The distribution of loss in two-treatment biased-coin designs

Abstract

The paper compares randomized rules of the biased-coin type for the sequential allocation of treatments in a clinical trial. An important characteristic is the loss, which measures the increase in the variance of parameter estimates due to the imbalance caused by randomization. Simulations are used to find the small-sample distribution of loss. For some rules a simple chi-squared approximation to the asymptotic distribution holds well down to very small sample sizes.