topic Re: Proc Mixed with Group Level outcome in Statistical Procedures

Proc Mixed with Group Level outcome

smrose — Sun, 08 Apr 2018 21:32:13 GMT

I'm trying to model a group level attribute that is only measured once during study using individual level predictor that was measured anywhere from 1 to 111 times/per person with 660 observations and 60 people. How do I specify that my outcome is a group level attribute that has only been measured once per subject? I think with code my outcome is being analyzed as if it were dis-aggregated across all time points.

proc mixed data = analysis.labshmp9 noclprint empirical ;
class subjectid ;
model sspg = shannon | shannon /solution ;
repeated / subject = subjectid ;
run;

Re: Proc Mixed with Group Level outcome

smrose — Wed, 11 Apr 2018 12:28:36 GMT

So after thinking about it some more I realized the repeated statement did not make sense in the context because my outcome was not repeated. That meant to specify my subgroups I needed to use the random statement. The only thing that could randomly varies in the model are the level 1 observations. So I ended up fitting a model like this:

proc mixed data = dataset noclprint namelen=32 covtest plots = all ;
class subjectid ;
model SSPG = shannonc /solution ddfm = bw cl ;
random shannonc /subject = subjectid;
run;

the degrees of freedom were reasonable.