Re: Kerberosaurus manakini

[Mickey Mortimer <Mickey_Mortimer111@msn.com>]

Virginia Friedman wrote-

> Just for the record, what is this "Sereno-esque must stop"?

By "The Sereno-esque "analyses" must stop!", I meant phylogenetic "analyses"
of the kind exemplified by Paul Sereno (e.g. 1999, 2000) should not be
presented as attempts to determine the true phylogeny from unbiased data.
Not to say that Paul is the only guilty party here- far from it.  He is
simply the best known in the dinosaur community and has really obvious
examples of it in his papers.
The basic problem can be easily outlined.  Phylogenetic analyses are
analyses.  They need to analyze something, in this case phylogenetic
relationships among a set of taxa.  They do this by weighing alternatives
using a set of predetermined criteria (parsimony, likelihood, etc.).  Each
alternative is a different hypothesis of relationships, and the one that
fits the criteria best is considered the best supported hypothesis.
Sereno's matrices and resulting topologies are not analyses in my mind
because there is a certain topology that was selected for a priori*.  They
are not testing anything.  There are very few characters and codings
included that would support any other topology.  This can be objectively
determined by the Consistancy Index (CI).  This can range from close to 0 to
1.00.  If every character only evolved once in your phylogeny, and never
reversed, you'd have a CI of 1.00.  If every character either evolved
convergently in two clades or reversed once after it evolved, you'd have a
CI of 0.5.  In other words, CI is calculated by dividing the number of times
a character changes state (the number of steps in the tree) by the number of
characters in the analysis.  The analyses in Sereno's 1999 paper had CI's
ranging from .89-.97.  Holtz's 2000 analysis has a CI of .44 for comparison.
The Theropod Working Group's latest analysis (Hwang et al., 2004) has a CI
of .43.  Mine has a CI of .28 right now.  You'll note Bolotsky and
Godefroit's had a CI of .92.  There is no reason to even run these matrices
through PAUP.  The resulting topology is obvious by just looking at the
matrix.

In Bolotsky and Godefroit's, characters 1-5 were all designed to group
hadrosaurids together to the exclusion of the outgroup.  Characters 6-8 are
designed to group hadrosaurines together.  9-12 and 13(1) are just
lambeosaurine apomorphies (so 9 out of 21 characters were useless, how about
that?).  Characters 13(2), 14 and 15 are designed to group
brachylophosaurins together, while character 16 groups the rest of the
hadrosaurines together to the exclusion of brachylophosaurins.  13(3) and 17
group saurolophins and edmontosaurins together.  21 is meant to group
saurolophins together, 18 and 19 to group Prosaurolophus with Saurolophus,
and 20 to group edmontosaurins together.  Only two codings disagree with
this.  Maiasaura lacks the hadrosaurine character 7, and lambeosaurines also
have saurolophin character 21.

Compare this to Hwang et al.'s (2004) matrix.  Of their first 21 characters,
characters 8,  9(2), 11, 16(2) and 20(2) have to evolve twice, characters 6,
7, 12, 14, 15, 18(2), 19 and 21 have to evolve three times, and characters
9(1), 10, 16(1), 17(1) and 20(1) four times.

The difference should be obvious.  If you have a phylogenetic hypothesis and
want to present the data, show a cladogram and describe the character
support for each clade, but don't go running it in PAUP as if your topology
was an unexpected result of testing unbiased data to find a most
parsimonious tree.  And certainly don't act as if your topology is equally
or better supported than alternatives, since you never tested it against
alternatives.

* Jaime informs me pers. comm. from Sereno indicates that in his case at
least, he runs analyses with a greater number of discordant characters, then
later prunes the discordant characters from the matrix before publication.
My statement still stands, as the result of the pruned analysis is known a
priori.  Additionally, I wonder why he would prune these characters, since
he is only misleading others into thinking his analyses are better supported
than they actually are, and robbing readers of the data these characters
would illustrate by being included in the matrix.  Also, I wonder why this
part of Sereno's methodology is never stated in his papers.

Mickey Mortimer
Undergraduate, Earth and Space Sciences
University of Washington
The Theropod Database - http://students.washington.edu/eoraptor/Home.html