The behaviour of the modified Levene's test when data are not normally distributed (Q638048)

The author considers tests for homogeneity of variances in one-way ANOVA problems. These tests include the original \textit{H. Levene} [Contrib. Probab. Stat., Essays in Honor of H. Hotelling, 278--292 (1960; Zbl 0094.13901)] test (LT) with absolute deviations from means, trimmed means and medians. LT with the \textit{T.K. Keyes} and \textit{M.S. Levy} [J. Educ. Behav. Stat., 22, 227-236 (1997)] correction for the deviations. The \textit{F.E. Satterthwaite} [Biomet. Bull., 2, 110-114 (1946)] correction of the degrees of freedom, the \textit{G.A. Milliken} and \textit{D.E. Johnson} [Analysis of messy data. Boca Raton, FL: CRC Press (2009; Zbl 1168.62001)] correction of the significance level and the Bartlett test. A bootstrap-type technique for p-values computation is discussed for some versions of the LT. Performance of these tests is compared via simulations of contaminated normal samples.