SAS code for Latin square design including repeated measures

[pdiff]

Perhaps I'm still fuzzy from New Years celebrations, but why is a "treatment*period" term present in Proc GLM code, but now Proc Mixed code?

TIA for any insights.

[sld]

Can we assume that the GLM and MIXED code snippets address the same experimental design? Posting the code would probably be helpful. Meanwhile....

If there are replicate squares with the same periods for each square, then you can estimate the treatment x period interaction; see this paper by Rob Tempelman. But that doesn't explain why one code includes the interaction and the other does not.

Maybe the GLM code incorporates period as fixed, and the MIXED code incorporates period as random? 

[pdiff]

Thank you for the reply. The example is apparently from a SE SUG meeting paper given here: https://analytics.ncsu.edu/sesug/2004/SD04-Yarandi.pdf .  It seems to me the interaction should not be present.

[Rick_SAS]

In the first examples in the paper (PROC GLM), the interaction term is not significant at the alpha=0.05 level. I suspect that is why the interaction term was dropped for the second example (PROC MIXED).

[sld]

Because sequence has zero df when the MODEL statement (for both GLIMMIX and the Type III tests in GLM) includes treatment*period, I'm thinking it could be something more fundamental--that sequence and treatment*period are confounded. The zero df result is really obvious in MIXED, less so in GLM because the TEST statement is used to assess sequence in the mixed model and the code in the paper specifies a Type I test for sequence (notably the Type III test does not work). Perhaps this is why the author of the paper changed code between PROCs....