[R-sig-Geo] Considering "boundary effect" in spdep or spgwr

[Roger Bivand]

On Thu, 28 Sep 2006, Hisaji ONO wrote:

> Hello.
> 
>  In some "geo-spatial analysis" text books & articles,
> "boundary effect" was referred.
> 
>  For spatial weight matrix and its related statistical
> methods, this issue is very important, I think.
> 
>  For this issue, I've looked for this in "spdep" and
> "spgwr" in their "manual" pdf files, but no luck.
> 
>  Current version of these packages consider "boundary
> effect"? 
> 

Yes, you are right that boundary (or edge) effects can be important. They 
concern the extent to which inferences may change if we change the edges 
of the study area, and/or the configuration effect that occurs when some 
observations are fully surrounded (in the centre of the study area) and 
others are on the edge. 

There is an indirect reference to this in the joincount.test() example in
spdep - it uses different settings of the spatial weights style= argument
to show how weights configuration can change inference. So in spdep, the
only treatment is through the style= argument.

In spgwr, fitting adaptive bandwidths seems to be one way of preventing
observations far from their neighbours being impacted by a single global
distance bandwidth. The GWR book doesn't have edge effects in its index.

Best wishes,

Roger

> 
> 
>  Regards.
> 
> 
> 
> 
> 
> 
> 

-- 
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