Date:         Mon, 2 Aug 2004 14:53:24 -0500
Reply-To:     GSnell@DataSavantConsulting.com
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Gregg Snell <gsnell@DATASAVANTCONSULTING.COM>
Organization: Data Savant Consulting
Subject:      Regression as a data step
Content-Type: text/plain; charset="us-ascii"

I need a little perspective from those of you a bit more stat-informed than
I.

I have a client that wants me to write code to create forecasted values such
that:

    VAR1=LAG1(VAR2);
   VAR3=<single regression with VAR1 as the regressor>;
   VAR2=VAR1+VAR3;

I want to use AUTOREG or some other PROC to calculate VAR3, but that
requires all values of VAR1 to already exist in a dataset.  However, the
dependency of VAR1 on VAR2 precludes this.  The only way I can see to
accomplish this is by doing the regression in a data step where I can
control the order each variable is created.  Has this (doing regression in a
data step) already been done or such a thing even feasible?

Please assume I know nothing about regression when responding ;-)

Thanks,

Gregg P. Snell
Data Savant Consulting
(913) 638-4640
(208) 977-1943 efax
GSnell@DataSavantConsulting.com