topic HPlogistic variable selection in Statistical Procedures

HPlogistic variable selection

shutfle1 — Mon, 29 Jul 2019 16:58:54 GMT

I am using a selection procedure within hplogistic on partitioned data. I have set the entry to .001 and the stay to 0.01 and choose=validate. However, I get some variables (that were not forced in) that have p>0.01. Why is this happening?

PROC HPlogistic DATA=inpatient_outcomes;
PARTITION ROLEVAR = PART(TRAIN="1" VALIDATE="0");
model outcome (event='1')= var1 var2.....var30/ include=4;

SELECTION METHOD=stepwise (CHOOSE=VALIDATE sle=0.001 sls=0.01);
run;

Re: HPlogistic variable selection

Ksharp — Tue, 30 Jul 2019 12:36:01 GMT

Don't use STEPWISE selection . That would get you bigger Type I ERROR.

Try LASSO or ADPTIVE via PROC HPGENSELECT .

Re: HPlogistic variable selection

StatDave — Wed, 31 Jul 2019 15:26:57 GMT

This can happen because a step in the selection process consists of adding or dropping a single variable. Note that if a variable is added it is entirely possible that one or more of the other variables could have their p-values change to be larger than the SLS= criterion. Since the models at all of the steps are evaluated by the CHOOSE= criterion and one is selected as the final model, that model could be one where one or more of its variables has p-values greater than the SLS= criterion. If you want to ensure that this doesn't happen, you need to remove the CHOOSE= option so that the model in the last step is chosen.