Understanding Spatial Regression Models from a Weighting Perspective in an Observational Study of Superfund Remediation
Understanding Spatial Regression Models from a Weighting Perspective in an Observational Study of Superfund Remediation
Superfund sites are locations in the United States with high levels of environmental toxicants, often resulting from industrial activity or improper waste management. Given mounting evidence linking prenatal environmental exposures to adverse birth outcomes, estimating the impact of Superfund remediation is of substantial policy relevance. A widespread approach is to fit a spatial regression, i.e., a linear regression of the outcome (e.g., birth weight) on binary treatment (e.g., indicator for Superfund site remediation) and covariates, along with a spatially structured error term to account for unmeasured spatial confounding. Despite this common practice, it remains unclear to what extent spatial regression models account for unmeasured spatial confounding in finite samples and whether such adjustments can be reformulated within a design-based framework for causal inference. To fill this knowledge gap, we introduce a weighting framework that encompasses three canonical types of spatial regression models: random effects, conditional autoregressive, and Gaussian process models. This framework yields new insights into how spatial regression models build causal contrasts between treated and control units. Specifically, we show that: 1) the spatially autocorrelated error term produces approximate balance on a hidden set of covariates, thereby adjusting for a specific class of unmeasured confounders; and 2) the error covariance structure can be equivalently expressed as regressors in a linear model. We also introduce a new average treatment effect estimator that simultaneously accounts for multiple forms of unmeasured spatial confounding, as well as diagnostics that enhance interpretability. In a study of Superfund remediation, our approach illuminates the role of design-based adjustment for confounding and provides guidance for evaluating environmental interventions in spatial settings.
Sophie M. Woodward、Francesca Dominici、Jose R. Zubizarreta
环境科学理论环境污染、环境污染防治环境科学基础理论
Sophie M. Woodward,Francesca Dominici,Jose R. Zubizarreta.Understanding Spatial Regression Models from a Weighting Perspective in an Observational Study of Superfund Remediation[EB/OL].(2025-08-27)[2025-09-11].https://arxiv.org/abs/2508.19572.点此复制
评论