R help - Aug 2006 - Handling realisations in geoRglm

Dear R users:

I want to model mosquito count data based on landcover attributes and 
meteorological variables using a Poisson GLSM in the geoRglm package.  I 
have monthly mosquito counts over more than 20 years with repeated 
observations from individual trap sites over time.  I have used 
as.geodata() to successfully read my dataset into the geodata format 
utilized by geoR and geoRglm, including specification of the 
realisations.  (I know the realisations have been specified correctly 
because I can get data summaries by realisation from the geodata 
object.)  However, after reading the packages' documentation and 
searching the mailing lists and other sources, it seems that the 
functions in geoRglm do not acknowledge the existence of multiple 
realisations per site.  I see that the likfit() function in geoR has an 
argument for realisations, but I cannot find anything similar in 
geoRglm.  Is it possible to model data with repeated realisations from 
the same sites using geoRglm?  If so, how can this be done and is there 
a way to model dependence among the realisations over time?

Thank you,
Chris Barker
Ph.D. Candidate
University of California, Davis
cmbarker at ucdavis.edu

Chris Barker <cmbarker <at> ucdavis.edu> writes:
> 
However, after reading the packages' documentation and
> searching the mailing lists and other sources, it seems that the 
> functions in geoRglm do not acknowledge the existence of multiple 
> realisations per site.  I see that the likfit() function in geoR has an 
> argument for realisations, but I cannot find anything similar in 
> geoRglm.  Is it possible to model data with repeated realisations from 
> the same sites using geoRglm?  If so, how can this be done and is there 
> a way to model dependence among the realisations over time?
> 
  I don't know; I would be interested in the conclusions you reach.
I, too, found the support for multiple realisations to be relatively
skimpy in geoR (only likfit, and not any of the variogram-based
methods, support realisations); the code for likfit.glsm in the
geoRglm packages also reveals no hint of realisation code.

  Even so, what there is in geoR says quite strongly that realisations
are expected to be independent.  My guess is that for complex space/time
dependence, and wanting an answer to a practical problem, you would be
better off with a GEE approach ...

Writing the package maintainer directly
would probably be more useful.

  Ben Bolker