Date:         Wed, 4 Feb 1998 16:30:41 -0500
Reply-To:     shapiro@underground.irhe.upenn.edu
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Dan Shapiro <shapiro@UNDERGROUND.IRHE.UPENN.EDU>
Organization: Institute for Research on Higher Education
Subject:      RMSE using a weighted Proc Reg
Content-Type: text/plain; charset=us-ascii; x-mac-type="54455854";
              x-mac-creator="4D4F5353"

I've found that when I run a weighted regression that the RMSE produced
is too large.  It seems as if the Error Sum of Squares is calculated on
the weighted data, but the MSE error is then computed by taking ESS/df.
But the Degrees of Freedom are unweighted yielding a MSE which is too
large.  The RMSE is similarly large as a result.
 
Is there any way I can obtain a reasonable RMSE using Proc Reg and a
weight?
 
thank you,
 
Dan