Fundamental contradictions among observational and experimental estimates of non-trophic species interactions

The difficulty of experimentally quantifying non-trophic species interactions has long troubled ecologists. Increasingly, a new application of the classic "checkerboard distribution" approach is used to infer interactions by examining the pairwise frequency at which species are found to spatially co-occur. However, the link between spatial associations, as estimated from observational co-occurrence, and species interactions has not been tested. Here we used nine common statistical methods to estimate associations from surveys of rocky intertidal communities in the Northeast Pacific Ocean. We compared those inferred associations with a new data set of experimentally determined net and direct species interactions. Although association methods generated networks with aggregate structure similar to previously published interaction networks, each method detected a different set of species associations from the same data set. Moreover, although association methods generally performed better than a random model, associations rarely matched empirical net or direct species interactions, with high rates of false positives and true positives, and many false negatives. Our findings cast doubt on studies that equate species co-occurrences to species interactions and highlight a persistent, unanswered question: how do we interpret spatial patterns in communities? We suggest future research directions to unify the observational and experimental study of species interactions, and discuss the need for community standards and best practices in association analysis.