Covatiance parameter estimate in Glimmix

Yukosmile — Tue, 07 Aug 2012 19:37:35 GMT

Hello,

I've been caught up in an issue in Proc Glimmix. For the same data, the same model and the same procedure, but I get different covatiance parameter estimate tables.

The treatment has 3 levels.

The model is :

proc glimmix data=analysis initglm method=laplace;

class trial rep treatment;

model event/group_total=treatment;

nloptions tech=newrap maxit=400;

random trial rep(trial);

run;

Covariance parameter estimates
Cov Parm Estimate Standard error

trial	0	.
rep(trial)	0.3120	.

Type III Test of Fixed effect

effect Num DF Den DF F value

treatment	2	24	53.62	<.0001

However,if I sort the data by treatment first and re-run the glimmix procedure, I have different above tables:

Covariance parameter estimates
Cov Parm Estimate Standard error

trial	0	0.2204
rep(trial)	0.3109	0.4130

Type III Test of Fixed effect

effect Num DF Den DF F value

treatment	1	24	162.42	<.0001

The df for numerator is 1. But the treatment has 3 levels, df should be 2.

Can anybody explain what causes these different results? Is it because the model is over-specified? Thanks a lot!

Re: Covatiance parameter estimate in Glimmix

SteveDenham — Wed, 08 Aug 2012 12:33:34 GMT

Get a ticket opened with Tech Support right away. Without access to your data, I can't really say what is going on, but that dropping of the treatment df makes me wonder if something has gone awry with the sorting, so check the log for that. Also check the stuff that GLIMMIX reports early on, such as number of observations, number of columns in the X and Z matrices, etc.

I admit, something really odd is happening.

Steve Denham

topic Re: Covatiance parameter estimate in Glimmix in Statistical Procedures

Covatiance parameter estimate in Glimmix

Re: Covatiance parameter estimate in Glimmix