A fast classificatory algorithm for large problems and the Bray-Curtis measure

Abstract While the Bray-Curtis measure has been attractive to ecologists as an appropriate approach to double-zero matches when performing numerical classifications of site-by-species data, it has been dis-advantaged by the absence of fast algorithms for large data sets and of suitable diagnostic tools. This paper addresses these deficiencies. A polythetic divisive algorithm is presented with extensive use of re-allocation to overcome the usual poor performance of divisive techniques. The algorithm appears to provide solutions of comparable merit to those of a widely-used agglomerative approach. Measures to assess an attribute's contribution to a split and its ability to discriminate between all groups are also presented.