Re: variable selection in logistic regression

From: Richard Ulrich (Rich.Ulrich_at_comcast.net)
Date: 11/16/04 

Date: Tue, 16 Nov 2004 15:10:19 -0500

On Tue, 16 Nov 2004 14:53:20 +0200, "Tommi" <t_kujanpaa@hotmail.com>
wrote:

> Hello!
> I made some experiments with SPSS and logistic regression.
> I am trying to explain binary coded variable with the help of some binary,
> some continuous explanatory variables.
> I used the method called forward (Wald) in SPSS for variable selection, and
> some of my explanatory variables were discarded and some were taken in the
> final model.
>
> If I had to justify why the Forward (Wald) method works/doesn't work for my
> case, is there any help in Web, or can you give me literature references
> where I can get some more light on the subject (which I am not yet so
> acquinted with)?
> I would be thankful if you could also help me with these other selection
> methods, for example LR forward and so on.

You can see my stats-FAQ for comments by several folks,
previously posted to the sci.stats.* groups, about why
step-wise selection is generally a bad idea.

Assuming you can justify steps -- Stepping DOWN from <all>
works better if you want to keep most of the variables; stepping
UP from <best> works better if you want just a few.
What "better" means remains problematic. 

-- 
Rich Ulrich, wpilib@pitt.edu
http://www.pitt.edu/~wpilib/index.html

Relevant Pages

  • variable selection in logistic regression
    ... I made some experiments with SPSS and logistic regression. ... I used the method called forward in SPSS for variable selection, ... some of my explanatory variables were discarded and some were taken in the ... If I had to justify why the Forward (Wald) method works/doesn't work for my ...
    (sci.stat.math)
  • Re: variable selection in logistic regression
    ... >>I made some experiments with SPSS and logistic regression. ... >>I would be thankful if you could also help me with these other selection ... mentioned the issue of Wald vs Likelihood ratio tests. ...
    (sci.stat.math)