Proposta per un nuovo test di omogeneità delle varianze

A new test for the homogeneity of variances hypothesis H0:s1 2 =...=s k2of k normal independent populations with unknown means versus the general alternative H1:s1 2 <>s h2 for some j<>h (j, h = 1,.... k) is proposed. The exact null distribution of the new test, based on a limited statistic, is obtained by means of inverse Mellin integral transformation when the k sample sizes are equal odd integers. A Monte Carlo study comparing the new test with direct competitors such as Bartlett's and Cochran's tests is also performed. Results show the superiority in power of the new test against the Cochran's test in almost any investigated cases. The proposed test is also more powerful than the Bartlett's test especially when one variance among the k tested is smaller than others.