[R-sig-ME] Which model to use?

[Benjamin Gillespie]

Hi there,

I'd greatly appreciate some advice on the following:

I have 5 indicies for 70 spatially correlated sites within a river catchment. The indices vary in nature: some are counts (i.e. species richness) and others are continuous scores. I also have a number of explanatory factors which I wish to include in my model. My models look roughly like this: index1~explanatory_factor1+explanatory_factor2+explanatory_factor2.

I've been reading Zuur et al 2009 to figure out which may be the most appropriate model to use for analysis. Initially, I thought it would be best to use a mixed model with appropriate family (e.g. poisson /quasipoisson for count indicies) and treat site ID (1, 2, 3,...70) as a random factor to account for the spatial correlation between sites. However, this didn't work (Error in R: "Number of levels of a grouping factor for the random effects must be less than the number of observations").

Alternatively, I have used GLS with spatial correlation and variance structures introduced to account for the spatial dependence and heterogeneity where appropriate - this approach seems to work well and I get results as expected, however, is GLS appropriate for count data? (I can't find anything from a google search or in Zuur et al 2009).

Many thanks in advance, please let me know if you would like further information,


Ben Gillespie, Research Postgraduate
o-------------------------------------------------------------------o
School of Geography, University of Leeds, Leeds, LS2 9JT
o-------------------------------------------------------------------o
http://www.geog.leeds.ac.uk/
o-------------------------------------o
@RiversBenG
o--------------o