A primary objective of environmental policies worldwide is to protect human health. Optimal policy design, however, is typically hampered by limited information regarding both the benefits and the costs associated with regulation. Benefits assessments frequently rely on translating laboratory findings to uncontrolled settings, extrapolating from high- to low- concentration exposures within and across societies, and drawing inferences from observational analyses that do not account for the endogeneity of pollution. Economic assessments have typically focused on the costs of compliance to firms. Efforts to improve societal welfare clearly depend on a strong understanding of both elements. Although the health- pollution relationship largely remains the pursuit of epidemiologists, the focus of economics on casual identification along with valuation techniques consistent with utility maximization has helped to reframe these relationships in a manner that facilitates policy choice and environmental rule setting.
Early epidemiological investigations of the impacts of extreme pollution events were some of the first compelling studies to suggest a causal relationship, with one of the most famous focused on the ‘killer fog’ in London, England in December, 1952. A temperature inversion combined with windless conditions led to a sudden and dramatic increase in air pollution. Because residents were used to winter fogs, there was little, if any, changes in behavior, leading to a rather clean measure of pollution impacts in this case. The dramatic rise in mortality that precisely coincided with the timing of the fog had been a driving force behind the federal regulations aimed at air pollution control.
The pollution levels experienced under this and similarly studied extreme events, however, had been dramatically higher than those that nearly all people in developed countries face today. Moreover, most exposures do not conveniently arrive as a ‘surprise’ under which causal impacts can be easily assessed, and it is on this front that economists have made their most significant contributions. In particular, economic studies have typically focused on quasi-experimental settings in order to synthesize the ‘surprise’ of pollution. Besides improving the causal understanding of these relationships by minimizing threats from confounding, it has also identified important compensatory behaviors undertaken by individuals to mitigate exposure.
These behavioral responses are often nontrivial because many pollutants are observable, and even those that are not easily detectable by the public, such as ozone and particulate matter, are forecast and publicized through a broad range of media outlets. If optimizing individuals compensate for changes in ambient pollution levels by reducing their exposure, estimates that do not account for these responses will understate the biologic relationship between ambient pollution levels and health. This problem is potentially severe because the more an individual is likely to suffer under pollution, the more they have to gain from reduced exposure. Indeed, emerging empirical evidence finds that behavioral responses are largest for more vulnerable individuals (Neidell, 2009; Graff Zivin and Neidell, 2009; Graff Zivin et al., 2011) and that individuals are more responsive to higher levels of pollution (Neidell, 2009; Mansfield et al., 2006). Equally important, these behavioral responses are costly and thus ignoring them will also understate the welfare effects of pollution (or the regulation thereof). Although the costs of spending additional time indoors, rescheduling activities, or even relocating to areas with better environmental quality are often difficult to enumerate, they can represent a substantial fraction of the total costs of pollution.
In the remainder of this article, a basic economic framework for evaluating environmental health impacts is presented, followed by a discussion of the core empirical challenges that researchers face in estimating the relationship between pollution and health. A selective review of significant contributions from the literature that focus on the effects of air pollution is then provided, concluding with some suggestions for fruitful lines of future research.
Estimation of the relationship between pollution and health is typically focused on the following health production function:
where h is a measure of an individual’s health, P is pollution levels assigned to the individual, and A is avoidance behavior. E are other environmental factors that directly affect health, such as weather and allergens, and S are all other behavioral, socioeconomic, and genetic factors affecting health. Given that meteorological elements can play an important role in pollution formation and can also affect health (e.g., cold weather increases asthma exacerbation), E is defined separately because it represents an important source of environmental confounding.
Two main approaches are taken to eqn , with the difference stemming from the treatment of avoidance behavior. The first, or ‘reduced-form’ approach does not directly control for avoidance behavior. As health impacts will depend on ambient pollution levels and avoidance behavior that determines exposure to those pollution levels, the health relationship can be expressed as the following total derivative: dh/dP=δh/δP+δh/δA*δA/δP. The second, or ‘production function’ approach directly controls for avoidance behavior to obtain the partial derivative: δh/δP.
The importance in separating these two approaches is to relate each to the benefit calculation, or willingness to pay (WTP) for a reduction in pollution (Harrington and Portney, 1987; Cropper and Freeman, 1991; Deschenes and Greenstone, 2011). In the reduced-form approach, welfare is typically expressed as: WTP =dh/dP*Ch+pA *δA/δP, where Ch is the ‘full’ cost associated with a change in health, and pA is the price of avoidance behavior. In the production function approach, welfare is typically expressed as: WTP= pA *[(δh/δP)/ (δh/δA)]. Although the production function approach appears more data hungry because of the need to control for avoidance behavior when estimating eqn , the reduced-form approach must also control for avoidance behavior in order to estimate δA/δP, although this can be done separately from estimating eqn . Furthermore, as these expressions demonstrate, all forms of avoidance behavior must be accounted for at some point in order to obtain a proper estimate of WTP.
One advantage of the reduced-form approach is that the econometrician does not need to properly specify the functional form of eqn  with respect to P and A. This is particularly helpful because data limitations often necessitate the use of proxy measures for avoidance behavior, and economic theory provides little guidance on how these proxy measures should enter into eqn .
The value of the production function approach is that it provides estimates of the biological effect of pollution. Because avoidance behavior is likely to vary across socioeconomic and cultural environments, but the biology is considerably less context specific, it facilitates generalizations across settings. Moreover, focusing on the biological effect enables one to potentially identify important nonlinear effects, such as threshold effects, and heterogeneous effects based on individual susceptibility, both of which can play an important role in defining the feasible set of policy interventions. Interested readers should consult Graff Zivin and Neidell (2013) for more elaboration on this framework.