

Abstract

Rerandomization is a powerful tool for experiment-based causal inference because it can better balance covariates than classic randomized designs, thereby leading to more accurate causal effect estimation. However, basic rerandomization and some of its extensions do not prioritize covariates that believed to be strongly associated with potential outcomes. To address this limitation, and thereby create more efficient rerandomization procedures, the quantification of covariate heterogeneity is appealing. We propose a Bayesian criterion for rerandomization that addresses this issue. Both theoretical analyses and numerical studies suggest that rerandomization procedures using Bayesian criterion can outperform existing procedures for balancing covariates. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Keywords

Adaptive design, Causal inference, Covariate adjustment, Randomization inference

Enable JavaScript for citation tools and additional paper interactions.