To investigate the theoretical properties of consensus trees obtained from large numbers of gene trees evolving at different loci, we construct consensus trees from independent gene trees that occur in proportion to their probabilities from coalescent theory. We consider majority-rule, rooted triple (R*), and greedy consensus both asymptotically as the number of loci approaches infinity and for finite numbers of loci. We investigate the effects of species tree branch lengths and find different consistency results for the three methods: majority-rule consensus trees can be increasingly likely to be unresolved, although not misleading; greedy consensus trees can be misleading for all species tree topologies with at least 5 taxa; and R* consensus is statistically consistent, i.e., guaranteed to return the species tree topology given enough loci, for any binary species tree topology with any set of branch lengths. We also investigate the performance of consensus trees as species tree estimators when there are finite numbers of loci.
view more