Date:  Mon, 21 Mar 2005 20:32:04 0800 
ReplyTo:  Dale McLerran <stringplayer_2@YAHOO.COM> 
Sender:  "SAS(r) Discussion" <SASL@LISTSERV.UGA.EDU> 

From:  Dale McLerran <stringplayer_2@YAHOO.COM> 
Subject:  Re: ZIP models 
InReplyTo:  6667 
ContentType:  text/plain; charset=usascii 

 Jeffrey Stratford <stratja@AUBURN.EDU> wrote:
> Hi. Two questions regarding ZIP models.
>
> 1. Some of my coefficients are negative for the covariates included
> in
> the zeroinflation part of the model. I understand a positive
> coefficient  increasing the covariate means an increased probability
> of
> occurring in the zero class. But how do I translate a negative
> coefficent?
>
> For example, I'm looking at the effect of urban cover on avian
> density
> and 12/14 species have the zeroinflation possitively associated with
> urban cover. So I'm translating this to mean that the excess of
> zeros
> in my data are explained by urban cover and it turns out that birds
> are
> missing from highly urbanized sites. However, two species have a
> negative coefficient for urban cover but what does that suggest?
>
First of all, the zeroinflation parameters need to be interpreted
in light of the Poisson mean expectation model. Assuming that
the mean expectation model for the urban cover areas have equal
or greater mean than the mean expectation for the nonurban
areas, then the observation that the urban effect is negative
means that urban areas have fewer areas in which there are
nonPoisson zero values. Loosely, that would suggest that these
species might flourish in urban areas. I believe that there are
some bird species which do flourish in urban areas. Pigeons come
to mind, but it seems to me that I have heard of other species
too. Certainly there are mammals which are not adverse to human
presence but in fact thrive in the presence of human activity.
I don't know what species you are studying, but you might have
to either reconsider your model (collinearity, left out variables,
etc.) or your assumptions.
> 2. How does one evaluate model fit when SAS only gives a
> 2loglikelihood? Is there a way to convert this to get a chisquare?
>
I don't know of a likelihood ratio goodnessoffit test for the
ZIP model. You could compare the likelihood of the ZIP model
with the likelihood of the Poisson model. The Poisson model is
nested in the ZIP model. Thus, you can test whether the zero
inflation assumption results in an improved fit compared to a
simple Poisson model.
> Thanks,
>
> Jeff
>
> ****************************************
> Jeffrey A. Stratford, Ph.D.
> 331 Funchess Hall
> Department of Biological Sciences
> Auburn University
> Auburn, AL 36849
> 3343299198
> FAX 3348449234
> http://www.auburn.edu/~stratja
> ****************************************
>

Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@NO_SPAMfhcrc.org
Ph: (206) 6672926
Fax: (206) 6675977

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