Hi Nicolai,
Nicolai Cryer wrote on 06/01/2012 01:14:
Hi all,
I am under the impression that, given two continuous variables that can be
shown (Pagel's Lambda or Gittleman& Kot's autocorrelation) to be dependent
on phylogeny, calling gls(variable1 ~ variable 2, correlation = corStruct)
returns a regression on derived variables that have been transformed given
the known correlation structure (under some evolutionary model).
Yes, your impression is correct: these are the phylogenetically
independent contrasts.
My question is: is there any point to plotting a regression on this
internally transformed data?
Yes, see ?pic.
I have some continuous variables that are quite meaningful, but when I
manually perform the transformation I get values that are meaninglessly
high (like a forearm length of 1,2 meters, compared to 6,5 centimeters
originally).
I might be doing the manual transformation the wrong way, in which case I'd
appreciate a comment.
I guess you want to ask a quite different question that is: given the
phenotypes for species 1, species 2, ..., what would be the phenotype of
species 2 if it were /not/ evolved from a common ancestor with species
1? I think it's a more complicated issue that cannot be solved with a
simple back-transformation.
Best,
Emmanuel
For variables X and Y
varmatrix<- vcv.phylo(phy = tree, correlation = corGrafen(1, tree))
transmatrix<- chol(varmatrix)
X.transformed<- transmatrix %*% X
Y.transformed<- transmatrix %*% Y
mylm<- lm(Y ~ X)
plot(Y ~ X)
abline(coef(mylm))
The resulting estimates are different than from a normal gls, gls(y ~ x,
correlation = corGrafen(1, tree)), so I would be happy to know what it is
I'm looking for.
Thanks a lot,
Best,
NC
[[alternative HTML version deleted]]
_______________________________________________
R-sig-phylo mailing list
R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
--
Emmanuel Paradis
IRD, Jakarta, Indonesia
http://ape.mpl.ird.fr/
_______________________________________________
R-sig-phylo mailing list
R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo