CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Time-varying step change detection and forecasting in spatio-temporal areal data models Authors:  Gavino Puggioni - University of Rhode Island (United States) [presenting]
Abstract: New methods are proposed that address some common issues in epidemiological studies: time varying step change detection in spatial autocorrelation, and short and medium term forecast stability. The goal is to provide a flexible framework to identify and target areas with increased risk, and to inform early warning systems for risk surveillance. We present a flexible two stage space-time CAR model with Bayesian model averaging on the set of predictors, and a stochastic process that models variations in step-change boundaries. After testing the method in a simulation study, we apply it to real data, and compare its forecasting performance with other commonly used methods.