04-08-2015 11:10 AM
I'm looking at longitudinal data with proc mixed and random intercepts. Each subject has 3 measurements (baseline, short term, long term). There are two independent variables (time and surgery type), along with an interaction term (see code below).
When I run the model with random intercepts, the Solution for Fixed Effects shows an estimate, standard error and t-value for the intercept, but no degrees of freedom and (obviously) no p-value.
When I run the same model without random intercepts, then the SFE reports the appropriate degrees of freedom and p-value.
There are 21 subjects in the dataset, each with 3 measurements. Is this an overparameterization issue?
proc mixed data=work.mydata order=data ;
class surgery time;
model outcome_score = surgery|time /solution ;
repeated /subject=mrn type=un r rcorr;
With random int, estimate for intercept = 57.2, SE = 7.1, DF = 0, t Value = 7.99, p-value = .
Without random int, estimate for intercept = 57.2, SE = 4.9, DF = 19, t Value = 11.66, p < .0001
Again, I think this is an overparameterization issue, but I'm not sure. Any input would be appreciated.