Stata vs. SAS Differences in Repeated Measures Anova

[David Mangen]

I have uncovered an interesting issue, and am curious if anybody has an explanation for what I am seeing.

I am running a repeated measures model with two time periods, one factor, and three continuous level covariates. For a number of reasons I have estimated these models in both SAS and Stata. Here is what is curious. For all of my within-subject effects, the SAS Type III SS and associated F-tests are identical to the Stata reported Partial SS. This is as expected -- all is fine.

The between-subject effects are, however, a different matter. While the SS associated with the factor variable is identical, and the residual SS is also an exact match, the SS attributed to the covariates differ considerably.

Has anybody else encountered this difference?

Thank you.

[Joseph Coveney]

Has anybody else encountered this difference?

Yeah. Stata is the odd one out, here.

I looked into it back then, and you can match the sums-of-squares manually, but it's not ideal. If you've got continuous between-subject covariates, you might be better off with -mixed . . . , reml dfmethod()-.

[David Mangen]

Thank you Joseph. I appreciate your response and your recommendation.