Quantitative phylogenetic analysis in the 21^st century

We review Hennigian phylogenetics and compare it with Maximum parsimony, Maximum likelihood, and Bayesian likelihood approaches. All methods use the principle of parsimony in some form. Hennigian-based approaches are justified ontologically by the Darwinian concepts of phylogenetic conservatism and cohesion of homologies, embodied in Hennig's Auxiliary Principle, and applied by outgroup comparisons. Parsimony is used as an epistemological tool, applied a posteriori to choose the most robust hypothesis when there are conflicting data. Quantitative methods use parsimony as an ontological criterion: Maximum parsimony analysis uses unweighted parsimony, Maximum likelihood weight all characters equally that explain the data, and Bayesian likelihood relying on weighting each character partition that explains the data. Different results most often stem from insufficient data, in which case each quantitative method treats ambiguities differently. All quantitative methods produce networks. The networks can be converted into trees by rooting them. If the rooting is done in accordance with Hennig's Auxiliary Principle, using outgroup comparisons, the resulting tree can then be interpreted as a phylogenetic hypothesis. As the size of the data set increases, likelihood methods select models that allow an increasingly greater number of a priori possibilities, converging on the Hennigian perspective that nothing is prohibited a priori. Thus, all methods produce similar results, regardless of data type, especially when their networks are rooted using outgroups. Appeals to Popperian philosophy cannot justify any kind of phylogenetic analysis, because they argue from effect to cause rather than from cause to effect. Nor can particular methods be justified on the basis of statistical consistency, because all may be consistent or inconsistent depending on the data. If analyses using different types of data and/or different methods of phylogeny reconstruction do not produce the same results, more data are needed.