[R-sig-phylo] tests of correlated evolution
Alejandro Gonzalez Voyer
alejandro.gonzalezvoyer at ebc.uu.se
Thu Apr 10 11:28:27 CEST 2008
>I interpret these results as: 1) support for the lamba-gls model
>(although the ols model outperforms the others, *because* there is
>phylogenetic signal in the data set we should account for phylogeny,
>and thus shouldn't compare the ols with gls models),
--Rholf(2006) suggests one must be cautious about trying to interpret
differences in the strength of correlations obtained from OLS vs GLS.
>2) the traits
>remain correlated in lambda-gls model, meaning that their covariance
>is not determined solely by their phylogenetic association.
--I think this can be sort of misleading. As mentioned by Rohlf (2006) there
is some misunderstanding on what the use of comparative methods can be
expected to accomplish. Basically you are addressing a statistical problem:
covariation of the elements in the error term. Phylogenetic methods use the
covariance matrix between the observations to compute the correlations
between variables taking into account the lack of independence in the
observations (Rohlf 2006).
By applying phylogenetic comparative methods you do not partial out the
effect of common ancestry.
>My questions are:
>--> do these interpretations make sense?
>--> can I interpret this as 'minimal effect of phylogeny' on the
>observed correlation between these traits, because the correlation
>doesn't break down upon accounting for phylogeny?
--My answer to your second question would be no. I don't see why you would
expect the correlation to break down after accounting for phylogeny. I am
interested to read what others think about this.
From: r-sig-phylo-bounces at r-project.org
[mailto:r-sig-phylo-bounces at r-project.org] On Behalf Of Katie Wagner
Sent: den 10 april 2008 04:56
To: r-sig-phylo at r-project.org
Subject: [R-sig-phylo] tests of correlated evolution
I'm interested in the effect of phylogeny on the correlation between
two continuous variables. I've tested for phylogenetic signal using
Pagel's lambda, and it is significantly non-zero (0.8). I have fitted
GLS models using BM (both straight BM, and BM modified by the
estimated lambda value) and OU, and also OLS. The likelihoods, AIC
scores, and correlations are pasted below.
ols: logLik = 6.218705; AIC = -6.437409, r= -0.7682 (p < 0.01)
lamba gls: logLik = 4.609906; AIC = -3.219813, r= -0.6312 (p < 0.01)
straight BM gls: logLik = -18.57622; AIC = 43.15244, r = -0.1777 (p >
ou gls: logLik = 1.078378; AIC = 5.843244, r = -0.9625 (p < 0.01)
I interpret these results as: 1) support for the lamba-gls model
(although the ols model outperforms the others, *because* there is
phylogenetic signal in the data set we should account for phylogeny,
and thus shouldn't compare the ols with gls models), 2) the traits
remain correlated in lambda-gls model, meaning that their covariance
is not determined solely by their phylogenetic association.
My questions are:
--> do these interpretations make sense?
--> can I interpret this as 'minimal effect of phylogeny' on the
observed correlation between these traits, because the correlation
doesn't break down upon accounting for phylogeny?
Thanks in advance for your interpretations and comments!
Department of Ecology & Evolutionary Biology
Ithaca NY 14853
Ph: (607)254-4240 email: cew35 at cornell.edu
[[alternative HTML version deleted]]
R-sig-phylo mailing list
R-sig-phylo at r-project.org
More information about the R-sig-phylo