Semiparametric combinations of two dependent tests

Abstract

A semiparametric estimator of the p-value for a class of combinations of two dependent test statistics is proposed. It is based on resampling techniques and knowledge of the marginal distributions of the two test statistics to be combined. The semiparametric estimator is shown to be asymptotically not less efficient than the nonparametric counterpart, only based on the estimation by resampling techniques of the joint null distribution of the two test statistics.