PubMed1998
Thorpe (1984) and Chappill (1989) argued that selection of a coding method should be [figure: see text] based on the purpose of coding. In our view, the purpose of coding is dictated by the principles of phylogenetic systematics. The foundation of phylogenetic systematics is the observation that monophyletic groups can be recognized if homologous character states, shared evolutionary novelties, can be identified (Hennig, 1966). Unfortunately, characters do not have labels indicating their homology. Instead, a systematist must propose a hypothesis of homology and evaluate its congruence with independent hypotheses based on other traits. In this context, the purpose of coding is to represent those hypotheses. The major obstacle to coding is that the a priori groups under analysis (i.e., taxa) often have ranges of variation that overlap to some degree. This is true whether traits are described qualitatively or quantitatively. One advantage of quantitative description is that it permits a more detailed analysis of how much the ranges of variation overlap. It may seem appropriate to use statistical methods to summarize the amount of overlap and even to decide objectively (on a priori grounds) whether taxa are similar or different. Above, we demonstrated some of the problems resulting from these uses of statistical analysis. In our view, the most important problem is the implication that similarity of the feature across taxa is the basis for inferring homology. The similarity that is relevant to phylogenetic analysis is not proximity in morphospace, but shared novelty. Statistical methods can describe proximity, but they cannot recognize novelty. The method of coding we recommend uses graphical displays of individual values. Coding decisions are based on all of the individuals in each taxon, not on summaries derived from models of expected distributions. Then, the evidence for inferring divergence is independently evaluated for each pair of overlapping taxa. Coding decisions are not based on a priori rules that have no bearing on recognition of evolutionary novelty. This is the same approach that is used to code qualitatively described traits.