Sequential tests for an unstable response variable

Abstract

The class of sequential tests investigated by Robbins & Siegmund (1974) for deciding which of two treatments is superior is considered. For normal responses, it is shown how these tests can be modified to take into account linear time-trends in the data or serial correlation between responses. A number of allocation rules designed to reduce the number of patients who receive the inferior treatment are compared. Simulation indicates that the error probabilities for the modified tests are generally no larger than those for the original tests with independent and identically distributed responses.