Process partitions, congruence, and the independence of characters: inferring relationships among closely related Hawaiian Drosophila from multiple gene regions.

As data from multiple sources (morphology, different genes) become available for inferring relationships among taxa, the problem of how these data should be analyzed has emerged as a topic of renewed theoretical discussion (Miyamoto, 1985; Kluge, 1989; Barrett et alv 1991; Bull et al., 1993; de Queiroz, 1993; Lanyon, 1993; Chippindale and Wiens, 1994; de Queiroz et al., 1995; Miyamoto and Fitch, 1995; Brower et al., 1996; Hedges and Maxson, 1996; Huelsenbeck et al., 1996; Nixon and Carpenter, 1996; Page, 1996). Philosophically, the problem is straightforward. Phylogenetic inference depends on evidence from characters that imply hierarchical patterns of grouping. Not all features of organisms are of equal evidential value for inferring relationships, so some form of discrimination between "good" and "bad" characters is necessary. However, such differential character weighting requires ad hoc assumptions about the nature of empirical evidence. Many of the current disputes in systematics revolve around the advantages and disadvantages of alternative weighting schemes (even the debate between advocates of maximum likelihood and advocates of cladistic parsimony can be viewed as a difference in willingness to make assumptions about differential weighting of character types and character state transformations). In our view, the goal of systematics is to produce phylogenetic hypotheses with the greatest explanatory power (in the Popperian sense). This explanatory power is reduced by the hypothesis' contingency on supernumerary background assumptions. Although no