I have significant results for Levene’s equality of error variance when running a MANOVA. I have seven groups in my predictor variable, and all but one has a significant Levene test result. How does this affect what I can do? I have three statistics books in front of me, all of which report non-significant results in the examples they use, and therefore carry on without suggestion. Does this invalidate my MANOVA? I have a non-significant result for Box’s test of equality of covariance matrices, and small and unequal group sizes (range 5-13).
you could break your analysis into multiple ANOVAs correcting for alpha inflation. you could break it into a polynomial logistic regression but that would mean asking your research question differently. you could also use a several regression analysis with dummy variables if need be. You may want to also consider combining group so that you have fewer groups with larger sizes. check your correlation matrix and see what variables are highly correlated and you can consider combining those groups. personally I would use regression analysis as this is recommended by some authors when you have a small sample with unequal groups. You can assess interaction effects with regression. You may simply be working with to many variables at once. consider the most parsimonious analyses.
I was hoping for a “report the XX statistic, instead” in the way that Welch/Brown-Forsythe is reported in ANOVA when the Levene equality of variance assumption is violated.
I fear that by doing multiple ANOVAs or regression, i leave myself open to the obvious question “why didn’t you run MANOVA” – my take is that problems identified in MANOVA are problems with the data, not problems with the test statistic. Correct me if i’m wrong.
Viewing 3 posts - 1 through 3 (of 3 total)
The forum ‘Default Forum’ is closed to new topics and replies.