Date: Thu, 21 Sep 2006 23:10:01 -0700
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: difference between regression coefficients in non-nested
In-Reply-To: <200609201334.k8KAn5Dw011878@mailgw.cc.uga.edu>
Content-Type: text/plain; format=flowed
bigdoctor2004@GMAIL.COM wrote back:
>IK wrote:
> > ok let me take another shot at defining better. These are two earnings
> > measures - one is mandated and the other is not. Proponents of the
> > earnings
> > measure that is not mandated argue that it is provided because it
> > provides
> > relatively more information than the mandated earnings measure with
> > respect
> > to the association with returns or price. I want to examine this
> > claim.
> > This is my definition of better. This therefore necessitates a
> > comparison
> > of both measures.
> >
> > >>>>
> >
> >
> > OK, but by 'more information' do you mean a higher Rsquare?
> > or something else?
> >
> > once you've figured that out, write back
> > and there may be
> > a solution.
> >
> > Peter Flom wrote:
> > > Peter L. Flom, PhD
> > > Assistant Director, Statistics and Data Analysis Core
> > > Center for Drug Use and HIV Research
> > > National Development and Research Institutes
> > > 71 W. 23rd St
> > > http://cduhr.ndri.org
> > > www.peterflom.com
> > > New York, NY 10010
> > > (212) 845-4485 (voice)
> > > (917) 438-0894 (fax)
> > >
> > >
> > > >>> IK <bigdoctor2004@GMAIL.COM> 09/19/06 7:50 AM >>> wrote
> > > <<<
> > > I want to be able to make statements regarding which variable X1 or x2
> > > is a better predictor of Y. I tried to use both variables in a nested
> > > regression but cos of multicoll. issues I cant.
> > >
> > > i forgot to add - my data is market data - returns and earnings and
>for
> > > the most part most of the underlying assumptions have not been
> > > violated. i used proc reg initially but then when i couldnt compare
> > > the slopes i included a predicted variable for Y in model 1 and used
> > > proc autoreg and did the same for model 2 i.e. a davidson and
>mackinnon
> > > test. am i out to lunch on all this?
> > >
> > > thanks
> > > >>>
> > >
> > > There are several issues:
> > > First, you have to define 'better'.
> > > Second, if X1 and X2 are strongly collinear, it means they do a
>similar
> > > job
> > > Third, a lot will depend on the distributions of x1 and x2, and
>whether
> > > your data is over the full range of both
> > >
> > > and probably a lot of other things
> > >
> > > Peter
>Not only the adj. R squared. For the adj. R squared measure, I intend
>to use a Vuong test which is another question. but for this particular
>test, I predict that the slope coefficient on X1 will be greater and
>more significant than that for X2 which is what I am hoping I can do
>(i.e. is valid) and that i can appropriately test. Any input greatly
>appreciated.
>
>thanks
>
I don't recommend Vuong tests. They have low power for small
n. They are generally outperformed by Clarke's distribution-free
test, even though that is harder to compute. And they are
sensitive to things like outliers and leverage points and other
divergences from the underlying assumptions of the models.
HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330
_________________________________________________________________
Add fun gadgets and colorful themes to express yourself on Windows Live
Spaces
http://clk.atdmt.com/MSN/go/msnnkwsp0070000001msn/direct/01/?href=http://www.get.live.com/spaces/features
| 
