statistics colloquium
Event Type:
Colloquium
Speaker:
Erik Erhardt
Event Date:
Friday, November 13, 2015 -
3:30pm to 5:00pm
Location:
SMLC 120
Audience:
General PublicFaculty/StaffStudentsAlumni/Friends
Sponsor/s:
Math and Stat Department
Event Description:
Title: A Bayesian stable isotope mixing model for covariate selection
Abstract: You are what you eat, conditional on selected covariates. We developed an extended Bayesian mixing model with variable selection to infer proportional contributions of diet sources to a consumer animal's diet conditional on covariates, such as season and location. Our model performs model selection to arrive at a parsimonious explanation of the diet. Bayesian methods apply for arbitrary numbers of isotopes and diet sources but existing models are somewhat limited as many can not condition on covariates. The current state of the art is to fit many separate models for each factor covariate combination and compare the separate posterior estimates. Our model uses stable isotope ratios and concentrations of carbon, nitrogen, and sulfur, isotopic fractionations, elemental assimilation efficiencies, as well as prior information (expert opinion) to inform the diet parameters within a single comprehensive model. Our model appropriately accounts for parameter correlation, uncertainty, and prior information at all levels of the analysis.