[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]
Post-SICB Report: a couple more (long) thoughts
Lastly, some thoughts on various aspects of the dinosaur-bird relationship
as seen by various workers at the SICB Feathers conference.
Trying to make an effort to understand the "other side", if you will, I
found some common complaints. Some have more to do with the personalities
involved than the science, so I'll have to let those pass. Others are more
A) The "hype" behind the Chinese feathered forms. Many of the folks there
thought that _Caudipteryx_ and _Protarchaeopteryx_ MUST have been flightless
birds and that Currie & company were grossly mistaken about their
I was asked "what derived feature shows that _Caudi._ was a
pre-_Archaeopteryx_ form". This is actually almost a nonsensical question:
it asks what derived feature shows that it lacks the derived features
uniting Archie and more advanced birds. It is like asking "what derived
features show that lungfish are a pre-tetrapod form?". It isn't derived
features which demonstrate that lungfish lie outside Tetrapoda: it is their
lack of the derived features uniting all tetrapods. Better questions would
be "are their features which unite _Caudi._ with birds more advanced than
Archie?" (as suggested by Martin and some others) or "are their features
which unite _Caudi._ with *previously known* non-avian theropod taxa?" (for
which see the Ostrom symposium... :-).
However, I (and some others on the net) have pointed out that the hype
around the already described featherd forms from China are not
ground-shattering if they lie in the phylogenetic position found in the
analysis presented by Ji et al. in their Nature paper. As they presented
it, _Caudipteryx_ is only one branch further out: with no known taxa in
between it and known feather forms, it isn't a surprise (or shouldn't be, in
a phylogenetic context. I suppose if some people still think that Archie
itself was THE first form to have feathers, it would be a surprise...).
Of the three different positions for _Protarchaeopteryx_, one (_P_ + (_C_ +
(Archie on up))) is again non-interesting: it doesn't optimize feathers on
any form previously known, but not known to be feathered. It is only the
other two (with _P_ as the sister group to velociraptorines, or as the
sister group to the velociraptorine + bird clade) that it becomes
interesting, because it supports the concept of feathered velociraptorines.
However, since velociraptorines were supposed to be the OUTGROUP in the
analysis, and _P_ was found in two of the three trees to be outside the
ingroup, there really isn't very strong support for the position of this
taxon anywhere within the tree.
The implications of feathered theropod fossils will only become
"interesting" when a) a feathered specimen of a previously known non-avian
group is discovered (like an oviraptorosaur or a dromaeosaur or a
tyrannosaur or an abelisaur...) (Net Alert: THESE ARE HYPOTHETICAL
EXAMPLES!! DON'T GO CITING THIS AS EVIDENCE!!) or b) a feathered form is
found on strong evidence to be closer to a previously known non-avian group
than to birds. These situations would be interesting because they would be
evidence that feathers extend deeper into the theropod tree than previously
supported (or that feathers evolved more than once...).
So, there are legitimate gripes about the hype around the Chinese fossils
*as described at present*. Indeed, these gripes are at least as stong (if
not stronger) within a cladistic context than outside of it.
B) Imagine the following hypothetical situation. Say someone (I don't know,
maybe a young charming theropod worker... :-) We'll call him "TH") proposed
that tyrannosaurids had proportionately longer metatarsi than typical
theropods of the same body size. Say that someone plotted this out
morphometrically. Say someone else (call him "JH") comes along and says "I
don't believe this result".
TH: "Okay, then, show how I am wrong."
JH: "Okay, but I'm not going to do a morphometric plot. I don't believe them."
TH: "Huh? Why?"
JH: "Because they keep on giving the same result. Even when different
inidividuals are plotted, and different workers to the plotting."
TH: ----- (semi-stunned silence)
JH: "Anyway, I found that these data points" [points to them] "are measured
incorrectly. That shows the analysis is wrong."
TH: "Well, fair enough in the absolute sense. The best-fit lines are going
to be different when replotted, as will the confidence intervals and
r-squared. Of course, it doesn't mean that the rest of the plot isn't
incorrect, or that the main pattern isn't maintained. What did you get when
you corrected those data points?"
JH: "I don't see why I should have to plot them myself. Showing those data
points are measured incorrectly are enough to show that the whole analysis
is flawed, and therefor wrong. And besides, I don't understand you
morphametricians obsession with "best-fit lines". Why should we prefer the
"best-fit lines" more than the "next best-fit"? Who ever said Nature was
TH: "No one ever did. Still, as an analytic tool, the preference is to
prefer the simplest explanation. If not the "best-fit", what criterion
should we use? The one that "feels best"?"
JH: "Well, I still think that these points" [points to a different set]
"are more important than all the others. Why can't you make them more
important in the analysis?"
TH: "We could, mathematically. Still, why that set? I know there the ones
you worked on, but JF worked on these ones over here and probably thinks
they might be more significant, and RM on these. Why should I favor yours?"
JH: "Because they will give you the right answer. The other data points
will mislead you. Look at how often morphometric plots give you totally
TH: "I thought you said the method was flawed because it always gives you
the same result, even when different data were plotted."
JH: "Anyway, those data points are probably all measured incorrectly. And
besides, why plot them at all? You morphometricians just plot up some data
points and stop with your plots. You don't try to understand the whole
animal, and its ecology, and its evolution."
TH: "Funny, I thought I was plotting these data in order to better
understand tyrannosaur locomotion and ecology."
JH: "Well, whatever. Beer?"
TH: "I can agree to beer".
Well, as you might imagine, the above is not entirely hypothetical (whoooo,
big surprise). Of course, the type of analysis involved wasn't
morphometrics, but cladistics. JH is a compiliation of various comments I
heard, often from many individuals, which I've tried to put together to try
and make some sense. Here are the main complaints about cladisitcs I heard
from multiple individuals:
1) If a character is coded incorrectly, then the whole cladogram is wrong.
- Okay, in an absolute sense the cladogram would be "incorrect" in that
the topology *might* be different and the metrics almost certainly *will*
when re-run with corrected data. Too bad there's no way to check it... Oh,
wait, there is. Re-run the data.
Furthermore, an incorrect character or characters does not guarantee
that the topology is incorrect. For example, the discovery that the
furcula is present in more theropods than just oviraptorosaurs and birds
does not in fact alter the tree topology: instead, it alters the position of
that character state change on the tree.
2) I shouldn't be forced to have to do my own cladogram.
- Okay. Still, running PAUP or Hennig86 does not hurt physically; it
does not damage the ozone; it does not contribute to crime; and it is no
more stressful than using DeltaGraph or PowerPoint or CorelDraw. It is
widely found already installed on computers in museums and universities in
various labs. So, why not do it? Why the reluctance? I still find it as
puzzling as the hypothetical example (i.e., I know that morphometric plot is
wrong, but I don't want to be forced to plot the data to show that it is wrong).
3) Cladograms always give the same answer even if each analysis uses many
different characters from each other, therefore they are flawed. However,
cladograms are always wildly different from each other, therefore they are
- I have heard both comments from the same individuals (also from
different individuals at different conferences, on this list, on sci.bio
groups, etc.). In response to the second question I would say: show that
this is the case. We can then examine the discrepencies. In response to
the first sentence I would suggest that perhaps the convergence of results,
even with different data matricies, may perhaps reflect that the method is
actually working, and that different workers are observing different suites
of characters within the same nested hierarchical pattern of nature. But
then, that's just me...
4) Why the "obsession" with parsimony?
- We've gone over that one a lot on this list: see the archives. In
brief, parsimony is equivalent to the best-fit of the line: the assumption
isn't that Nature is necessarily 'clean', but that as an analytical tool we
should prefer the explanation that fits the most data. In the case of
plotted data, we want some simple way of describing an equation that is
"close to" the most points under some metric: that's why we have best-fit
lines of various sorts. For cladistics, we want a descriptor that gives us
the fewest number of evolutionary changes required to explaine the
distribution of characters in the taxa examined. Under some assumptions
(primarily concerning types of molecular changes within genomes) we might
prefer a maximum likelihood model; for morphology, most workers use some
form of parsimony (Wagner, Dollo, etc.).
It came to my attention how little some of the people making these
comments have read up on cladistics as a method. Weighting schemes,
different models of parsimony, Bremer support, etc., seemed foreign to them.
As with a lot of stuff, I would suggest people try to look into the details
(i.e., not just the results and their implications, but the methods of
getting there) before rejecting it.
5) My character complex is better than your character complex.
- It would be great if we knew before hand which characters or suites of
characters gave the "true" answer, and which were misleading. However,
people tend to disagree on which ones will give the "true" answer. That's
why, personally, I like to see large data matricies with a fair number of
polymorphic characters rather than small 'clean' ones: the former are more
likely to capture the actual diversity of form among the taxa, whereas the
latter might be more easily swayed by a particular complex of characters of
interest to the worker who created it (a complaint I have heard about my
1994 paper, and one I've given about Gauthier's 1986 paper).
6) Data matricies are full of miscoded data.
- Some might well be. And, you know what? If you can justify the
miscoded info, you can tell the person who compilied it. They will likely
be a) happy to change it or b) show you the reason they coded it such.
Either way, somebody (maybe both) will learn something. Is this bad?
Futhermore, the claim that most data matricies are full of incorrect
data should be backed up with data. I have not seen such forthcoming.
7) We can agree on beer.
- Okay, maybe not everybody, and it doesn't have to do with cladisitics,
but I wanted to end on a positive note.
Other complaints (that "the cladists" stop with cladograms and do no more;
confusion over the difference between phylogenetic reconstruction and
phylogenetic taxonomy; equation of the personalities of some of the
individuals involved with flaws in the methodology; etc.) have been covered
elsewhere ad nauseum.
So, I've found a lot of misunderstanding about the method and a lot of
reluctance (which personally I don't understand) to learn more about it or
to try it themselves. This was by no means universal, but I heard each of
the claims above from more than one individual.
Thomas R. Holtz, Jr.
Vertebrate Paleontologist Webpage: http://www.geol.umd.edu
Dept. of Geology Email:firstname.lastname@example.org
University of Maryland Phone:301-405-4084
College Park, MD 20742 Fax: 301-314-9661