Skip to content

Zero Inflated Poisson

3 messages · Sven Adler, Edzer Pebesma, Virgilio Gomez Rubio

#
Dear all,
I want to model bird distribution at see. The data consists 10000
samples but 90% are zeros. I try first a gamm (mgcv)with quasipoisson
but it might bee problematic with so many zeros. I look for a (spatial)
model that can use Zero Inflated Poisson distribution or similar. Is
there any?

Thank you
sincerely yours

Sven
#
*I'm not sure if you're aware of the following publication:

*

*2005*: Mapping sea bird densities over the North Sea: spatially
aggregated estimates and temporal changes
<http://www.citeulike.org/article/3505281>
Pebesma Edzer J. and Duin Richard N. M. and Burrough Peter A.
/Environmetrics/

*The analysis done there is mostly reproduced by:
*

*library(gstat)
demo(fulmar) # Fulmaris glacialis
*

*Another direction you might want to look at is package geoRglm.
*

*Best regads,
--
Edzer
*
Sven Adler wrote:

  
    
#
Sven,

In addition to what Edzer has pointed out, I have been trying to fit
some zero-inflated Binomial models recently, but I have found that
including spatial random effects may not be a good idea when the number
of zeros is large. 

In our case, we have 126 observations with only 17 non-zero
observations. We used package VGAM to fit zero-inflated models and
WinBUGS to fit models with random effects. In general, including random
effects is problematic and convergence was terrible in some cases. My
feeling is that the zero-inflated part of the model may clash with
random effects. 

If you have several observations per site this may change. We also
observed a better fit when we consider several observations per site (by
splitting the observations into 3 age groups).

I do not think that there is a way of fitting zero-inflated spatial
models in R (but I may be wrong). I asked Paulo Ribeiro some months ago
about whether geoR/geoRglm can handle that and the answer was no.

Hope this helps.

Virgilio

El mi?, 11-03-2009 a las 16:21 +0100, Edzer Pebesma escribi?: