2016 Progress Report: A Causal Inference Framework to Support Policy Decisions by Evaluating the Effectiveness of Past Air Pollution Control Strategies for the Entire United StatesEPA Grant Number: R835872C004
Subproject: this is subproject number 004 , established and managed by the Center Director under grant R835872
(EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).
Center: Regional Air Pollution Mixtures
Center Director: Koutrakis, Petros
Title: A Causal Inference Framework to Support Policy Decisions by Evaluating the Effectiveness of Past Air Pollution Control Strategies for the Entire United States
Investigators: Zigler, Corwin , Barrett, Steven , Dominici, Francesca , Mickley, Loretta J. , Schwartz, Joel
Institution: Harvard University , Massachusetts Institute of Technology
EPA Project Officer: Keating, Terry
Project Period: December 1, 2015 through November 30, 2020
Project Period Covered by this Report: December 1, 2015 through November 30,2016
RFA: Air, Climate And Energy (ACE) Centers: Science Supporting Solutions (2014) RFA Text | Recipients Lists
Research Category: Air , Climate Change , Air Quality and Air Toxics , Airborne Particulate Matter Health Effects
The overall objective of Project 4 is to develop a new methodological framework rooted in principles of causal inference to investigate the effectiveness of specific control strategies on impacting the largest power-generating units in the United States. In Project 4, we combine state-of-the-art atmospheric modeling, causal inference methods, and national data sets to conduct accountability research; that is, research that characterizes causal effects of well-defined regulatory actions at power plants on: 1) emissions; 2) air quality across distant locations in accordance with atmospheric fate, transport, and other factors; and 3) health outcomes. Project 4 has 4 specific objectives.
Objective 1 is to develop a national databaseon emissions control technologies employed at a large number of power-generating units in the US linked with: continuous emissions monitoring, ambient air quality monitoring, weather, population demographics, and Medicare hospitalization and mortality outcomes for the period 1995 to 2015. Objective 2is to estimate and compare the causal effects of past control strategies implemented at the largest power-generating facilitieson SO2, NOx, CO2, and PM2.5 emissions and population exposure to criteria pollutants (PM2.5 and O3) for the entire US for the period 2000 to 2015. This requires integrating new statistical methods for causal inference with atmospheric chemistry models of how changes in emissions impact ambient exposures across distant locations in accordance with atmospheric fate, transport, and other factors. Objective 3is to estimate the causal effects of past control strategies implemented at the largest power-generating facilities on mortality and morbidity in the entire US both locally and nationally, and compare the differential health impact of different control strategies. And Objective 4is to develop approaches for mediation analysis that will quantify the extent to which causal effects of regulatory actions on health outcomes can be attributable to changes in targeted modifiable factors (e.g., emissions, targeted pollutants), as opposed to being driven by co-benefits to other factors.
On Objective 1, we have made significant progress during Year 1. We have compiled a nationally-representative database linking (in time and space) data on power plant emissions from continuous emissions monitors, auxiliary information on EGUs from the Energy Information Administration, ambient pollution monitoring data, data-fusion based estimates of ambient pollution, population demographics, climatological factors, health outcomes among US Medicare beneficiaries, and air movement trajectories modeled with the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT). We have successfully implemented and scaled computation for the HYSPLIT simulation procedure to simulate mappings between power plants and population locations (e.g., zip codes). Existing HYSPLIT implementations would have required computational time on the order of years for our required purposes. These tools are now serving as the basis for many ongoing statistical methods development projects and analyses, and have been developed (along with postGIS tools) within an R interface for reproducible workflow.
In support of Objectives 2 and 3 during Year 1, we have focused our efforts on proposing, developing, and deploying new methods for causal inference for evaluating causal effects of air quality policies. We have prepared and submitted a number of manuscripts to peer-reviewed publications during Year 1, including papers discussing adjustment for unmeasured spatial confounding, adjustment for confounding in analysis of multivariate exposures, and adjustment for misclassified treatment alignment. A full list of Project 4 submitted manuscripts is provided below. In addition, we have made conference and academic presentations of our work relevant to evaluating effectiveness of interventions on power plants, focusing on development and application of causal inference methods. A list of representative presentations is also provided.
The HYSPLIT capabilities described above represent a first approximation to long-range pollution transport that a) is computational feasible and b) can serve as the basis for ongoing statistical methods development. The next year will be spent augmenting the sophistication of our capabilities to incorporate knowledge of chemical transport into statistical methodology. We will explore use of other techniques (e.g., adjoint modeling) that can balance sophistication and refinement against computational scalability. In addition, we will continue to deploy the methods developed during Year 1 to epidemiological analyses of the causal impacts of power plant emissions and interventions.
Journal Articles on this Report : 1 Displayed | Download in RIS Format
|Other subproject views:||All 7 publications||1 publications in selected types||All 1 journal articles|
|Other center views:||All 60 publications||51 publications in selected types||All 51 journal articles|
||Braun D, Gorfine M, Parmigiani G, Arvold ND, Dominici F, Zigler C.Propensity scores with misclassified treatment assignment: a likelihood-based adjustment.Biostatistics2017;18(4):695-710.||
Accountability assessment, power-generating sector, intervention evaluation
- An R Package for EPA data retrieving and processing Exit
- PM2.5-Nonattainment Exit
- Harvard Dataverse Air Quality Regs Exit
- Distance Adjusted Propensity Score Matching R Package Exit
- DAPSm-Analysis Exit
Progress and Final Reports:Original Abstract
Main Center Abstract and Reports:R835872 Regional Air Pollution Mixtures
Subprojects under this Center: (EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).
R835872C001 Project 1: Regional Air Pollution Mixtures: The Past and Future Impacts of Emission Controls and Climate Change on Air Quality and Health
R835872C002 Project 2: Air Pollutant Mixtures in Eastern Massachusetts: Spatial Multi-resolution Analysis of Trends, Effects of Modifiable Factors, Climate and Particle-induced Mortality
R835872C003 Project 3: Causal Estimates of Effects of Regional and National Pollution Mixtures on Health: Providing Tools for Policy Makers
R835872C004 A Causal Inference Framework to Support Policy Decisions by Evaluating the Effectiveness of Past Air Pollution Control Strategies for the Entire United States
R835872C005 Project 5: Projecting and Quantifying Future Changes in Socioeconomic Drivers of Air Pollution and its Health-Related Impacts