Permutation tests for mixed paired and two-sample designs

Abstract

Permutation tests based on previously developed statistics are proposed for the case of mixed paired and two-sample designs. Different weighting schemes of previous tests are explored to understand the strengths and weaknesses of each test. A simulation study compares the power and Type I error rates of the new tests with those previously developed. Rank-based statistics generally performed as well as or better than parametric statistics, particularly for nonnormal distributions.