Separating Sampling Bias From Abundance Shows That Different Methods Catch Different Wild Bees

Ecological community sampling methods have taxonomic biases, producing samples where relative abundances of taxa may differ from the underlying sampled community. Evaluating sampling methods' relative biases is therefore necessary for accurately interpreting community data. Wild bees (Hymenoptera: Apoidea) have been the focus of intensive community sampling and many studies have compared the properties of samples collected by different methods. However, comparative studies have often conflated differences in sampling bias with differences in effort and absolute abundance between methods, potentially obscuring methods' true biases.Here, we compare wild bee communities in the northeastern United States as sampled by pan traps, vane traps, and hand netting. Using a dataset of simultaneous sampling by different methods, we compare sample richness and composition between pairs of methods while accounting for differences in the overall number of bees sampled by each.For a given number of individuals sampled, hand netting captured more bee species than pan traps, which captured more species than vane traps. Pan traps sampled a different pool of species than either of the other two methods. Of 21 bee genera analyzed, 8 were overrepresented in pan trap samples relative to hand netting, whereas 7 were relatively underrepresented in pan traps. When compared against vane traps, 4 genera of 20 were relatively overrepresented in pan traps, whereas 6 were relatively underrepresented. Pan traps poorly represented very large-bodied genera as compared with the other methods.We find pervasive biases in bee community sampling methods, with most genera showing significant differences in relative abundance in at least one methodological comparison. At times, genera were relatively underrepresented even by methods that collected them in higher absolute abundance. Since bias is unavoidable in community sampling, studies must measure taxon-specific biases in the context of their system and evaluate the robustness of analytical results.