Keselman et al. [8], Conover et al. [9], and Balakrishnan and Ma [10]. O'Brien [11] proposed a modification of Levene's test (OB50) which has been recommended over the W50 test for lighter-tailed distributions [11,12]. Marozzi [13] considered the W50 and OB50 tests, as well as permutation versions of these tests, and found that the permutation versions of these tests tended to be more robust and have higher power. They recommended the permutation W50 test as a computationally simple robust test, but also the permutation version of OB50, which had higher power for symmetric and lighter-tailed skewed distributions. Richter and McCann [14] found that the permutation RMD test due to Higgins [15] was generally superior to W50 and OB50, especially for heavier-tailed distributions. In this paper, we extend the RMD, W50 and OB50 tests to the problem of simultaneous pairwise comparison of scale. The method of Richter and McCann [16] is used to control the familywise error rate (FWER), and estimated Type I error rates and power of the methods are compared. 2. Methods Consider a one-way layout with t treatments and observations per treatment. We assume a location-scale model, = + , = 1, … , , = 1, … , where and are the location and scale parameters, respectively, of treatment i, and are independent and identically distributed with median 0. It is desired to test 0 ∶ = versus ∶ = for all pairs ≠. 2.1. LEV and W50 tests Levene [6] proposed a robust test (LEV) to compare scale parameters, using the ANOVA Fstatistic computed on the absolute deviations from the treatment means, = � − �. Brown and Forsythe [7] suggested a statistic (W50) that instead used absolute deviations from the treatment medians, = � − �, which we will refer to in the remainder of the paper as deviances. Conover et al. [9] found W50 to have better power and size properties than LEV. Both Levene [6] and Brown and Forsythe [7] suggested that p-values be based on the Fdistribution with − 1 and − degrees of freedom. Marrozzi [13], however, examined permutation versions of these tests, and found the permutation versions to be more robust and more powerful than those based on the F-distribution. 2.2. OB50 tests O'Brien [11] proposed a modification to LEV, suggesting the scores () =