[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]
Re: Query on so-called "correlated" characters, reply
On May 22, 2009, at 4:18 PM, David Marjanovic wrote replies to my
Would the following characters be considered co-related and therefore
one would be unusable?
1. Height of last sacral neural spine greater than acetabulum
diameter. Height of first caudal spine greater than centrum
Tricky. I would expect a strong correlation between the height of
the last sacral neural spine and the first caudal neural spine
(though I haven't actually looked into that). The question is: how
much correlation is there (in your dataset) between acetabulum
diameter and centrum height?
These were just pulled out of imagination. No data set. And they were
meant to be tricky. Note that these two have nothing in common but an
articulation, yet seem to be correlated by virtue of proximity. On
the same subject, is scapula size correlated to humerus size? Now
think about Effigia.
2. Quadratojugal process of jugal present or absent.
Quadratojugal present or absent.
The first must be scored as unknown (which, for PAUP*, is the same
as "inapplicable") in all taxa that lack a quadratojugal.
Consider taxa that have a portion of the quadratojugal ossified only
on the quadrate and not on the jugal, whether or not a quadratojugal
process is present on the jugal.
3. Femur absent. Number of tarsal bones zero.
The second must be scored as unknown in all taxa that lack a femur
(and thus automatically also the rest of the hindlimb).
Besides, the second character has problems of its own. First, it is
correlated to certain kinds of aquatic lifestyle (like axial-based
swimming) and therefore most likely correlated to a couple other
characters in the matrix. Second, it's correlated to ontogenetic
age: for taxa known from adults, score the adults and ignore all
other specimens; for taxa not known from adults, score the highest
observed number of tarsals _or higher_ (partial uncertainty).
Together, these issues might make the character useless -- try to
figure that out (I can't do it without your dataset).
Again, no dataset. Just imagination. I'm simply showing that two
bones widely separated can be correlated. Given this, does vertebral
count also correlate with loss of limbs? I've read a paper that says
so. And if so, which character gets the boot? Do you see how it
becomes very unworkable and crazy-making if you allow yourself to
make correlation after correlation? Tail bend correlated with limb
transformation into fins. Primary feather development correlated with
And there's the opposite case:
Antorbital fenestra with or without a maxillary fossa. 'Without a
fossa' appears on chroniosuchids, pterosaurs and Proterosuchus, which
are not related. The fossa is not necessarily correlated to the
presence of a fenestra. It either is present, or it is not.
4. Naris larger/smaller than antorbital fenestra. Antorbital
fenestra present or absent.
The first must be scored as unknown in all taxa that lack an
Yet, if no antorbital fenestra is present, isn't the naris, by
default, larger than the antorbital fenestra? Why not score it that
way? How does an antorbital fenestra phylogenetically begin? I've
seen foramina that might be the start of an antorbital fenestra. Do
you ignore those?
I note that in certain recent papers and reviews this has become an
increasingly sticky subject.
Why? It's straightforward, except for my extra remarks about tarsal
Straightforward? Not if you consider skull size versus neck length.
Pubis orientation versus tooth shape and predentary presence. It gets
crazy real fast.
Point is: I've had experience separating legless squamates from one
another even though the various characters for leglessness were left
in. There remains enough phylogenetic data to separate them according
to their leggy sisters. Parsimony wins every time, even though it
would seem the odds would be stacked against it. I say, throw a ton
of characters and scores at a taxon and see how it comes out. The
sister are going to, more or less, have the same situation. Parsimony
has weight and should prevail.
IF it doesn't (and I'm just trying to see all sides of this
argument), can you reply with an example of how deleting correlated
characters and scores brings greater resolution and insight to a