[R-sig-Geo] Choice of Spatial weights

[Roger Bivand]

On Fri, 29 Feb 2008, stefan lhachimi wrote:

> Dear all,
>
> as a matter of curiosity does anybody know literature which discusses
> what spatial weight to choose (e.g. k-nn, single or double
> contiguity)? Or has anybod a good rule of thumb?

There are papers by Daniel Griffith, and one by Francois Bavaud, both 
prefer sparse to denser. This one is online:

http://www.unil.ch/webdav/site/imm/users/fbavaud/private/spatialweights_bavaud.pdf

> I found an article which proposes to try several specification and
> take either the one with the highest lambda  or the one with the best
> overall model fit (in a regression). But I think that is statistically
> speaking that is not very satisfactory, in particular if the
> application does not give any indication what weight to use.

Yes, this is a grey area, where different disciplines approach things 
differently. Looking at the data many times is of course a problem. Not 
infrequently, the "accepted" type of weights for a discipline is the one 
chosen, but this isn't a satisfactory answer. Opinions?

Roger

>
> Any hints?
> Best,
> Stefan
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
> 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