Hello,

If I'm correct, under "options" for Repeated Measures procedure in SPSS you can select "homogeneity tests". In the output, Box M represents the homogenity of variance for all variables taken together.

When I perform my analysis however, instead of box M, I get the following output message:
"Box's Test of Equality of Covariance Matrices is not computed because there are fewer than two nonsingular cell covariance matrices."

My question, what does this mean? Why can't I get a homogeneity of variance test for all variables taken together?

Thanks in advance for taking the time!
Pieter

Pieter,

Did you finally find out what the cause of that message was? I'm having the same problem, with a dataset that has 3 IV's with 2 levels each, and 2 DV's (questionnaire items, 32 items/levels total).

The message is not actually an "error" message, saying there is something wrong with the data. It seems to indicate that the test cannot be run with the type of data I (and you) have, but it doesn't tell you what the problem is in plain English (what is a "nonsingular cell covariance matrix" anyway?). It does run the rest of the tests (Levene, univariate, etc.), but without Box's, you cannot really claim (according to some people) that your data is valid.

I checked the SPSS website and they don't actually provide statistical support, only technical support (to install a program, uninstall it, etc.). So they wouldn't clarify what that message means :(

Can anyone recommend another package that would take a .sav dataset (i.e., the SPSS format) and run these kinds of tests (a MANOVA including a Box's test)?

Thanks,

Roberto

I think using the pooled covariance matrix would solve the problem.If I recollect my memories correct, this won't affect the rest of the analysis

Given that all the other tests (Levene's, post-hocs, etc.) are being computed, someone suggested that SPSS must be using the pooled (total) variance covariance matrix. If I could demonstrate (with any method, Box's or not) that those values are valid/reliable, then I could go with those tests. But as I mentioned, Box's (or an equivalent) is expected as part of this analysis.

Hello. I was wondering if there is any more follow up with the Box M issue. I am getting the following message as well: "Box's Test of Equality of Covariance Matrices is not computed because there are fewer than two nonsingular cell covariance matrices." Can anyone help with this? I know if was an issue for people a few years ago. Thanks!