Using Mean Similarity Dendrograms to Evaluate Classifications

Abstract

Mean similarity dendrograms are introduced as a new graphical tool for evaluating classifications based on sample data from replicate objects within each of the proposed classes. The dendrograms compare the mean similarities between objects within the same class to the mean similarity between objects in different classes. They were designed to complement multidimensional scaling plots and permutation tests of class structure. The dendrograms offer a concise picture of the overall strength of a classification as well as the compactness and isolation of individual classes. Although broadly applicable, the dendrograms were motivated by a need for easily communicated assessments of land classifications that are intended to serve as geographic frameworks for environmental research and management. The dendrograms and other similarity-based tools are applied to a single-factor classification of fish communities sampled along a 281-km section of the Willamette River in the state of Oregon (U.S.). In a second example, the tools are used to evaluate a two-factor classification of fish communities sampled in wadeable streams of Oregon's Cascade Mountains and Willamette Valley. The dendrograms help to assess the relative classification strengths of the two factors, factor interactions, and an alternative classification derived from cluster analysis.