Date: Fri, 26 Jan 2001 16:15:48 EST
ReplyTo: Hongjiew@AOL.COM
Sender: "SAS(r) Discussion" <SASL@LISTSERV.UGA.EDU>
From: Hongjie Wang <Hongjiew@AOL.COM>
Subject: suppressor in regression
ContentType: text/plain; charset="USASCII"
Suppressor is defined as a predictor x such that is related to another
predictor, but does not relate (or weakly) to the target variable. However,
by including this variable, the R^2 improves. In other words, the
contribution(x)< contribution(x given others).
In developing a complex model(hundreds of variables.), one usually start with
individual variable screening.
This is not the optimial way, but a good and practical way to select
variables.
However, such approach will not double overlook suppressors.
Can someone comment on the following?
1. In general, shoudl special efforts be spent to uncover suppressors?
and should suppressors be included in the final model?
What is the impact of the intrepretability of the model given the inclusion
of suppressors?
thanks a lot!
Hongjie Wang
hongjiew@aol.com
