Date:         Wed, 16 Jul 2008 16:30:55 -0400
Reply-To:     Kevin Viel <citam.sasl@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Kevin Viel <citam.sasl@GMAIL.COM>
Subject:      Re: Simulated ZIP: SAS for Mixed Models, 2nd Ed.

On Wed, 16 Jul 2008 12:15:40 -0500, Robin R High <rhigh@UNMC.EDU> wrote:

>If you have no predictors other than the intercept, then lambda_i = lambda
>for the count part of the model for all observations; then overdispersion
>is even more likely. However, by adding predictors (getting different
>lambdas for different combinations of the x values) you hope to explain
>variation in the data through the Poisson part of the model and thus
>reduce overdispersion.  If not, then a negative binomial ZI model might be
>of value.  And as I experienced last week, in some situations excess 0's
>can be modeled more appropriately with a "hurdle" model (ZAP or ZANB),
>though that goes beyond the scope of your question.  And also for
>reference if you have random effects, that feature is found with adding an
>error term to the x_i*B term.  There are examples of that model in the
>SAS-L archives.

Doh!  Of course. IOTT.  I started with the trivial case of intercept only
simulations.  Good to know I can think ahead, Ahem!

Interesting about the hurdle models.  I was thinking the inflated models
would suffice without checking at this point.  I should have known better.
Having you had any convergence or Hessian problems?  Are you using COUNTREG?

Thanks again,

Kevin