Matched Pairs in Sequential Trials for Significance of a Difference between Proportions

Abstract

Two methods of arranging a sequential trial for matched pairs are discussed. Method 1 assumes, in all strata, a constant 0 M, while Method 2 assumes a proportional change in the incidence of successes or failures. The number of pairs required to reach a decision is in general smaller for a matched-pair trial than for the usual method of unrestricted pairing. Computations involved in arranging a matched-pair trial are explained by means of an example, and detailed proofs of all propositions are given in the text. The results of experiments set up to test the working of the two methods agree closely with theoretical expectations.