Abstract: Sample size calculations require specification of the treatment effect, but what if this is not known? Two-stage tests use the first stage to estimate the treatment effect and modify the sample size accordingly. The purpose of this paper is to unify the theory of two-stage testing based on treatment effect. The conditional error function approach of Proschan and Hunsberger (1995) is shown to be a useful way to evaluate the properties of any two-stage test. The connection between two-stage tests and positive quadrant tests is exploited to motivate certain conditional error functions.
Key words and phrases: Double sampling, type I error rate, p-value, positive quadrant tests.