R help - May 2004 - variation explained by the non-spatial and spatial component using variogram models fitted to residuals of a non-spatial model (GeoR)

Hello all!

 

I fitted a variogram model using GeoR to regression model residuals. I would
like to assess how much of the variation 

is explained by the non-spatial regression model and the spatial model. I’ve
got the log likelihood, AIC and deviance respectively residual SS and R2 for
the non-spatial regression model, but only the log likelihood for the
spatial model. Are there any possibilities to get from the log likelihood to
the deviance or SS and thereby R2 of the spatial model? 

 

I tried to get from the deviance to the log likelihood for models where I
had both quantities, but failed by using the formulas given in text books.
This might be due to that R does not report the scaled deviance. By dividing
by the reported scaling parameter, I did not get the same value either. A
problem however is that I did not have the likelihood for the full model. I
tried to calculate it from the sample variance as proposed by McCullagh and
Nelder: -0.5*log(2*pi*sigma2). Is that possible?

 

 

Any suggestions how to solve my problem?

 

Swantje

 

 


	[[alternative HTML version deleted]]