Truth or Consequences: Effects of Phylogenetic Accuracy on Two Comparative Methods

Abstract

The insight that species cannot be treated as independent points for analysis has revolutionized comparative biology. Biologists now face choices among several new procedures for removing phylogenetic non-independence from their data. This paper compares two very different new methods, one an independent contrasts approach, the other based on phylogenetic autocorrelation. We discuss issues of performance when the topology and branch lengths of the phylogeny are known in full, and how the two approaches stand up to incompletely resolved trees and incorrect branch length information—common situations for comparative biologists. Simulations show independent contrasts to be valid when branch lengths are known, whether or not the topology is known fully. Type I error rates are higher for phylogenetic autocorrelation, especially when the topology is known only very incompletely. Incorrect branch lengths affect independent contrasts more adversely than phylogenetic autocorrelation. We discuss how the results reflect the different assumptions of the two methods, and stress the importance of assessing how well real data fit the internal assumptions of methods used to analyse them.