Date:         Fri, 21 Oct 2005 23:01:45 -0700
Reply-To:     David L Cassell <davidlcassell@MSN.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David L Cassell <davidlcassell@MSN.COM>
Subject:      Re: PLS logistic and Foster-Stine
In-Reply-To:  <200510202246.j9KLA6i9022544@malibu.cc.uga.edu>
Content-Type: text/plain; format=flowed

meir_sebag@HOTMAIL.COM wrote:
>I am looking for 2 things implemented in SAS, is anybody has some
>ressources:
>
>1. I am looking for an implmentation of the logistic PLS regression in SAS
>2. I am looking for an implemnjation of the Foster/Stine approach to
>variables selection without using SEM v5 and the new GSForward node.

Interesting.  Two very different approaches.  Let me guess.  You're doing
data mining and you have umpteen thousand variables to wade through.

Okay, as far as I know, there are no SAS implementations for either.  And
let
me fly off the handle before you press the delete key.

Foster/Stine is just doing forward selection with a different rejection
rule.  I
think their methodology is suspect, since the same old problems with the
theory
underlying forward/stepwise/backward selection still hold.  The tests people
use at each stage are simply wrong, and using an information-theoretic
cutoff
doesn't fix that.  Furthermore, as Stine's own stuff on the Penn website
shows,
deviations from the basic underlying assumptions (multivariate normal, no
outliers,
no heteroskedasticity, no leverage points, no measurement error problems,
multicollinearity, suppressor variables, and so on and so on) can muck up
the
results beyond repair.  So - at a bare minimum - you end up having to check
all the accepted and rejected regresison fits anyway to make sure things
didn't
go kablooey in the middle of your computations.  Oh, and applying such
selection methods to logistic regressions just makes things worse.

Next, as far as I know, there is no single "logistic PLS" method.  There are
a
couple of things being called logistic regression with PLS features.
There's
Kernel Logistic PLS.  There's penalized logistic regression with PLS.
There's..
well, you get my point.  So I don't know *which* method you're after.
Sorry.  Whichever it is, I don't have a SAS implementation.  But you'd
probably do just as well using the SAS %TREEDISC macro instead.

HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
Express yourself instantly with MSN Messenger! Download today - it's FREE!
http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/