|
Abstract:
|
The exposure pathways models that relate biomarker concentrations of toxic metals to local and global environmental exposures, while hierarchical in nature, often involve thousands of parameters. Because closed form expressions for the joint posterior distributions are unavailable, we sample from these distributions using Markov chain Monte Carlo (MCMC). We will discuss the computational methods we use to fit complicated hierarchical Bayesian models using MCMC. A key ingredient in making this work is the Battelle MCMC library, which allows a statistician to focus on the modeling, rather than the intensive data management required in large problems. We demonstrate methods for diagnosing and assessing the fit of these complicated models and consider sensitivity analyses that describe the impact of different parts of the model on the posterior distributions of the parameters.
|
