Prediction = Parsimony or Partitions?

Abstract

The predictive value of classifications is one of several criteria which might be used to evaluate the goodness various types of classifications, however constructed, and perhaps might be used for choosing a particular method for constructing general classifications or for devising a new method of classification construction. It is not the only criterion that might be used and I will not advocate it as a criterion for choosing classifications — merely for evaluating them. A review of the literature on predictive value, although imprecise, yields conclusions that can be used to formulate a working definition of predictive value useful empirically for evaluating various numerical approaches to its study. The hierarchical nature of biological classifications, it will be seen, is one of the most important phenomena that must be incorporated into a definition and the evaluation of numerical methods or previous studies.