Testes de comparações múltiplas baseados na distribuição da midrange estudentizada externamente

Abstract

This paper proposes four types of multiple comparison tests, to evaluate the performance of proposed tests in large experimental situations via Monte Carlo simulation, to compare the experimentwise error rate and power of the proposed tests with the tests in the literature, mainly with the of Tukey, SNK and Scott-Knott tests, and finally, a R package was developed, midrangeMCP, to facilitate the users in applying the tests. The proposed tests are: Midrange Tukey test (TM), Midrange Student-Newman-Keuls test (SNKM), Mean Grouping based on the Midrange test (MGM) and Mean Grouping based on the Range test (MGR). The MGM, MGR and TM tests showed superior performance to the original Tukey and Skott-Knott tests, except when the tests were conducted under partial H0. The SNKM test was the only one that has a low performance when compared to the original test (SNK test) for all the evaluated situations. The MGM test obtained greater prominence among the proposed tests when evaluated its performance, adding the advantage of not introduce ambiguity in the results.