Concatabominations: Identifying Unstable Taxa in Morphological Phylogenetics using a Heuristic Extension to Safe Taxonomic Reduction

For a variety of reasons, some phylogenetic data sets are replete with missing entries. Attitudes toward abundant missing data, specifically concerns over its potential to mislead or confound phylogenetic inferences, are varied. Thus, there is a current debate on the impact of missing entries upon the accuracy of phylogenetic inferences (Wiens 2006; Lemmon et al. 2009; Philippe et al. 2011; Wiens and Morrill 2011; Roure et al. 2013). Perhaps less controversial is that individual taxa may sometimes be relatively phylogenetically unstable by virtue of limited data and extensive missing data (e.g., Wilkinson 1996; Sanderson and Shaffer 2002; Wiens 2003; Wilkinson 2003). Wilkinson (1995) developed an approach for diagnosing taxon instability due to missing data a priori termed safe taxonomic reduction (STR). STR allows the identification of “rogue” taxa that can be removed from a data set safe in the knowledge that their removal will not impact upon the interrelationships that will be inferred among the remaining taxa under the parsimony criterion. The potential benefits of such deletion are reductions in numbers of optimal trees and run times and better resolved consensus summaries.
STR has been fairly widely used, mainly by paleontologists confronted with relatively incomplete fossil taxa (see Anquetin 2012; Graf 2012; McDonald 2012; for some recent examples), and also in the context of the matrix representation with parsimony (Baum 1992; Ragan 1992) approach to supertree construction (e.g., Cardillo et al. 2004). Nonetheless STR is not always as effective as one might hope (e.g., Mannion et al. 2013). Here, we present a simple heuristic method for identifying potentially unstable taxa that may be useful in cases where STR does not succeed in ameliorating all the problems caused by missing data. We illustrate the approach through application to the saurischian data of Gauthier (1986), which was previously used to illustrate STR and thus is particularly appropriate for demonstrating the ability of the new method to achieve more than STR alone