Do model-based phylogenetic analyses perform better than parsimony? A test with empirical data

The use of model-based methods to infer a phylogenetic tree from a given data set is frequently motivated by the truism that under certain circumstances the parsimony approach (MP) may produce incorrect topologies, while explicit model-based approaches are believed to avoid this problem. In the realm of empirical data from actual taxa, it is not known (or knowable) how commonly MP, maximum-likelihood or Bayesian inference are inaccurate. To test the perceived need for sophisticated model-based approaches, we assessed the degree of congruence between empirical phylogenetic hypotheses generated by alternative methods applied to DNA sequence data in a sample of 1000 recently published articles. Of 504 articles that employed multiple methods, only two exhibited strongly supported incongruence among alternative methods. This result suggests that the MP approach does not produce deviant hypotheses of relationship due to convergent evolution in long branches. Our finding therefore indicates that the use of multiple analytical methods is largely superfluous. We encourage the use of analytical approaches unencumbered by ad hoc assumptions that sap the explanatory power of the evidence.