-------- Original Message --------
Subject: Re: classification of unknown individuals
Date: Tue, 29 Nov 2011 07:51:24 -0500
From: Ondřej Mikula <onmik...@gmail.com>
To: morphmet@morphometrics.org
Dear Claudia,
it could be helpful to work with a symmetric component of shape
variation only.
Take the known pairs of bones, for each pair take average of their
Procrustes shape coordinates and run PCA on these mean shapes. Project
bones of unknown affiliation on these principal components and pair
them according to Euclidean distances in this PC space. The projection
is achieved as follows:
- take the grand mean of symmetric mean shapes, subtract it from shape
coordinates of each bone to be projected and store the resulting shape
coordinates as rows of the matrix X
- calculate X * P, where P is matrix whose columns are loadings of
subsequent principal components.
Best
Ondra
On 28 November 2011 23:12, morphmet
<morphmet_modera...@morphometrics.org> wrote:
-------- Original Message --------
Subject: classification of unknown individuals
Date: Mon, 28 Nov 2011 14:27:23 -0500
From: Garrido-Varas, Claudia <c.garrido-va...@tees.ac.uk>
To: 'morphmet@morphometrics.org' <morphmet@morphometrics.org>
Dear Morphometricians,
I have the following problem. I have analyzed 32 pairs of bones, each pair
belonging to a single individual. I have reflected the lefts and calculated
the asymmetry between sides and between individuals using the ANOVA in
MorphoJ:
Shape, Procrustes ANOVA:
Effect SS MS df F P (param.)
Pillai tr. P (param.)
Individual 0.05135962 0.0001183401 434 2.85 <.0001
9.73 <.0001
Side 0.00034556 0.0000246831 14 0.59 0.8707
0.44 0.4744
Ind * Side 0.01805056 0.0000415911 434 24.25 <.0001
10.95 <.0001
Error 1 0.00153700 0.0000017154 896 1.38 <.0001
5.38 <.0001
Residual 0.00223224 0.0000012457 1792
When I do PCA the bones (Right and Left from a same individual) are
displayed with no clear association between them, for example the right bone
of individual 1 is closer to other individuals than to its left pair. But
when I do a regression of Procrustes coordinates and centroid size, and
correct for allometry each individual clusters very clearly.
So, my question is how I can use this information to classify unknown pairs.
How for example I could do the following, I have 10 pairs that I know they
belong to single individuals and I have 4 pairs (assuming they belong in 2
pairs) that are mixed among them. How can I see them in a graph and see that
they cluster with the right pair?
Based in that the differences between individuals is greater than intra
individual, I have been trying, unsuccessfully to show this graphically.
Thank you very much for your advice.
Claudia
--
Institute of Animal Physiology and Genetics
Academy of Sciences of the Czech Republic
Veveri 97, 60200 Brno, Czech Republic
Institute of Vertebrate Biology
Academy of Sciences of the Czech Republic
Studenec 122, 67502 Konesin, Czech Republic
