[R-sig-Geo] spatial regression with generalized least squares

[Paulo Justiniano Ribeiro Jr]

Dear G Allegri

I'v e being running through old email here and gound this message from
yours posted a while ago

well, you could for instance, model the covariance matrix using a
correlation function of the distances using nlme() with a
spatial covariance structed or likfit() in package geoR.
Both will estimate mean (regression) and covariance jointly
(instead of alternating between OLS and veriogram fits)


Paulo Justiniano Ribeiro Jr
LEG (Laboratorio de Estatistica e Geoinformacao)
Universidade Federal do Parana
Caixa Postal 19.081
CEP 81.531-990
Curitiba, PR  -  Brasil
Tel: (+55) 41 3361 3573
Fax: (+55) 41 3361 3141
e-mail: paulojus AT  ufpr  br
http://www.leg.ufpr.br/~paulojus

-------------------------------------------------------------------------
53a Reuniao Anual da Regiao Brasileira da Soc. Internacional de Biometria
14 a 16/05/2008, UFLA, Lavras,MG
http://www.rbras.org.br/rbras53
-------------------------------------------------------------------------


On Thu, 31 Jan 2008, G. Allegri wrote:

> I have a question about using GLS estimation within the Regression
> Kriging framework.
> In Rossiter and Hengl texts it is stated that it makes not so much
> difference, in many practical situations, using OLS rather then GLS.
> I'd like to test it in my work.
> What's the more feasible way to adobt it in R?
> The RK method suggests to use iteratively the regression coefficient
> estimates using the covariance matrix derived from the residual
> covariance modelling.
>
> 1 - How to automate this iteration scheme? I'm not so expert in R scripting...
> 2 - Could gls, from the nlme package, be used?
>
> Giovanni
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>