Hierarchical Dynamic Dispersion Models for Discrete-Valued Spatio-Temporal Data

• Activity Number: 410
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Guohui Wu*+ and Scott Holan and Christopher Wikle and Mark Wildhaber
• Companies: University of Missouri and University of Missouri and University of Missouri and U.S. Geological Survey
• Keywords: Bayesian hierarchical model ; Discrete-valued ; Dispersion ; Dynamic ; Spatio-temporal

Abstract

Modeling discrete-valued spatio-temporal data can often be challenging due to lack of routine model specification, high dimensionality, and potential over/under dispersion. Additionally, in many real-world applications spatial over/under dispersion, exhibited in the data, varies as a function of time. Motivated by the need to accommodate time-varying dispersion, while flexibly incorporating different spatio-temporal dependence structures, we propose a hierarchical Bayesian model that incorporates a versatile likelihood specification that facilitates dynamic dispersion. Our approach allows the likelihood function to readily adapt to different nuances of the discrete-valued data. Simultaneously, by taking a hierarchical Bayesian approach, our approach naturally quantifies various sources of uncertainty. The effectiveness of our methodology is demonstrated through simulation and through a real-ecological application.