PROC GLIMMIX not giving AIC after adding random effect

raheleh22 — Mon, 17 Jun 2024 19:39:22 GMT

Hi,

I've been using Proc Glimmix to predict the effect of several predictor variables on my outcome where the distribution is negative binomial. for the purpose of model comparision when I run the following code I receive the AIC tests:

proc glimmix data=modelupdated;
class metcat(ref="2") age_new(ref="1") sexnum (ref="2") race(ref="2");
model Rate_12= age_new sexnum race metcat/ dist=nb link=log s;
run;

However when I add the random effect of counties, I do not receive the AIC in fit stati table in the output.

proc glimmix data=modelupdated1;
class metcat(ref="2") age_new(ref="1") sexnum(ref="2") race(ref="2");
model Rate_12 = age_new sexnum race ave_uninsured metcat / dist=nb link=log s;
random intercept /type=un subject=fips_all_digit_num ;
run;

any thoughts on this and how I can get AIC becuse I need to compare the two different models vs eacg other.

Thank you,

Re: PROC GLIMMIX not giving AIC after adding random effect

jiltao — Mon, 17 Jun 2024 23:17:59 GMT

PROC GLIMMIX uses the Pseudo-likelihood estimation method to fit the model when there is a random effect, so no AIC is computed because no true likelihood is computed. To get the AIC, you can use a different estimation method, for example, method=quad or method=laplace in the PROC GLIMMIX statement.

Hope this helps,

Jill

topic Re: PROC GLIMMIX not giving AIC after adding random effect in Statistical Procedures

PROC GLIMMIX not giving AIC after adding random effect

Re: PROC GLIMMIX not giving AIC after adding random effect