Date:         Wed, 14 Jul 2010 18:40:03 -0400
Reply-To:     Joe Whitehurst <joewhitehurst@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Joe Whitehurst <joewhitehurst@GMAIL.COM>
Subject:      Re: Step-Wise Methods re-evaluated
Comments: To: Sigurd Hermansen <HERMANS1@westat.com>
In-Reply-To:  <FE10F31634E7F34B87AA143D596085415B95326D@EX-CMS01.westat.com>
Content-Type: text/plain; charset=ISO-8859-1

What a curious combination--rigid _and_ rational.

On Wed, Jul 14, 2010 at 6:03 PM, Sigurd Hermansen <HERMANS1@westat.com>wrote:

> Step-wise model selection, viewed by some as an automated Ouija Board,
> continues in use by statisticians despite many attempts to discredit the
> method. In a recent Statistics in Medicine article, Ryan E. Wiegand sums up
> his evaluation of step-wise variable selection methods:
>
> ".... To conclude, stepwise agreement is often a poor strategy that gives
> misleading results and researchers should avoid using multiple SVS
> algorithms to build multivariable models."
>
> See http://www3.interscience.wiley.com/journal/123472582/abstract .
>
> Many SAS procedures support step-wise selection. I don't see SI working
> actively to deprecate the use of step-wise methods.
>
> Has anyone seen a good defense of step-wise methods as an aid, along the
> lines of exploratory graphics, to selecting a good predictive model? One
> supposed feature or another must appeal to otherwise rigidly rational
> statisticians.
> S
>