CONDITIONAL APPLICATION OF MEAN COMPARISON TESTS TO A SIGNIFICANT RESULT OF THE OVERALL F TEST ON ANALYSIS OF VARIANCE

The purpose of this study was to compare the result of the overall F test on analysis of variance with the result of the following multiple comparison procedures: Tukey, Duncan, Fisher’s Least Significant Difference, Student-Newman-Keuls and Scheffe. For this were simulated in the completely randomized design, by Monte Carlo method, 2,000 experiments by scenario, which are in a total of 128 cases, formed by the combination of the number of treatments of the experiments, the number of repetitions for treatments, the coefficient of variation and two combinations for the treatment effects. It was observed, in part of the simulated experiments, divergent results between the overall F test and the mean comparison procedures included in the survey. Still, in most cases the change of the number of treatments and coefficient of variation of the experiments led to significant changes in contradictory results, which didn’t occur when the number of repetitions for treatments was changed. Finally, the contradictory results don’t appear to be linked to the deviation of at least one of the assumptions of analysis of variance.