Quantitative fatty acid signature analysis (QFASA) is a recently developed diet estimation technique that depends on statistical methods. QFASA has been used successfully to obtain point estimates of the diet of predators such as seals and seabirds. In this talk, inference for a population of predators is considered. We discuss the issues involved in modelling this data and the interpretation of parameters. As the data is compositional and often with zeros, special techniques are required for the construction of confidence intervals for the diet estimates. We propose a mixture model approach to manage the zeros followed by a specially constructed bootstrap. Our methods are assessed through simulations using pseudo-seals. Finally, a real-life data set involving seals will be presented to illustrate the usefulness of our methods in practice.

Joint work with C. Stewart (Department of Computer Science and Applied Statistics, University of New Brunswick St.John)