Hi,
I am a doctoral student in Social Epidemiology and I am applying a variety of spatial statistical methods in my dissertation using R. I am conducting a cross-sectional analysis of neighborhood correlates of health outcomes. I am concerned about modeling two of my outcomes in linear models, as two of my outcomes are really count data. Im familiar with how to run non-OLS models for my data including Poisson and zero-inflated Poisson models in R. Additionally, I know that there are packages for generalised linear spatial models (geoRglm) and packages for spatial count regression (spatcounts). To my knowledge, while the spdep package can test the residuals for global spatial autocorrelation and the lagrange multiplier diagnostic for spatial dependence in OLS models, it can not for other models. Previously, Roger Bivand wrote: "While lm.morantest() can be used on glm output objects, no work has been done to establish whether this is a sensible idea. It remains problematic to simulate spatially dependent discrete variables" ( http://www.mail-archive.com/r-sig-geo@stat.math.ethz.ch/msg07991.html). I am unsure if there are R packages that can test the residuals for global spatial autocorrelation and the lagrange multiplier diagnostic for spatial dependence based on the results for non-OLS models (e.g. Poisson models). ***Are there R packages that can test global spatial autocorrelation in Poisson models or Negative Binomial models and the lagrange multiplier diagnostic for spatial dependence based on the results for these non-OLS models?*** Since I am finding global spatial autocorrelation in my OLS models, I am ultimately running spatial linear error regression models. The geoRglm and spatcounts packages seem to be based on Bayesian spatial methods (which I dont know but want to learn during my postdoc). So ***Are there R packages that allows fitting spatial Poisson error regression models?*** In summary, I want to run a Poisson model or another type of non-OLS model for two outcomes then evaluate global spatial autocorrelation for the residuals of the model. If the residuals are significant for spatial autocorrelation, I then want to fit the spatial form of that model (e.g. spatial Poisson error model). I know how to do this all with OLS models using the spdep package, but am unsure which R packages (if they exist) allow me to do so using non-OLS models. Do you know if such packages exist? Thanks, Dustin [[alternative HTML version deleted]]
_______________________________________________ R-sig-Geo mailing list R-sig-Geo@stat.math.ethz.ch https://stat.ethz.ch/mailman/listinfo/r-sig-geo
