Empirical Challenges
In this section, three primary challenges confronted by empiricists when estimating the relationship between pollution and health are outlined. Although weather is a potential confounder, this is not discussed at length because it is directly observable (often at a finer scale than pollution data), so that any threat can be obviated through the careful control of relevant variables.
Measurement Of Health
The measurement of health outcomes and how to place monetary values on them is a persistent challenge. A frequently used measure is mortality, which is objectively measured and can be readily monetized using estimates of the value of a statistical life (VSL). One concern with using mortality is that it is an extreme outcome that misses more subtle outcomes that may be more commonplace. Furthermore, using VSL to monetize these impacts may be misleading if the loss only represents short-term mortality displacement, commonly referred to as ‘harvesting.’
Measures of morbidity have also been examined using data on hospitalizations for various conditions, largely respiratory related. Although hospitalizations clearly capture events less severe than death, they may introduce sample selection. Those who have a relationship with a primary care physician (PCP) and receive regular care may never experience a hospitalization, and access to a PCP is clearly endogenous. Furthermore, the economic valuation of hospitalizations is particularly difficult as hospital charges (which are all that is typically available) do not capture the costs associated with the pain and suffering experienced by sickened individuals or their family members.
Birth outcomes are another metric that has some of the desirable properties of both mortality and morbidity endpoints, albeit for a select population. Like mortality and hospitalizations, birth outcomes are a census and not a sample, hence offering large sample sizes for analysis. Unlike mortality, birth outcomes can capture more subtle impacts, and unlike hospitalizations, they do not introduce sample selection because any birth that files for a birth certificate is reported. Valuation approaches can be used when the birth outcome studied has been linked to monetizable events – for example, birth weight has been linked with education and earnings (Black et al. 2007) – although these links may not capture all relevant costs.
An emerging area of focus is on indirect ‘health’ outcomes at school or the workplace, principally absenteeism and performance. Such outcomes offer terrific promise for capturing rather subtle health impacts that might be broadly disseminated throughout society. They are also generally straightforward to monetize, particularly for performance. Limited data availability, especially for representative samples, is a formidable obstacle to the conduct of credible empirical work in this area.
Assignment Of Local Pollution Levels
Most studies focus on air pollution because of the availability of data from ambient air pollution monitors, which typically measure air concentrations at an hourly scale at a fixed location. Although this frequency of measurement generates data at a fine temporal scale, the limited number of monitor locations relative to the size of a country and the geographic distribution of the population leads to data that are rather coarse on a spatial scale. As a result, studies often approximate contemporaneous pollution levels based on an individual’s general location and the location of the monitor. This crude approach leads to measurement error that increases with an individual’s distance from the monitor and the degree to which pollutants disperse nonuniformly. This measurement error will typically bias estimates downward, but with a large enough dataset, researchers can use data from multiple monitors and various weighting techniques to obtain more precise assignments of localized pollution levels. A finer level of geographic disaggregation for individuals, such as a residential address, also allows for better assignment of relevant pollution levels and hence is more likely to provide precise estimates.
The usual mobility of individuals in their everyday life (not in response to pollution, discussed below), both within a day and over time, can also present a measurement issue. Individuals spend their time not only at home, but at work, school, and other possible locations that are not typically recorded. Although the use of personal monitors is designed to overcome this, two issues remain: (1) the high costs of personal monitoring often result in the use of a small, unrepresentative sample without a clearly defined control group; and (2) the link to policy is less clear because indoor sources also contribute to pollution but are subject to different regulatory rules. Mobility over time also presents a significant measurement issue in assigning cumulative exposure over longer periods of time. Focusing on children, and in particular infants, whose parents are typically less mobile, can greatly limit this concern (Joyce et al. 1989; Chay and Greenstone, 2003).
Behavioral Responses To Pollution
Optimizing individuals may respond to pollution with permanent changes, such as relocating (i.e., sorting), and temporary changes, such as spending less time outside. As argued above, it is crucial to understand the role of these behavioral responses both to allow generalizations from one setting to another and to account for the full welfare costs of pollution. Although careful quasi-experimental designs can address permanent behavioral changes by exploiting exogenous shocks to pollution levels, short-run changes pose greater challenges because many of these responses involve nonmarket behaviors that are difficult to observe. For example, simply spending less time outside on a polluted day is a highly effective means for reducing exposure, but such an activity is rarely recorded. Clearly, the degree to which such short-run behavioral responses will be important depends on the ‘visibility’ of pollution, either literally, through information dissemination, or through health feedbacks that allow individuals to infer it on the basis of physiological responses.