Rerandomization to Improve Covariate Balance in Randomized Experiments
Ph.D. Dissertation, Harvard University, Cambridge, MA
133 Pages Posted: 15 Jul 2011
Date Written: May 1, 2011
Randomized experiments are the "gold standard" for conducting research and establishing a causal effect, yet often in practice chance imbalances exist in covariates. If covariate data are available before an experiment is conducted, these chance imbalances can be prevented by checking covariate balance before the experiment takes place. Provided an objective definition of imbalance has been specified in advance, imbalanced randomizations can be rerandomized, and this rerandomization process can continue until a randomization yielding comparable groups is achieved. By improving covariate balance, rerandomization provides more precise and trustworthy estimates of a treatment effect.
Keywords: Randomization, Treatment Allocation, Experimental Design, Clinical Trial, Causal Effect, Mahalanobis Distance, Hotelling's T²
JEL Classification: C90
Suggested Citation: Suggested Citation