It looks quite OK; I suspect however that instead of using
mean(s.v$gamma) for SSTot, for this mean the same weighting scheme
should have been used.
--
Edzer
On 09/02/2010 02:59 PM,
kili at grf.rs
wrote:
Hi list,
I'm not sure in my calculation of R2 coefficient of determination for
variogram. Could someone check it, or give the better way.
Here is an example:
library(sp)
library(gstat)
data(meuse)
coordinates(meuse) <- ~x+y
s.v<- variogram(log1p(zinc)~sqrt(dist), meuse)
vr.fit <- fit.variogram(variogram(log1p(zinc)~sqrt(dist), meuse), vgm(0,
"Exp", 300, 1))
## SSErr is it summ of weighted squares of residuals ?
(SSErr<-attr(vr.fit,"SSErr"))
## SStot total sum of weighted squares
weig<-s.v$np/s.v$dist^2 #fit.method fitting method, used by gstat. The
default method uses weights $N_h/h^2$ with $N_h$ the number of point pairs
and $h$ the distance.
(SStot<- sum(weig*(s.v$gamma-mean(s.v$gamma))^2) )
(R2<-1-SSErr/SStot ) # R2 coefficient of determination
Best regards,
Kili