-------- Original Message --------
Subject: Re: Multiple Partial Least Squares analyses
Date: Thu, 5 Feb 2009 06:47:30 -0800 (PST)
From: Philipp Mitteroecker <[email protected]>
To: [email protected]
References: <[email protected]>
Dear Brendan,
"If I have several different sets of shape variables, say N, and I want
to find out which shapes co-vary in a significant manner"
I think it is not really a meaningful question whether parts covary in a
"significant manner". Most major components of higher organisms, such as
vertebrates, covary in size and/or shape, especially if you think of
parts of the cranium. It may be more interesting to address "how" and to
what extent they covary, or how this covariation affects evolution.
"It seems like I should also do something akin to a Bonferroni
correction" Once you have a reasonable 0-hypothesis and you perform
several independent tests, you should use a correction. This does not
depend on sample size.
When applying two-block PLS to several blocks of variables, as in Bastir
& Rosas 05, there is the problem that the covariances between the PLS
scores (the singular values) depend on the sample size relative to the
number of variables. The covariance (and also correlation) would be
expected to be larger between blocks with few variable, as compared to
blocks with more variables measured on the same sample. We pointed this
out in Mitteroecker & Bookstein 2007 Syst Biol. However, the patterns of
shape variation (the singular warps) are not affected by this problem.
In the end, PLS is an exploratory tool or for multivariate "calibration".
Best,
Philipp
On Mi, 28.01.2009, 21:49, morphmet wrote:
-------- Original Message --------
Subject: Multiple Partial Least Squares analyses
Date: Wed, 28 Jan 2009 12:29:17 -0800 (PST)
From: Brendan McCane <[email protected]>
To: [email protected]
G'day Morphers,
I have a question about performing PLS on multiple shape variable sets.
If I have several different sets of shape variables, say N, and I want
to find out which shapes co-vary in a significant manner, can I just
perform N(N-1)/2 separate PLS analyses using permutation tests to test
for significance? It seems like I should also do something akin to a
Bonferroni correction, but I haven't noticed anyone else do that.
For example, in the following paper:
Markus Bastir, Antonio Rosas, "Hierarchical Nature of Morphological
Integration and Modularity in the Human Posterior Face", American
Journal of Physical Anthropology, 128:26-34, 2005.
The authors test the covariation between the midline cranial base,
middle cranian fossa, and the mandibular ramus by performing 3
independent PLS analyses.
Should a Bonferroni-like correction be applied in cases like this, or is
the permutation solid even for large N? Any advice would be appreciated.
--
Cheers,
Brendan.
----------------------------------------------------------------------------
Brendan McCane, Head of Dept, Email: [email protected]
Department of Computer Science Phone: +64 3 479 8588/8578.
University of Otago Fax: +64 3 479 8529
Box 56, Dunedin, New Zealand.
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
--
Dr. Philipp Mitteröcker
Konrad Lorenz Institute for Evolution and Cognition Research, Austria
Department of Theoretical Biology, University of Vienna, Austria
http://www.virtual-anthropology.com/Members/philippm
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
