On group sequential tests based on robust location (Q1979098)

The famous Princeton study of \textit{D.F. Andrews, P.J. Bickel, F.R. Hampel, P.J. Huber, W.H. Rogers} and \textit{J.W. Tukey} [Robust estimates of location. Survey and advances. (1972; Zbl 0254.62001)] had a strong influence on further research on robustness. Much recent research concentrates on robustness properties of estimators for a fixed sample size, e.g., the behaviour of the influence function, breakdown point, maximal bias curve, and efficiency considerations. There are different strategies based on robust estimators to identify outliers. In some areas of applied statistics, e.g. in planning and analyzing clinical trials, group sequential plans play an important role. Such plans can reduce the average sample size number, i.e. the expected sample size when the test stops, which is attractive from ethical, time and financial aspects. In contrast to the fixed sample size case, much less research has been published on the application of robust estimators to group sequential plans. The aim of this paper is to study the behaviour of group sequential two-sample tests for location difference if one replaces the classical non-robust estimators in the \(t\)-test statistic by modern robust estimators for location and scale. Four criteria are considered: the actual level and power of the test, the average sample size number, and the bias of the naive estimated standardized treatment difference. It is investigated, whether commercial software can be used to apply these tests.