[R-sig-Geo] Problem with BCtransform?
Ruben Roa
RRoa at fisheries.gov.fk
Tue May 3 20:32:50 CEST 2005
> -----Original Message-----
> From: r-sig-geo-bounces at stat.math.ethz.ch
> [mailto:r-sig-geo-bounces at stat.math.ethz.ch]On Behalf Of Ruben Roa
> Sent: 03 May 2005 10:44
> To: R-sig-Geo at stat.math.ethz.ch
> Subject: [R-sig-Geo] Problem with BCtransform?
>
>
> Hi:
>
> In geoR, when using BCtransform to back-transform the beta
> parameter into the space of the original data, like in
>
> >mean(BCtransform(rnorm(5000,mean=likfitobject$beta,sd=sqrt(li
> kfitobject$beta.var)),lambda=XXX,inverse=TRUE))
>
> i have noticed (in two applications) that the value of this
> mean is much smaller (2.3 and 3.4 times smaller) than
> the mean of the kriged predictions, obtained with
>
> >mean(krig.object$predict)
>
> even though geoR performs the back-transformation before kriging.
>
> I would have thought that the two means (from
> back-transformed beta and from kriged predictions) would be
> fairly similar. Of the two means the one that is closer to
> the mean of the raw data is the mean of the kriged
> predictions so i wonder maybe there is some problem with the
> BCtransform function when called by the
> user? Or maybe there is something fundamental that i don't
> understand about the beta parameter (and thus its
> back-transform)?
>
> Thanks for clarifications.
Because the kriging predictions are close to the actual values of
the raw data, and these kriged predictions have been computed by back-
transforming using the estimated lambda in the Box-Cox distribution,
i must conclude that the problem of the difference between the back-
transform of beta and the mean of the kriged predictions lies in the
value of beta itself. Why is beta substantially lower in my two
applications in relation to the value that it should be if its back
transform were to be close to the mean of the raw data and to the mean
of the kriged predictions?
R.
More information about the R-sig-Geo
mailing list