topic Re: Does proc glmselect support random effects in Statistical Procedures

Does proc glmselect support random effects

RobertF — Sat, 07 Apr 2018 18:24:28 GMT

I'm building a predictive model that predicts future hospital length of stay. The dataset is episode level, including multiple episodes for many of the members. I'd like to use a penalized GLM (such as a lasso) while at the same time including a member-level random effect.

I know proc glmselect does not have a "random" statement. Is there a feature I'm missing or a workaround that allows random effects to be included in a predictive model?

Re: Does proc glmselect support random effects

Ksharp — Sun, 08 Apr 2018 12:51:30 GMT

Answer is no.

But you could use interact term to simulate RANDOM effect.

model y=x1 x2 x1*x2 x1*x1 x2*x2 ;

Re: Does proc glmselect support random effects

RobertF — Sun, 08 Apr 2018 21:13:27 GMT

Thank you Ksharp - I could treat the Member ID # as a fixed effect and model it's interactions, but wouldn't that mean having thousands of Member ID # dummy variables (and using up lots of degrees of freedom)?