|
Abstract:
|
Randomization balances covariates between treatment groups on average, but in any particular randomization, covariate imbalance between groups is possible, and even likely. Rather than proceeding with observed imbalance, if the experiment has not been conducted yet, units can and should be rerandomized to treatment groups. Criteria for adequate covariate balance should be specified in advance, but these criteria, both in how covariate balance is measured and in how stringently balance is enforced, are flexible and can be specified in the context of a particular experiment. The general idea, rerandomizing units when covariates are imbalanced, is applicable to a wide variety of experimental settings.
|
