Recently, likelihood-based methods have enabled more rigorous investigations of food-web data with models. Two models that have recently been formulated in this framework are the group model, where all possible links between each pair of groups are assigned the same probability, and the probabilistic niche model, where each species is assigned a location in the niche space determining who eats it, and a center and range determining whom it is likely to eat. We present a hybrid Bayesian model that clusters species into groups while retaining feeding-niche structure, so that species in the same groups tend to have similar niche locations and feeding ranges. We employ several variants of the model, including one that allows feeding range to increase with niche location within groups. We test the model variants against food-web data, including a food web from the Serengeti, to determine whether the combination of group and niche structure is supported, and whether feeding range increases with niche value.
Results/Conclusions
The hybrid model with both niche and group structure substantially improves model fit over a niche model without groups in the Serengeti data set. Increasing feeding range with niche location within groups is also supported in carnivore and herbivore groups, contrary to conclusions made with the probabilistic niche model without groups for other data sets. Furthermore, Bayesian hierarchical modeling, enabled by Markov-chain Monte Carlo methods, provides an effective framework for exploring general patterns across food webs as well as the detailed structure of specific food-web data sets.