The attitrion of taxonomic expertise

Alan Harvey aharvey at AMNH.ORG
Fri Oct 18 14:14:21 CDT 1996

James Lyons-Weiler wrote

>...I fail to see, however, why morphology can or should be excused
>from statistical tests that are now available.

I agree. If the test is available, appropriate and useful, it should be
used regardless of the type of data.

>>To identify a character state as a synapomorphic because it is not
>homoplastic on the tree it helped you choose is entirely circular.  What
>the statistical test I described avoids is relying on that circularity.
>It does this by measuring a predictable consequence of the presence of
>phylogenetic information in a matrix without using trees at all.  It
>does all of this with one of the simplest of tests (student's t-test
>for two means). This is helpful because cladistic parsimony will give you
>a set of optimal trees even for random data, and prevents the application
>of spurious, non-informative data.  I hope it's found to be helpful.

I'm intrigued; you have a method for identifying the synapomorphies within
a data set without reference to a phylogeny? Aren't synapomorphies
_defined_ in a strictly phylogenetic framework?

>> Synapomorphy relates directly to homology. We test homology by congruence or
>> 'best fit' in a context of evolutionary parsimony.
>> How can that be circular? [from Thomas Pape]

>The "best fit" part is circular; the data that are used to choose the tree
>that provides the best fit are used to test homology (the data).

This seems overstated. Proposals of homology are made on one character at a
time, not the entire data set. The above statement thus should read "The
algorithm chooses the tree that best fits the entire data set; this tree
results in hypotheses/tests [another sticky point] of homology for each
character." At worst this is a weak case of circularity, because the
character under consideration is only one of many characters in the data
set used in the evaluation. Compare this to the circularity in the
unfortunate practice of ratio-denominator correlations, in which one value
figures prominently in both the dependent and independent variable of every
data point. Or consider the problems of assigning transformation series to
a character and then using the resulting tree to study the evolution of
that character on the tree. That seems like a circularity issue of note.




Alan W. Harvey (aharvey at
Assistant Curator of Invertebrates
American Museum of Natural History
Central Park West at 79th Street
New York, NY 10024
(212) 769-5638; fax (212) 769-5783

More information about the Taxacom mailing list