A Novel Method to Improve Estimates of Predator Diet Compositions

Food web studies require estimating relative contributions of prey to predator diets. In aquatic ecology, a typical source of this information is stomach content data, the analysis of which is often confounded by several statistical problems. Two key issues are independence among samples collected in time/space, as well as covariance between consumption of prey and prey type. Existing analysis methods do not currently address these challenges, nor do they create an appropriate likelihood function to permit formal model selection, likelihood-based parameter estimation and foster Bayesian analyses. We developed a Bayesian model that addresses these challenges to quantitatively estimate prey contributions to predators. Our model accounts for covariance between prey type and consumption rates, and can be fitted to data using standard numerical methods. By extending the model to include random effects, we are also able to directly estimate independence among samples. We apply our model to multiple stomach content datasets to compare the resulting prey contribution estimates to traditional diet estimation methods. Results of both simulation testing and applications to actual diet data reveal that our novel method not only removes bias from the aforementioned issues, but also drastically improves precision.