The mean and standard deviation of distributions with five different levels of assimetry and kurtosis were used to simulate 6,000 experiments, in a randomized complete block design with five treatments and six replications, considering two levels of precision and three levels of block effects. The tests of homogeneity of variance proposed by Han and Shukla, for randomized block design, and those of Bartlett, Cochran and Hartley, developed for one-way classification were applied to the simulated data. The last three tests were also used after correcting for block effects. The amount of rejection of the null hypothesis increased with the levels of assimetry and kurtosis in all tests. The percentage of rejection of the hypothesis of homogeneity of variances using Shukla's test and those of Bartlett, Cochran and Hartley, without block correction, were strongly affected by non-normality mainly when the block effect was small. The rejection of the null hypothesis when used Han's test or those of Bartlett, Cochran and Hartley tests, adjusted for block effects, were larger than 5% when the coefficients of assimetry and kurtosis were higher than 1.11 and 4.04 respectively.
tests of homogeneity of variances; assimetry; kurtosis