Multiple Comparison Procedure Based on Mallows 1-Distance Bootstrap

摘要

The Mallows distance is a strong distance measure between probability measures.In this work the Mallows distance is studied as a measure between random samples, and the basic theory of estimation of tail probabilities for the distribution of this distance is presented. Efron's bootstrap is used as the method of estimation. The asymptotic consistency of the estimation is studied by simulation experiments in small sample cases. As an application of the Mallows 1-distance bootstrap the classical multiple comparison problem is solved by using the mallows 1-distance as the distance measure, and a new grouping method, the separate groups test is presented.