[R-sig-Geo] R-sig-Geo Digest, Vol 55, Issue 11

[Nicholas Lewin-Koh]

Hi Jeff,
I think you can fit a spatial error model using a glmm in nlme or lme4.
The negative binomial
is not implemented, but a gamma model (if your counts are "well
behaved") should be a good approximation.
Another posibility is to use the VGAM package or mgcv and assume a
smooth surface for your x and y coordinates.
This of course is not the same as a spatial lag model, but does capture
the spatial trend (if any).

If overdispersion is the problem you can probably use the Poisson model
and try to bring the extra variation in as a random effect. You haven't
given us the
structure of your data. Is it the number of bird species at each point?
Is it 
counts of birds along a gradient? How the sampling was done is
important.

Nicholas

> Date: Wed, 12 Mar 2008 09:46:42 -0400
> From: "Stratford, Jeffrey" <jeffrey.stratford at wilkes.edu>
> Subject: [R-sig-Geo]   spatial verions of negative binomial
> To: <r-sig-geo at stat.math.ethz.ch>
> Message-ID:
> 	<72B2B2BDFE51E64DAC63814956149800248553 at STAFFEXCHVS3.wilkesu.wilkes.edu>
> 	
> Content-Type: text/plain;	charset="US-ASCII"
> 
> Hi group,
> 
> Does anyone have code for spatial error and spatial lag models for
> negative binomial distributions?  There are spatial error and lag models
> in the Dormann et al. Ecography paper (a very nice read) but these fit
> linear models.  Or is there a way to modify the Poison models?  
> 
> If code doesn't exist and you want to develop it, I would be happy to
> give you authorship.  I know that WinBugs can be modified but I have no
> experience with programming it.  
> 
> It's a very straightforward data set and a very important paper looking
> at the effects of urbanization on avian species richness.  The original
> ms was accepted by one reviewer and rejected by the second, who wanted
> to incorporate spatial autocorrelation.  
> 
> Cheers,
> 
> Jeff
> 
> 
>