Online Program

WITHDRAWN - Who is most vulnerable? Estimating heterogeneous causal effects of air quality regulations with a novel principal stratification-based Bayesian machine learning approach (307862)

Keywords: Air Quality Regulations, Pollution, Health Outcomes, Causal Inference, Principal Stratification, Machine Learning, Bayesian Causal Trees

As many studies have shown that air pollution is a major environmental risk for health, various policies have been enacted in the US to reduce the level of the pollutants. However, as regulatory actions are becoming prohibitively expensive, many stakeholders are calling for targeted policies aimed at reducing exposure in the most vulnerable groups. The existing literature lacks robust methods for evaluating heterogeneity in the health effects of air pollution regulations across population subgroups. We address this shortcoming by developing a novel approach combining causal inference principles and machine learning methods to assess which subgroups are most vulnerable to the non-implementation of regulatory policies. To do so, we introduce two major innovations in principal stratification: we rely on machine learning methodologies for the imputation of the missing potential outcomes for the pollution levels; we employ data-driven techniques to discover the sub-populations with the highest associative and dissociative effects. In the motivating application we estimate the heterogeneity in the causal effects of a 2005 policy for the reduction of pollution on later health outcomes.