Date:         Wed, 22 Dec 1999 23:47:35 -0800
Reply-To:     Kattamuri Sarma <KSSarma@WORLDNET.ATT.NET>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Kattamuri Sarma <KSSarma@WORLDNET.ATT.NET>
Subject:      Re: a question in regression
Comments: To: Paige Miller <paige.miller@KODAK.COM>
Content-Type: text/plain; charset="iso-8859-1"

In my comments I have assumed that the coefficients are statistically
significant,
the fit is good. In addition to these statistical tests, we should make
tests
to see whether the magnitudes are meaningful. Let me give an example from
economics.
Suppose you are fitting an equation to some economic data,
 your coefficients may be statistically significant, you may have
an overall good fit. But the equation itself may not be consistent with
economic theory.
It happens all the time. You get statistically good fit, but meaningless
from economic theory.
I am sure it happens in other fields as well.

For this you do additional tests, sensitivity tests, multiplier tests, etc.
Examining the
derivatives is one such test.  It is not a statistical test  ,
but a plausibility test from the
point of economic theory.

Kattamuri Sarma
  -----Original Message-----
From: Paige Miller <paige.miller@KODAK.COM>
Newsgroups: bit.listserv.sas-l
To: SAS-L@LISTSERV.UGA.EDU <SAS-L@LISTSERV.UGA.EDU>
Date: Wednesday, December 22, 1999 5:02 AM
Subject: Re: a question in regression


>Kattamuri.Sarma@RESPONSEINSURANCE.COM wrote:
>>
>>  Take the partial derivative of Y w.r.t X,     which will be something
like
>> a +2bX,  ( I am assuming
>>      that you have fitted Y = constant + aX + b*X**2 ) and see how this
>> behaves
>>     at each plausible value of X.   Only from the  partial derivative
>> evaluated at each x,  can you tell whether you
>>    have a meaningful equation or not...
>
>I would have advised, since we are talking about a regression, which is
>by nature empirical, that the derivatives aren't the only thing you
>might want to consider, but also how well your fitted model predicts the
>data. In fact, I would put a lot more emphasis on model fit than on the
>derivatives.
>
>--
>Paige Miller
>Eastman Kodak Company
>paige.miller@kodak.com
>"It's nothing until I call it!" -- Bill Klem, NL Umpire