How robust is the multivariate analysis to the assumptions of normality and equality of variances?


Let me to expose this questions with a simple example: A repeated measures ANOVA with two between-subject factors (Grade -3 levels-, and Sex -2 levels-), and a within-subject factor (Source -2 levels: teachers and parents) was conducted. N = 258. Data were collected using 3 subescales, i.e. 6 DV: IN, HIP, and TOT subescales for teachers and parents.

The assumption of normality is not met for any of the 6 DVs and, in addition, Levene's is significant for all of them. Does this mean that I must use nonparametric procedures or can I follow the multivariate path? (it should also be noted that the subsamples are very similar in all pairwise comparison -in no case> 1.5).




Fortran must die
It depends on the numbers of cases. Once you have several hundred normality and equal error variance are normally not going to matter.
Thanks for the reply. The total sample is 258 participants, but when divided by grade and sex, the number of participants in each cell is much smaller. To be more specific, this is the distribution by cells:


Is it the multivariate analysis (i.e. source x grade x sex by 6 DV ) still robust with this sample?