Issues with Proc GLIMMIX Intercept/LSmeans

Nebulus — Wed, 07 Mar 2018 01:55:01 GMT

I'm running a GLMM with a binomial error distribution on proportion parasitism data values which range 0-1 and are not transformed.

I'm having a problem where I'm getting a negative intercept and thus my LSMEANS are negative, which shouldn't be the case. Unless it should be the case due to the link function...?

The model code looks like this

proc glimmix data=final ic=q;
class interval climate season site;
model parasitism = interval(season) season climate tot_nymprev / dist=bin link=logit s;
random intercept / subject=site;
random interval / subject=site residual;
random _residual_;
lsmeans season /cl diff plot=meanplot adjust=tukey;
lsmeans climate /cl diff plot=meanplot adjust=tukey;
run;

Proc means gives me the correct values.

Below is the paritial output showing the issue with the intercept and LSMEANS.

Can anyone explain what is happening. I'm completely perplexed at this point. I would expect to get positive LSMEANS for parasitism rates..

Re: Issues with Proc GLIMMIX Intercept/LSmeans

ChrisNZ — Wed, 07 Mar 2018 02:41:06 GMT

Moved to "SAS Procedures" forum.

Re: Issues with Proc GLIMMIX Intercept/LSmeans

PaigeMiller — Wed, 07 Mar 2018 12:00:45 GMT

These are on the logit scale. You need to un-transform the results.

topic Re: Issues with Proc GLIMMIX Intercept/LSmeans in Statistical Procedures

Issues with Proc GLIMMIX Intercept/LSmeans

Re: Issues with Proc GLIMMIX Intercept/LSmeans

Re: Issues with Proc GLIMMIX Intercept/LSmeans