Hi Krishna,
> (1) I'd like to examine the links between group size (cont. outcome) and
> various social network measures (cont. predictors) across 10 species of
> nonhuman primates. I first tried implementing the PGLS code in Caper
> package that incorporates random-sampling of intraspecific
Hi James,
One key issue is that you’ve got multiple components in the GLS model - the
data and the covariance matrix. If you change the covariance matrix, then
changes in residuals in the model aren’t comparable.
What caper does internally is to fit the intercept-only model for the data
using
Dear Priya,
Simon Blomberg posted on this back in 2008:
https://stat.ethz.ch/pipermail/r-sig-phylo/2008-June/000128.html
His point is that scaling the variables before the analysis using their simple
standard deviations doesn�t correctly account for the phylogenetic structure. I
think there
Hi Ben,
I don’t think growTree() will help, I’m afraid. You might be able to get the
inheritance rules to get the model right but there isn’t a way in growTree to
simulate characters along an existing tree - the traits and topology are all
simulated at the same time.
David
On 20 Oct 2014, at
Hi Luke and all,
Nice explanation! Hadn’t thought it through but that makes sense that you’ll
get them for species that are less similar to their sisters than you’d expect.
With an edit to remove the line of code that objects to negative lambda, pgls()
can estimate likelihood for negative
Dear Xavier,
Also - they're two slightly different things. The F test on pgls() does look
like a bug - it isn't handling df for categorical variables correctly in
calculating the F value - the previous thread is to do with treating branch
length transformations as parameters.
Cheers,
David
Hi Adam,
Thanks for the quick explanation - and you're right, there is a bug in the
summary method for pgls that is improperly including the intercept in numerator
degrees of freedom. I've fixed it and will roll a new caper version out onto
CRAN soon.
Cheers,
David
On 27 Jan 2013, at 15:28,
Hi Adam,
At present, branch length transformations are not treated as coefficients
inside of the analysis of variance in pgls. This is something I've pondered
recently so it would be good to get opinions. My brain is a bit fried this
evening so I'm not going to weigh in with any actual
Hi,
There are a couple of Pagel references that discuss the various transformations
in pgls():
Pagel, M. D. Inferring the historical patterns of biological evolution. Nature
401, 877884 (1999).
Pagel, M. D. Detecting correlated evolution on phylogenies: a general method
for the comparative
Hi Simon,
Slightly off the top of my head, but I'd use the distribution of I' over the
1000 trees and test if it is different from the expectation of 0.5. That
ignores the variation in the expectation, so simulate 1000 random trees of the
same number of tips, and then do a t.test to compare
Hi Yanthe,
Just to second what Graham says - it sounds like a crunch problem. The crunch()
and brunch() functions in caper do both take the raw data and then calculate
contrasts in all the variables in the model simultaneously. The pic() function
in the ape package does the same for one
Hi Jay,
There isn't a function but you can get the plots directly from models. The
avPlot function can be reproduced quite simply:
library(car)
mod1 - lm(prestige ~ income+education+type, data=Duncan)
mod2 - lm(prestige ~ education+type, data=Duncan)
mod3 - lm(income ~ education+type,
attached. As you can also see,
PIC returns a very different result than PGLS.
Many thanks,
Karin
Von: Orme, David d.o...@imperial.ac.ukmailto:d.o...@imperial.ac.uk
An: Karin Schneeberger k.schneeber...@yahoo.demailto:k.schneeber...@yahoo.de
Cc: r-sig-phylo@r
Dear Karin,
Liam is right - this shouldn't happen unless there is a problem with
optimisation. It is difficult to optimise on very flat likelihood surfaces and
I've found you do sometimes get inconsistent results between different packages
with lambdas close to zero. However, getting one and
Hi Edwin,
That is a bug - I've fixed it in the developmental version on R-forge but
haven't yet rolled out a new version to CRAN. Many thanks for reporting it!
Phylogenetic independent contrasts, where comparisons need to be made between
pairs of branches, require methods adjustments for
Hi all,
I forgot to copy my reply to Luke to the list - here's another solution in case
the question crops up again.
Cheers,
David
Begin forwarded message:
Date: 20 July 2011 11:57:46 GMT+01:00
To: Luke Matthews matth...@mednetworks.commailto:matth...@mednetworks.com
Subject: Re:
16 matches
Mail list logo