turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- SAS Programming
- /
- SAS Procedures
- /
- Automated Model Selection Question

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

03-12-2009 11:05 AM

I had posted this in the SAS Macro board, and was told this might've been a better place for this question. It seems like my question was answered, but I wanted to see if this board had any other input.

With Automated Model Selection (best subsets, forward selection, etc.), I'm having trouble with datasets that have higher order terms or interaction terms. How can I manipulate SAS into taking these variables into consideration? (ie, if Variable 'x' discarded in backwards elimination, then it has to discard Variable 'x^2' and Variable 'x*y')

With Automated Model Selection (best subsets, forward selection, etc.), I'm having trouble with datasets that have higher order terms or interaction terms. How can I manipulate SAS into taking these variables into consideration? (ie, if Variable 'x' discarded in backwards elimination, then it has to discard Variable 'x^2' and Variable 'x*y')

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to deleted_user

03-12-2009 01:39 PM

First, I'd like to make a recommendation on another statistical technique that many find preferable to stepwise regression or best subsets. The problem is that Least Squares just isn't that good a modelling technique when you have many X variables and they are highly correlated with one another. There are many critiques of stepwise and best subsets on the Internet (and elsewhere) in these situations. Here are some. A better technique is Partial Least Squares regression, which in SAS is PROC PLS.

If you absolutely have to use PROC REG and want interactions and polynomial terms in the model, first create an X matrix in PROC GLMMOD to represent your main effects, interactions and polynomial terms, and then run that through PROC REG. I will note that I have never had much success doing things this way, hence the PROC PLS recommendation. Message was edited by: Paige

If you absolutely have to use PROC REG and want interactions and polynomial terms in the model, first create an X matrix in PROC GLMMOD to represent your main effects, interactions and polynomial terms, and then run that through PROC REG. I will note that I have never had much success doing things this way, hence the PROC PLS recommendation. Message was edited by: Paige