A recent increase in the application of phylogenetic data to community ecological research has seen a rapid proliferation of new analytical methods. This is particularly true of studies which use the phylogenetic structure of communities to draw inferences about the ecological processes that determine community assembly. In these studies, phylogenetic structure is typically quantified using a metric based on pairwise distances between co-occurring species derived from an ultrametric phylogenetic tree. The observed value is then compared against a null distribution, typically derived via a randomization procedure, to test whether it deviates significantly from chance expectation. It is commonly assumed that these metrics are independent of species number. This is not the case, however, owing to the distinctive, non-normal frequency distributions shown by the pairwise phylogenetic distances upon which they are based. In this paper, I explore the properties of the distributions of pairwise distances derived from topologies of different size and shape, investigate their implications for various types of analyses, and explore solutions to the problem.
Results/Conclusions
The non-normal frequency distribution of pairwise phylogenetic distances invalidates comparisons between communities of different size, and the use of some combinations of metrics and null models. The problem extends beyond community phylogenetics.
