I'm trying to set up a model which consists of two measurements over time nested within patients nested within providers nested within clinics. The analysis is a time by condition analysis with clinics randomized to a study arm. I've set up a 3 level model with time, patient, provider, following Murray's GRT book as:
proc mixed data=MixedCBIUni info order=internal covtest noclprint;
class time patient provider arm;
model a1clast=arm time arm*time / ddf=38,38,38 ddfm=res solution;
repeated time/ type=cs subject=patient(provider*arm);
random int time/subject=provider(arm);
lsmeans arm*time / slice=arm slice=time cl e;
I haven't seen applications with a 4th level (in this case: clinic) and am unsure how to set up the subject part of the repeated and random statement. The code above does not seem intuitive to me, but results agree exactly with a run in MLWin of this same model.