Background/Question/Methods Detection error can bias observations of ecological processes, especially when species are never detected during sampling. In many communities, the identity of these missing species is known from previous research and natural history collections, but this information is rarely incorporated into subsequent models. Here, I present prior aggregation as a method for including information from external sources in Bayesian hierarchical detection models. Prior aggregation combines information from two or more prior distributions— in this case, an ecologically informed, species-level prior and an uninformed community-level prior— into a single aggregate. Using simulated data supplied to a multi-species occupancy model, I tested whether the prior aggregation method improves model estimates of missing species and the full community. Specifically, I tested the effects of 1) the accuracy of the species-level prior and 2) the relative contribution of the species-level prior to the final aggregate on the precision and accuracy of posterior estimates. Finally, I applied the prior aggregation method to a real dataset of small mammals in Vermont.
Results/Conclusions Prior aggregation improves estimates of undetected species and the full community, provided the information included in the prior is correct. Increasing the relative contribution of the informed prior to the aggregated distribution increases the precision of model estimates, but may reduce accuracy when the informed prior is inaccurate. When applied to a dataset of small mammals in Vermont, the prior aggregation method yielded more precise estimates of the eastern cottontail Sylvilagus floridanus, a species observed at several sites in the study region but never captured. While care must be taken to ensure accurate information is supplied to the model, the prior aggregation approach successfully incorporates external information into the model without sacrificing the advantages of modeling all species in the context of the community.
