Date: Thu, 17 Jul 2003 09:01:10 -0700
Reply-To: Dale McLerran <stringplayer_2@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Dale McLerran <stringplayer_2@YAHOO.COM>
Subject: Re: Multiple Regression?
Content-Type: text/plain; charset=us-ascii
That should do the job pretty well. In theory, the REG
procedure just as you have outlined below is the correct
approach. My only concern would be if X, Y, and MarkerElev
are collinear such that there is significant ill-conditioning
in your data, then the REG procedure may not work well. You
might want to use the procedure ORTHOREG instead of REG.
ORTHOREG is designed to produce more accurate parameter
estimates in the case where the predictor variables are
highly correlated, or other ill-conditioning of the data
Chances are that PROC REG is just fine. However, I would at
least fit the model using the ORTHOREG procedure and compare
the parameter estimates from the two procedures. If the
parameter estimates arising from ORTHOREG are different at
the third or fourth significant digit, then you should
probably use ORTHOREG.
ORTHOREG will not output the predicted and residual values
as does REG. If you find that you should not be using REG
but should instead be using the ORTHOREG procedure, then
you would have to construct the predicted values from the
estimated parameters applied to your predictor variables.
--- Kevin Myers <WHMyers@CABLEONE.NET> wrote:
> Sorry for following up my own post, but after digging through docs
> following seems simple code using PROC REG seems to do what I need:
> proc reg data=WBMarkers;
> model Marker2Elev=X Y MarkerElev;
> output out=WBMarkers predicted=Marker2Est residual=Marker2Error;
> run; quit;
> In the above, Marker2Elev is the surface that I'm wishing to fit and
> based on the values of X, Y, and MarkerElev (my other surface).
> ----- Original Message -----
> From: "Kevin Myers" <WHMyers@CABLEONE.NET>
> Newsgroups: bit.listserv.sas-l
> To: <SAS-L@LISTSERV.UGA.EDU>
> Sent: Wednesday, July 16, 2003 8:35 PM
> Subject: Multiple Regression?
> Dear SAS Gurus,
> I have a couple of sets of data each representing a different three
> dimensional surface. Each of these surfaces is very roughly planar,
> with superimposed undulations. If I use x and y to represent the
> coordinates in the horizontal plane, and let f(x,y) and g(x,y)
> represent the
> elevation of the two surfaces above the horizontal plane, then based
> on the
> natural mechanisms involved I believe that the following relationship
> approximately hold true:
> g(x,y) = c1 * f(x,y) + c2 * x + c3 * y + c4
> I have a reasonably large number of sample data points representing
> each of
> the two surfaces. Based on these data points, I would like to find
> values c1, c2, c3, and c4 for use in the above equation which
> minimize the
> error in the predicted values of g(x,y) versus the observed values.
> I'm familiar with simple linear regression, and this problem appears
> to be
> quite similar. I'm guessing that this is a "multiple regression"
> since it involves more variables and more constants that need to be
> best fit
> by the regression model. However, I know very little about multiple
> regression or about the various SAS procedures that deal with it.
> someone out there give me some tips as to how I can best use SAS to
> this problem?
> Thanks in advance for any help!
Fred Hutchinson Cancer Research Center
Ph: (206) 667-2926
Fax: (206) 667-5977
Do you Yahoo!?
SBC Yahoo! DSL - Now only $29.95 per month!