Parameter estimation following group sequential hypothesis testing

Abstract

Parameter estimation techniques which fail to adjust for the interim analyses of group sequential test designs will introduce bias in much the same way that the repeated use of single sample hypothesis testing causes inflation of the type one statistical error rate. Methods based on the duality of hypothesis testing and interval estimation require definition of an ordering for the outcome space for the test statistic. In this paper, estimation following a group sequential hypothesis test for the mean of a normal distribution with known variance is investigated. A proposed ordering of the sample space based on the maximum likelihood estimate of the mean is found to result in estimates which compare favourably with estimates computed from orderings investigated by Tsiatis, Rosner & Mehta (1984) and Chang & O'Brien (1986) for a variety of group sequential designs. The proposed ordering is then adapted for use when the sizes of groups accrued between analyses is random.