Empirical evaluation of community similarity metrics

Increased recognition of the complexity of ecological systems has fueled a demand for more sophisticated descriptors of community structure.  Quantitatively-oriented researchers have met this need with a proliferation of new metrics and tools, many beyond the grasp of the average ecologist.  However, the development of user-friendly software has made these metrics accessible.  While invaluable, these new tools pose two major hazards:  First, their proliferation could jeopardize comparisons among related studies because each uses a different metric.  Second, their misapplication by inexperienced users could result in improper analyses and incorrect conclusions.  To address these concerns I have carried out an analysis of how different community similarity metrics perform using nine empirical datasets collected along a single environmental gradient.  The metrics chosen include classic (Søresen’s and Jaccard’s), derived (Chao-Søresen’s and Chao-Jaccard’s abundance-based), and alternative (multivariate dispersion) indices for assessing community similarity.  The datasets, each representing a different taxonomic group, include different numbers of species, numbers of individuals, and completeness of sampling, providing a range of different scenarios likely to be encountered by empiricists.  Estimates of community similarity differed among the indices, highlighting differences in their structure –  Each seems to focus on certain aspects of community composition.  Users are urged to consider the ecological questions they have in mind, and the nature of their dataset, before deciding on a metric.  Nevertheless, it may be worthwhile to report values for a standard set of metrics so that comparisons can be made among related studies.