topic Re: GLIMMIX macro and overdispersion in Statistical Procedures

GLIMMIX macro and overdispersion

Blase — Wed, 26 Mar 2014 12:19:09 GMT

Dear All,

Earlier I have used the GLIMMIX macro in an analysis and I was confident that it automatically dealed with overdispersion, so I did not checked whether our data was overdispersed or not. However, now a colleague of mine asked whether our data was overdispersed and if we have dealt with it. I tried to find a reference saying that GLIMMIX macro automatically deals with overdispersion, but I failed. Can anyone tell me such a reference?

Thanks in advance,

Blase

Re: GLIMMIX macro and overdispersion

SteveDenham — Wed, 26 Mar 2014 16:24:14 GMT

I do not believe that a reference such as that exists. It has been a long time since I used the macro version of GLIMMIX, but I believe the only way to handle overdispersion was through the use of a REPEATED statement.

Steve Denham

Re: GLIMMIX macro and overdispersion

Blase — Wed, 26 Mar 2014 16:31:39 GMT

How can one control for overdispersion with the repeated statement? We had one measurement for each subject.

Re: GLIMMIX macro and overdispersion

SteveDenham — Wed, 26 Mar 2014 16:45:13 GMT

In PROC GLIMMIX, you would add:

random _residual_/group=groupvariablename;

But that is not the issue, right now. Go back to your original request. You need to get SAS for Mixed Models, and be sure it is the FIRST edition. Chapter 11 deals with %GLIMMIX. On page 457 there is a section called "Failing to Account for Overdispersion", where it tells you how to fix the extra-dispersion parameter at a fixed value. The first sentence of this section says:

"To see what happens if you do NOT accunt for overdispersion, you can override the automatic estimate of the extra-dispersion parameter 'phi'."

So, trust in your prior belief and cite:

Littell, R. C. Milliken, G. A., Stroup, W. W. & Wolfinger, R. D. (1996). SAS System for Mixed Models, Cary, NC: SAS Institute Inc.

Steve Denham

Re: GLIMMIX macro and overdispersion

Blase — Wed, 26 Mar 2014 21:39:55 GMT

Thank you very much. This is what I needed.

Blase