Partition of Information: the Centperc Problem

Abstract

Dale's CENTPERC-0 statistic for the classification of part-quantitative ecological data is briefly described and the problem of its partition into qualitative and quantitative components is examined. It is shown that the statistic cannot be partitioned in this way in its present form. A new, more general, statistic is defined which permits such partitioning. It is shown by examples that the CENTPERC-0 statistic is biased towards the quantitative component when the number of occurrences is low. This explains why this statistic tends to give unsatisfactory results in certain circumstances.