BLOMBERG, S.P.; IVES, A.R.; GARLAND, T.: Detecting Phylogenetic Signal in Comparative Data

Most species resemble their nearest relatives, yet a general comparative method that measures such phylogenetic “signal” is not yet available. We present new procedures for measuring phylogenetic signal in continuous-valued characters, which use information on both topology and branch lengths. The procedures can be implemented either with independent contrasts or with generalized least-squares methods. We describe a simple permutation (randomization) method that can be used to test the null hypothesis of no similarity among relatives. The test has correct Type I error rates and good power (e.g., 0.8 with alpha = 0.05 and N = 18). The method can also be used to quantify the fit of any candidate tree to any set of trait data. The fit of a tree to data can often be improved by branch length transformations. Therefore, we also provide two biologically based transformations, one based on the Ornstein-Uhlenbeck model of stabilizing selection, and another based on a model of accelerating/decelerating evolution. Hypothesis tests of parameters in these models provide a way to detect the presence of these different modes of evolution. Moreover, by comparing the fit of one or more candidate trees to the fit of a star phylogeny (no heirarchical structure), one can determine whether conventional statistical methods, which assume a star phylogeny, may be acceptable. Application of the methods to real data sets obtained from the literature reveals that most traits (with reasonable sample sizes) exhibit significant phylogenetic signal, including behavioral and ecological traits that are thought to be relatively evolutionarily maleable. In addition, the transformation of branch lengths can be useful when finding the tree that best fits the data for purposes of statistical analyses.