A new nonparametric approach for multiplicity control: Optimal subset procedures (Q2488431)

A new test procedure is proposed for multialternative hypothesis testing. It can be used in the case where the space of the considered hypotheses consists of all possible intersections of some ``elementary'' hypotheses. The idea is to select a hypothesis with minimal \(p\)-level w.r.t. some bi-alternative test, and then to adjust the obtained \(p\)-level for multiplicity. Permutation procedures are considered for such adjustment and step-up procedures for approximately best selections. Stepwise regression is considered as example. Results of simulations and application to neuroimaging data are considered.