[R-sig-Geo] GWR

[Roger Bivand]

On Mon, 4 Dec 2006, CARMEN RODRIGUEZ wrote:

> Attached you will find the scripts and results from R and a comparation of 
> the coefficients from both R and GWR 3.0. Hope you can help me to figure out 
> the reason of the big differences.

Two reasons:

1) The GWR adaptive proportion to include is stated to be about 23 
observations of 50, the R one was about 19. 

2) You were using the arguments in gwr() wrongly. When doing adaptive GWR,
you should pass the output of gwr.sel() through the adapt= argument, and
should not include bandwidth=. For your data, with adapt=col.dd and
adapt=(23/50), you get results much more like GWR3 - for 23/50 agreeing
pretty closely.

Hope this helps,

Roger

> 
> Thank you very much.
> 
> 
> >From: Roger Bivand <Roger.Bivand at nhh.no>
> >Reply-To: Roger.Bivand at nhh.no
> >To: CARMEN RODRIGUEZ <carm_rodrig at hotmail.com>
> >CC: r-sig-geo at stat.math.ethz.ch
> >Subject: Re: [R-sig-Geo] GWR
> >Date: Wed, 29 Nov 2006 11:33:56 +0100 (CET)
> >
> >On Wed, 29 Nov 2006, CARMEN RODRIGUEZ wrote:
> >
> > > I insist on GWR. To run a GWR model I'm using both R and GWR 3.0. 
> >Comparing
> > > the estimated coefficients for an adaptative bisquare kernel in R with 
> >the
> > > results for an adaptative kernel in GWR 3.0 (which is supposed to be a
> > > bisquare function too), differences are very big. I check the bandwidths 
> >and
> > > they are not that different (as I understand, in R the adaptative 
> >bandwidth
> > > is expressed as a percentage of observations, while in GWR 3.0 the 
> >bandwidth
> > > is the number of observations itself). But anyway, the estimated
> > > coefficients shouldn't be so different, should they?
> >
> >Well, without documentation, it is impossible to tell. So please provide
> >documentation, access to data and scripts of what you are doing that
> >worries you. We cannot see over your shoulder.
> >
> >Roger
> >
> > >
> > > Thank you.
> > >
> > > _________________________________________________________________
> > > Descubre la descarga digital con MSN Music. Más de un millón de 
> >canciones.
> > >
> > > _______________________________________________
> > > 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
> >
> 
> _________________________________________________________________
> Acepta el reto MSN Premium: Protección para tus hijos en internet. 
> Descárgalo y pruébalo 2 meses gratis. 
> http://join.msn.com?XAPID=1697&DI=1055&HL=Footer_mailsenviados_proteccioninfantil
> 

-- 
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