Wednesday, July 15, 2009
Innovative stable isotope mixing models -- new paper in PLoS One!
Since our 2008 paper in Ecology Letters on stable isotope mixing models, there has been an explosion of interest in improving the analytic methods for these tools (our tool MixSIR is not the first mixing model by any means, but it is the first fully Bayesian mixing model).
What are stable isotope mixing models? In essence, they are quantitative tools for estimating the contribution of different isotope sources (e.g. prey items) to a mixture (e.g. predator) based on the similarities and differences of the isotopic signatures of sources and mixtures. In other words, these models rest on the logic "if a predator has a similar isotopic signature to a prey item, it is likely the predator is eating that prey item." Of course, this assumes that stable isotope signatures are transferable -- there is a huge body of literature demonstrating that they are, although there are rules governing this transfer (isotope fractionation based on the metabolic costs of respiring heavier vs. lighter isotopes).
Because these models ultimately are estimating proportional contributions of sources to a mixture (where proportions sum to unity), the quantitative methods required are non-normal and non-trivial.... until now! In a paper recently published in PLoS One, my colleagues (Eric Ward, Jon Moore, Chris Darimont) and I made use of a data transformtion in order to normalize the proportionality of paramters in the mixing model. Put simply, this simple trick places complicated non-normal mixing models into a general linear mixed model framework. This approach is advantageous becuase researchers can now explicitly parameterize variabilility in the diets of individuals, or groups of individuals, through random effects. These random effects parameters can be seen as estimates of niche width -- and that's the other really cool thing about the analytic approach -- researchers now have to tools to estimate the ecological niche widths of individuals, groups and populaitons based on stable isotope signatures!
Should be intesting to see how the models are recieved and applied. The paper is free to download, and since it is PLoS one, other researchers and interested parties can leave "cyber" comments direclty on the "paper". Go check it out!