A New Look at the Predictive Value of Numerical Classifications

Publisher Website
Google Scholar
Abstract
The concept of the predictive value of biological classifications is reviewed and 2 indices are introduced which can be used to evaluate and compare the goodness of classifications (with regard to their predictive value) constructed using different classification methods. These indices incorporate the unique relationship between hierarchical classifications and the expectations of prediction success for character states. The predictive value of a classification is the degree to which states of characters are constant within and restricted to taxa in the classification. Previous studies used phenetic correlations, taxonomic congruence, parsimony and partitions of the taxa to measure predictive value. These approaches are reviewed and found deficient. To have high predictive value for new characters or new taxa, a classification must be constructed to have maximal predictive value fit for the characters initially used to construct it. Predictions made with regard to the observation of new characters vs. new taxa are the same, in contradiction to previous discussions.

This publication has 7 references indexed in Scilit: