ABSTRACT
Suppose that we want to compare the response rates between an experimental arm and a control arm. Often in a between-arm comparison, the characteristics of study patients may be heterogeneous. In this case, the heterogeneity is characterized by some known predictors, and a stratified method is applied to the final analysis. When the distribution of the stratification factors is identical between two arms, unstratified testing ignoring the population heterogeneity controls the type I error rate but loses statistical efficiency. If the distribution of the predictors is different between two arms, however, unstratified testing does not maintain the type I error rate accurately, which can be more serious than losing statistical efficiency. In order to balance the distribution of the factors, we usually randomize the patients by stratifying for the important predictors.
