Comments on optimizing the selection of the number of groups in a classification tree

Abstract

In a recent issue of this journal Guidi et al. (2009) proposed a Random Simulation Test (RST) to assist with the determination of the optimal number of clusters to be extracted from a hierarchical classification tree. This short note illustrates the potential influence that the richness of a simulated new site has on the determination of the optimal level to cut the dendrogram. This is discussed in the context of the need for clear objectives and criteria against which a classification tree can be evaluated. A simple modification of the RST is proposed that the simulated new site of the matrix has an appropriate number of positive entries that lies within the range of the observed matrix. The impact of this modification is then evaluated using an ecological community site-by-species matrix.