How robust are popular beta diversity indices to sampling error?

Beta diversity (beta) is used in biogeography, ecology, and conservation to assess the heterogeneity of local communities. Ideally, researchers could include sensitivity to error in the list of reasons to choose a beta index. However, only numerical undersampling has been rigorously studied. This study compared multiple beta indices to determine which are most robust to geographic undersampling, numerical undersampling, and taxonomic error using simulated landscapes. For these landscapes, eight beta indices were chosen to represent families of beta and used to measure real and errant data. Six indices used both presence-absence and abundance data, while two more used only abundance data. Six of the abundance-based indices had adjusted versions for individual undersampling, and these versions were also evaluated (total = 14 indices). Presence-absence-and abundance-based indices were comparable in sensitivity to total method error. Numerical undersampling and taxonomic error generally caused more error in beta than randomly distributed geographic undersampling. Among presence-absence-based indices, Jaccard's dissimilarity was the most robust to error overall, while b-3 was the most robust among narrow-sense measures. Among abundance-based indices, Bray-Curtis and BDTOTAL were the most robust to error. Some commonly used beta indices (e.g., Sorensen, Simpson) are relatively unreliable given errors of taxonomy or numerical undersampling. Future studies of beta should focus on using more robust indices (Jaccard, Bray-Curtis, BDTOTAL), and past studies based on error-sensitive indices should be considered with caution. Studies of beta should emphasize adequate numerical sampling and taxonomic accuracy to minimize errors in beta.