COMPARING THE RELATIVE EFFICACY OF BAYESIAN, MAXIMUM-LIKELIHOOD, AND PARSIMONY METHODS FOR THE ESTIMATION OF TOPOLOGY FROM MORPHOLOGY THROUGH THE USE OF BOTH SIMULATED AND EMPIRICAL DATA

The widespread adoption of total-evidence phylogenetic topology and dating methods relies upon Markov models to encapsulate the process of substitution between discrete morphological character states. Currently only one such model is widely implemented, the single parameter Mk-model. This is troubling since the efficacy of the Mk model has not been the subject of validation or development in the fifteen years since its introduction.

We characterise the relative accuracy and resolution afforded by several competing methods for estimating topology from morphological data; the Bayesian implementation of the Mk-model, the maximum-likelihood implementation of the Mk-model, equal-weights parsimony, and implied-weights parsimony. This is achieved through the analysis of replicate simulated morphological matrices in which the distribution of homoplasy matches that seen in empirical data. We find that the Bayesian implementation of the Mk-model may be considered preferable due to a posterior distribution of trees allowing for a satisfactory level of accuracy in results without the false precision introduced by more deterministic approaches. Building on our findings, we re-analysed a number of empirical morphological matrices for which evolutionary hypotheses have been tested exclusively using parsimony methods. We show that many of these hypotheses cannot be accepted when tested against Bayesian consensus trees. Our results suggest that a widespread adoption of Bayesian methods for the analysis of morphological data may redefine our understanding of many taxonomic relationships.