Re: Transformational and taxic approaches to character coding [was Re: Philosophies for Character Ordering]

["Jonathan R. Wagner" <jonathan.r.wagner@mail.utexas.edu>]

David Marjanovic wrote (quoting me):

> I'm among those who don't like getting a cc of everything, because this
> means getting each e-mail twice within the same minute.

    We've been through this all before on the list. I consider it a
necessary courtesy, especially for those of us who receive the List in
digest format, because we can't spend our days sorting through the
individual postings (most of which seem to come from you, I notice ;).


> > Note that I specifically
> > code for the presence of cotylar caudals as a similarity among these
> > organisms, even though the picky anatomical details (i.e., which end is
> > cotylar) are different.
>
> Is it really a picky detail which end is cotylar?

    That was a reference to the phrasing of my earlier post, not a value
judgement. The point being, should a difference in which end is condylar
keep you from coding the fact that taxa share caudals with condylar ends?


> Actually it must be coded as inapplicable for those that have 1(0).

    Oh, fer cryin' out loud... Yes, you are correct, it is inapplicable...
for the 99% of us who use PAUP, it is treated as the same thing as
inapplicable. It is best to maintain a difference when coding. You CAN code
characters as inapplicable for PAUP, it is just treated as unknown.


> NONA is able to deal with inapplicable codings.

    I was not aware of this! Can you direct me to a reference on how it
handles them algorithmically.


> Hennig86 claims to be, but just treats them as unknown...

    Hennig86 has more problems than that... If you are still using it, you
might want to consult the literature.


> In PAUP* at least, this doesn't matter, because PAUP* makes unrooted trees
> that are a posteriori rooted with the outgroup (if you don't tell it
> something else).

    This is true, unless you use the "easy out" coding that Wilson
advocates. In such cases, the cost to transform between states is different
in different directions, and the tree MUST be rooted a priori of analysis
(i.e., it is not time-reversable). I do not know of any theoretical
discussion of what this can do to the analysis, say, vs. unrooted trees.
Offhand, I'd say that (unlike with unrooted trees), you had better get your
outgroup right or it could significantly affect ingroup topology.


> Why not:
>
> 1. Proximal articulation: (0) cotylar, (1) flat, (2) condylar
> 2. Distal articulation: (0) cotylar, (1) flat, (2) condylar

    Sure... I suppose... but why? I tend to limit coded states to those that
are observed in the data (since you could concieve of an infinite number of
alternate state to code, some of which could easily be interpolated between
observed states in an ordered matrix.
    And I would still throw in a character for presence of condylar
articulations.

> This allows for biconvex (22), amphiplatyan (11) and amphicoelous (00)
> vertebrae as well as procoelous (02) and opisthocoelous (20) ones. These
two
> characters are (almost?) independent, because only (01) and (10) don't
occur
> in reality (or do they?),

    There is a saying that, whatever vertebral articulation you desire, it
can be found in a turtle.


> and they're certainly logically independent. --

    Actually, not true. Arguably, presence of a cotyle on one end of a
vertebra suggests we should logically expect to find a condyle on the other.
This is not always the case, but that is a statement of *practical*
independence, not logical independence.


> This way of coding does however bring up the problem of whether to order
> these characters. And here we should really know something about their
> developmentary genetics.

    And, to go back to my original post: WHY? Why should we know something
about their "developmental genetics?" Why does it matter? :)

>
> > 1. Caudal vertebrae: 0) procoelous, 1) amphicoelous, 2) opisthocoelous
> > [ORDERED]
>
> ...(3) biconvex, (4) amphiplatyan [FANCY STEP MATRIX]? Sure, if you're
> lucky, you don't have taxa with those morphologies in your matrix, but...

    Well, I meant this example mostly as a response to Tim Williams' point,
not as a suggestion. Besides, youstill have to come up with a justifcation
of the weights in your step matrix.

>
> > You can also selectively unorder this character to
> > see if the relevant parts of the analysis are sensitive to that
> > "assumption." In my limited experience, such sensitivity is rare.
>
> But to do this test usually increases the impressiveness factor of your
> results (by showing how robust they are),

I never said you shouldn't do the test, just that I've found that analyses,
even with relatively small matrices, usually don't change much in response
to unordering one character.


> and therefore perhaps the impact factor of your paper, too. =8-)

    Not likely... I'd say that has more to do with journal readership,
subdiscipline, and scientific politics. But maybe I'm just cynical... :)


    Wagner