Extended-Paulson sequential selection (Q1091724)

Sequential methods for choosing the best out of k treatments have already been studied under the assumption that the corresponding observations are normally distributed having the same known variance. The author proposes a more general procedure which allows to drop the normality assumption. Moreover, he points out that in some cases the new procedure makes a stronger confidence without (asymptotically) inflating sample sizes whereas in other cases it retains the same confidence with asymptotic sample sizes being substantially superior to existing procedures.