byLinkachu04-15-201607:47 AM - edited 08-23-201602:35 PM
It would be great if the SELECTION= option in PROC PHREG gave you a choice of using the p-values from model-based variance, sandwich variance (if specifying COVS(AGGREGATE) with the ID statement), or from a frailty model (if using RANDOM statement).
Currently, the SELECTION= option only uses the model-based variance estimates. This can be quite misleading if you have also specified the COVS(AGGREGATE) option with the ID statement, because the model-based variance p-values (which are being used in the selection process) are not displayed unless you also specify the DETAILS option in the MODEL statement! I recently used PROC PHREG to do a stepwise model selection and this issue was the source of confusion over why a factor with P=0.10 remained in my final model when I had set SLE= and SLS= to 0.05 (the p-value for the model-based variance estimate was 0.02).
It would also be great to allow selection with frailty model (specifying the RANDOM statement).