UC Berkeley

How can we effectively mitigate the public health damages associated with air pollution, particularly particulate matter (PM)? While the government can initiate efforts by reducing PM emissions, a critical question emerges: do the public health responses to PM vary depending on its originating locations? If that is the case, the most effective approach would involve prioritizing the abatement of emissions from the origin with the most significant impacts on deteriorating public health outcomes due to PM exposure. This dissertation addresses this question head-on in its first chapter, investigating whether PM impacts exhibit heterogeneity across origins, encompassing transboundary anthropogenic emissions from different countries, non-anthropogenic emissions from desert dust and wildfires, and domestic anthropogenic sources. The second chapter delves into the implications of these findings, exploring the disparities in estimated PM impacts when utilizing different types of transboundary PM contributions as instrumental variables, which have been traditionally considered valid exogenous variations for identifying the impacts of PM. Returning to the core objective of alleviating public health burdens, the third chapter of this dissertation delves into the role of information disclosure. By inducing public awareness of air pollution levels and promoting avoidance behavior through advisories or warnings, governments seek to reduce the health burdens. This chapter examines the efficacy of air quality alert systems in achieving this goal, weighing the potential reduction of healthcare costs against the operating expenses of warnings.

More detailed abstracts of the three chapters are as follows. The first chapter starts by recognizing that a large portion of emitted PM crosses borders, damaging health outside of its originating jurisdiction. As a result, there is substantial interest in determining the origins of imported PM and measuring its impact(ApSimon and Warren 1996; Jazil and Brown 2012; Crawford et al. 2021; Du, Guo, et al. 2020; Du, Jin, et al. 2020a; Jordan et al. 2020; National Research Council 2010). Current approaches to estimating health damages from PM assume that pollution originating in different jurisdictions causes the same per-unit harm (Dedoussi et al. 2020; Lim et al. 2020; Chen, Li, et al. 2021; Sergi et al. 2020; Chen, Lin, et al. 2022; Diao et al. 2021; Heo, Ito, and Kotamarthi 2023; S.-B. Han et al. 2021; Liu et al. 2020; Gu and Yim 2016). However, because the chemical and physical features of PM originating from different locations vary dramatically,(Harrison and Yin 2000; Kelly and Fussell 2012; Strak et al. 2012; Schmid and Stoeger 2016) it is widely hypothesized that PM from distinct origins may generate differing health effects.(Gilmour et al. 2007; Li et al. 2019; Seagrave et al. 2006; Chen, Hoek, et al. 2022; Strak et al. 2011; Wu et al. 2021; Achilleos et al. 2017; Thurston, Chen, and Campen 2022). It has remained challenging to test this theory because only total PM (i.e., transboundary PM + domestic PM) is measured by the observational network, preventing analyses from disentangling harms that originate from different jurisdictions but impact a single population concurrently. Thus, PM originating domestically has confounded the measurement of harm from transboundary PM, and vice versa. Here, we provide the first direct and unconfounded evidence that the health impacts of PM depend on its origin. We simultaneously measure harm over time from both transboundary and domestic PM within a single population located at the nexus of the world’s most contentious transboundary air pollution disputes(Jia and Ku 2019; Shapiro and Yarime 2021; Lim et al. 2020; Jung, Choi, and Yoon 2022). We use an atmospheric model to decompose the origins of PM individuals are exposed to at each location in South Korea every day during 2005–2016. We then link these data to universal healthcare records in an econometric analysis that simultaneously measures and accounts for harms from seven types of PM, each from a distinct origin. We discover that the toxicities of transboundary PM from North Korea and China are approximately 5× and 2.6× greater, respectively, than PM originating domestically in South Korea, and that the health response to natural dust differs from anthropogenic sources. Because toxicity differs by origin, we compute that transboundary sources contribute only 43% of South Korea’s anthropogenic PM load but generate over 70% of its associated respiratory health costs. Our results directly validate the longstanding hypothesis(Gilmour et al. 2007; Li et al. 2019; Seagrave et al. 2006; Strak et al. 2011; Wu et al. 2021; Thurston, Chen, and Campen 2022) that PM should not be treated as if it is a single pollutant, but instead should be considered a mixture of pollutants of distinct origin, each with a unique measurable impact on human health.

The second chapter investigates the implications of the results of the first chapter from the perspective of the applied econometrics. When using instrumental variables (IV), many studies in applied econometrics have focused on the conventional requirements for the use of IVs–the first-stage condition and the exclusion restriction–to get estimates of the causal impacts and extrapolate those coefficients to counterfactual policy settings. An issue is that the results of IV regressions could differ considerably depending on which sets of IVs are in use, even if these IVs are deemed appropriate in terms of those two conventional requirements. Identifying variations induced by a particular set of IVs can be associated with certain groups of people or channels that determine the influence of the endogenous variable on the dependent variable. These groups and channels may have different characteristics compared to those associated with other IV sets, thus leading to heterogeneous responses. (Angrist and Krueger 2001; Mogstad and Torgovitsky 2018; Mogstad, Santos, and Torgovitsky 2018; Dunning 2008) As a result, it is possible that the IV or two-stage least-square (2SLS) estimates may diverge, even though these estimates are assumed to represent one ‘causal impact,’ and only depict a part of the impacts associated with target variations (or policy-relevant variations (PRV)) one aims to analyze in a counterfactual analysis. However, it is still unknown to what extent the different sets of identifying variations can result in 2SLS estimates which are not qualitatively similar to those produced when other sets of IVs are used, in real-world empirical settings.

The last chapter investigates the welfare impacts of disclosing information on extreme levels of air pollution. Though air-quality alert systems (AQAS) cover more than 1.7 billion people worldwide, there has been little welfare analysis of these systems. This chapter presents a theoretical framework for deriving lower bounds on the net benefits of an AQAS and applies it to a South Korean system currently covering over 51 million people. Estimating a regression discontinuity design, we find that an alert issuance reduced youth respiratory expenditures by 30% and adult cardiovascular expenditures by 23%. The overall system reduced externalized health expenditures by 28.6 million dollars during 2016−2017, with a minimum benefit-cost ratio of 7.1:1. Including dynamic impacts of alerts increases the minimum benefits (benefit-cost ratio) to 36.7 million dollars (9.2:1). Our findings imply that the AQAS generates significant net benefits and suggests that manipulation of air quality data, which has been observed in other contexts, may negatively impact social welfare.