Rerandomization to improve covariate balance in experiments

摘要

Randomized experiments are the "gold standard" for estimating causal effects,yet often in practice, chance imbalances exist in covariate distributionsbetween treatment groups. If covariate data are available before units areexposed to treatments, these chance imbalances can be mitigated by firstchecking covariate balance before the physical experiment takes place. Provideda precise definition of imbalance has been specified in advance, unbalancedrandomizations can be discarded, followed by a rerandomization, and thisprocess can continue until a randomization yielding balance according to thedefinition is achieved. By improving covariate balance, rerandomizationprovides more precise and trustworthy estimates of treatment effects.