A1844
Title: Bayesian variable selection for censored spatial responses with application to PFAS concentrations in California
Authors: Suman Majumder - Indian Statistical Institute (India) [presenting]
Abstract: Per- and polyfluoroalkyl substances (PFAS) are persistent contaminants of public health concern due to their resistance to degradation, widespread occurrence, and potential health effects. Statistical analysis of PFAS concentrations in groundwater is complicated by censoring from detection limits, strong spatial dependence, and high-dimensional covariates. While PFAS levels are believed to be influenced by diverse sociodemographic, industrial, and environmental factors, their relative contributions remain unclear, motivating statistical approaches that can isolate key predictors from a large candidate set. A Bayesian hierarchical framework is developed that embeds censoring within a spatial process model via approximate Gaussian processes and employs a global-local shrinkage prior for effective high-dimensional variable selection. To refine inference, three post-selection strategies are compared based on their predictive accuracy, robustness to censoring, and stability of variable inclusion. Application of this framework to PFOS concentrations in California groundwater identifies a scientifically meaningful set of drivers including demographic factors, industrial sources, proximity to airports, traffic density, and environmental features such as herbaceous cover, elevation, and ozone concentration. The proposed framework provides interpretable inference while simultaneously offering actionable insights into the factors driving PFAS contamination in groundwater.
