topic Re: proc logistic model selection for categorical variable in Statistical Procedures

proc logistic model selection for categorical variable

zzzyyy — Tue, 30 Jan 2018 14:47:40 GMT

my variable pool contains both categorical variables and continuous variable. I plan to use /selection=stepwise to do the stepwise variable selection.

For categorical variables, is it enough to specify them in class statement? will the selection process treated different levels of a categorical variable as one variable? Or I need to specify categorical variable like 'groupnames = 'Height' 'Age' in the proc reg variable selection?

Re: proc logistic model selection for categorical variable

Reeza — Tue, 30 Jan 2018 16:15:56 GMT

will the selection process treated different levels of a categorical variable as one variable

Yes, it will. If you want to treat them each individually you need to make your own dummy variables and in general it's not statistically recommended anyways.

@zzzyyy wrote:

my variable pool contains both categorical variables and continuous variable. I plan to use /selection=stepwise to do the stepwise variable selection.

For categorical variables, is it enough to specify them in class statement? will the selection process treated different levels of a categorical variable as one variable? Or I need to specify categorical variable like 'groupnames = 'Height' 'Age' in the proc reg variable selection?

Re: proc logistic model selection for categorical variable

Rick_SAS — Tue, 30 Jan 2018 19:13:53 GMT

Can you clarify? Your title says 'logistic' but you mention "PROC REG."

For linear regression models, look at PROC GLMSELECT.

For logistic regression models, look at PROC HPGENSELECT.

Re: proc logistic model selection for categorical variable

zzzyyy — Tue, 30 Jan 2018 20:33:26 GMT

I don't want to treat them individually, I want different levels be treated as a whole group. So is proc logistic class statement enough to

restrict the variable selection method so that a group of variables enters or leaves the model together?

Re: proc logistic model selection for categorical variable

zzzyyy — Tue, 30 Jan 2018 20:38:31 GMT

since I used to use proc reg and groupname option to restrict the effect selection method so that a group of variables enters or leaves the model together. Now I'm going to use proc logistic, I'm wondering will class statement reach the same result as groupname during variable selection process.

Re: proc logistic model selection for categorical variable

Rick_SAS — Tue, 30 Jan 2018 20:42:00 GMT

Unless you specify the SPLIT option on the CLASS statement in PROC GLMSELECT or PROC HPGENSELECT, the whole variable will either be included or excluded. PROC LOGISTIC does not support the SPLIT option, so again the whole variable is selected or not.

Re: proc logistic model selection for categorical variable

zzzyyy — Wed, 31 Jan 2018 13:38:10 GMT

got it, thanks!