05-13-2015 03:27 PM
I notice that proc PHREG can output the estimated survival rate or the cumulative hazard rate; however, both calculations requires a specification of the baseline, i.e. lambda0(t) or S0(t), while neither of them is pin down by the partial likelihood function.
The description of the likelihood used in PHREG can be found here (The section of "Partial Likelihood function" in "Details about the proc PHREG")
The computation for the survival rate is here (The section of "Output Statement" in proc phreg):
As you can see, S0 shows up in the survival rate calculation, but I doubt if it matters for the optimization or estimation.
Can someone give me more details on what is going on behind the scene.