Given the output you presented, I can only say to try the following. You can fix the fixed effects in both MIXED and GLIMMIX by using the workaround I proposed earlier of subtracting x'beta from the observed value, where you set the coefficients in beta to generate your profiles. I suspect you will need to fit 20 or more of these datasets to get a decent plot. The MODEL statement would be for an intercept only model (e.g. MODEL ynew=;, with no fixed effects specified). You probably do not want to profile the random effects at the same time (well, perhaps you do, but that is another step), so you will need to add a noiter option to the random statement, and input the values you wish to do the profiling at. If you do want both at the same time, you'll get a higher dimensional response surface. Use ODS OUTPUT to save the final log likelihood for each of the runs, reshape the data for plotting, and you should have something close to what you want.
SteveDenham
... View more