07-09-2015 02:21 PM
In case of Proc Lifereg modeling, is there a way of scoring a validation data set using the parameter estimates obtained from a training data set, something similar to Proc score?
And how do we compare the model performance between the validation and training data sets, something along the lines of gains/lift charts/tables?
08-13-2015 04:36 PM
Proc lifereg does include the distribution (_DIST_) when it writes out coefficients, but I don't think proc score can use _DIST_ to create estimated values from coefficients from LIFEREG.
Looking at statistics references it is not easy to figure out how to simulate models created with PROC LIFEREG. That suggests it may be simpler to use something else. I tried to simulate an example from a SAS website (SAS/STAT(R) 13.2 User's Guide) in Excel using information about the Weibull distribution to verify that I could match predicted and actuals from their example, but I could not do so.
Note to SAS Support: This person is right. This is someting not in SAS that is needed for PROC LIFEREG to be useful.
THIS IS SOMETHING MISSING FROM SAS WITHOUT A FIX DOCUMENTED BY SAS!!!!!!!!!