[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

Re: Post-SICB Report: a couple more (long) thoughts

In a message dated 1/14/99 2:21:08 PM EST, Edels@email.msn.com writes:

<< Because his intuition is so often right, JH would be reluctant to give
 up a position just because you used mathematics to show he was wrong.  For
 some people, mathematical analysis is not enough.  (And for some others, it
 is often questionable - or suspect).  The contradictory idea that an
 analysis can be run by multiple people and - a) get the same answer all the
 time, or b) be wildly different - is merely a defensive posture by those who
 are uncomfortable with mathematics, and is a feeble attempt to discredit the
 viable of the mathematical analytical method. >>

Mathematics certainly doesn't make >me< uncomfortable with cladistics!

Although Tom's SICB posts are most enjoyable, he doesn't quite hit the nail on
the head in descerning the key problem with cladistic analysis. He may be
fairly representing the views of others, but these seem largely to be straw
men. For example, it is by no means a point against cladistics that one may
run analyses of several different data matrices for a particular group and
obtain the same cladogram from each. In fact, this is actually a >good< thing,
and shows that one has very likely discovered the correct phylogeny! I can't
understand how this can be an objection against cladistic analysis itself.

The key problem with cladistic analysis is that it is >inherently<
unfalsifiable. You have a character matrix, you have a computer with
appropriate software, you plug and chug, and voila, a cladogram appears. How
do you >know< that this cladogram is the True Cladogram for the group? Or to
put it in Popperian terms, how can you tell whether it is wrong? Suppose it
doesn't accord with, say, biogeography or stratigraphy or functional anatomy
or physics? The cladist will say that the biogeography or the stratigraphy or
the functional anatomy or the physics is wrong, and must be fitted, like one
of Procrustes's victims, to conform to the cladogram. The (imperious) attitude
of the cladist is that the cladogram and the corresponding phylogeny must come
first, that phylogeny will illuminate biogeography and functional anatomy, but
not vice versa.

Cladistic analyses using morphological data are frequently in conflict with
analyses using molecular data. In any such instance, which method is to be
preferred? Why? The fact that these questions have not yet been answered--or
have been "answered" only in an unsatisfactorily subjective manner--casts a
shadow across the whole of cladistic methodology.

To me, when a cladogram doesn't conform with what we know of biogeography or
stratigraphy or functional anatomy or physics, it is a signal that there is a
problem--and the problem might very well lie in cladistics, not necessarily in
the other aspects. If, in particular, one might move a couple of low-level
branches on a cladogram in order to get it to conform with the other aspects,
then I would say it is almost certain that the cladogram is wrong, or highly
suspect, and it is time to return to the drawing board.

I think the principal problem that many paleontologists have with cladistic
analysis is that cladistics is not reliable enough to command priority over
biogeography, stratigraphy, functional analysis, physics, and the other
methodologies in the paleontologists' toolkit that may be brought to bear on
discovering life's One True Phylogeny. Just because you can arithmetically
calculate the reliability of a particular cladogram to ten decimal places
doesn't make the cladogram itself likewise correct (or incorrect, for that

A recent paper summarized the results of computer experiments to recover a
simple, previously constructed four-node cladogram (sorry, my copy of the
paper is buried somewhere, so I can't provide a citation here) through
cladistic analyses, and there was always a certain percentage (something like
10-18%) of instances when the analysis found an incorrect cladogram.
(Different kinds of analyses found different sets of bad cladograms, but the
percentage of "badograms" was about the same.) To me, this means that for any
choice of four nodes on a larger cladogram, there is a 10-18% chance that they
have been placed incorrectly. Think of the possibilities for error when there
are dozens of nodes. Now try to figure out how one might distinguish the
wrongly placed nodes from the right ones.

One can maintain that, because the fossil record is so spotty, we haven't
enough specimens or data to produce a correct cladogram, and that as the
fossil record improves, so will our cladograms. There are two clear objections
to this. First, who says when the fossil record is "good enough"? If it isn't,
then why are we doing these analyses from insufficient material? And second,
if cladistic analysis is worth anything, it should at least place the
specimens and taxa that we already know into their proper mutual
relationships. New material and new taxa should rest comfortably on the tree
that prior cladistics has constructed. If the discovery of new material
>overturns< a phylogeny, is it a signal that the previous analysis was
incorrect, or is it merely a signal that the new analysis is incorrect? How
might we know that this can't or won't happen again and again?

The cladist's response to correcting a questionable cladogram is frequently to
run more cladograms, or to merge the different analyses into a "consensus
tree." The latter response is particularly insidious, because eventually one
runs out of characters, and the consensus tree can no longer be checked even
with further cladistic analyses. And if the error is systematic, or inherent
to cladistics, then running more cladograms only compounds the problem. Nobody
has ever shown that a series of more detailed cladistic analyses must converge
to a correct phylogeny.

So this is what I mean when I say that cladistic analysis is inherently
unfalsifiable: We need to go outside cladistic analysis to falsify it, but if
all other kinds of analysis are subordinated to cladistics, then we lose
falsifiability, and with it, the science of the methodology.