Circularity and testing.

A. Contreras-Ramos atilano at IBIOLOGIA.UNAM.MX
Fri Oct 18 17:10:24 CDT 1996


James:
>Congruence itself cannot provide a critical test for
>homology because conditions exist where 1) congruence through convergence
>will be misleading, and 2) incongruence through relatively simple
>geneaological processes (i.e., coalescence) results from more than one
>true phylogenetic history for different characters.  The accumulation of
>more and more data itself is an entirely inductive exercise; how do you
>know when you have enough data?  How do know you wouldn't find
>incongruence over the next hill?  Such is the natural of pure empiricism,
>but tests exist to tackle these problems, and more are under development.

It may not be a critical test, but under most working conditions is the
best we can do. Each character is a hypothesis of homology, and often a
best guess.  Again, we don't know (and will never know for sure) what is
synapomorphous and what is convergent.  But we can accumulate evidence in
favor (or against) of a hypothesis of relationships, and so for certain
characters to be plesiomorphies, synapomorphies, homoplasies, and so on,
but this can change with further evidence.  There is never enough data,
hypotheses gain support but never become statements of truth.

>naturally, we are apt to come to disparate conclusions when we change our
>assumptions.  Where is the test?
What I tried to say is that redoing an analysis, for instance by changing
what characters we use or their weight, as well as by deleting or
incorporating new characters and using new sources of data (e.g., adding
larval characters to adult ones or molecular data), is a way of testing a
previous hypothesis of relationships (and so for the shynapomorphies
supporting such relationships). If a larval tree is congruent with an
adult tree (or some of the groupings), I take it as if the characters
supporting the structure of those trees receive mutual support, and that
would be the test (each tree or analysis tests the other).

>So if I have 80 characters, homology assumed, get a tree (chosen* by the
>criterion of maximum parsimony), then I have a phylogeny.  Then I look at
>synapomorphies on the tree.  I check out my original assessment of
>homology by adding more data; I get a new tree; this is a phylogeny; I
>look again at the synapomorphies on the tree, to check my original
>hypothesis of homology.  Somewhere in there is a critical test?  BTW,
>can't homoplasious character states be homologous?  So how can a
>hypothesis of synapomorphy test a hypothesis of homology?  When it comes
>down to it, the outcome of the test depends upon which data are collected
>first.
We have to be clear that we are working within certain conceptual
assumptions, let's say the cladistics or a homology and parsimony
paradigm, then we obtain* a tree or set of trees.  Maybe there is no
critical test, but each time you perform your analysis (e.g., with more
data) you are testing a hypothesis of relationships and of the characters
that support it, but you need something to change or one would be doing
the same exercise over and over. You are probably right if you mean that
this is not statistical testing.  I guess parsimony tests our first hand
proposals of homology.  What might be to us a perfectly good homology
(and potential synapomorphy) may turn out into a convergence after an
analysis, but that character could potentially reemerge as a synapomorphy
if more data is incorporated. These are somewhat unordered thoughts,
please don't take them too serious.  ;-) Bye now.


Atilano Contreras-Ramos
Inst. Biologia, Depto. Zoologia; U.N.A.M.;
Apdo. Postal 70-153; 04510 Mexico, D.F.; MEXICO.
(525) 622-5705,06,12; exts. 286,287 (oficina)
(525) 550-0164 (fax); (525) 568-0213 (casa)
atilano at ibiologia.unam.mx
species at servidor.unam.mx




More information about the Taxacom mailing list