Seminars

We consider a study starting with two dose levels: A, B and a placebo control C.  A loser between A and B will be dropped at the end of Stage I, and the winner will be compared with C at the end of Stage II. To control alpha, one may use numerical integration to directly evaluate the critical value for the test statistic. Fortunately, normal approximation to this statistic is good enough for most practical situations and fine-tuning of the approximation requires only a very simple modification of the p-values.  Power evaluation is more complicated when A and B have different treatment effects. We will demonstrate that with appropriate modifications, normal approximation gives very accurate estimates for power and sample size evaluation.