Date:         Wed, 7 May 1997 10:07:43 -0400
Reply-To:     Ismail Parsa <iparsa@EPSILON.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Ismail Parsa <iparsa@EPSILON.COM>
Subject:      Re: chaid on sas
Comments: To: ecm@NWU.EDU

|>  From: Edward Malthouse <ecm@NWU.EDU>
|>  Subject:      Re: chaid on sas
|>
|>  I have used SPSS' version of CHAID and found it to be very dangerous.  The
|>  problem I have with it is that it is prone to overfitting the data -- I get a
|>  nice tree and decent fit statistics, but when I apply it to another data set
|>  the predictions are not so good.
 
Tighten your stopping criteria and don't opt for a large/long tree.
 
|>  I have had much better luck with Splus' CART (classification and regression
|>  trees is an improved version of chaid) functionality.  It includes options
 
The two are not technically related. CART uses purity measures, like GINI index
or entropy whereas CHAID uses chi-sqaure criterion in building trees.
 
|>  that automate the process of validating the model with holdout samples.
 
You can certainly score a validation sample using CHAID.
 
Regards.
 
*-----------------------------*
|        Ismail Parsa         |
|   Epsilon Data Management   |
|     50 Cambridge Street     |
|   Burlington MA 01803 USA   |
|                             |
| E-MAIL: iparsa@epsilon.com  |
| V-MAIL: (617) 273-0250*6734 |
|    FAX: (617) 272-8604      |
|                             |
|  The Usual Caveat Applies   |
*-----------------------------*
 
 
|>
|>  I sincerely hope that SAS does not follow SPSS' model for CHAID.
|>
|>  Ed Malthouse
|>
|>
|>  In article <19970430.161151.580850.NETNEWS@AUVM.AMERICAN.EDU>,
|>     "Zhang, Alex - O&MD NY" <OGILVY!OMDNY1!ZHANGAL@OMDNY.ATTMAIL.COM> wrote:
|>  >I was told at a SAS seminar about a month ago that CHAID and neural network
|>  >are only available through SI's Professional Service as part of the
|>  >customized packages.  SI is planning to release a data mining product which
|>  >will include CHAID at the end of this year.  A beta version may be available
|>  >as early as this Summer.
|>  >
|>  >Alex Zhang
|>  >Ogilvy & Mather Direct
|>  >alex.zhang@OGILVY.COM
|>  > ----------
|>  >From: Philip Holland
|>  >To: Multiple recipients of list SAS-L; ZHANGAL
|>  >Subject: Re: chaid on sas
|>  >Date: Wednesday, April 30, 1997 4:54PM
|>  >
|>  >skmxx <skmxx@ibm.net> wrote:
|>  >>Hi! everyone.
|>  >>
|>  >>I need your advice.  I want to run CHAID on sas, but, I cannot find
|>  >>where. In sas/stat manual I could not find any word CHAID on the index.
|>  >>I know there is CHAID option on spss. On sas where is this option or
|>  >>comparable option?  I really appreciate your advice.
|>  >>
|>  >>Soo Hong Kim
|>  >>Graduate Student
|>  >>Syracuse Univ.
|>  >
|>  >I believe there are some SAS macros for CHAID stats
|>  >processing, supplied by the SAS Institute.  I suggest you
|>  >contact your local SAS office, but be warned that they may not
|>  >be free...
|>  >
|>  >HTH..............Philip R Holland <phil.holland@bcs.org.uk>
|>  >                 Holland Numerics Ltd
|>  >                 Royston, Herts, UK
|>  >                 mobile: +44-(0)850-295556
|>
|>  Edward Malthouse
|>  Assistant Professor
|>  IMC Department
|>  Medill School of Journalism
|>  Northwestern University
|>  Evanston, Illinois
|>  ecm@nwu.edu
|>