The paper provides an explicit justification for the principle that a
uniform taxon should contribute only one datapoint in comparative anal
yses with discrete variables. The justification is that phylogenetic p
atterns in variables unincluded in the proposed test vitiate the assum
ption of independence, both at the level of species and at the level o
f branch segments. The consequence is that a uniform taxon cannot safe
ly be counted as more than one datapoint. The arguments use a branchin
g discrete Markov process in continuous time, with the new feature tha
t the tested variables are only a subset of the evolving characters. T
his model is proposed as a useful criterion for measuring the merit of
proposed tests, and illustrates the necessity for models in evaluatin
g comparative methods. (C) 1997 Academic Press Limited