A sequential elimination procedure for choosing the best population(s) based on multiple testing (Q1116602)

A general sequential procedure for the elimination of inferior elements from a set of m populations is proposed. The procedure consists of applying simultaneously all pairwise comparisons. A population is eliminated, when it is rejected as being inferior in an appropriate individual test, e.g. in a repeated significance test. The \(\alpha\)-level for the individual tests can be enlarged after the elimination of a population. Under fairly general conditions, the procedure controls the probability of erroneously eliminating at least one of the best populations, irrespective of how many populations are the best ones. The procedure can be improved considerably, if only one best population is known to exist, or the set of populations not having been eliminated should contain at least one (but not necessarily all) of the best populations with probability 1-\(\alpha\).