Date: Sat, 17 Jul 2010 17:23:02 -0400
Reply-To: Wensui Liu <liuwensui@GMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Wensui Liu <liuwensui@GMAIL.COM>
Subject: Re: Genmod and zero inflated models
Content-Type: text/plain; charset=ISO-8859-1
before jumping from poisson model to zip, it might be worth to try NB
model with a more flexible variance assumption but more parsimonious
as a practical rule, you might use vuong test to compare ZIP and NB.
On Fri, Jul 16, 2010 at 2:18 PM, Jerry Davis <firstname.lastname@example.org> wrote:
> Several months ago I was helping an entymology grad student and his data
> (insect counts) had a lot of zeros. I guess the insects weren't
> attracted to his treatments and avoided the traps.
> Here is a summary:
> count freq percent
> 0 6437 87.16
> 1 651 8.82
> 2 215 2.91
> 3 50 0.68
> 4 19 0.26
> 5 4 0.05
> 6 5 0.07
> 7 1 0.01
> 10 1 0.01
> 11 1 0.01
> 17 1 0.01
> This is a *lot* of zeros for agriculture data, in my experience. I had
> a copy of the data and one afternoon thought it would be a good exercise
> to fit a ZIP model with GENMOD and check out the output. SAS told me
> that I didn't have zero inflated data and ignored my ZEROMODEL statement!
> Actually, a regular poisson model fit pretty well; the deviance
> parameter was .88.
> Anyway, new stuff comes in every day or two and I never got back to this
> exercise, but I wonder what a zero inflated distribution looks like.
> Jerry Davis
> Experimental Statistics
> UGA, CAES, Griffin Campus