Significance tests of consensus indices

Abstract

Monte Carlo methods were used to examine the sampling distribution of eight consensus indices based on either of the following two models: all bifurcating trees are equally likely; or all trees (including both bifurcating and multifurcating trees) are equally likely. Ten different consensus-tree methods were applied before computing consensus indices. The strictness of a consensus-tree method and the recognition ratio of a consensus index are the two main factors determining the consensus-index distribution. The former factor influences the mean index values; the latter changes the shape of the distribution curves. We furnish significance test tables for consensus trees or indices based on randomly generated trees permitting multi-furcations. These tables can be used to test whether a given consensus tree or consensus-index value obtained from data on real organisms differs significantly from what one would expect if one were computing such quantities from randomly sampled trees.