Re: GLS

["Edzer J. Pebesma" <e.pebesma@geog.uu.nl>]

Julian Blanc wrote:
> I am trying to assess the relative importance of different spatial
> coordinates and various base functions in a spatial process. Using OLS
> regression, the standard errors for beta coefficients  are unreliable due to
> spatial autocorrelation, and generalised least squares must be used instead
> (i.e specifying "method: trend;" in the gstat command file). Now, my
> question is this: Is it possible to obtain the beta parameters (and their
> standard errors) for each coordinate/base function separately using GLS in
> GSTAT? Or is this simply made impossible by the fact that a covariance
> function is being specified?

Yes, use:

set debug = 33; # 1 (default) + 32 (GLS results)
set logfile = 'gstat.log'; # or whatever

and view the log file. It gives you both beta and Cov(beta).

> I'd also like to know whether it is possible to
> obtain the coefficient of determination (R-square), though I presume this
> can be obtained by regressing the predicted values on the sample data using
> an ordinary regression program.

It doesn't give R2, I even don't know how it's defined in GLS models. I
don't
think the solution is equal to that of OLS regression, so it may be not
so
easy (to get the right one). Tell me if you find out.

Good luck,
--
Edzer