Abstract
The large sample test for testing equality of variance ratios in 2 comparative experiments is found to be fairly insensitive to moderate non-normality. The comparision of upper 5% points calculated by asymptotic theory results with the exact values of Schuman and Bradley (1959) test based on normal theory results indicates that the percentage points do not seem to be very much affected by moderate values of Kurtosis, the usual measure of the effect of non-normality. In practical situations, Fisher''s z-transformation can be applied to test the equality of variance ratios without making normality assumptions about the usual random effects linear models.

This publication has 1 reference indexed in Scilit: