Message-ID: <Pine.LNX.4.44.0701272207070.7109-100000@reclus.nhh.no>
Date: 2007-01-27T21:11:53Z
From: Roger Bivand
Subject: test the OLS presumption by a GWR model?
In-Reply-To: <20070126215608.35190@gmx.net>
On Fri, 26 Jan 2007, Marco Helbich wrote:
> Dear list members,
>
> i have calculated a GWR model and concerning to this, i have a question:
> should i test the 'local' gwr output with the standard 'global' test
> procedures of a OLS model (e.g. Breusch-Pagan-Test)? If yes, which one?
The authors of the GWR book suggest some tests, but since Wheeler &
Tiefelsdorf (2005) in Journal of Geographical Systems, I would suggest
that caution in inference from standard (non-Bayesian) GWR is sensible.
You could do BP tests on each iteration if the tests are well-defined for
weighted regression. Most tests are however not set up to do this, for
example bptest in the lmtest package has no weights= argument.
Roger
>
> best regards
> Marco
> __________________
>
> Marco Helbich
> ISR
> Postgasse 7/4/2
> A-1010, Vienna
>
>
--
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