A serious matter with character-taxon matrices

Abstract

Knowing the structure and branching order of the evolutionary tree is essential to many studies of the evolutionary process and the history of life. Most attempts to reconstruct evolutionary trees are based on rectangular matrices, in which coded states are assigned for each character and each taxon under study. Everyone who has tried to infer the pattern of evolution from such character-taxon matrices is confronted with a great deal of “noise” and uncertainty. Often, there are multiple “most parsimonious trees” (those requiring the fewest evolutionary steps), which often differ significantly in topology; and many nodes in such trees are poorly supported; that is, there are contradictions and inconsistencies in the data. Some of this noise arises for reasons well known to phylogeneticists. Character states may arise multiple times, evolve in parallel, or reverse (homoplasy); taxa vary insufficiently among themselves; and diversification may be so rapid that data are insufficient to detect the order of branching. Polytomies are regarded as a nuisance rather than as real information. Characters may be coded inappropriately, so that there are too many redundancies, inapplicable character states, and multistate characters, as well as dependence among characters. Moreover, it is highly debatable whether evolution is parsimonious, and there are conflicts about how to establish the polarity of characters (primitive vs. derived). (For reviews and discussion see Sober 1988; Huelsenbeck and Crandall 1997.)