Don <[EMAIL PROTECTED]> wrote in message [EMAIL PROTECTED]">news:[EMAIL PROTECTED]... > Greetings, > > I'm aware that there are various assumptions underlying the use of an > ANOVA 1 model e.g. homogenity of variance and normality within each > group. What is the best way to test these?
I probably wouldn't test either formally. I'd at least look at a normal probability plot (a q-q plot or a p-p plot), and look at the within group variances to get some idea of the relative deviations from assumptions. Some people do use something like Levene's test for the homogeneity of variance assumption (there are other tests). I haven't seen much discussion on the effect on type I error and power of the full procedure "test homogeneity of variance, if not rejected, do anova, otherwise do <whatever else>" which is presumably the idea most people have when doing the test. For that matter, it'd be useful to see studies of the effects of the various approaches advocated in different books. Glen . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
