A Monte Carlo study of nonparametric multiple-comparison tests for a two-way layout

Abstract

A Monte Carlo study of two overall tests, the Friedman and Doksum, and five nonparametric multiple-comparison tests, the McDonald-Thompson, Nemenyi, Wilcoxon, Rhyne-Steel, and sign test, was done to determine the tests’ Type I error rates in a two-way layout with k = 3, 5, and 7 treatments and n = 8, 11, and 15 blocks. It was found that (1) the Friedman test was superior to the Doksum test as an overall test, and (2) the Wilcoxon signed-ranks test, when protected by a significant Friedman test, was the best pairwise multiple-comparison procedure. In a second Monte Carlo study, a modified sign test, termed the stepped-down sign test, was found to be superior to the Wilcoxon test as a protected test; it is recommended for pairwise comparisons in a two-way layout. © 1979, Psychonomic Society, Inc.. All rights reserved.