Significant output in Levene's test for equality of variances in MANOVA; what to do?

Question

I want to perform a MANOVA in SPSS as follows: i have a independent variable that consists of three groups (- of which the group sizes are not equal), and six dependent variables.

When checking the Levene's test for equality of error variances, there are two dependent variables in which the test is significant; so, the error variances are not equal. Because of the significant Levene's test output, my thoughts are that I can't perform this analysis, and have to alter my type of analysis. I know that in a univariate ANOVA, we have the Brown-Forsythe's F and Welch's F which adjust the F-test for inequality of variance. But, in the MANOVA, there seems no such option.

My question is what options I do have. Should I instead perform several ANOVA's (and by doing so omitting the power of a MANOVA)? Or should I eliminate the two variables from the MANOVA and run it with only four dependent variables?

Answer 1

First, make sure you look at the boxplots of the residuals instead of just using Levene's tests. The significant result could be due to outliers, a bimodal distribution, or skewness that you may need to address.

To get equal variances, try log-transforming (ln) your dependent variables before you run MANOVA.

Also, there are four test statistics that can be used in MANOVA. Hotelling's trace and Pillai's criterion are the least affected by violations in assumptions, but Wilk's is the most commonly used.