Re: MorphoJ question

[morphmet_moderator]

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

Date: Sat, 06 Oct 2012 15:55:43 -0700
From: morphmet_modera...@morphometrics.org
Reply-To: morphmet_modera...@morphometrics.org
Subject: Re: MorphoJ question
To: morphmet@morphometrics.org

----- Forwarded message from Chris Klingenberg -----

Date: Fri, 5 Oct 2012 18:35:34 -0400
From: Chris Klingenberg
Reply-To: Chris Klingenberg
Subject: Re: MorphoJ question
To: morphmet@morphometrics.org

Dear Daniela

I'm afraid, omitting one half of the skulls is not going to increase the sample size. If you include the landmarks in the median plane (and average over reflected and relabeled copies of each specimen), using those and the paired landmarks of one side is going to be exactly the same as using the symmetric component from the analysis with object symmetry. The reason for this is that the paired landmarks on one side perfectly predict what their matches on the other side are doing...

The dimensionality of the symmetric component of variation is roughly (exactly for 2D applications) that of the total variation (or an analysis ignoring the symmetry of the skulls). So it is pretty much the same as omitting half the skull, except for the fact that it does not just ignore the other half. In particular, the dimensionality of just one side plus median landmarks (after Procrustes fit for object symmetry, averaged over reflections) is the same as that of the symmetric component with all landmarks. (If you don't do the averaging over reflections, the dimensionality of one side plus median landmarks will be larger than that of the symmetric component.)

There is only a difference if you use software that does not deal with singular covariance matrices well. All standard morphometrics programs should do reasonably well, because ill-conditioned or singular covariance matrices are very frequent. Programs like SAS and the likes will also handle this similarly.

So, on second thought, throwing out the paired landmarks from one side won't help. Because it only creates additional trouble (with visualizing shape changes, etc.), it;'s better not to do it.

I hope this clarifies the issue.

Best wishes,
Chris

On 05/10/2012 20:13, morphmet_modera...@morphometrics.org wrote:

----- Forwarded message from Daniela Sanfelice -----

Date: Thu, 4 Oct 2012 10:54:42 -0400
From: Daniela Sanfelice
Reply-To: Daniela Sanfelice
Subject: MorphoJ question
To: morphmet@morphometrics.org

Hey dear morphometricians!

To improve my samples size, I will recalculate all my projects and results about morphological integration in pinnipeds skulls using just one skull side.

So... no more object symmetry! However,  superposition was done considering object symmetry...

Can I just exclude the landmarks anyway, and after include the new specimens (with only the left side information)?

I suppose not... but if I can´t do this way, how can I do it?

Should I exclude the landmarks from all the raw data file by hand or there is another easier way?

Many thanks for your help and contributions!

Cheers, Daniela Sanfelice.

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

--

***********************************
Christian Peter Klingenberg
Faculty of Life Sciences
University of Manchester
Michael Smith Building
Oxford Road
Manchester M13 9PT
United Kingdom

E-mail: c...@manchester.ac.uk
Phone: +44 161 2753899
Skype: chris_klingenberg
***********************************

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

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