Philosophical views of phyl. reconst.

Ronald DeBry histone at acpub.duke.edu
Tue May 16 07:09:37 EST 1995


In article <wfl1-140595101149 at 132.236.3.65>,
Warren Frank Lamboy <wfl1 at cornell.edu> wrote:
>Can somebody please define or point to a reference that defines the three
>philosophical views underlying phylogenetic inference that were recently
>characterized by Joe Felsenstein as:
>
>
>
>         hypothetical-deductive
>
>         logical parsimony
>
>         statistical?
>
>
>I am particularly interested in the features of each that distinguish them
>from one another, and, of course, what assumptions are inherent in each. 
>
>Thank you.
>-- 
>Warren F. Lamboy                            "It's easy if you know how to
>                                             do it, but it's impossible if
>                                             you don't know how to do it."




I'll let someone who actually subscribes to the cladist school of
philosophical justification describe the hypothetico-deductive and
logical parsimony approaches.

The statistical approach dates back to the work of
Cavalli-Sforza and Edwards in the early-mid sixties.  In essence, they
suggested that we view the phylogenetic tree as an unknown parameter
that we are trying to estimate.  The data are usually traits
(morphology, DNA sequences, etc) of extant taxa.

While some methods are explicitly "statistical", i.e. maximum
likelihood, any method can be viewed in a statistical framework.  For
example, in parsimony analysis the tree length can be viewed as a test
statistic.  A major part of parsimony analysis is finding the shortest
tree, and that does not change.  In the statistical approach, we do not
just focus on one shortest tree.  The shortest tree is an estimate of
the phylogeny, just like the sample mean is an estimate of a population
mean.  In almost every case, there are alternative trees that are nearly
as well supported as the most parsimonious tree.  A goal of the
statistical approach is to estimate which alternative trees cannot be
rejected as being significantly worse compared to the best tree.

Even in the statistical approach, the ultimate goal would be to reduce
the confidence set of non-rejected trees down to just a single tree.
Such a small confidence set is not justified with most current datasets,
and in the statistical approach we are willing to accept a level of
uncertainty in our estimates of a phylogeny.

Because any method can be viewed as a statistical method, the
assumptions of the statistical approach are the assumptions of the
actual method used.  Thus parsimony has one set of assumptions, while
maximum likelihood has another.

I hope this helps some.

Ron DeBry
Dept. of Medicine
Box 3380
Duke Univ. Medical Center
Durham, NC  27710

histone at acpub.duke.edu

D
D
D
D
I'll let someone who actually uses the hypotheico-deductive



More information about the Mol-evol mailing list