A general strategy for the simultaneous classification of variables and objects in ecological data tables

Abstract

A general iterative algorithm is proposed for the arrangement of data matrices by optimizing X 2 , sum of squares or pooled entropy for blocks determined by clusters of variables and objects. The method includes k ‐means clustering and constrained block clustering as special cases. Possibilities for evaluating the resulting matrix include calculation of relative contributions to measures of block sharpness, comparison of partitions and construction of consensus tables. A new measure for comparing rearranged data matrices is suggested. The distributional properties of final results produced under different starting conditions are examined. The performance of the algorithm is tested on binary vegetation data from the Italian Alps.