Calcite | Level 5

Glimmix did not converge using countdata with a lot of zero values and right-censored.

Dear all

I ran into problems with my Glimmix-syntax if the dependent variable is right-censored and there are a lot of zero-values (a histogram is attached). For example, that's the case if the dependent variable is the number of nursing days in a year.

The current syntax (which works perfectly for other dependent variables):

``````Proc glimmix data=FINALEIGHT oddsratio  noclprint ;
Class  Complexity (ref="1") H18 Gender (ref="1") Age_category2 (ref="1");
nloptions maxiter=50;
Model  nursing_N = Age_category2 Complexity  Gender year2/ddfm=satterthwaite solution link=log dist=negbin cl;
random _residual_ / subject=H18 type=cs;
estimate 'gender female vs male' Gender 1 -1/ilink cl;
estimate 'complexity severe vs moderate' Complexity -1 1 0/ilink cl;
estimate 'age +40 vs 30-39' Age_category2 -1 1 0/ilink cl;
run;``````

I also attached a screenshot of the output. The convergence process stops already after a few iterations.

Is it possible to solve this problem? I have been searching all over the internet but with little result.

Thank you all.

Kind regards

Ruben

2 REPLIES 2
SAS Super FREQ

Re: Glimmix did not converge using countdata with a lot of zero values and right-censored.

Hello,

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

Paper 332-2012

Tips and Strategies for Mixed Modeling with SAS/STAT® Procedures

Kathleen Kiernan, Jill Tao, and Phil Gibbs, SAS Institute Inc., Cary, NC, USA

SECTION II: TROUBLE SHOOTING CONVERGENCE FAILURES IN MIXED MODELS

pp. 8-9

Cheers,

Koen

Calcite | Level 5

Re: Glimmix did not converge using countdata with a lot of zero values and right-censored.

Thanks for your reply Koen. Apparantly, I had to divide the values of the dependent variable so they were all between 0 en 10. Then it converged perfectly.

Kind regards

Ruben

Discussion stats
• 2 replies
• 531 views
• 0 likes
• 2 in conversation