[R-sig-Geo] Geostatistical model with uncertainty in response variable
sq210 at cam.ac.uk
Wed Sep 13 15:19:13 CEST 2006
I wonder whether someone could advise on my problem:
I want to use AIC to conduct model selection for a geostatistical model in
which the response variable is roughly normally distributed. But each
y-value is an estimate, and so I would like to take into account variation
in the uncertainties of the estimates. Also, I want to use the Matern
structure to model spatial correlations.
I have found several options, each of which has one or more drawbacks:
1. Use gls() in nlme. I think this can do all I want, except there is no
Matern corStruct class.
2. Use likfit() or krige.bayes() in geoR. But these can't take into
account uncertainty in the response variable
3. Use another way of direct ML estimation of parameters (Hoeting et al.
2006. Ecological Applications 16:87-98). But this can't take into account
uncertainty of the response, and also doesn't give variance estimates of
the regression parameters.
Does anyone have any suggestions about how I might proceed? I'd like to
avoid having to construct a new corStruct class if possible (don't really
have the necessary expertise).
Suhel Quader, PhD
Department of Zoology
University of Cambridge
More information about the R-sig-Geo