I have run a mixed model in SAS using Proc Glimmix and I have obtained different results depending upon whether I use G-side or R-side random effects in the model. The random effect I fitted was ‘study’ for a meta analysis. What I would like to know is why choosing between two types of random effect variances would yield such large differences in the least squares means in the output? Also if someone could explain the implications of choosing the G matrix over the R matrix for the variances of the random effects I’d be grateful. How would one know which type of random effect to fit to the model in advance?
Many thanks,
Lucy Wright
---------------------------------
Yahoo! Answers - Get better answers from someone who knows. Tryit now.
|