Abstract: A sequential test is considered in which two treatments are compared and treatment allocation is data-dependent. Brownian motion approximations are obtained for the bias and variance of the maximum likelihood estimator of treatment difference at the end of the trial. For normal responses, simulation indicates that the approximations work well for several data-dependent allocation rules.
Key words and phrases: Brownian motion, clinical trial, correction for overshoot, data-dependent allocation rule, Gittins index, stopping time.
<
/BODY>