Ajuste do procedimento de agrupamento de médias Scott-Knott para experimentos não-balanceados

Abstract

This paper presents a review about the evolution of the most broadly disseminated means comparison procedures and of the clustering means procedure proposed by Scott & Knott (1974) with a solution to use it even under unbalanced experiment designs. Since the original test was developed to be used specifically under complete designs, the loss of a single plot can increase the Type I Error Rate. To avoid inefficient Error control in the missing plot scenario we propose an adjustment that should maintain a similar protection of the original. The proposed solution was validated over more than 40 million of simulated experiments following the Monte Carlo method. As results a minimal loss of power and a satisfactory Type I Error control was observed with the original procedure features maintained. In order to make easier the implementation of the adjustment, a SAS System macro with minor hardcoding requirement is provided as well.