topic PROC GLIMMIX in Statistical Procedures

PROC GLIMMIX

AJ8 — Thu, 20 Apr 2017 13:21:04 GMT

Running PROC GLIMMIX on two different occasions with the same data yield two different estimates for "Fixed Intercept", while eveyrthing else (Covariance estimates/OR/ CL etc.) stayed the same.

This is the second time it happened, the first time was a year ago with a different set of data.

I would really APPRECIATE if someone tells me what is going on.

Re: PROC GLIMMIX

Rick_SAS — Thu, 20 Apr 2017 14:01:28 GMT

I assume that you upgraded to a newer version of SAS/STAT between runs?

Re: PROC GLIMMIX

AJ8 — Thu, 20 Apr 2017 14:06:17 GMT

Hi there!

Thank you so much for your reply! I don't think that I upgraded, but I am running SAS server. Does that automaticlaly upgrade my SAS? Even so why would the estimate varies?

Re: PROC GLIMMIX

Rick_SAS — Thu, 20 Apr 2017 14:37:15 GMT

You can find your SAS version by running

%put &=SYSVLONG;

on both systems.

You can get different answers for ill-fitting problems when you change releases or operating systems (Windows <--> Linux). The problem usually indicates that the likelihood function is very flat and/or that the model does not fit the data. For a discussion, see this thread:

Re: PROC GLIMMIX

AJ8 — Thu, 20 Apr 2017 14:32:02 GMT

THANK YOU SO MUCH! I can't open the thread though, it is invalid.Do you mind providing me the link again? THANK YOU!

Re: PROC GLIMMIX

Rick_SAS — Thu, 20 Apr 2017 14:38:30 GMT

I think I had a 'cut-and-paste' error. I have edited and corrected the link in my previous response.

Re: PROC GLIMMIX

AJ8 — Thu, 20 Apr 2017 16:59:23 GMT

THANK YOU!