Date:         Wed, 18 Apr 2007 09:19:26 -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: Genmod procedure
In-Reply-To:  <Pine.GSO.4.58.0704181039050.28969@leukothea>
Content-Type: text/plain; charset=iso-8859-1

--- Kevin Roland Viel <kviel@EMORY.EDU> wrote:

> On Tue, 17 Apr 2007, Dale McLerran wrote:
>
> > Now, this does raise one other point about the use of likelihood
> > ratio tests for testing parameter constraints.  You can only use
> > a likelihood ratio test to compare models which have exactly the
> > same set of response values!
> >
> > What does this mean in practice?  Suppose that you want to test
> > the null hypothesis H0:b3=0 where b3 is the parameter associated
> > with variable X3.  Now, suppose that X3 has some missing values.
> > For ease of presentation, we will assume that the response and all
> > other predictors are fully observed.  So, let's say that
> observation
> > 29 has a missing value for X3.  Now, when X3 is included in the
> > model, observation 29 is not used to fit the model.  If you do
> > not take care to exclude observation 29 when fitting the model
> > in which X3 has its effect fixed at zero (when X3 is dropped from
> > the MODEL statement), then you cannot use a likelihood ratio test
> > to compare the two models.  You must make sure that you use the
> > same set of observations when fitting all models in order to
> > properly compute the likelihood ratio test statistic.  You can
> > use a WHERE statement (or data step option) to excluded
> observations
> > for which X3 is missing when fitting the model which does not
> > include X3.  Thus, to perform a likelihood ratio test for b3, you
> > would want to submit code something like the following:
>
> Dale,
>
>   Thanks for pointing that out.  My impression is that this is not
> well
> known in my experience.  I think it come to those that have a bit of
> data ranger in them :)
>
> Kevin
>

Kevin,

That is my impression, too.  It is awfully easy to overlook the
need to make sure one uses the same response set in each model.  I
wasn't certain whether to include all that verbiage in my response.
Thanks for your reply to confirm that it was warranted!

Dale

---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@NO_SPAMfhcrc.org
Ph:  (206) 667-2926
Fax: (206) 667-5977
---------------------------------------

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around
http://mail.yahoo.com