-------- Original Message --------
Subject: Re: common allometric components and residual shape components
Date: Fri, 9 Mar 2012 09:22:20 -0500
From: [email protected]
To: [email protected]
Hi all,
concerning Drake and Klingenberg regression scores
vector (S).....
using CAC.... I get it when considering my data as ONE
group; I used the rcc() function from the package
"CCA". This function can manage ill-conditioned X'X
(and/or Y'Y) matrices. Using "1" for both lambda1 and
lambda2 returns results IDENTICAL to MorphoJ;
thanks to Dean Adams for pointing out this in his
previous message
Dealing with multiple groups, MorphoJ offers the
possibility to perform a "pooled within groups"
regression; in R, clearly, I tought reasonable to
perform SEPARATE CCAs for any group and then to append
(rbind() in R) the Y scores of any CCAs. But when
plotted against my X variable (i.e. log(centroid
size)) I DID NOT GET the same results of "pooled
within groups" regression scores performed in MorphoJ.
Can anyone say me if I miss something?
Thankyou in advance
Paolo
-------- Original Message --------
Subject: Re: common allometric components and residual
shape components
Date: Thu, 8 Mar 2012 11:12:06 -0500
From: Dean Adams <[email protected]>
To: [email protected]
Eric,
Actually, if your specimens comprise only a single
group, the CAC of
Mitteroecker et al. (2004) and the regression score
(S) from Drake and
Klingenberg (2008) will be identical. The reason is
that both are based
on size-regressions of mean-centered shape variables,
which for 1 group
are obtained from the same mean (i.e. the 'global'
mean and group mean
are the same in this case). You can confirm this
quite simply in R by
obtaining both the CAC and the S values for the same
data and
correlating/plotting them. Of course, for multiple
groups the CAC and S
will be different, as CAC is found from
group-mean-centered data.
This is something that should have been made explicit
in the previous
literature, but was not pointed out (or perhaps not
appreciated).
Dean
--
Dr. Dean C. Adams
Associate Professor
Department of Ecology, Evolution, and Organismal
Biology
Department of Statistics
Iowa State University
Ames, Iowa
50011
www.public.iastate.edu/~dcadams/
phone: 515-294-3834
On 3/8/2012 9:56 AM, morphmet wrote:
-------- Original Message --------
Subject: Re: common allometric components and
residual shape components
Date: Wed, 7 Mar 2012 17:44:06 -0500
From: [email protected]
To: [email protected]
thanks, and i've got my figures now.
I'm also comparing appending the coordinate data
with centroid size and
running the prcomp() in R to the output from
MorphoJ, in which I
regressed
PC1 (from residual data) against the regression
scores of shape
coordinates vs centroid size. If they are the same
or similar, then the
regression scores are the same or similar to the so
called CAC, and
PC1 of
size corrected data is equivalent to the so called
RSC1.
eric
-------- Original Message --------
Subject: Re: common allometric components and
residual shape
components
Date: Wed, 7 Mar 2012 11:29:04 -0500
From: Aki Watanabe <[email protected]>
To: [email protected]
Hi Eric,
For R, you can use prcomp(), instead of princomp().
The former uses
spectral decomposition, so it doesn't give you an
error when you have
more variables than specimens.
Cheers,
Aki
On Wed, Mar 7, 2012 at 10:54 AM, morphmet
<[email protected]
<mailto:[email protected]>>
wrote:
-------- Original Message --------
Subject: common allometric components and
residual shape components
Date: Wed, 7 Mar 2012 07:27:41 -0500
From: [email protected]
<mailto:[email protected]>
To: [email protected]
<mailto:[email protected]>
Hi,
I have a question about common allometric
components and residual
shape
components, or CAC and RSC, and how RSCs relate
to PCs generated
from
size-corrected data.
So, the CAC is a regression line, calculated
using a pooled within
group
regression of coordinate data on size. And if
so, is this line also
referred to as a pooled allometric vector? And
will MorphoJ allow
one to
use this vector in a regression with another
variable, such as the
RSC? Or
is it a better bet to obtain the allometric
vector in another
program,
such as R.
Using the residuals from that regression and
doing a PCA, the PC1
is the
same as the RSC? Is this correct? If so,
MorphoJ will certainly do
this.
One more issue: doing PCA using R...I have
problems with an error
about
having too many variables (relative to the
number or rows, I
suppose). How
do I work around this in R?
Eric
U of Calgary
--
Aki Watanabe
Department of Biological Science
Florida State University
King Life Science Building
319 Stadium Drive
Tallahassee, FL 32306-4295
University of Chicago - AB '09
Biological Sciences and Geophysical Sciences
Website:
http://sites.google.com/site/akinopteryx/home
Weblog: http://akiopteryx.blogspot.com/
--
Paolo Piras
Center for Evolutionary Ecology
and
Dipartimento di Scienze Geologiche, Università Roma Tre
Largo San Leonardo Murialdo, 1, 00146 Roma
Tel: +390657338000
email: [email protected]
