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

Re: Cladistics (was Sci. Am. - present)

In a message dated 98-02-21 03:02:23 EST, cbrochu@fmppr.fmnh.org writes:

<< How would you "improve" a parsimony analysis?  I'm not trying to be
 patronizing - this is a serious question.  Given the principle of total
 evidence, the hypothesis that best matches all the available data is the
 preferred one - are you advocating some sort of significance test for
 character data?  If so, on what objective basis would this be based?  And
 bear in mind, switching to molecular data won't necessarily "improve" the
 analysis, except by adding additional relevant information. >>

This carries the walloping untested assumption that nature is parsimonious. I
don't believe this for a minute. At best, nature may be "not wildly

In particular, therefore, I do not believe that nature necessarily strictly
generates a cladogram with the least number of steps, only--in most cases--a
cladogram with something fairly close to the least number of steps, and every
so often, a cladogram with a considerably larger number of steps. The
phylogenetic patterns found by cladistic analysis are all very nice, but
nobody has shown that they correspond to reality--only that they sometimes
correspond, to varying degrees (almost never exactly), with other patterns
generated by other means or (shudder!) other cladistic analyses. Maybe they're
real, or maybe they're all way off in some systematic way that we haven't yet
discovered (such as consistently biased choice of outgroup or choice of
representative taxa, or hidden assumptions about character-state polarity). In
this sense, they're no better than the phylogenies generated by "old
fashioned" methods, and not particularly more reliable. A cladist can no more
show that his or her cladograms are correct than I can show that they're not;
thus the whole exercise is pointless.

It strikes me that "easy cladograms," in which the lineages show nice
progressions of character states from primitive to derived are likely to be
correct. But "hard cladograms," in which the patterns of derived states differ
in an apparently random way from group to group, and which the cladistic
analysis essentially relates by "majority vote," are very unlikely to be
correct. Hard cladograms are probably insoluble by cladistic analysis.

Of course there is presently no objective way to assign specific weights to
specific characters, which also means that assigning an >equal< weight to
>all< characters is just as baseless as assigning unequal weights to them. On
what would you base the assumption that all characters should be weighted
equally? That nature does this? How do you know? How can you assert that a
skull character is (or is not) worth the same as a vertebra character? Perhaps
what cladists should be doing is amassing a gigantic Ultimate Character Matrix
that contains all known states of all known characters, run one Giant
Cladistic Analysis to generate the one Ultimate True Phylogeny, and declare
their work finished. Sort of like the Human Genome Project.

Plugging and chugging character matrices--or DNA/protein sequences--through
computers to generate cladograms may be lots of fun, but without an objective
way to evaluate the accuracy of the output >independently< of doing more
plugging and chugging, it's just wasted computer time. Saying it's "the best
we can do from the available data" and so forth is not particularly
reassuring, and not even close to a guarantee that the method works correctly.
"Garbage in, garbage out" is always a problem, perhaps an insoluble one.
Meanwhile, the literature is filling up with flash-in-the-pan computer-
generated phylogenies that do little to illuminate but certainly play havoc
with the convenience and nomenclatural aspects of taxonomy.