I have a ms. in preparation about this. Basically, it shows that there
does not seem to be any statistical problems using partial warp scores
as a multivariate data matrix in such analyses. That may be why the ms.
is delayed - it is more interesting when I can find problems!

The mulreg module in NTSYSpc can be used to regress partial warps onto
one or more independent variables taking phylogeny into account (using
generalized least squares). The "problem" is that you then have
correlations among partial warps which are not that interesting in
themselves. What should one do with the fact that, for example, partial
warp 2x is correlated with partial warp 3y either with or without taking
the lack on independence of species into account? What is useful about
partial warps is that they span the tangent space and thus (for small
variation in shape) capture all possible variation. The variables
themselves are not very interesting.

Note that there seems to be some confusion in the literature about what
comparative methods such as independent contrasts actually do. They do
not really "correct for phylogeny". In generalized least squares models
(for which the method of independent contrasts is a special case) the
estimates of the correlations between variables or regression
coefficients are unbiased whether or not you take the lack of
independence of observations (due to phylogeny) into account. What the
methods do is to down weight correlated observations to take into
account the fact that the effective sample sizes are often much smaller
than n. There is a change in their variances making the use of these
methods a statistically more conservative approach.

If there is interest, I can produce new versions of tpsRegr and tpsRelw
that take phylogenetic covariances into account. The tpsTree program
already produces the necessary phylogenetic covariance matrix. The
required changes are not that complicated.

-----------------------
F. James Rohlf
State University of New York, Stony Brook, NY 11794-5245
www: http://life.bio.sunysb.edu/ee/rohlf 

> -----Original Message-----
> From: [EMAIL PROTECTED] 
> [mailto:[EMAIL PROTECTED] On Behalf Of 
> [EMAIL PROTECTED]
> Sent: Wednesday, January 21, 2004 8:35 AM
> To: [EMAIL PROTECTED]
> Subject: Shape variables and comparative methods
> 
> 
> Dear Morphmeters,
> 
> Are there any problems (statistical, procedural, 
> philosophical) with using shape variables (such as relative 
> warps scores) as input variables in comparative analyses 
> (such as independent contrasts) to examine the correlation of 
> certain aspects of shape, with other variables?  If this has 
> been addressed before, are there references for articles 
> where this was discussed?  I haven't been able to find 
> anything in the morphmet archives (but I'm not sure if I was 
> using them correctly).
> 
> Thanks in advance,
> 
> C. Tristan Stayton
> ==
> Replies will be sent to list.
> For more information see 
> http://life.bio.sunysb.edu/morph/morphmet.html> .
==
Replies will be sent to list.
For more information see http://life.bio.sunysb.edu/morph/morphmet.html.

Reply via email to