The benefit of stratification in clinical trials revisited