[R-sig-phylo] Measuring Phylogenetic Signal

[Alyson Brokaw]

Hello Everyone,

I am working with a comparative dataset using bat morphometrics. As part of
my analysis, I want to estimate the phylogenetic signal of my variables. I
understand how to do this using R. My question is more specifically about
what kind of data I should be using when calculating the estimates.

For the purposes of my other analyses (linear regressions), I have
log-transformed my data to meet assumptions for normality. When estimating
phylogenetic signal, should I use my non-transformed, raw variables or the
transformed variables? I get slightly different outputs if I run both on
the same measure. My intuition is that using the raw values is more
interpretable, but figured I would ask some people with more experience.

Thank you for your time.

-Alyson
____________________________________________________________

Alyson Brokaw
M.S. Candidate: Biology, Humboldt State University
Cornell University '11, Ecology and Evolutionary Biology

LinkedIn Profile <http://www.linkedin.com/pub/alyson-brokaw/3a/704/820>
Follow my research journey here! <http://afbrokaw.wordpress.com/>

[[alternative HTML version deleted]]

_______________________________________________
R-sig-phylo mailing list - R-sig-***@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-***@r-project.org/

[Liam J. Revell]

Dear Alyson.

There is no general rule about this; however, my suggestion would be to
use log-scaled values. This is because on a log-scale proportional
changes in the trait are equal, independent of the magnitude of the
trait. That is, a change of 1% in mass of whale is the same as a change
in 1% in mass of a mouse. If your analysis includes both mice and
whales, then on the original scale mice may appear to be changing very
little in mass, while whales change a great deal - even if (relative to
their sizes) both groups are changing just as much. On the other hand,
if your analysis is of only whales or only mice it will make relatively
little difference whether you use log-scaled data or the original values.

I hope this is of some help. All the best, Liam

Liam J. Revell, Associate Professor of Biology
University of Massachusetts Boston
web: http://faculty.umb.edu/liam.revell/

Post by Alyson Brokaw
Hello Everyone,
I am working with a comparative dataset using bat morphometrics. As part of
my analysis, I want to estimate the phylogenetic signal of my variables. I
understand how to do this using R. My question is more specifically about
what kind of data I should be using when calculating the estimates.
For the purposes of my other analyses (linear regressions), I have
log-transformed my data to meet assumptions for normality. When estimating
phylogenetic signal, should I use my non-transformed, raw variables or the
transformed variables? I get slightly different outputs if I run both on
the same measure. My intuition is that using the raw values is more
interpretable, but figured I would ask some people with more experience.
Thank you for your time.
-Alyson
____________________________________________________________
Alyson Brokaw
M.S. Candidate: Biology, Humboldt State University
Cornell University '11, Ecology and Evolutionary Biology
LinkedIn Profile <http://www.linkedin.com/pub/alyson-brokaw/3a/704/820>
Follow my research journey here! <http://afbrokaw.wordpress.com/>
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo

_______________________________________________
R-sig-phylo mailing list - R-sig-***@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-***@r-project.org/

[Theodore Garland]

I agree with everything that Liam wrote -- right on.

Another point is that if you are looking at morphometric traits, then most
of them are probably highly positively correlated with body size. In that
case, testing for phylogenetic signal in, say, wing length, is going to be
largely redundant with testing for signal in body mass. Hence, for your
traits other than body size, you may want to analyze size-corrected values,
as described here:

Blomberg, S. P., T. Garland, Jr., and A. R. Ives. 2003. Testing for
phylogenetic signal in comparative data: behavioral traits are more labile.
Evolution 57:717–745.

Cheers,
Ted


Theodore Garland, Jr., Distinguished Professor

Department of Evolution, Ecology, and Organismal Biology (EEOB)

University of California, Riverside

Riverside, CA 92521

Office Phone: [phone]

Facsimile: [phone] (not confidential)

Email: ***@ucr.edu

http://www.biology.ucr.edu/people/faculty/Garland.html

http://scholar.google.com/citations?hl=en&user=iSSbrhwAAAAJ


Director, UCR Institute for the Development of
<http://idea.ucr.edu/>Educational
Applications <http://idea.ucr.edu/>


Editor in Chief, *Physiological and Biochemical Zoology
<http://www.press.uchicago.edu/ucp/journals/journal/pbz.html>*


Fail Lab: Episode One

http://youtu.be/c0msBWyTzU0

Post by Liam J. Revell
Dear Alyson.
There is no general rule about this; however, my suggestion would be to
use log-scaled values. This is because on a log-scale proportional changes
in the trait are equal, independent of the magnitude of the trait. That is,
a change of 1% in mass of whale is the same as a change in 1% in mass of a
mouse. If your analysis includes both mice and whales, then on the original
scale mice may appear to be changing very little in mass, while whales
change a great deal - even if (relative to their sizes) both groups are
changing just as much. On the other hand, if your analysis is of only
whales or only mice it will make relatively little difference whether you
use log-scaled data or the original values.
I hope this is of some help. All the best, Liam
Liam J. Revell, Associate Professor of Biology
University of Massachusetts Boston
web: http://faculty.umb.edu/liam.revell/

Post by Alyson Brokaw
Hello Everyone,
I am working with a comparative dataset using bat morphometrics. As part of
my analysis, I want to estimate the phylogenetic signal of my variables. I
understand how to do this using R. My question is more specifically about
what kind of data I should be using when calculating the estimates.
For the purposes of my other analyses (linear regressions), I have
log-transformed my data to meet assumptions for normality. When estimating
phylogenetic signal, should I use my non-transformed, raw variables or the
transformed variables? I get slightly different outputs if I run both on
the same measure. My intuition is that using the raw values is more
interpretable, but figured I would ask some people with more experience.
Thank you for your time.
-Alyson
____________________________________________________________
Alyson Brokaw
M.S. Candidate: Biology, Humboldt State University
Cornell University '11, Ecology and Evolutionary Biology
LinkedIn Profile <http://www.linkedin.com/pub/alyson-brokaw/3a/704/820>
Follow my research journey here! <http://afbrokaw.wordpress.com/>
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

[[alternative HTML version deleted]]

_______________________________________________
R-sig-phylo mailing list - R-sig-***@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-***@r-project.org/

[Manabu Sakamoto]

Following on from what Ted just said about size-correction - one can use a
phylogenetic regression (GLS) with the trait of interest as the dependent
variable and size as the independent variable, while simultaneously
estimating lambda. The program BayesTraits can do this (
http://www.evolution.rdg.ac.uk/BayesTraitsV3.0.1/BayesTraitsV3.0.1.html)
and I think the pgls function in the caper R package should be also able to
do this if I recall correctly.

thanks,
Manabu

Post by Theodore Garland
I agree with everything that Liam wrote -- right on.
Another point is that if you are looking at morphometric traits, then most
of them are probably highly positively correlated with body size. In that
case, testing for phylogenetic signal in, say, wing length, is going to be
largely redundant with testing for signal in body mass. Hence, for your
traits other than body size, you may want to analyze size-corrected values,
Blomberg, S. P., T. Garland, Jr., and A. R. Ives. 2003. Testing for
phylogenetic signal in comparative data: behavioral traits are more labile.
Evolution 57:717–745.
Cheers,
Ted
Theodore Garland, Jr., Distinguished Professor
Department of Evolution, Ecology, and Organismal Biology (EEOB)
University of California, Riverside
Riverside, CA 92521
Office Phone: [phone]
Facsimile: [phone] (not confidential)
http://www.biology.ucr.edu/people/faculty/Garland.html
http://scholar.google.com/citations?hl=en&user=iSSbrhwAAAAJ
Director, UCR Institute for the Development of
<http://idea.ucr.edu/>Educational
Applications <http://idea.ucr.edu/>
Editor in Chief, *Physiological and Biochemical Zoology
<http://www.press.uchicago.edu/ucp/journals/journal/pbz.html>*
Fail Lab: Episode One
http://youtu.be/c0msBWyTzU0
http://youtu.be/c0msBWyTzU0

Post by Liam J. Revell
Dear Alyson.
There is no general rule about this; however, my suggestion would be to
use log-scaled values. This is because on a log-scale proportional

changes

Post by Liam J. Revell
in the trait are equal, independent of the magnitude of the trait. That

is,

Post by Liam J. Revell
a change of 1% in mass of whale is the same as a change in 1% in mass of

a

Post by Liam J. Revell
mouse. If your analysis includes both mice and whales, then on the

original

Post by Liam J. Revell
scale mice may appear to be changing very little in mass, while whales
change a great deal - even if (relative to their sizes) both groups are
changing just as much. On the other hand, if your analysis is of only
whales or only mice it will make relatively little difference whether you
use log-scaled data or the original values.
I hope this is of some help. All the best, Liam
Liam J. Revell, Associate Professor of Biology
University of Massachusetts Boston
web: http://faculty.umb.edu/liam.revell/

Post by Alyson Brokaw
Hello Everyone,
I am working with a comparative dataset using bat morphometrics. As part of
my analysis, I want to estimate the phylogenetic signal of my

variables. I

Post by Liam J. Revell

Post by Alyson Brokaw
understand how to do this using R. My question is more specifically

about

Post by Liam J. Revell

Post by Alyson Brokaw
what kind of data I should be using when calculating the estimates.
For the purposes of my other analyses (linear regressions), I have
log-transformed my data to meet assumptions for normality. When

estimating

Post by Liam J. Revell

Post by Alyson Brokaw
phylogenetic signal, should I use my non-transformed, raw variables or

the

Post by Liam J. Revell

Post by Alyson Brokaw
transformed variables? I get slightly different outputs if I run both on
the same measure. My intuition is that using the raw values is more
interpretable, but figured I would ask some people with more experience.
Thank you for your time.
-Alyson
____________________________________________________________
Alyson Brokaw
M.S. Candidate: Biology, Humboldt State University
Cornell University '11, Ecology and Evolutionary Biology
LinkedIn Profile <http://www.linkedin.com/pub/alyson-brokaw/3a/704/820>
Follow my research journey here! <http://afbrokaw.wordpress.com/>
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at

--
Manabu Sakamoto, PhD
***@gmail.com

[[alternative HTML version deleted]]

_______________________________________________
R-sig-phylo mailing list - R-sig-***@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-***@r-project.org/

[Theodore Garland]

Yes, certainly, but if you want something like the K statistic of Blomberg
et al. (2003), which will let you compare with a lot of other traits in
their database, then you need to do the univariate calculations on
size-corrected data. Also, if you are worried about saying "signal was
significant for this trait but not that trait," then you want to make sure
your power is comparable. It is likely to be different if you use
different approaches for different traits.
Cheers,
Ted

Post by Manabu Sakamoto
Following on from what Ted just said about size-correction - one can use a
phylogenetic regression (GLS) with the trait of interest as the dependent
variable and size as the independent variable, while simultaneously
estimating lambda. The program BayesTraits can do this (
http://www.evolution.rdg.ac.uk/BayesTraitsV3.0.1/BayesTraitsV3.0.1.html)
and I think the pgls function in the caper R package should be also able to
do this if I recall correctly.
thanks,
Manabu

Post by Theodore Garland
I agree with everything that Liam wrote -- right on.
Another point is that if you are looking at morphometric traits, then

most

Post by Theodore Garland
of them are probably highly positively correlated with body size. In

that

Post by Theodore Garland
case, testing for phylogenetic signal in, say, wing length, is going to

be

Post by Theodore Garland
largely redundant with testing for signal in body mass. Hence, for your
traits other than body size, you may want to analyze size-corrected

values,

Post by Theodore Garland
Blomberg, S. P., T. Garland, Jr., and A. R. Ives. 2003. Testing for
phylogenetic signal in comparative data: behavioral traits are more

labile.

Post by Theodore Garland
Evolution 57:717–745.
Cheers,
Ted
Theodore Garland, Jr., Distinguished Professor
Department of Evolution, Ecology, and Organismal Biology (EEOB)
University of California, Riverside
Riverside, CA 92521
Office Phone: [phone]
Facsimile: [phone] (not confidential)
http://www.biology.ucr.edu/people/faculty/Garland.html
http://scholar.google.com/citations?hl=en&user=iSSbrhwAAAAJ
Director, UCR Institute for the Development of
<http://idea.ucr.edu/>Educational
Applications <http://idea.ucr.edu/>
Editor in Chief, *Physiological and Biochemical Zoology
<http://www.press.uchicago.edu/ucp/journals/journal/pbz.html>*
Fail Lab: Episode One
http://youtu.be/c0msBWyTzU0
http://youtu.be/c0msBWyTzU0

Post by Liam J. Revell
Dear Alyson.
There is no general rule about this; however, my suggestion would be to
use log-scaled values. This is because on a log-scale proportional

changes

Post by Liam J. Revell
in the trait are equal, independent of the magnitude of the trait. That

is,

Post by Liam J. Revell
a change of 1% in mass of whale is the same as a change in 1% in mass

of

Post by Theodore Garland
a

Post by Liam J. Revell
mouse. If your analysis includes both mice and whales, then on the

original

Post by Liam J. Revell
scale mice may appear to be changing very little in mass, while whales
change a great deal - even if (relative to their sizes) both groups are
changing just as much. On the other hand, if your analysis is of only
whales or only mice it will make relatively little difference whether

you

Post by Theodore Garland

Post by Liam J. Revell
use log-scaled data or the original values.
I hope this is of some help. All the best, Liam
Liam J. Revell, Associate Professor of Biology
University of Massachusetts Boston
web: http://faculty.umb.edu/liam.revell/

Post by Alyson Brokaw
Hello Everyone,
I am working with a comparative dataset using bat morphometrics. As

part

Post by Theodore Garland

Post by Liam J. Revell

Post by Alyson Brokaw
of
my analysis, I want to estimate the phylogenetic signal of my

variables. I

Post by Liam J. Revell

Post by Alyson Brokaw
understand how to do this using R. My question is more specifically

about

Post by Liam J. Revell

Post by Alyson Brokaw
what kind of data I should be using when calculating the estimates.
For the purposes of my other analyses (linear regressions), I have
log-transformed my data to meet assumptions for normality. When

estimating

Post by Liam J. Revell

Post by Alyson Brokaw
phylogenetic signal, should I use my non-transformed, raw variables or

the

Post by Liam J. Revell

Post by Alyson Brokaw
transformed variables? I get slightly different outputs if I run both

on

Post by Theodore Garland

Post by Liam J. Revell

Post by Alyson Brokaw
the same measure. My intuition is that using the raw values is more
interpretable, but figured I would ask some people with more

experience.

Post by Theodore Garland

Post by Liam J. Revell

Post by Alyson Brokaw
Thank you for your time.
-Alyson
____________________________________________________________
Alyson Brokaw
M.S. Candidate: Biology, Humboldt State University
Cornell University '11, Ecology and Evolutionary Biology
LinkedIn Profile <http://www.linkedin.com/pub/

alyson-brokaw/3a/704/820>

Post by Theodore Garland

Post by Liam J. Revell

Post by Alyson Brokaw
Follow my research journey here! <http://afbrokaw.wordpress.com/>
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at

--
Manabu Sakamoto, PhD
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

[[alternative HTML version deleted]]

_______________________________________________
R-sig-phylo mailing list - R-sig-***@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-***@r-project.org/

[David Bapst]

Alyson-

Following off of what Liam said, one thing to consider is as most measures
of phylogenetic signal aren't relative to the units of the traits
considered, any transformation of the data should be about equally
interpretable. To take a spin with Liam's example, if , if the log-scale
trait had high phylogenetic signal, you would infer that large things
(Liam's whales) really are more likely to be related to other large things
(and the same for small things), in the case where you reduce variation
among the bigger things (whales), and treat the smaller things (mice) as if
they had more variance.

For example, you might think the large size of 'whales' in your dataset
reflects signal (evolutionary conservatism - simply that they are all big
partly because they share common ancestors that had a big size), but if the
sizes of large things (whales) vary much more than all the variation among
your small things ('mice'), a measurement of signal on a raw scale might
think that maybe that is not very good signal, as much more evolution
change had to occur along the branches linking large/'whale' species than
the branches linking related 'mice'. This is a pretty typical situation.

I suppose one might not want to log scale if they thought that there was
not much evolutionary difference among big things, but a lot among small
things, but this just reflected higher rates of change between small
things. So, to defend not log-scaling, you'd basically need to argue that
evolutionary size change *doesn't* scale with size (or doesn't scale
positively, at least), but evolution-scales-with-size seems like one of
those things we generally assume prima facie is true in biology, so I
suppose you'd need to have a pretty good explanation why that would be.

Uh, I hope that philosophizing made sense.

(And yeah, I'm sure someone in the peanut gallery will point out that our
choice of a log scaling is entirely arbitrary, because who really knows
what the proper scaling of biological data along size gradients are
anyway....)

-Dave Bapst
Geology & Geophysics, Texas A & M University

Post by Alyson Brokaw
Hello Everyone,
I am working with a comparative dataset using bat morphometrics. As part of
my analysis, I want to estimate the phylogenetic signal of my variables. I
understand how to do this using R. My question is more specifically about
what kind of data I should be using when calculating the estimates.
For the purposes of my other analyses (linear regressions), I have
log-transformed my data to meet assumptions for normality. When estimating
phylogenetic signal, should I use my non-transformed, raw variables or the
transformed variables? I get slightly different outputs if I run both on
the same measure. My intuition is that using the raw values is more
interpretable, but figured I would ask some people with more experience.
Thank you for your time.
-Alyson
____________________________________________________________
Alyson Brokaw
M.S. Candidate: Biology, Humboldt State University
Cornell University '11, Ecology and Evolutionary Biology
LinkedIn Profile <http://www.linkedin.com/pub/alyson-brokaw/3a/704/820>
Follow my research journey here! <http://afbrokaw.wordpress.com/>
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-

--
David W. Bapst, PhD
Asst Research Professor, Geology & Geophysics, Texas A & M University
Postdoc, Ecology & Evolutionary Biology, Univ of Tenn Knoxville
https://github.com/dwbapst/paleotree
Google Calendar: https://goo.gl/EpiM4J

[[alternative HTML version deleted]]

_______________________________________________
R-sig-phylo mailing list - R-sig-***@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
Searchable archive at http://www.mail-archive.com/r-sig-***@r-project.org/