03-08-2016 04:47 PM
I am attempting to run a nominal model with Glimmix (SAS 9.3). My data structure has each observation corresponding to one minute of physical activity with many observations for each person. The nominal outcome is the location of that activity (e.g. home, road, park) with 10 categories. I am trying to look for differences in locations of activity by various sociodemographic characteristics (e.g. gender), controlling for the fact that each person has multiple minutes of activity. Here is my basic set up
proc glimmix data=mvpa104aim1c2 method=laplace;
class gender2 newID main3;
model main3 (REF="home")=gender2 /dist=MULT link=glogit solution cl ;
RANDOM intercept / SUBJECT=newID type=vc group=main3;
When I run this, I get a note that the initial estimates did not yield a valid objective function. I tried running binary models for each location vs home and inputting those as starting values in the parms statement. When I run this model, I receive a note that values given in the parms statement are not feasible. Any ideas? The only other note I see is that there are n=196 missing subject effects however no records have a missing value for newID. There are very large sample sizes (thousands) for each location category for the outcome. I have tried a couple of other covariance structures (UN, ar(1)) and increased the iterations too. Any advice is appreciated!