I´m relatve unexperienced using proc mixed.
My problem is, that running lsmeans in my reduced model results in the message Non-est. instead of an estimate.
Don´t understand why.
Can anyone help, please ??
Repeated measurement of the variable region at mri_id, wbvcalc as covariante.
Proc Mixed Data=kerstin1 method=ml covtest scoring=2;
Class mri_id sex0F1M region hemi agegroup ;
Model response = region sex0f1m hemi age WBVCalc
region * age
region * wbvcalc
/noint solution ;
repeated / Subject=mri_id type=cs group = region ;
lsmeans region / pdiff;
Nonestimability generally implies that at least one of the cells of the 2nd level interactions or higher is unpopulated. Even if these interactions are not included in the model, this can occur. I would go back to your output from fitting the full model, and examine which combinations of your class variables are not included in the lsmeans statement.
If there are "missing" combinations, the usual approach is to fit a means model (see Milliken and Johnson, Analysis of Messy Data, vols 1 and 3. From there, using ESTIMATE or LSMESTIMATE statements you can construct the appropriate means and differences in which you are interested.
You may also want to check your REPEATED statement. The way I read it currently, region is NOT identified as a repeated factor, but as a group factor, giving separate variance estimates for each region. I think you might want:
REPEATED region/subject=mri_id type=cs;
If you truly want different variances for each region, you might try: