Re: Combining geometric and traditional morphometric datasets

[morphmet_moderator]

----- Forwarded message from morphmet_modera...@morphometrics.org -----

Date: Wed, 19 Mar 2014 09:38:05 -0700
From: morphmet_modera...@morphometrics.org
Reply-To: morphmet_modera...@morphometrics.org
Subject: Re: Combining geometric and traditional morphometric datasets
To: morphmet@morphometrics.org

----- Forwarded message from Joseph Kunkel <jkunk...@une.edu> -----

Date: Sat, 8 Mar 2014 15:39:59 -0500
From: Joseph Kunkel <jkunk...@une.edu>
Reply-To: Joseph Kunkel <jkunk...@une.edu>
Subject: Re: Combining geometric and traditional morphometric datasets
To: "morphmet@morphometrics.org" <morphmet@morphometrics.org>

Carmelo,

You are dealing with an Analysis of Dispersion issue that is dealt with in Rao (1965).

Rao, C. R. (1965). “Linear Statistical Inference and Its Applications.” John Wiley, New York NY.

You have a general linear model situation:

Y = XB ,

where Y is nXp with n samples and p observations on each sample including your 10 pairs of x,y landmarks plus k other linear observations making p = 20 + k.

You then have some hypothetical taxonomic or phylogenetic design matrix, X,   You want to know if the p columns of Y can be explained by the design matrix X and whether the 20 landmarks or the k linear measures afford any additional information to explain the taxonomic or phylogenetic design.  The question of correlation of the landmark data with the linear dimension data is not amenable to a bivariate correlation but each individual linear measure could have a partial correlation coefficient calculated with each landmark variable.

Analysis of Dispersion was designed to answer those questions.

Analysis of Dispersion is a multivariate extension to Analysis of Covariance.

I have R-scripts that will allow you to answer the questions I have described above which will take the aligned landmarks from landmark analysis and appended linear columns to make the Y matrix and you provide the suitable X design matrix to answer your question.

It does require that each record includes landmarks and linear measures from a single specimen.   I am not sure what you mean about the landmarks and the linear measures being in a separate partition.  The Y matrix contains all the data, landmark and linear measures, and you hypothesize the partitions in your analysis.

If you are interested I have the R-scripts available for online download.

Joe Kunkel

On Mar 8, 2014, at 4:08 AM, morphmet_modera...@morphometrics.org wrote:

----- Forwarded message from morphmet_modera...@morphometrics.org -----

    Date: Wed, 05 Feb 2014 22:32:35 -0800
     From: morphmet_modera...@morphometrics.org
     Reply-To: morphmet_modera...@morphometrics.org
     Subject: Re: Combining geometric and traditional morphometric datasets
     To: morphmet@morphometrics.org

----- Forwarded message from Carmelo Fruciano <c.fruci...@unict.it> -----

Date: Sat, 1 Feb 2014 02:35:23 -0500
From: Carmelo Fruciano <c.fruci...@unict.it>
Reply-To: Carmelo Fruciano <c.fruci...@unict.it>
Subject: Re: Combining geometric and traditional morphometric datasets
To: morphmet@morphometrics.org

morphmet_modera...@morphometrics.org ha scritto:

----- Forwarded message from Kara Feilich  -----

    Date: Fri, 17 Jan 2014 16:22:49 -0500
     From: Kara Feilich
     Reply-To: Kara Feilich
     Subject: Combining geometric and traditional morphometric datasets
     To: morphmet@morphometrics.org

Hi all,

I'm fairly new at this, so I hope this question makes sense: 
I'm trying to look for covariation and/or modularity among four  
datasets (all taken from the same individuals, with a phylogeny),  
where one dataset has Procrustes coordinates for body landmarks, and  
the other datasets use linear measures. Is there a way to look for  
(even just two-way) covariation among the datasets? I would like to  
use a partial least squares approach, but I'm not sure if the single  
dimension linear measures will play with the two dimensional  
landmarks. 

Though, if the landmark coordinates are broken down so that the x  
and y components of the coordinates are considered independent (i.e.  
if you have 10 landmarks, it's considered 20 variables), I should be  
able to just append linear measures as long as I consider them a  
separate partition, maybe? I hope?

Any ideas on how to work with geometric and traditional measures in  
tandem would be greatly appreciated. 

Hi Kara,
I'm not quite sure about what you mean in the "appending" part.
In general, however, you can use partial least squares and the  
Escoufier RV coefficient to see how and how "strongly" your blocks of  
variables (datasets) covary.
Best,
Carmelo

--
Carmelo Fruciano
Marie Curie Fellow - University of Konstanz - Konstanz, Germany
Honorary Fellow - University of Catania - Catania, Italy
e-mail c.fruci...@unict.it
http://www.fruciano.it/research/

----- End forwarded message -----

----- End forwarded message -----

-·.  .· ·.  .><((((º>·.  .· ·.  .><((((º>·.  .· ·.  .><((((º> .··.· >=-       =º}}}}}><
Joseph G. Kunkel, Research Professor
112A Marine Science Center
University of New England
Biddeford ME 04005
http://www.bio.umass.edu/biology/kunkel/

----- End forwarded message -----

----- End forwarded message -----