Abstract
Many fields of biology employ cross-species comparisons. However,
because species descend with modification from common ancestors, and rates
of evolution may vary among branches of an evolutionary tree, problems
of nonindependence and nonidentical distributions may occur in comparative
data sets. Several phylogenetically based statistical methods have
been developed to deal with these issues, but two are most commonly used.
Independent contrasts attempts to transform the data to meet the i.i.d,
assumption of conventional statistical methods. Monte Carlo computer
simulations attempt to produce phylogenetically informed null distributions
of test statistics. A disadvantage of the former is its ultimate
reliance on conventional distributional assumptions, whereas the latter
may require excessive information on biological parameters that are rarely
known. We propose a phylogenetic permutation method that is akin
to the simulation approach but requires less biological input information.
We show that the conventional, equally likely (EL) randomization model
is a special case of our phylogenetic permutations (PP). An application
of the method is presented to test the correlation between two traits with
cross-species data.