[R-sig-Geo] negative local R2 in gwr

[Roger Bivand]

On Wed, 27 Jan 2010, Jorge Fernando Saraiva de Menezes wrote:

> Dear all,
>
> I have started to use gwr in data of species counts in islands of the
> western hemisphere, but  the gwr() output contains negative locals R2s,
> especially when bandwith value are low. This happen also when I use the
> optimal bandwidth choosed by gwr.sel() with CV. Does anybody knows what
> could cause this problem?
>
> Here the code that I used (all variables continuos):
>
>> attach(islands)

No, please don't use attach(), it is irrelevant here (as would coords= be 
if you use an sp class as is recommended).

>> bnd = gwr.sel(RICHNESS ~ log(AREA) + DISTANCIA, data=islands,
   coords=cbind(islands$LONG,islands$LAT))
>> gwr(RICHNESS ~ log(AREA) + DISTANCIA, data=islands,
  coords=cbind(islands$LONG,islands$LAT), bnd)
>

(do use spaces in quoted commands to reduce mailer mangling).

Please document this using a dataset that others have access to 
(preferably a built-in data set, such as meuse in sp). Please state what 
version of spgwr you are trying to use, local R2 has changed fairly 
recently, and continues not to be well founded (GWR3's local R2 cannot be 
reproduced, the current ones are from the unpublished GWR4).

With a very tight bandwidth (longlat coordinates maybe, why not use GC 
distances?), the sum of weights (reported - always store the output 
object) may be little more than unity - the local R2 could go almost 
anywhere. Have you looked at pairs() of the local coefficients?

Roger

>
> Thanks,
> Jorge Menezes
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> 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