COS 122-5 - Quantifying individual diet specialization using Bayesian hierarchical models

Thursday, August 10, 2017: 9:20 AM
D133-134, Oregon Convention Center
Kyle Coblentz, Integrative Biology, Oregon State University, OR, Adam E. Rosenblatt, School of Forestry and Environmental Studies, Yale University and Mark Novak, Integrative Biology, Oregon State University, Corvallis, OR
Background/Question/Methods

Increasing evidence suggests that individuals within populations often differ from one another in ways that influence their impact on community dynamics and ecosystem functioning. In particular, the specialization of individual generalist predators on subsets of the population’s diet has the potential to alter our understanding of food webs and predator-prey interactions. To fully understand the causes and consequences of this diet specialization, we need to be able to quantify individual diets accurately. Currently used methods quantify the diets of individuals by simply using the proportion of different prey types that are observed in the individual’s diet. Although, without other information, this measure corresponds to the maximum likelihood estimate of the individual’s diet, maximum likelihood estimates can be severely biased when based on low sample sizes, which are common in studies of individual diets. Here we instead estimate individual diets by constructing Bayesian hierarchical models. We apply both methods to simulated and empirical data sets to determine the accuracy of the methods in estimating diet specialization and illustrate the differences in the estimates between the two methods.

Results/Conclusions

The currently used observed proportions approach consistently overestimates diet specialization, particularly when sample sizes per individual are low or are heterogeneous among individuals, two common features of empirical data. In contrast, the Bayesian hierarchical method estimates remain largely unbiased. This is due to the Bayesian hierarchical method providing shrinkage estimators which pull individual estimates towards population means. The amount of shrinkage is dependent on the sample size in a manner that discounts observations from individuals with very low sample sizes and thus prevents bias in estimates of diet proportions. The Bayesian hierarchical approach also permits the simultaneous estimation of both prey proportions and their variability within and among levels of organization (e.g. individuals, experimental treatments, populations) which is illustrated with the analysis of empirical data sets. Overall, the Bayesian hierarchical approach represents a useful framework for improving the quantification and understanding of diet specialization and predator foraging more generally.