topic Re: PROC GLIMMIX did not converge in SAS Programming

PROC GLIMMIX did not converge

bknight — Tue, 19 Nov 2019 22:03:59 GMT

Hi,

I am attempting to run logistic regression using glimmix to account for clustering by individual school level. My code is:

proc glimmix data=wcc;
class school;
model lloc_4r = perception sex county schooltwo schooltwo*perception/link=logit dist=binomial;
random intercept / subject= school;
run;

When i run the code, the output tells me the model did not converge. This only happens when school is added. There are 21 different schools and I expect clustering effects for them. Does this mean something is wrong with my syntax or the dataset, and is there anything I can do about it? My assumption is that it might be related to how the schools are coded/formatted in the dataset which is by their school code (i.e. 0008, 0250, 0915, etc.). Any help would be appreciated.

Brad

Re: PROC GLIMMIX did not converge

Ksharp — Wed, 20 Nov 2019 11:59:07 GMT

Maybe you don't have a good initial / start value for parameter estimator.

Try using proc logistic to get the parameter estimator and feed it into proc gliimmix.

Or @Rick_SAS might have good idea.

Re: PROC GLIMMIX did not converge

Rick_SAS — Tue, 26 Nov 2019 15:48:26 GMT

Is there a note in the log that says why the convergence did not occur? If so, please post.

I wrote a blog post the discusses one reason for non-convergence (estimate of G matrix not positive definite), but the article links to two excellent papers that explain additional reasons why a mixed model might not converge...and what to do about it.