B1275
Title: Dynamic ICAR spatiotemporal factor models
Authors: Marco Ferreira - Virginia Tech (United States)
Hwasoo Shin - Virginia Tech (United States) [presenting]
Abstract: A novel class of dynamic factor models is proposed for spatiotemporal areal data. This novel class of models assumes that the spatiotemporal process may be represented by some latent factors that evolve through time according to dynamic linear models. As the dimension of the vector of latent factors is typically much smaller than the number of subregions, the proposed class of models may achieve substantial dimension reduction. At each time point, the vector of observations is linearly related to the vector of latent factors through a matrix of factor loadings. Each column of this matrix may be seen as a vectorized map of factor loadings relating one latent factor to the vector of observations. Thus, to account for spatial dependence, it is assumed that each column of the matrix of factor loadings follows an intrinsic conditional autoregressive (ICAR) process. Hence, the class of models is called the Dynamic ICAR Spatiotemporal Factor Models (DIFM). A Gibbs sampler is developed for the exploration of the posterior distribution. In addition, model selection through a Laplace-Metropolis estimator of the predictive density is developed. Two case studies are presented: the first is on simulated data demonstrating that DIFMs are identifiable and that the proposed inferential procedure works well, whereas the second case study demonstrates the utility of the DIFM framework with an application to the drug overdose epidemic in the United States.
