Bayesian Inference for Temporal Gradients from Regionally Aggregated Space-Time Data

• Activity Number: 502
• Type: Invited
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Sudipto Banerjee*+ and Harrison Quick and Bradley P. Carlin
• Companies: University of Minnesota and University of Minnesota and University of Minnesota
• Keywords: Bayesian inference ; Gradients ; Markov random fields ; Stochastic processes ; Spatial-temporal modeling ; Spatial statistics

Abstract

Advances in Geographical Information Systems (GIS) have led to a burgeoning of spatial-temporal databases. Accounting for associations across space and time in analysing such data sets is routinely called for. Here we consider the less usual setting where space is discrete (e.g. aggregated data over regions) and time is continuous. This can be envisioned when, for instance, we have a collection of functions of time over regions, but these functions are spatially associated. A major objective of our current project is to carry out inference on gradients of the temporal process, while at the same time accounting for spatial similarities across neighbouring regions. We employ a flexible stochastic process for time embedded within a dynamic Markov Random Field framework and we subsequently carry out inference on temporal gradients in a posterior predictive fashion. We analyse a dataset comprising monthly county level asthma hospitalization rates in California where we are concerned with temporal changes in the residual and fitted rate curves after accounting for seasonality, spatial-temporal ozone levels, and several spatially-resolved important socio-demographic covariates.