EFFICIENT CALCULATION OF THE PERMUTATION DISTRIBUTION OF ROBUST 2-SAMPLE STATISTICS

Polynomial-time algorithms are presented for calculating the permutation distribution of several robust test statistics in the two-sample setting. The class of statistics considered are differences of trimmed means, medians, Winsorized means and bi-weight means. This approach of applying permutation theory to robust statistics ensures both robustness of validity and efficiency. The asymptotic permutation distribution is found since reliance on asymptotic results may be desirable in large-sample situations when our time bounds may be unacceptable.