An Operational Method for Evaluating Classifications

An information theoretic optimality criterion measuring how well any classification preserves the information in a set of char- acters is applied to 12 classifications of the Ranunculus hispidus com- plex. Eleven of these classifications are derived from a new taxonomic study of this complex and one is taken from the most recent previous studies by Benson and by Lourteig. By this criterion, the most recent previous classification is shown to be less effective in summarizing the set of characters used than most of the classifications proposed, which are part of one hierarchical series. The final choice of one classification depends, in part, on the level of morphological distinctness required by the species concept employed. The values for the optimality crite- rion for each of the 12 classifications and for 20 random classifications are compared. It is argued that the values for the 12 nonrandom clas- sifications are extremely unlikely to occur at random. When a new classification is proposed, it is usually the intent of its author to provide a better treatment of the group of organisms than those of previ- ous workers. However, no new classification proposed is universally adopted by other workers. The decision to use one of several available treatments for any particular group depends upon recognizing one of these taxonomic treatments as best. Thus the criteria for determining which of several classifications for a given group of organisms is best become critical. Here we describe an operational procedure for evaluating the effectiveness of a classification for preserving information in a set of characters and provide an example of its use in evaluating several classifications proposed for the