Hi all, I am running a survival analysis in PHREG in which I am interested in predicting survival from the interaction of 2 categorical variables, each with two levels (treatment = control or experimental; size class = small (S) or large (L)). I am using version 9.4 I first ran the code: proc phreg data=males;
class treatment size_class;
model exposuredays*event(0)=treatment*size_class/ties=exact;
run; I then decided I wanted to set the reference level for each class variable to be able to make more intuitive comparisons and ran the following code: proc phreg data=males;
class treatment (ref='control') size_class (ref='S');
model exposuredays*event(0)=treatment*size_class/ties=exact;
run; I thought the p-values, AIC values, etc would be same from each, but they are not even close. I found some lecture notes online which said that without reference coding the model "estimates the difference in the effect of each level compared to the average effect over all four levels" and that with reference coding the model "estimates the difference in the effect of each level compared to the reference level." (http://www.misug.org/uploads/8/1/9/1/8191072/bgillespie_phreg.pdf) However, I don't fully understand 1) what this means, or 2) which specification is most appropriate for my analysis. It seems to me that with two categorical levels for each effect, using referencing would be more appropriate, but as I don't fully understand why there is a difference in the first place, I am not very confident about that. I feel like this must have come up before but I didn't see it in previous posts. Can anyone help me to understand? Thanks in advance for any suggestions!
... View more