You probably have data entry errors. Did you check for that? David Booth

BTW MANOVA is almost never used for anything today.

David Booth

Hello Hanan,

If your six DVs comprise five subscores and one overall / total score, then you have a linear dependency (e.g., the total score is a composite of the individual subscores), and any statistical package that attempts to invert the covariance or correlation matrix for that set of scores should note the same problem. In such instances, just drop the total score from the set, as it is redundant of information that is contained in the subscores.

If the dependency isn't obvious to you (as in the case above), you can always evaluate collinearity indices for the set of six scores, by entering them as IVs in a regression analysis (use something else as a DV...doesn't matter what it is, so long as it has some variation). VIF or tolerance values will help you identify the specific culprits among the six scores which are too dependent on/redundant with the other scores.

Good luck with your work.

Thank you very much to you both !

David Morse Seeing as the total score is important for my analysis would there be any other way to still be able to incorporate it without affecting the homogeneity of variances? If not, would Pillai's Trace be an acceptable criterion for this as it is supposed to robust against violations of the assumptions of MANOVA?

Further to David Eugene Booth's comment that MANOVA is not used much nowadays, see this nice article by Francis Huang.

Article MANOVA: A Procedure Whose Time Has Passed?

The classic article by Huberty & Morris (1989) is also one that people who think they want to use MANOVA should read, IMO.

Article Multivariate Analysis Versus Multiple Univariate Analyses

HTH.

Hello Hanan,

In order to run a multivariate analysis, and include the total score, you'll have to be willing to omit one of the five subscores in the analysis. The only other option is to analyze the total score by itself.

Good luck with your work.