COMMENTS ON THE 
Note: The first Smithsonian Botanical
Symposium "Linnaean Taxonomy in the 21st Century" was held on 3031
March 2001 in
The two terminating branches and one basal
free branch attached to any internal cladogram branch can be arranged in
three ways giving ((AB)C) or ((AC)B) or ((BC)A). There is commonly support
(shared traits) for one or both of the two arrangements alternative to the
optimal. If the conflicting units of support (steps) are equal evidence of
shared relationships, then the best assurance (without additional information
on branch reliability) that parsimony analysis can give us is that there is a
little better than 33% probability that the optimum branch represents the
correct arrangement, not the two closest alternatives. Bootstrapping and
decay index are not direct measures of branch support. The null hypothesis in cladistics is a
bush. Any shortest resolved tree is best evidence of relationship. But best
for what use? Consider: Flip a coin 100 times to see if it is
(phylogenetically) loaded. Cladistic philosophy: 50 heads and 50 tails means
the null cannot be falsified, and we cannot hypothesize the coin is loaded.
But, for instance, 54 heads and 46 tails is taken as evidence of loading for
the head side up, being the "best explanation." Statistically,
however, one could do a nonparametric test with the null being
"equiprobable and randomly distributed" (54 or more heads would
only occur randomly 24% of the time, and the null cannot be rejected at, say,
the .95 confidence level), therefore there is no evidence of loading that one
would act on. We are left with the unimpressive probabilistic proportion
54/100 for the chance of loading on heads. Only recently have statistical
tests (other than subsampling) been introduced for gauging the reliability of
individual branch arrangements. While many obvious or
"uncontested" phylogenies are supported by parsimony analysis, a
PhyloCode implies additional resolution and reliability. Over the past 30
years, however, published resolved branch arrangements that are less than
acceptably probable or which are not distinguishable from a random
distribution have not been identified as such though doubtless common. An
additional problem exists with molecular studies  differential lineage
sorting of genes may produce well supported gene trees that are different
from the species tree. One needs a minimum (by exact binary calculation) of
three identical gene trees (with no contrary trees) for probabilistic
reconstruction of the species tree, five if introgression is suspected (in
ms.). Thus, identifiable probabilistic reconstructions of absolute branch
orders are still in the future, and a PhyloCode based on past cladistic
studies is not an acceptable alternative to standard nomenclature. For more
discussion, see "Deconstructing
Reconstruction" and various reprints. Originally published as: Richard H.
Zander. 2001. Invited comments, First Smithsonian Botanical Symposium. Plant
Press 4(2): 15. 
