Date:         Fri, 18 Oct 2002 12:42:57 GMT
Reply-To:     "Jerry W. Lewis" <post_a_reply@NO_E-MAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "Jerry W. Lewis" <post_a_reply@NO_E-MAIL.COM>
Organization: AT&T Broadband
Subject:      Re: Least Squares Linear Regression
Content-Type: text/plain; charset=us-ascii; format=flowed

Using Excel for statistical analysis is NOT for the novice, because so
many obvious approaches are poorly implemented.  However, Excel does do
some things better than SAS.  For instance

http://groups.google.com/groups?selm=cEe%254.29665%24Gj5.531879%40news-east.usenetserver.com

gives an extremly ill-conditioned polynomial fit problem.  As noted in

http://groups.google.com/groups?selm=3D81E207.6000506%40no_e-mail.com

the chart based polynomial trendline computes all coefficients correctly
to 9 figures, which is far better than I know how to do in SAS 8.2.

Please note that this is a discussion of numerical capabilities, not the
wisdom of fitting a high order polynomial to a limited number of data
points over a narrow range.

Jerry

David L. Cassell wrote:

> First, *never* use Excel to do statistical analysis, unless
> you can afford to get the occasional drastically wrong answer.
> [If you are commanded to do so for a homework assignment, then
> the failure of Excel clearly won't be counted against you.]
> If you're asking in a SAS newsgroup/list, you should be using
> SAS to make sure that ill-conditioned data don't cause your
> analysis software to fail in embarrassing ways.
>
> Second, your so-called 'Multiple R' is really just the square
> root of your typical R-squared value from the regression.
> That's all.  You can get the formula for R-squared from any
> intro textbook.
>
> Go Anteaters!
>
> David
> --
> David Cassell, CSC
> Cassell.David@epa.gov
> Senior computing specialist
> mathematical statistician