On Wed, 10 Dec 2008 10:03:29 -0500, Peter Flom
>You will probably want to look into Zero Inflated Poisson (ZIP) and Zero
Inflated Negative Binomial (ZINB) models.
>DAle has discussed how to program these in NLMIXED, here on SAS-L; a
number of authors have written about them. IIRC, J. Scott Long's
book "Regression models for categorical and limited dependent variables"
has a particularly good section on these.
>Essentially, you run two models:
>A) What is the chance of a zero vs. other?
>B) Given nonzero count, model count
With a low lambda, you expect many zeroes. Peter offers good advice. I
will just add that you may also want to consider the hurdle versions of
those zero-inflated analyses, too.
The archives are your best friend at the moment :)