Sensitivity analysis of different methods of coding taxonomic polymorphism: an example from higher-level bat phylogeny

New information concerning strengths and weaknesses of different methods of coding taxonomic polymorphisms suggests that results of some previous studies may have been unintentionally biased by the methods employed. In this study, we demonstrate that a form of sensitivity analysis can be used to evaluate the effects of different methods of coding taxonomic polymorphisms on the outcome of phylogenetic analyses. Our earlier analysis of higher-level relationships of bats (Mammalia: Chiroptera) employed superspecific taxa as terminals and scored taxonomic polymorphisms using ambiguity coding. Application of other methods of dealing with polymorphisms (excluding variable characters, inferring ancestral states, majority coding) to the same data yields phylogenetic results that differ somewhat from those originally reported based on ambiguity coding. Monophyly of some clades was supported in all analyses (e.g., Microchiroptera, Rhinopomatoidea, and Nataloidea), while other groups found to be monophyletic in the original study (e.g., neotropical Nataloidea) appeared unresolved or nonmonophyletic when other methods were used to code taxonomic polymorphisms. Several groupings that were apparently refuted in the initial study (e.g., Noctilionoidea including Mystacinidae) were supported in some analyses, reducing some of the apparent incongruence between the trees in our earlier analysis (which were based principally on morphology) and other trees based on molecular data. Perceived support for various groupings (branch support, bootstrap values) were in some cases significantly affected by the methods employed. These results indicate that sensitivity analysis provides a useful tool for evaluating effects of different methods of dealing with taxonomic polymorphism in superspecific terminal taxa. Variation in results obtained with different methods suggests that it is always preferable to sample at the species level when higher-level taxa exhibit taxonomic polymorphism, thus avoiding methodological biases associated with different methods of dealing with taxonomic polymorphisms during data analysis.