Re: [MORPHMET] relative positions of landmark partitions in integration tests - how important?

[David Katz]

Now if only archaeological and museum skeletal samples would stop
disarticulating and stay in anatomical position.

On Wed, Jan 24, 2018 at 11:29 AM, David Katz  wrote:

> No argument here.
>
> On Wed, Jan 24, 2018 at 11:12 AM, Adams, Dean [EEOBS]  > wrote:
>
>> David,
>>
>>
>>
>> Yes, fixing the angle first allows for a single GPA, which then retains
>> relative size information across the substructures. And yes, it puts things
>> in a common orientation. Both have major advantages for downstream PLS, and
>> its interpretation. Many PLS-based integration analyses in GMM examine
>> covariation in substructures from a single larger object subjected to a
>> single GPA.  I point to the classic paper of Bookstein et al. 2003 as an
>> exemplar.
>>
>>
>>
>> I would argue that it is in such cases where integration patterns are
>> most interpretable, as with a single GPA one is able to characterize
>> covariation patterns among sets of variables whose spatial relationships
>> have been retained throughout the analysis. This is rather important. When
>> this isn’t the case, it is much more challenging to derive interpretable
>> estimates. This is what I was alluding to in the last part of my previous
>> post.
>>
>>
>>
>> Dean
>>
>>
>>
>> Dr. Dean C. Adams
>>
>> Professor
>>
>> Department of Ecology, Evolution, and Organismal Biology
>>
>>Department of Statistics
>>
>> Iowa State University
>>
>> www.public.iastate.edu/~dcadams/
>>
>> phone: 515-294-3834 <(515)%20294-3834>
>>
>>
>>
>> *From:* katz.w...@gmail.com [mailto:katz.w...@gmail.com] *On Behalf Of *David
>> Katz
>> *Sent:* Wednesday, January 24, 2018 11:48 AM
>> *To:* Adams, Dean [EEOBS] 
>> *Cc:* Christy Anna Hipsley ; MORPHMET <
>> morphmet@morphometrics.org>
>> *Subject:* Re: [MORPHMET] relative positions of landmark partitions in
>> integration tests - how important?
>>
>>
>>
>> I read Christy's question a little differently, and requiring some
>> clarification.
>>
>>
>>
>> First, Dean, doesn't Dean & Felice fix the angle between jaw and cranium
>> so that you can subject a craniomandibular dataset to a common GPA, which
>> at the PLS step has the benefit of preserving relative size relationships
>> between the two structures? On the other hand, it's not clear if Christy
>> obtained landmarks in a way that allows her to do this as easily. For
>> instance, her sample may have had crania and mandibles dissected out and
>> landmarked separately, in which case she subject the cranium and mandible
>> to separate GPAs. Christy?
>>
>>
>>
>> If GPA was performed separately for the cranium and mandible, then
>> couldn't Christy's issue simply be that cranium and mandible have
>> differently oriented principal axes (as determined by gpagen)?
>>
>>
>>
>> Also, Christy, are you using the "plot.pls" function (warpgrids=TRUE) for
>> plotting? If so, then I think one option would be to make shapes=TRUE in
>> that function. This will give you an output that includes the coordinates
>> of the extreme shapes (I assume as *p**3 matrices). After that, you just
>> have to choose which two columns from crania extremes and which two columns
>> from mandible extremes provide you with common cranial and mandibular
>> views. You can then make 2D plots your landmarks using the base R plotting
>> functions. Unfortunately, you won't have warpgrids. But you can overlay two
>> (e.g., mandible) configurations to show the difference between extremes. Or
>> you can plot one extreme as the ball-and-stick wireframe and the change to
>> the other extreme as displacement vectors emanating from the wireframe.
>>
>>
>>
>> David
>>
>>
>>
>> On Wed, Jan 24, 2018 at 10:15 AM, Adams, Dean [EEOBS] <
>> dcad...@iastate.edu> wrote:
>>
>> Christy,
>>
>>
>>
>> That data example contained variation in relative jaw position among
>> specimens, which could affect shape estimates, as well as down-stream shape
>> analyses.  Several approaches have been proposed for dealing with such
>> rotational variation (see Adams 1999; also Bookstein’s Orange book). One
>> approach is to rotate one subset of landmarks so the angle between subsets
>> is invariant across specimens. The geomorph function ‘fixed.angle’ performs
>> this operation for 2D landmark datasets.
>>
>>
>>
&

Re: [MORPHMET] relative positions of landmark partitions in integration tests - how important?

[David Katz]

No argument here.

On Wed, Jan 24, 2018 at 11:12 AM, Adams, Dean [EEOBS] 
wrote:

> David,
>
>
>
> Yes, fixing the angle first allows for a single GPA, which then retains
> relative size information across the substructures. And yes, it puts things
> in a common orientation. Both have major advantages for downstream PLS, and
> its interpretation. Many PLS-based integration analyses in GMM examine
> covariation in substructures from a single larger object subjected to a
> single GPA.  I point to the classic paper of Bookstein et al. 2003 as an
> exemplar.
>
>
>
> I would argue that it is in such cases where integration patterns are most
> interpretable, as with a single GPA one is able to characterize covariation
> patterns among sets of variables whose spatial relationships have been
> retained throughout the analysis. This is rather important. When this isn’t
> the case, it is much more challenging to derive interpretable estimates.
> This is what I was alluding to in the last part of my previous post.
>
>
>
> Dean
>
>
>
> Dr. Dean C. Adams
>
> Professor
>
> Department of Ecology, Evolution, and Organismal Biology
>
>Department of Statistics
>
> Iowa State University
>
> www.public.iastate.edu/~dcadams/
>
> phone: 515-294-3834 <(515)%20294-3834>
>
>
>
> *From:* katz.w...@gmail.com [mailto:katz.w...@gmail.com] *On Behalf Of *David
> Katz
> *Sent:* Wednesday, January 24, 2018 11:48 AM
> *To:* Adams, Dean [EEOBS] 
> *Cc:* Christy Anna Hipsley ; MORPHMET <
> morphmet@morphometrics.org>
> *Subject:* Re: [MORPHMET] relative positions of landmark partitions in
> integration tests - how important?
>
>
>
> I read Christy's question a little differently, and requiring some
> clarification.
>
>
>
> First, Dean, doesn't Dean & Felice fix the angle between jaw and cranium
> so that you can subject a craniomandibular dataset to a common GPA, which
> at the PLS step has the benefit of preserving relative size relationships
> between the two structures? On the other hand, it's not clear if Christy
> obtained landmarks in a way that allows her to do this as easily. For
> instance, her sample may have had crania and mandibles dissected out and
> landmarked separately, in which case she subject the cranium and mandible
> to separate GPAs. Christy?
>
>
>
> If GPA was performed separately for the cranium and mandible, then
> couldn't Christy's issue simply be that cranium and mandible have
> differently oriented principal axes (as determined by gpagen)?
>
>
>
> Also, Christy, are you using the "plot.pls" function (warpgrids=TRUE) for
> plotting? If so, then I think one option would be to make shapes=TRUE in
> that function. This will give you an output that includes the coordinates
> of the extreme shapes (I assume as *p**3 matrices). After that, you just
> have to choose which two columns from crania extremes and which two columns
> from mandible extremes provide you with common cranial and mandibular
> views. You can then make 2D plots your landmarks using the base R plotting
> functions. Unfortunately, you won't have warpgrids. But you can overlay two
> (e.g., mandible) configurations to show the difference between extremes. Or
> you can plot one extreme as the ball-and-stick wireframe and the change to
> the other extreme as displacement vectors emanating from the wireframe.
>
>
>
> David
>
>
>
> On Wed, Jan 24, 2018 at 10:15 AM, Adams, Dean [EEOBS] 
> wrote:
>
> Christy,
>
>
>
> That data example contained variation in relative jaw position among
> specimens, which could affect shape estimates, as well as down-stream shape
> analyses.  Several approaches have been proposed for dealing with such
> rotational variation (see Adams 1999; also Bookstein’s Orange book). One
> approach is to rotate one subset of landmarks so the angle between subsets
> is invariant across specimens. The geomorph function ‘fixed.angle’ performs
> this operation for 2D landmark datasets.
>
>
>
> As to your comment on whether or not such positional variation makes a
> difference, yes it can. PLS examines the degree of covariation between
> blocks of variables and estimates of the between block covariation will
> differ if one set of variables is rotated relative to the other. Whether
> this results in a large or small difference in r-PLS values is
> data-dependent, but the values will not be the same.
>
>
>
> For this reason, prior to any PLS analysis for evaluating integration and
> covariation patterns, one must first carefully consider what type of
> variables are being utilized, how they were generated, and whether it

RE: [MORPHMET] relative positions of landmark partitions in integration tests - how important?

[Adams, Dean [EEOBS]]

David,

Yes, fixing the angle first allows for a single GPA, which then retains 
relative size information across the substructures. And yes, it puts things in 
a common orientation. Both have major advantages for downstream PLS, and its 
interpretation. Many PLS-based integration analyses in GMM examine covariation 
in substructures from a single larger object subjected to a single GPA.  I 
point to the classic paper of Bookstein et al. 2003 as an exemplar.

I would argue that it is in such cases where integration patterns are most 
interpretable, as with a single GPA one is able to characterize covariation 
patterns among sets of variables whose spatial relationships have been retained 
throughout the analysis. This is rather important. When this isn’t the case, it 
is much more challenging to derive interpretable estimates. This is what I was 
alluding to in the last part of my previous post.

Dean

Dr. Dean C. Adams
Professor
Department of Ecology, Evolution, and Organismal Biology
   Department of Statistics
Iowa State University
www.public.iastate.edu/~dcadams/<http://www.public.iastate.edu/~dcadams/>
phone: 515-294-3834

From: katz.w...@gmail.com [mailto:katz.w...@gmail.com] On Behalf Of David Katz
Sent: Wednesday, January 24, 2018 11:48 AM
To: Adams, Dean [EEOBS] 
Cc: Christy Anna Hipsley ; MORPHMET 

Subject: Re: [MORPHMET] relative positions of landmark partitions in 
integration tests - how important?

I read Christy's question a little differently, and requiring some 
clarification.

First, Dean, doesn't Dean & Felice fix the angle between jaw and cranium so 
that you can subject a craniomandibular dataset to a common GPA, which at the 
PLS step has the benefit of preserving relative size relationships between the 
two structures? On the other hand, it's not clear if Christy obtained landmarks 
in a way that allows her to do this as easily. For instance, her sample may 
have had crania and mandibles dissected out and landmarked separately, in which 
case she subject the cranium and mandible to separate GPAs. Christy?

If GPA was performed separately for the cranium and mandible, then couldn't 
Christy's issue simply be that cranium and mandible have differently oriented 
principal axes (as determined by gpagen)?

Also, Christy, are you using the "plot.pls" function (warpgrids=TRUE) for 
plotting? If so, then I think one option would be to make shapes=TRUE in that 
function. This will give you an output that includes the coordinates of the 
extreme shapes (I assume as p*3 matrices). After that, you just have to choose 
which two columns from crania extremes and which two columns from mandible 
extremes provide you with common cranial and mandibular views. You can then 
make 2D plots your landmarks using the base R plotting functions. 
Unfortunately, you won't have warpgrids. But you can overlay two (e.g., 
mandible) configurations to show the difference between extremes. Or you can 
plot one extreme as the ball-and-stick wireframe and the change to the other 
extreme as displacement vectors emanating from the wireframe.

David

On Wed, Jan 24, 2018 at 10:15 AM, Adams, Dean [EEOBS] 
mailto:dcad...@iastate.edu>> wrote:
Christy,

That data example contained variation in relative jaw position among specimens, 
which could affect shape estimates, as well as down-stream shape analyses.  
Several approaches have been proposed for dealing with such rotational 
variation (see Adams 1999; also Bookstein’s Orange book). One approach is to 
rotate one subset of landmarks so the angle between subsets is invariant across 
specimens. The geomorph function ‘fixed.angle’ performs this operation for 2D 
landmark datasets.

As to your comment on whether or not such positional variation makes a 
difference, yes it can. PLS examines the degree of covariation between blocks 
of variables and estimates of the between block covariation will differ if one 
set of variables is rotated relative to the other. Whether this results in a 
large or small difference in r-PLS values is data-dependent, but the values 
will not be the same.

For this reason, prior to any PLS analysis for evaluating integration and 
covariation patterns, one must first carefully consider what type of variables 
are being utilized, how they were generated, and whether it even makes sense to 
interpret the results biologically.

Dean

Dr. Dean C. Adams
Professor
Department of Ecology, Evolution, and Organismal Biology
   Department of Statistics
Iowa State University
www.public.iastate.edu/~dcadams/<http://www.public.iastate.edu/~dcadams/>
phone: 515-294-3834

From: Christy Anna Hipsley 
[mailto:chips...@unimelb.edu.au<mailto:chips...@unimelb.edu.au>]
Sent: Tuesday, January 23, 2018 9:34 PM
To: MORPHMET mailto:morphmet@morphometrics.org>>
Subject: [MORPHMET] relative positions of landmark partitions in integration 
tests - how important?

Hi all,

I'm trying to

Re: [MORPHMET] relative positions of landmark partitions in integration tests - how important?

[David Katz]

I read Christy's question a little differently, and requiring some
clarification.

First, Dean, doesn't Dean & Felice fix the angle between jaw and cranium so
that you can subject a craniomandibular dataset to a common GPA, which at
the PLS step has the benefit of preserving relative size relationships
between the two structures? On the other hand, it's not clear if Christy
obtained landmarks in a way that allows her to do this as easily. For
instance, her sample may have had crania and mandibles dissected out and
landmarked separately, in which case she subject the cranium and mandible
to separate GPAs. Christy?

If GPA was performed separately for the cranium and mandible, then couldn't
Christy's issue simply be that cranium and mandible have differently
oriented principal axes (as determined by gpagen)?

Also, Christy, are you using the "plot.pls" function (warpgrids=TRUE) for
plotting? If so, then I think one option would be to make shapes=TRUE in
that function. This will give you an output that includes the coordinates
of the extreme shapes (I assume as *p**3 matrices). After that, you just
have to choose which two columns from crania extremes and which two columns
from mandible extremes provide you with common cranial and mandibular
views. You can then make 2D plots your landmarks using the base R plotting
functions. Unfortunately, you won't have warpgrids. But you can overlay two
(e.g., mandible) configurations to show the difference between extremes. Or
you can plot one extreme as the ball-and-stick wireframe and the change to
the other extreme as displacement vectors emanating from the wireframe.

David

On Wed, Jan 24, 2018 at 10:15 AM, Adams, Dean [EEOBS] 
wrote:

> Christy,
>
>
>
> That data example contained variation in relative jaw position among
> specimens, which could affect shape estimates, as well as down-stream shape
> analyses.  Several approaches have been proposed for dealing with such
> rotational variation (see Adams 1999; also Bookstein’s Orange book). One
> approach is to rotate one subset of landmarks so the angle between subsets
> is invariant across specimens. The geomorph function ‘fixed.angle’ performs
> this operation for 2D landmark datasets.
>
>
>
> As to your comment on whether or not such positional variation makes a
> difference, yes it can. PLS examines the degree of covariation between
> blocks of variables and estimates of the between block covariation will
> differ if one set of variables is rotated relative to the other. Whether
> this results in a large or small difference in r-PLS values is
> data-dependent, but the values will not be the same.
>
>
>
> For this reason, prior to any PLS analysis for evaluating integration and
> covariation patterns, one must first carefully consider what type of
> variables are being utilized, how they were generated, and whether it even
> makes sense to interpret the results biologically.
>
>
>
> Dean
>
>
>
> Dr. Dean C. Adams
>
> Professor
>
> Department of Ecology, Evolution, and Organismal Biology
>
>Department of Statistics
>
> Iowa State University
>
> www.public.iastate.edu/~dcadams/
>
> phone: 515-294-3834 <(515)%20294-3834>
>
>
>
> *From:* Christy Anna Hipsley [mailto:chips...@unimelb.edu.au]
> *Sent:* Tuesday, January 23, 2018 9:34 PM
> *To:* MORPHMET 
> *Subject:* [MORPHMET] relative positions of landmark partitions in
> integration tests - how important?
>
>
>
> Hi all,
>
>
>
> I'm trying to run an integration test in geomorph, or rather
> phylo.integration on 2 sets of Procrustes aligned coordinates for cranium
> and jaws of lizards, landmarked on both sides. When I plot the results I
> get graphs of the landmarks for the positive and negative extremes of
> PLS1&2, but for the cranium they are in lateral view while for the
> mandibles they are in frontal view.
>
> I'm wondering if this is an issue for the estimation of r-PLS, since in
> Adams & Felice 2004 (PLoS ONE: Assessing Trait Covariation and
> Morphological Integration on Phylogenies Using Evolutionary Covariance
> Matrices), they write "the position of the jaw was standardized relative to
> the skull by rotating the jaw to a common articulation angle among
> specimens".
>
>
>
> If it is an issue, how do I rotate the position of one of the partitions
> to be in the same orientation as the other?
>
>
>
> Thanks for any advice!
>
> Christy
>
>
>
> *Dr Christy Anna Hipsley | ARC DECRA Fellow *
>
> School of BioSciences, University of Melbourne/
>
> Museums Victoria
>
> GPO Box 666
>
> Melbourne, Victoria 3001 Australia
>
> *T:* +61 3 8341 7423 <+61%203%208341%207423> *E:*
> christy.hips...@unimelb.edu.au; chips...@museum.vic.gov.au
>
> http://blogs.unimelb.edu.au/hipsleylab/
>
> --
> MORPHMET may be accessed via its webpage at http://www.morphometrics.org
> ---
> You received this message because you are subscribed to the Google Groups
> "MORPHMET" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to morphmet+unsu

RE: [MORPHMET] relative positions of landmark partitions in integration tests - how important?

[Adams, Dean [EEOBS]]

Christy,

That data example contained variation in relative jaw position among specimens, 
which could affect shape estimates, as well as down-stream shape analyses.  
Several approaches have been proposed for dealing with such rotational 
variation (see Adams 1999; also Bookstein’s Orange book). One approach is to 
rotate one subset of landmarks so the angle between subsets is invariant across 
specimens. The geomorph function ‘fixed.angle’ performs this operation for 2D 
landmark datasets.

As to your comment on whether or not such positional variation makes a 
difference, yes it can. PLS examines the degree of covariation between blocks 
of variables and estimates of the between block covariation will differ if one 
set of variables is rotated relative to the other. Whether this results in a 
large or small difference in r-PLS values is data-dependent, but the values 
will not be the same.

For this reason, prior to any PLS analysis for evaluating integration and 
covariation patterns, one must first carefully consider what type of variables 
are being utilized, how they were generated, and whether it even makes sense to 
interpret the results biologically.

Dean

Dr. Dean C. Adams
Professor
Department of Ecology, Evolution, and Organismal Biology
   Department of Statistics
Iowa State University
www.public.iastate.edu/~dcadams/
phone: 515-294-3834

From: Christy Anna Hipsley [mailto:chips...@unimelb.edu.au]
Sent: Tuesday, January 23, 2018 9:34 PM
To: MORPHMET 
Subject: [MORPHMET] relative positions of landmark partitions in integration 
tests - how important?

Hi all,

I'm trying to run an integration test in geomorph, or rather phylo.integration 
on 2 sets of Procrustes aligned coordinates for cranium and jaws of lizards, 
landmarked on both sides. When I plot the results I get graphs of the landmarks 
for the positive and negative extremes of PLS1&2, but for the cranium they are 
in lateral view while for the mandibles they are in frontal view.
I'm wondering if this is an issue for the estimation of r-PLS, since in Adams & 
Felice 2004 (PLoS ONE: Assessing Trait Covariation and Morphological 
Integration on Phylogenies Using Evolutionary Covariance Matrices), they write 
"the position of the jaw was standardized relative to the skull by rotating the 
jaw to a common articulation angle among specimens".

If it is an issue, how do I rotate the position of one of the partitions to be 
in the same orientation as the other?

Thanks for any advice!
Christy

Dr Christy Anna Hipsley | ARC DECRA Fellow
School of BioSciences, University of Melbourne/
Museums Victoria
GPO Box 666
Melbourne, Victoria 3001 Australia
T: +61 3 8341 7423 E: 
christy.hips...@unimelb.edu.au; 
chips...@museum.vic.gov.au
http://blogs.unimelb.edu.au/hipsleylab/
--
MORPHMET may be accessed via its webpage at http://www.morphometrics.org
---
You received this message because you are subscribed to the Google Groups 
"MORPHMET" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to 
morphmet+unsubscr...@morphometrics.org.

-- 
MORPHMET may be accessed via its webpage at http://www.morphometrics.org
--- 
You received this message because you are subscribed to the Google Groups 
"MORPHMET" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to morphmet+unsubscr...@morphometrics.org.