OadisHa (1994) 10:415-423 PROBLEMS WITH ZERO-LENGTH BRANCHES

In this note we call attention to a possibly common situation in which frequently used phylogenetic analysis and diagnosis programs may report cladograms that do not have support at every node from the data matrix analyzed to produce them. The problem involves branches that may or may not be interpreted as having zero length, the ambiguity arising from equally applicable but mutually exclusive alternative optimizations. Platnick et al. (1991) and Wilkinson (1995) commented on manifestations of the problem that involved missing entries and "arbitrary resolutions", respectively. Wilkinson (1995) emphasized that programs can report topologies including zerolength branches even when no missing entries are present. However, Wilkinson posed the problem abstractly, a single character on one tree, addressed only one horn of the dilemma (branches ambiguously of zero or greater length were always assigned zero length), considered solutions only within the philosophical framework offered by PAUP, and offered no advice regarding what practicing systematists could do to detect the problem. In addition, the treatment of the zero-length branch problem by programmers is already more diverse than reported by Wilkinson (1995), and users of the most common packages should be aware of the different interpretations adopted by programmers. Wilkinson (1995) offered some algorithmic suggestions for a solution, but his solution, although helpful, strikes us as less than ideal. We also focus here on what practicing systemadsts can do now to detect and avoid the problem. Finally, the solution is not necessarily as simple as eliminadng "arbitrary resolutions". Most importandy, Wilkinson (1995) did not consider what happens if branches of ambiguous length are permitted to be non-zero rather than zero-length. A more fundamental issue is the meaning of "acceptable support" for phylogenedc hypotheses, and therefore which hypotheses are worth considering, as Platnick et al. (1991) discussed. That debate is just beginning, but should involve the user community in addidon to programmers and methodologists. Table 1 conveys the problem of zero-length branches more concretely than the example provided by Wilkinson (1995, Fig. 1). Taxa D and E are identical, as are F and G, but this does not affect the point and keeps the matrix simple. Five supports the ingroup BCDEFG, Four supports CDEFG, Two and Three support DEFG. One, however, is homoplasious, and permits two parsimonious expla-