Species or supraspecific taxa as terminals in cladistic analysis? Groundplans versus exemplars revisited.

Increasing thenumber of taxa in aphylogenetic analysis can have a profound effect on accuracy (Hendy and Penny, 1989; Wheeler, 1992; Hillis, 1996; Graybeal, 1998; Halanych, 1998; Soltis et al., 1998; but see alsoKim, 1996; Poe and Swofford, 1999). Although including all ormost constituent species in analyses attempting to resolve relationships among higher taxa is desirable, it may be impractical (Donoghue, 1994; Mishler, 1994; Nixon and Carpenter, 1996; Rice et al., 1997). The number of possible cladistic arrangements increases exponentially with the addition of taxa (Felsenstein, 1978), thus requiring more complex analyses, which cannot guarantee optimal solutions or may be intractable with available computer technology. Two alternative approaches for overcoming this constraint on the number of taxa included in an analysis differ in the use of species versus supraspeciŽc taxa as terminal entities. Bothhave the sameaim—to estimate the groundplan, or plesiomorphic states, of the higher taxa concerned. The choice between these approaches is intimately related to the problem of analyzing large data sets. As the number of large data sets has increased, so has the number of papers addressing this issue (e.g., Nixon and Davis, 1991; Donoghue, 1994; Mishler, 1994; Yeates, 1995; Nixon and Carpenter, 1996; Kron and Judd, 1997; Rice et al., 1997; Bininda-Emonds et al., 1998; Griswold et al., 1998; Wiens, 1998a). However, no apparent consensus has been reached, and empirical studies using either approach, or a combination thereof, continue to appear. On the basis of comparisons of real and hypothetical data matrices, several authors have argued in support of species as terminal taxa (Yeates, 1995; Kron and Judd, 1997; Griswold et al., 1998; Wiens, 1998a). In a recent simulation study,Wiens (1998a:411) concluded that “using species as terminals gives consistentlymore accurate estimates than do the other coding methods, even when only a few species are sampled from each higher taxon[and is] strongly recommended for empirical studies.” Other authors advocate the use of supraspeciŽc terminal taxa (Rice et al., 1997; Bininda-Emonds et al., 1998). For example, Rice et al. (1997:560) rejected the use of species as terminals because “if we conŽne ourselves to choosing among terminal taxa, we will inevitably lengthen branches, with all the attendant problems” and proposed the inferred ancestral states (IAS) method for coding supraspeciŽc terminals (see below) as a “more promising avenue.” In the present contribution, Iwill argue fromŽrstprinciples that the use of species as terminals is superior, both practically and philosophically, to the use of supraspeciŽc terminals. I will then provide criteria for selection of species, given the varying availability of data.