A Test and Review of the Empirical Performance of the Ontogenetic Criterion

The empirical performance of the ontogenetic criterion for determining character polarities was tested based on three published and one new data set originating from a study of larval Sepsidae (Diptera). The ontogenetic criterion polarizes about as many characters as does outgroup comparison. However, for the four data sets conflicting polarities were inferred for 4–45% of all characters for which both polarization methods were applicable. Four different versions of the ontogenetic criterion were identified from the literature and were empirically tested. The version used by de Queiroz (1985, Syst. Zool. 34:280–299) and Wheeler (1990, Cladistics 6:225–268) fared best, as determined by low levels of homoplasy and a small number of inferred equally parsimonious tree roots. For three of four empirical data sets, hypothetical ancestors based on outgroup comparison and the ontogenetic criterion resulted in identical tree roots. Hypothetical ancestors based on outgroup comparison implied less homoplasy than did those based on the ontogenetic criterion, so the ougroup method appears to be superior for character polarization. To test whether polarity decisions from both polarization methods should be combined into one hypothetical ancestor, two different “composite” ancestors were reconstructed. One comprised only the plesiomorphic states of the characters for which both criteria predicted the same polarity (consensus). It inferred multiple parsimonious roots without distinctly lowering the level of homoplasy. The second ancestor comprised the plesiomorphic states as predicted by one or both methods (composite). It tended to infer single roots, even for data sets for which very different rooting positions were identified by the two polarization criteria, thereby obscuring conflict between the two methods. High levels of homoplasy in hypothetical ancestors based on the ontogenetic criterion are not due to the inclusion of presence/absence characters in the data set, for which the application of the ontogenetic criterion would require hypotheses of homology for “absent” character states across ontogenetic stages of different species. Instead, the ratio of terminal deletions to terminal additions is a better predictor for the success of the ontogenetic criterion for any particular data set.