The Accuracy of the Maximum Parsimony Method for Phylogeny Reconstruction with Morphological Characters

Abstract

The accuracy of maximum parsimony for phylogeny reconstruction using morpho- logical data has never been adequately tested. In this study 128,000 known phylogenies of eight species and their associated discrete unordered character state data sets were generated by computer simulation. In order to mimic morphological data, values of the following seven simulation param- eters were varied: 1) number of characters: 2) maximum number of character states per character: 3) number of character state changes per speciation event, and 4) the probabilities of parallelisms, reversals, phyletic changes, and extinctions. Character state data for the known simulated phylog- enies were input into the program Hennig86, which found all equally parsimonious trees. The accuracy of maximum parsimony was judged primarily by the percent of time the strict consensus tree was found to be identical to the true tree. Parallelisms, reversals, phyletic changes, and ex- tinctions all significantly decrease maximum parsimony's ability to find the true phylogenetic tree. In addition, maximum parsimony is less accurate when there are fewer characters, more character states per character, and more character state changes per speciation event. For the parameter values considered, maximum parsimony did not do particularly well at finding the true phylogenetic tree. For 24.2% of the simulations, the chance that maximum parsimony found the true tree was less 25%; for 60.2% of the simulations, it was less than 50%; and for 85.2% of the simulations, it was less than 75%. In order to determine the relevance of these results to real data sets, values of the consistency index from the simulated phylogenies were compared to values reported in published plant mor- phological studies. For simulated data sets having values of the consistency index in the range reported in the literature, maximum parsimony found the true phylogenetic tree only from 0.7% to 27.8% of the time (mean = 10.3%). I advise caution, therefore, in using phylogenies reconstructed from morphological data to make inferences about biological processes or patterns that rely on knowledge of the true phylogeny. Additionally, I raise the question whether the great emphasis in systematics on phylogeny reconstruction using maximum parsimony and other cladistic methods is misplaced.