Eclecticism in cladistics: Convergence among opposing optimality criteria may be a hoax

Abstract

Currently, most authors gave up on justifying the choice of a single optimality criterion in phylogenetics. Instead, they frequently apply as many methods as feasible as an unofficial means to tackle the question of the uncertainty inherent to tree reconstruction. For example, we reviewed recent volumes of the journal Molecular Phylogenetics and Evolution (volumes 94–115) and observed that 311 of 545 articles used this informal strategy (i.e., more than one method). However, only two articles attempted to justify this approach, both stating that convergence provides some measure of robustness. Neither offered any references or supporting evidence for that claim. This should not be surprising since older but similar reviews have noticed a comparable pattern. Nevertheless, we interpret this pattern differently. While others have asserted that the frequent convergence in the literature (which in our case is reported in 295 of 311 publications) is an indication of the futility of comparing methods, we hypothesize it could be an indication of publication bias. Using 117 empirical cladograms, we simulated nine alignments per tree and performed 100 tree search experiments per dataset using parsimony, maximum likelihood, and posterior probability. We designed alignment simulations to improve the retention index in comparison to the original matrix and facilitate convergency. Still, the vast majority of results were divergent by measures of the match-split distance of unrooted binary trees. These results suggest that convergency may be rare and that it is a poor indicator of the distance between the results and the original tree. Furthermore, we analyzed our data under the light of Bayesian theory to argue that the comparison of opposing optimality criteria can, at least under certain experimental conditions, increase the false discovery rate in phylogenetics.