First question: Is this because...? My answer: Probably. Getting the right error term is critical. This statement says that there are two variance components to consider, due to the design. The first is block*cultivar*year and the second is trt_name*block*cultivar*year. You multiply the subject through the "before slash" terms. Using the subject= syntax imporves performance. Second question: This specification should make it clear that this is the AR + Random effect model. The subject to the repeated G side model is also a term in the random effects. See Littell et al. in the Journal of Animal Science (1998) 72:1216. Third question: I would look at the Poisson distribution as well, using AICC to compare. Also check the overdispersion ratio (chi square to df) in the fit stats. It should be near one. Fourth question: We can address multiple comparisons a lot of ways. The more I think of it, the 0.18 probably comes out of Tukey's method for the number of comparisons available, so that the lines option actually matches up with the adjusted p values. Get a copy of Walt Stroups's Generalized Linear Mixed Models. The preface says that hardest thing about GLMM's is "un-teaching" all of the things learned in the general courses. And I am starting to get it. Steve Denham
... View more