2017/04/25 by Abhirup Datta, Sudipto Banerjee, Datta, Abhirup +3 · 3 citations
Economics, Econometrics and Finance · Medicine · Social Sciences · #Data-Driven Disease Surveillance #FOS: Computer and information sciences #Health disparities and outcomes #Methodology (stat.ME) #Spatial and Panel Data Analysis
paper · pdf · doi:10.48550/arxiv.1704.07848
openalex publication_date 2017/04/25 · openalex created_date 2022/10/06 · openalex updated_date 2026/07/28
Hierarchical models for regionally aggregated disease incidence data commonly\ninvolve region specific latent random effects that are modeled jointly as\nhaving a multivariate Gaussian distribution. The covariance or precision matrix\nincorporates the spatial dependence between the regions. Common choices for the\nprecision matrix include the widely used ICAR model, which is singular, and its\nnonsingular extension which lacks interpretability. We propose a new parametric\nmodel for the precision matrix based on a directed acyclic graph (DAG)\nrepresentation of the spatial dependence. Our model guarantees positive\ndefiniteness and, hence, in addition to being a valid prior for regional\nspatially correlated random effects, can also directly model the outcome from\ndependent data like images and networks. Theoretical results establish a link\nbetween the parameters in our model and the variance and covariances of the\nrandom effects. Substantive simulation studies demonstrate that the improved\ninterpretability of our model reaps benefits in terms of accurately recovering\nthe latent spatial random effects as well as for inference on the spatial\ncovariance parameters. Under modest spatial correlation, our model far\noutperforms the CAR models, while the performances are similar when the spatial\ncorrelation is strong. We also assess sensitivity to the choice of the ordering\nin the DAG construction using theoretical and empirical results which testify\nto the robustness of our model. We also present a large-scale public health\napplication demonstrating the competitive performance of the model.\n