TESTS FOR HOMOGENEITY OF VARIANCE IN FACTORIAL-DESIGNS

The variance portion of P. A. Games's (See PA, Vol 62:2685) 3-factor model of inference on independent groups is extended. Six procedures that convert tests of spread into tests of location are reviewed and explored in a Monte Carlo study of how to test variances in a factorial design. The statistic In s-2 is shown to be a slight improvement over the procedure outlined by J. E. Overall and J. A. Woodward (see record 1975-04313-001). The dependence of these 2 tests on the normality condition is illustrated. Four robust alternatives of somewhat lower power are contrasted. The jackknife test is the most powerful and is only slightly sensitive to leptokurtosis if the ns are equal. The median test described by M. B. Brown and A. B. Forsythe (1974) is acceptable, but it uses average deviations rather than variances. The Box-Scheffé test is always robust. No single test is ideal. A 2-stage process is recommended. (31 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved). © 1979 American Psychological Association.