If you are getting into more advanced statistical methods you would be better off working out the small details of exchanging files in and out of ArcMap and then investing your time learning R. If you are competent in GIS and databases, R is a natural complement. You will soon be running into the same issues as you discover more advanced Geostatistical analyses that are intriguing and beyond ArcMap.
Thanks, John. This is what I have been doing. I think I should have specified that I'm trying to account for spatial autocorrelation in the dependent and in some of the IVs.
Yes, I figured that was where you were heading.
I don't know if there is a package allowing both zero-inflation and geographical weighting, but one thing you might consider as an intermediate step would be to include some geographic variables as predictors in your ZINB model and then test for spatial correlation in the residual process. Often when rich co-variate data are available, most of the apparent spatial correlation can be removed by inclusion of these underling mechanisms. In that case, the spatial weighting would be unnecessary.
The other advantage is having a mechanistic explanation for the spatial correlations. But in the end you may still need to account for the spatial variation that remains, so still need the software.
The following link has a good summary of your zero-inflated options in R: http://cran.r-project.org/web/packages/pscl/vignettes/countreg.pdf. For zero-inflated negative binomial models, see the zeroinfl function in the pscl package.
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics
- Geographic Information System