07-24-2017 03:33 PM - edited 07-24-2017 03:33 PM
I'm analyzing roughly 300 subjects with 36 clusters. Both my explanatory and dependent variable are continuously distributed. I'm using proc genmod to adjust for the repeated measurements within the clusters. In my baseline model of this association, I see that the algorithm converged but also a note/warning about convergence. I get parameter estimate and they are within my expectation so there doesn't seem to be anything unbelievably wrong with the results, but the warning still concerns me.
proc genmod data=data; class cluster; model outcome = average_score / dist=normal link=identity; repeated subject=cluster / type=cs; run; NOTE: Algorithm converged. NOTE: No convergence of GEE parameter estimates after 50 iterations. WARNING: Iteration limit exceeded.
After running this base model I add covariates in and have no issues with convergence or iterations. There are then two "NOTE: Algorithm converged." lines in my log. This is contrary to what I would think since I'm adding more parameters to the model to estimate.
I also have run some sensitivity analyses using a weighted average score and that model (with and without covariates) has no convergence issues.
Any ideas whats up with this?
07-25-2017 08:33 AM
Make the number of iterative bigger ?
proc genmod data=sashelp.class; class sex; model weight = height / dist=normal link=identity maxiter=100; repeated subject=sex / type=cs; run;