Rather than provide an exhaustive review of the economic literature examining the relationship between pollution and health, this section limits its attention to a selection of studies that offer key insights or introduce important methodological advances.
One of the earliest examples of a quasi-experimental approach to estimate an environmental health relationship in relatively recent times is found in a series of studies by Pope et al. (1992); Ransom and Pope (1992); Ransom and Pope (1995). The authors used changes in pollution that had resulted from the opening and closing of a steel mill, which was a major source of particulate matter, in the central Valley of Utah due to a labor strike. As the steel mill had closed due to a labor strike, the temporary changes in pollution were credibly exogenous and unlikely to lead to any immediate residential sorting. Furthermore, the authors selected a neighboring, unaffected community as a control group to account for time trends by estimating difference-in-differences models. When the steel mill was closed, the authors found significant declines in school absences, respiratory-related hospital admissions, and mortality. One potential concern with this study is that the steel mill closure has also led to a temporary change in income, which may affect one’s use of time and services. This does not seem likely to be an issue for school absences, hence at least some of the findings are credibly causal. A more significant concern with the design is that, as an ‘event study,’ the pollution variable is common to all members in a group for a given time period (despite the availability of individual level health outcomes as dependent variables). As a result, their standard errors are likely to be nontrivially understated, making the appropriate statistical inference in this setting particularly challenging (Donald and Lang, 2007).
One important study by Chay and Greenstone (2003) overcame this problem by focusing on the recession of the early 1980s. The dramatic change in manufacturing that had resulted from this recession induced considerable spatial variation in total suspended particulates (TSPs) throughout the US in a short period of time, with some areas experiencing as large as a 35% decline in 3 years. These changes in TSPs are unlikely to be related to other factors affecting health. Importantly, although income changed considerably at the same time, it did not show comparable spatial patterns as with TSP. Using this exogenous variation in levels of pollution at the county-year level to identify environmental health effects, they estimate that a one-unit decline in TSPs associated with the recession yields benefits of roughly US$14 billion, recognizing that this captures only one health outcome and only for a specific group.
Although the Chay and Greenstone results are nontrivial, the continued improvements in air quality since then suggest that the results also apply to a time period when pollution levels in the US are considerably higher. Currie and Neidell (2005) turn their attention to infant mortality in California during the 1990s, a period that is much more reflective of contemporary pollution levels across much of the developed world. They use zip code fixed effects to account for residential sorting, thereby exploiting the strong temporal variations in pollution levels in the short-run due to changes in plausibly exogenous ambient conditions (rather than anthropogenic sources) to identify health impacts. They find that reductions in carbon monoxide over the 1990s saved approximately 1000 infant lives in California, which translates into benefits of roughly US$4.8 billion.
Currie et al. (2009), like Currie and Neidell (2005), focus on infant outcomes in a more recent time period, but use the exact address of the mother to improve pollution assignment and estimate sibling fixed effect models to control for differences in family background and genetics. They find that a one-unit change in mean carbon monoxide (CO) during the last trimester of pregnancy increases the risk of low birth weight by 8%, and a one-unit change in mean CO during the first two weeks after birth also increases the risk of infant mortality by 2.5% relative to baseline levels. The authors calculate that the 15-year decline in CO from 1989–2003 translates into US$720 million in lifetime earnings from improvements in birth weight and US$2.2 billion from the reduction in infant mortality for the 2003 birth cohort. The use of sibling fixed effects increases estimates, suggesting the importance of accounting for maternal characteristics within neighborhoods. And the better assignment of pollution by using the mother’s exact address rather than zip code also increases point estimates, consistent with measurement error inducing a downward bias.
In a novel design, Lleras-Muney (2010) uses the relocation of military personnel to estimate the effect of various pollutants on children’s health. The relocation of personnel is entirely based on ‘the needs of the army’, which explicitly rules out the possibility of sorting and offers a plausibly exogenous source of variation in pollution. Using this design, LlerasMuney finds that a one standard deviation decrease in ground-level ozone exposure decreases the probability of a respiratory hospitalization for children by 8–23%. Her estimates suggest that lowering pollution levels nationwide to the levels experienced in ‘low’ pollution areas would save approximately US$928 million (US$1994) in direct medical expenditures alone.
All of the previously mentioned studies exploit ‘natural’ experiments that generate exogenous changes in ambient pollution in order to minimize concerns regarding residential sorting and other long-run behavioral responses to poor environmental quality. They generally ignore potential short-run adjustments that could also impact the environment-health relationship, and hence provide estimates of a reduced-form relationship between pollution and health. The key challenge in capturing these short-run behavioral responses is clearly the availability of data suited for the task, and researchers often follow creative paths for obtaining such data. One example is Neidell (2009), who uses attendance data from several outdoor facilities in Los Angeles to uncover significant behavioral responses to high ozone levels that are forecasted through smog alerts. As smog alerts are issued only when ozone is forecasted to exceed a particular threshold, he employs a regression discontinuity design to compare attendance on days just above the threshold to that just below. Although this paper does not provide estimates of the costs of avoidance behavior, in a closely related paper Graff Zivin and Neidell (2009) examine successive days of smog alerts to show that the costs of avoidance behavior, due to limited opportunities for intertemporal substitution, are increasing over time. Graff Zivin et al. (2011) identify substantial increases in the purchase of bottled water when local municipalities violate drinking water standards. As this type of avoidance behavioral is market-based, the authors have calculated the costs associated with it, and have found that water quality violations in 2005 induced roughly US$60 million worth of bottled water purchases nationwide.
Two notable studies attempt to produce estimates of the biological effect of ozone on health. In the paper discussed earlier, Neidell (2009) controls for smog alerts and ozone forecasts as a proxy for avoidance behavior when estimating the relationship between ozone and respiratory-related hospitalizations. Using zip code fixed effects and exploiting the strong daily temporal variation in ozone, he finds that including these proxies significantly increases the estimated impact of ozone on health. Moretti and Neidell (2011) use daily boat arrivals and departures into the port of Los Angeles as an instrumental variable (IV) for ozone levels, which deals with both avoidance behavior and measurement error in pollution assignment. Boat traffic represents a major source of pollution for the Los Angeles region and, because of the extended length of travel and unpredictable conditions at sea, daily variation in boat traffic is arguably uncorrelated with other short-run determinants of health and is not included in the ozone forecasts used to encourage avoidance behavior. Similar to Neidell (2009), they find that using boat traffic as an IV leads to significantly larger estimates for the impacts of pollution on health.
Although the short-run behavior literature has generally assessed the costs associated with avoiding exposure, this again represents a partial characterization of social welfare; a complete calculation requires an assessment of both avoidance costs as well as the costs of those adverse health effects that are not avoided. To our knowledge, the only attempt to bring both pieces together in a quasi-experimental setting is from Deschenes and Greenstone (2011), who focus on the health effects of extreme temperatures, which are forecast to increase under climate change. They construct a WTP estimate to avoid extreme heat that includes the costs due to excess mortality as well as expenditure on energy consumption as a proxy for air conditioning usage to buffer individuals from exposure to that heat. Using county fixed effects to exploit the plausibly exogenous variation in temperatures in an area within a given year, they find that the avoidance costs are roughly 25% of the mortality costs.
Although most of the literature has focused on primary health endpoints, for example, mortality and hospitalizations, an emerging literature has begun to examine the manifestation of less visible health assaults on nonhealth outcomes. Although these impacts are referred to as secondary, it remains possible for them to exceed the costs of primary impacts depending on their prevalence. Almond et al. (2009) examine the impact from prenatal exposure to radioactive fallout from the 1986
Chernobyl accident on both birth and schooling outcomes for children in Sweden. Although Sweden is more than 500 miles away from Chernobyl, weather conditions forced some of the plume over Sweden, and local variation in rainfall levels led to stark geographic variation in the levels of fallout throughout the country. Their study reveals that radiation exposure exhibits latent effects that affect human capital development later in life. Although they find little evidence of health effects as measured by birth outcomes and childhood hospitalizations, they find significant decreases in several schooling outcomes that correspond to roughly US$510 million in lost annual earnings.
Graff Zivin and Neidell (2012) also follow a nontraditional approach by examining the impact of ozone on worker productivity. They use a unique dataset on agricultural workers who are paid by piece rate and whose labor supply is highly inelastic in the short run, hence limiting the scope for avoidance behavior and the need to value it. Using models with worker fixed effects to exploit plausibly exogenous daily variation in ozone levels, they find that a 10 ppb decrease in ozone concentrations increases worker productivity by 5.5%, which translates into productivity benefits to the agricultural industry of approximately US$700 million.