Repeated significance tests with biased coin allocation schemes (Q1070723)

Suppose that the responses to two treatments are from a single parameter exponential family. Treatments are allocated sequentially using a biased coin design (that is, the probability that the \((n+1)\)-st subject is assigned to treatment A is a function of the proportion of the first n subjects that have been assigned to treatment A). Asymptotic error probabilities are found for a repeated signifance test of the hypothesis that the two treatments elicit the same response.