Tony,
gls() fits the model by ML or REML (the default), and the parameters in
the correlation structure (e.g., lambda) are estimated if needed or
specified.
Cheers,
Emmanuel
Anthony R Ives wrote on 21/04/2011 17:45:
Emmanuel,
I don't know gls, so I'd have to look at this. Technically, th
Emmanuel,
I don't know gls, so I'd have to look at this. Technically, the ME
problem can't be solved with GLS. In the literature, people seem to
use GLS to describe models with any form of non-zero correlation
structure, while my understanding is that GLS is the estimation
technique
b =
Tony,
It's possible to specify a fixed variance function in gls(), eg:
vf <- varFixed(~ v)
gls(y ~ x, correlation = cor, weights = vf)
where v is the measurement error (can be heterogeneous among
abservations). Variance functions can also be combined:
vf <- varComb(varFixed(~ vx), varFixed(~
Scott,
I would love it if somebody would. If I were doing it now, I'd
simultaneously use an OU or maybe lambda transform. Several people
have code that does, for example, regression while assuming residual
variation has some non-Brownian phylogenetic structure (e.g., Lavins
et al. 2008).
Thanks Arne,
Apologies for not noticing the similar post.
Scott
On Wednesday, April 20, 2011 at 10:27 AM, Arne Mooers wrote:
> Hi Scott:
>
> Tony hasn't, Liam has estimates for the slope, but no c.i.s on that slope.
> I'm playing around with this too at the moment (just posted a query re.
>
Dear R users,
I am curious if anyone has re-written the Matlab code given in Ives et al. 2007
Syst. Biol. (for including within species measurement error into examining
relationships among traits) for use in R.
Thanks! Scott
[[alternative HTML version deleted]]
___