I have been trying to run a MANOVA with one IV (factor) where there are 6 categories and 24 summed questionnaire items (DVs) where responses all range from 4-16 (only two items were correlated above a .8 so I left them all in). I have 277 overall respondents and at least 28 in each of the individual categories. I'm using SPSS 20 to run the test. When I run the procedure, I get the descriptives, multivariate and univariate tests, etc., but for Box's I get this message: "Box's Test of Equality of Covariance Matrices is not computed because there are fewer than two nonsingular cell covariance matrices."

I have consulted Pallant's SPSS Survival Manual and conducted an online search which led me to conclude that this problem happens a lot and that many others have been frustrated by what to do next. In my search, I found a discussion on this discussion list from 2007 about something similar (http://www.talkstats.com/showthread.php/3574-MANOVA-and-Box-s-test) and saw an answer in another location online by Paul Swank where he said: "Box's test uses the determinants of each groups variance covariance matrix relative to the total variance covariance matrix, if I remember correctly. Thus, at least one of your eight matrices is not positive definite. Because the overall Manova uses the pooled matrix, you still get results. You may want to do a factor analysis of each groups matrix to identify which one is the culprit." Unfortunately, I don't really understand what he's suggesting might be done (the portion I've highlighted in bold). Can anyone interpret this for me? If I do this (run a factor analysis of each groups matrix), what do I look for (to determine the culprits) and what do I do with the culprits once I find them?

Thanks for whatever help you might have for me.

Stephanie