Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: stepwise regression


From   Richard Williams <Richard.A.Williams.5@nd.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: stepwise regression
Date   Tue, 13 Apr 2004 08:07:02 -0500

At 09:22 AM 4/13/2004 +0200, fabiopericolini@fastwebnet.it wrote:
Hi all,
I have performed the sw (stepwise) regress in Stata, my results with backward
or forward estimation is very different, but I know that shouldn't be so.
My question is: is the Stata sw procedure the real stepwise estimation??!!
It seems to me that the selection is based only on the significance and
not on the marginal explainatory power of every single variable.
There is nothing that says that results have to be the same. Indeed, you can come up with reasonable examples involving suppressor effects where none of the variables will enter if you use forward selection, and none will be dropped if you use backwards. And, if you are trying to choose from hundreds of variables, it is not at all surprising that you get different results.

As others have noted, stepwise is generally a poor way to go. Surely if you have 100s of variables many of them must tap the same concepts and can be combined into scales.

But, if you are determined to use an approach that goes both forwards and backwards (e.g. adds a variable, then goes back and drops something that is no longer significant, then goes forward again, etc.) SPSS and probably other packages offer it if you have access to them.

Another suggestion: if you use stepwise, cross-validate your results against a different sample. More practically, divide your sample in two. This reduces the likelihood that you are just capitalizing on chance in your SW analysis.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
FAX: (574)288-4373
HOME: (574)289-5227
EMAIL: Richard.A.Williams.5@ND.Edu
WWW (personal): http://www.nd.edu/~rwilliam
WWW (department): http://www.nd.edu/~soc

*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/




© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index