In applied settings, complex constraints and non-invertible transformations of variables can lead to hierarchical models in which it is impossible to write a likelihood model in closed form. We develop an approach for making marginal inference for such models in the case when sampling from a distribution related to the intractable likelihood is possible. This approach uses the weighted bootstrap within a data augmentation setting, and allows for straightforward inference based on multiple imputation. We apply this approach to a study of landscape-scale gene flow in mule deer in Colorado and Utah.
BYODrink, Sandwiches provided at 10:45am, before start of talk