[R-sig-Geo] SpatialFiltering function in the spdep package

[Roger Bivand]

On Wed, 1 Jun 2011, Jonathan Thayn wrote:

> I have a quick question regarding the SpatialFiltering function. I have 
> created a dataset wherein I control the amount of spatial 
> autocorrelation in the model residuals before submission to spatial 
> filtering. The SpatialFiltering function works really well when model 
> residuals are positively spatially autocorrelated, but it doesn't seem 
> to work when the residuals are spatially negatively autocorrelated. It 
> seems to me from the literature, that spatial filtering with 
> eigenvectors aught to be able to handle negatively spatially 
> autocorrelated model residuals. Is something about the way 
> SpatialFiltering is coded that excludes negative SAC? Does anyone have 
> any other ideas that might explain this? Thanks.

The comment at the head of the source for the function says:

# tol: tolerance value for convergence of spatial filtering (Moran's I).
# The search for eigenvector terminates, once the residual
# autocorrelation falls below abs(Moran's I) < tol. For positive
# spatial autocorrelation in the residuals of the basic unfiltered model,
# only those eigenvectors associated with positive autocorrelation are in
# the selection set. Vice versa, for negative autocorrelation in the
# regression residuals.
#
# zerovalue: eigenvectors with eigenvalues smaller than zerovalue
# will be excluded in eigenvector search. Allows to restrict the
# search set of eigenvectors to those with extreme autocorrelation levels.

so the intention was to take both sides symmetrically. Do you have a small 
example with say simulated autocorrelation (code) to demonstrate the 
problem? It may be that a comparison somewhere is not protected by abs().

Roger

>
>
> Jonathan B. Thayn, Ph.D.
> Illinois State University
> Department of Geography - Geology
> 200A Felmley Hall
> Normal, Illinois 61790-4400
>
> (309) 438-8112
> jthayn at ilstu.edu
> my.ilstu.edu/~jthayn
>
>
>
>
>
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no