10-09-2014 04:59 PM
I've a data set including two fixed factors (house, strain) with 3 levels of each. The response variable is a lameness score of birds with scores ranging from 0 to 4 (0, 1, 2, 3, 4). I wanted to know if there is a effect of house and/or strain for each score separately. I considered using glimmix with following code
proc glimmix data=work;
class house strain;
model score=house strain/dist=multinomial link=glogit ;
estimate 'A vs B' house 1 -1/ bycat;
estimate 'A vs C' house 1 0 -1/ bycat;
estimate 'B Vs C' house 0 1 -1/ bycat;
estimate 'D vs E' strain 1 -1/ bycat;
estimate 'D vs F' strain 1 0 -1/ bycat;
estimate 'E vs F' strain 0 1 -1/ bycat;
However Sas keeps giving me wrong degrees of freedom for main effects and i think the whole analysis was wrong. It would be of great help if somebody could tell me what i'm doing wrong or what would be the appropriate code...
10-14-2014 07:54 AM
Well, assuming no interaction, something odd is happening. By specifying link=glogit, you should be getting nominal data, and the estimates should be correct--if everything else is working. So some design questions: I assume 3 houses and 3 strains. How many observations are available? What degrees of freedom are coming up in the output? Maybe with those we can identify the source of the oddness.