You are here:
Use and abuse of mixing models (MixSIAR)
Citation:
Stock, B., B. Semmens, E. Ward, J. Moore, A. Parnell, A. Jackson, D. Phillips, S. Bearhop, AND R. Inger. Use and abuse of mixing models (MixSIAR). Ecological Society of America, Maryland, MD, August 09 - 14, 2015.
Impact/Purpose:
A consortium of scientists from U.S., Canada, Ireland, and United Kingdom academic institutions, as well as EPA, have developed a new model that can provide insight into the flow of nutrients through food webs. The model uses information on the ratios of different stable (non-radioactive) isotopes of an element in animals and in the foods that they eat. With these data the model quantifies the diet composition of each individual animal and how this varies within different segments of the population (e.g., by age, sex, geographic location). Details about choices of parameter values, use of prior data, error structures, number of sources, use of covariates, the number of isotopic tracers used, and the addition of fatty acid data as well as isotopic data are discussed to outline “best practices” for implementing the model in food web studies.
Description:
Background/Question/MethodsCharacterizing trophic links in food webs is a fundamental ecological question. In our efforts to quantify energy flow through food webs, ecologists have increasingly used mixing models to analyze biological tracer data, often from stable isotopes. While mixing models are based on simple concepts, incorporating important biological complexity complicates analysis. In order to spur the implementation of advances in mixing model theory, we recently developed MixSIAR, a GUI tool written in R. Here, we address common questions and pitfalls in mixing model analyses we have seen from our work on MixSIAR.Results/ConclusionsWe explain and suggest “best practices” to ecologists implementing mixing models on the following: 1) Markov Chain Monte Carlo (MCMC) parameters, 2) incorporating prior distributions and the influence of priors, 3) alternative error structures, 4) source aggregating/splitting, 5) incorporating fixed, random, and continuous covariates, 6) effects of the number of tracers included, and 7) application to fatty acid data as well as stable isotope data. We conclude by outlining unresolved research questions in the use of mixing models such as MixSIAR.