Dear list members, I have a large dataset with point data. Each data point has X, Y, number of species present and values for one single explanatory variable. Each point was generated from a polygon grid shapefile and they are at equal distances. I conducted GLS analyses introducing variance structures and correlation structures to improve the model. I plotted variograms before and after adding the correlation structures but I would like to be able to conduct a test that gives me actual values for spatial autocorrelation, so that I can show whether adding the correlation structures improved significantly the model. My variograms do not seem to improve after adding the correlation structures so I don't know if I am conducting the right analyses. Many thanks in advance Daniel -- View this message in context: http://r.789695.n4.nabble.com/Spatial-regression-tp4638512.html Sent from the R help mailing list archive at Nabble.com.
DanielFV <kalandru <at> hotmail.com> writes: [snip]> I conducted GLS analyses introducing variance structures and correlation > structures to improve the model. > I plotted variograms before and after adding the correlation structures but > I would like to be able to conduct a test that gives me actual values for > spatial autocorrelation, so that I can show whether adding the correlation > structures improved significantly the model. > My variograms do not seem to improve after adding the correlation structures > so I don't know if I am conducting the right analyses.[snip] One thing to watch out for is that you want to extract the residuals with type="normalized" (see ?residuals.lme) (or use resType="normalized" in ACF() or Variogram: see ?acf.GLS, ?Variogram.GLS) -- otherwise, your residuals will be "pearson" type, corrected for non-homogeneous variance but not for correlation (I have fallen into this trap myself). Ben Bolker
Thank you for your reply. I have used the normalized residuals for my variograms, and although the values of the semivariogram are lower, the shape and steepness of the curve is exactly the same. Any potential test that could help me justifying any improvement on the model other than merely reporting a lower AIC value? Cheers, Daniel -- View this message in context: http://r.789695.n4.nabble.com/Spatial-regression-tp4638512p4639030.html Sent from the R help mailing list archive at Nabble.com.