A general-purpose method for two-group randomization tests

Abstract

We outline a novel approach to calculate exact p-levels for two-sample randomization tests. The approach closely resembles permute in its applications, with the main difference being that the results are approximated only if the execution time needed to calculate exact p-levels would exceed a specified maximum. We demonstrate its use by deriving p-levels for the significance of Somers' D, the coefficient of variation, the difference in means and in medians, and the difference in two multinomials. © 2009 StataCorp LP.