topic Re: lasso logistic regression control variables in Statistical Procedures

lasso logistic regression control variables

Levi_M — Wed, 01 Feb 2023 16:55:05 GMT

I am conducting a logistic regression on variables that were selected via LASSO (hpgenselect). I have a question about my methodology.

(1) If I want to control for age and gender, do I exclude them from the lasso selection but include them in the logistic regression? Or do I include them in both the lasso and logistic regression?

(2) Is it appropriate to use hpgenselect and follow up with logistic regression

- Thank you -

Re: lasso logistic regression control variables

StatDave — Wed, 01 Feb 2023 17:11:03 GMT

Yes, you can use PROC HPGENSELECT with METHOD=LASSO in the SELECTION statement. Use the INCLUDE= option to specify the effects you want to keep in the model, for example INCLUDE=(AGE GENDER). If you want the final model to control for age and gender, then those variables need to be forced to stay in the model using the INCLUDE= option.

Re: lasso logistic regression control variables

Levi_M — Wed, 01 Feb 2023 19:17:16 GMT

thank you so much for your quick and informative response.

Re: lasso logistic regression control variables

Amyzlot1 — Tue, 01 Aug 2023 20:14:07 GMT

Can you provide SAS code. I'm getting errors when I add the include= statement.

ods graphics on / LABELMAX=1900;

proc hpgenselect data=cs.analysis_final;
class Hispanic_includes_all_races IDU HCV;
model cs_case = Hispanic_includes_all_races IDU HCV;
selection method=lasso (stop=none choose=bic) details=all
include=Hispanic_includes_all_races;
/* bicplot / plotfit=yes; */
ods output Coefficients=lassocoef;
run;

Thank you!

Re: lasso logistic regression control variables

StatDave — Tue, 01 Aug 2023 20:30:12 GMT

The INCLUDE= option goes in the MODEL statement, not the SELECTION statement. See the HPGENSELECT documentation.

Re: lasso logistic regression control variables

Amyzlot1 — Tue, 01 Aug 2023 21:14:57 GMT

Thank you! This worked!

ods graphics on / LABELMAX=1900;

proc hpgenselect data=cs.analysis_final ;
class Hispanic_includes_all_races age_dich IDU HCV;
model cs_case (event=last) = Hispanic_includes_all_races age_dich IDU HCV / include= (Hispanic_includes_all_races age_dich) dist=binomial link=log;
selection method=lasso (stop=none choose=bic) details=all;
output out=Out xbeta predicted=Pred;
run;

Can you generate r-squared statistics ang plots in proc hpgenselect ? (I've checked the documentation but I'm not finding example code.) Thanks!