Causal Inference Methods for Evaluating Air Quality Control Policies
*Corwin Zigler, Harvard School of Public Health
Keywords: causal inference, interference, SUTVA, air pollution, policy evaluation
Existing methods for causal inference and comparative effectiveness research (CER) are primarily tailored to clinical investigations of individual-level therapies, but have limited capacity to evaluate complex public health policies. One important example is the urgent need to evaluate the health benefits of policies that compel air pollution sources across the U.S. (e.g., power plants) to adopt measures to reduce harmful pollution emissions with the goal of cleaning ambient air, reducing population exposure to harmful pollutants, and improving public health. A defining feature of such policies is that air pollution travels across space, meaning health outcomes at a given location are determined not by a single action, but by collections of actions taken at multiple power plants. This leads to a challenge we refer to as bipartite causal inference with interference, whereby potential outcomes for a given location are determined by collections of treatments applied at many power plants. We outline new Bayesian methods to estimate causal effects in such settings, using regulatory policies targeting harmful emissions from U.S. power plants as an illustrative case study.