Fitting weighted nlme models

[JohnM]

Hi All

I've been fitting a number of models to a database consisting of
measurements of tree biomass data from a number of different studies.
Because this is a sort of meta analysis, I'm using mixed models with
random effects included to account for methodological and site
differences between studies (following a similar approach to Wirth et
al - Tree Physiology 2004). I've also wanted to compare models fitted
to transformed (logarithmic) data with those fitted on the original
arithmetic scale using weighted analysis.

I have two issues that have arisen in the course of this analysis:

1. I have encountered some problems with specifying weights in nlme
models. For most models, I would like to weight the observations by 1/
D2H (where D2H = DBH^2*Height); however, if I use the varFixed option
in an nlme model I get an error message, e.g.,

fit.eq11.lme<-nlme(Total.tree ~ b0*DBH.cm^b1, start=c(b0=0.2, b1=2),
fixed=b0+b1~1, random=b0+b1~1|Study, weights=varFixed(~D2H))

Error in recalc.varFunc(object[[i]], conLin) :
dims [product 165] do not match the length of object [737]
In addition: Warning message:
In conLin$Xy * varWeights(object) :
longer object length is not a multiple of shorter object length

The varFixed option works fine in weighted nonlinear models fitted to
the pooled dataset using gnls. While the varPower option in nlme works
fine, I cannot figure out how to fix the power and it seems that the
software calculates the power of the covariate that produces minimum
variance estimates of the parameters (which isn't a bad thing!).
However, I would like to have control over the weighting of
observations so that I can investigate different options. Am I just
doing something really dumb here??!!

2. Once I have fitted all these models, how can I compare them? Is
Furnival's index valid for mixed effects models?

Any help or suggestions would be gratefully received

Cheers,

John

John Moore
Edinburgh Napier University
j.m...@napier.ac.uk