proc glinmixed using poisson distribution

[lefzadra]

I'm trying to analyze count variables (poisson distribution)  with repeated measures using proc glinmix. I have observation of behavior of animals  observed during 3 days after applying a treatment. In this case treatment is "environmental enrichment"  for calves to improve their behavior (two classes, yes or no). Activities (y in my model) is measured by time doing certain activity. I have many "0" time observation in my file, then I'm trying to analyze with GLINMIX using Poisson distribution.

Model i'm using is:

proc glinmixed data=b method=quad;

class treatment day id;                 *id is the identification o animal *;

model eating = treatment + day / link=log s dist=poisson;   *eating is measured in minutes  *;

subject = id;

lsmeans  treatment;

run;

My doubt is: 1) how could I interpret least square means for treatment?; 2) in the case I have many "zero" values for y, sometimes the least square means is negative. Is it correct?

Many thanks,

Lenira

[PaigeMiller]

PROC NLMIXED or PROC GLIMMIX or something else?

[lefzadra]

PROC GLIMMIX

[PaigeMiller · Accepted solution]

Activities (y in my model) is measured by time doing certain activity. I have many "0" time observation in my file, then I'm trying to analyze with GLINMIX using Poisson distribution.

Time involved in a certain activity doesn't seem like a Poisson distribution to me. 

But probably the reason you are getting least squares means less than zero is that you are not using the ILINK option in the LSMEANS statement which should invert the log transformation and then no value can be less than zero.