topic Re: ordinal mixed effects model did not converge in PROC GLIMMIX in Statistical Procedures

ordinal mixed effects model did not converge in PROC GLIMMIX

Tiny_Kane — Sun, 15 Dec 2019 15:53:25 GMT

Hi there,

I was trying to fit an ordinal mixed effects model using PROC GLIMMIX procedure. The code is as below. The outcome Y has 3 ordered value (1,2,3). Y_baseline is the outcome measured at baseline time. The model includes random intercept and random-effect of time, and the baseline outcome as covariate. The data includes 53 subjects with 181 observations. However the model did not converge, even if I combined the three values into 2 values and run regular logistic mixed effects mode, it still did not converge. Is it due to that the sample size is too small? Is there any solution? Thank you in advance.

PROC GLIMMIX data=final1;
CLASS pt Y_baseline(ref=first);
MODEL Y(desc order=internal)= Y_baseline days/DIST=multinomial LINK=cumlogit SOLUTION CL;
RANDOM int days/subject=pt;
RUN;

Kane

Re: ordinal mixed effects model did not converge in PROC GLIMMIX

PaigeMiller — Sun, 15 Dec 2019 16:00:39 GMT

Please read this article by @Rick_SAS

https://blogs.sas.com/content/iml/2015/08/26/convergence-sim.html

Re: ordinal mixed effects model did not converge in PROC GLIMMIX

Tiny_Kane — Sun, 15 Dec 2019 16:22:38 GMT

Proc glimmix does not have OUTTEST= option.

And the ConvergenceStatus=CS did not output any information, only one row data, reason=Did not converge., status=1, pdG=1. That's it.

Re: ordinal mixed effects model did not converge in PROC GLIMMIX

PaigeMiller — Sun, 15 Dec 2019 16:52:42 GMT

http://support.sas.com/resources/papers/proceedings12/332-2012.pdf

https://communities.sas.com/t5/General-SAS-Programming/GLIMMIX-convergence-problems/td-p/119472