Felsenstein versus Ockham

R. Zander bryo at PARADOX.NET
Fri Jan 22 14:21:15 CST 1999


Nice question, Thomas. In my opinion, simplicity is justified to the extent that
nature is parsimonious; it is, but not optimally so. Both the parsimonists and the
statistical phylogeneticists err when they use optimality criteria to estimate a
point when an interval is better justified by contradictions in the data.

Here is how to choose: The only time a reconstruction is probabilistic (or
dependable or actionable -- some people don't believe that probability can apply
to single events) is when you have high Bremer support in parsimony analysis or
high posterior probability in Bayesian estimation. This is occasionally reached in
some papers, e.g. see Knox & Palmer's work in Syst. Bot. 23, 1998, see fig. 5,
which shows fairly good Bremer support throughout much of the tree, or Yang &
Rannala Mol. Biol. Evol. 17, 1997, page 722, where 95-99% post. probability is
reached.

Likelihood is a method of point estimation, and cannot deal with intervals. The
likelihood values (because they are calculated on different tree topologies)
cannot be used to generate posterior probabilities, and chi-square apparently does
not apply. Only consistency of the process seems to be a justification (aside from
philosophical justifications having to do with "best" explanations) and
consistency is a feature of the method, certainly not the data.

I've written a paper on this
Zander, R. H. Phylogenetic reconstruction, a critique. Taxon 47: 681-693. 1998.
and will send you a reprint.


Thomas Schlemmermeyer wrote:

>   Dear All,
>
>   This thread has been played on this list already some time ago. But
>   I haven't had that much theoretical trainig at that time, so, maybe,
>   it is worth to play it again, eventually with somewhat different points
> views and new insights.
>
>   Cladists, to my knowledge, apply the principle of maximum parsimony
>   (Ockham's razor) because of some rather complicated reasons, but in normal
>   everyday use simply because it is simple to use maximum parsimony.
>
>   Geneticists of the Felsenstein-Type, on the other hand, argue that there
>   are many simple reasons, why maximum parsimony can lead to the wrong tree.
>   I am now lazy and reading this famous book about comparative biology
>   by Harvey and Page. They promote a rather equilibrated approach between
>   gene statisticians of the Felsenstein-school and strict cladists of
>   the Farris, Nelson and Co-school.
>
>   It seems to me that the problem of reconstruction of evolutionary history
>   is (according to this famous book by Harvey) that the lab annotations
>   got lost. It is like reconstructing an experiment with bad protocols.
>
>   So, how good are statistical approaches to evolutionary history and how
>   good is maximum parsimony, and how to choose?
>
>        Cheers,    Thomas
>
> ------------------------------------------------------------------------------
> Thomas Schlemmermeyer
> Museu de Zoologia, Universidade de Sco Paulo
> Caixa Postal 42694
> CEP 04299-970
> Sco Paulo, SP, Brasil



--


Richard H. Zander, Curator of Botany
Patricia M. Eckel, Research Fellow in Botany
Buffalo Museum of Science
1020 Humboldt Pkwy, Buffalo, NY 14211 USA
bryo at paradox.net   voice: 716-896-5200 ext. 351




More information about the Taxacom mailing list