The precise physiological mechanism(s) by which air pollution could cause the health effects indicated in epidemiologic studies is not always fully understood. As a result causal inferences are generally developed based on consistent evidence across multiple epidemiological studies including different areas and study designs, and results from toxicological and human exposure studies in conjunction with the criteria of biological plausibility.

Often co-pollutants are included in integrated assessments separately and their health or economic consequences summed. This may underestimate or overestimate actual damages. The true harmful agent may not be the pollutant under study but a related pollutant or group of pollutants with similar sources and/or formation pathways. For example, O₃ can be considered a marker for an array of photochemical pollutants. Nitrates and sulfates are related to PM as they contribute to secondary particles. Interaction between multiple pollutants is not well understood, and most results are presented for an individual pollutant, although air pollution is experienced as a mixture.

While a substantial literature provides consistent evidence that particles are detrimental to health and a limited number of population-based studies have examined PM effects by chemical composition 60, the differential toxicity of various forms of the PM mixture is unidentified. Differential effects have been demonstrated based on particle size, however chemical composition also appears to play a role as the same size distribution provides different effect estimates based on region 6162. In current analysis of ancillary benefits, all particles of a given size (e.g., PM_2.5) may be treated with equivalent toxicity, however if for example sulfates are more harmful than other particles, technologies that reduce emissions of particles from coal combustion may result in greater health benefits than other technologies. If, for example, elemental carbon is identified to be more detrimental to health, transportation technologies may be more effective.

The vast majority of epidemiological studies applied in ancillary benefits studies use ambient monitoring data as a surrogate for individual or community-level exposure. The relationship between personal exposure and ambient monitoring data varies by pollutant, typically with better correlation for particles than for O₃ 6364. Use of ambient monitors increases the possibility of exposure misclassification, which if non-differential would generally drive effect estimates towards the null, resulting in underestimates. This issue has particular importance for the extrapolation of concentration-response functions from one area to another, as the relationship between ambient monitors and exposure, and thereby health, is a function of indoor pollution and indoor/outdoor activity patterns, which may vary widely across populations.

Many concentration-response functions applied in ancillary benefits studies assume a log-linear relationship between exposure and risk. If the true shape differs, incorrect estimates could be obtained. If the assessment includes pollutant levels above those used to generate the concentration-response function, results will be distorted if the log-linear or otherwise assumed function does not hold. If there exists a safe level below which pollution does not adversely impact health, calculations based on functions assuming no threshold would be incorrect for pollutant levels below the threshold value. Some studies have examined the shape of the concentration-response curve, however such analysis does not exist for all pollutants and health outcomes. Several recent US-based studies found no evidence of a threshold at typical concentrations for the relationship between mortality and O₃ 65 or PM 66.

Pollution and health relationships developed in one area may not be applicable in another location due to differences in the underlying population and pollutant characteristics 67. Efforts are often made to apply locally-derived studies 49, however concentration-response functions do not exist for many outcomes and pollutants for much of the world. Therefore US and European studies are generally employed, although a growing number of epidemiological studies are underway in Asia and Latin America 68697071. Uncertainties introduced by such extrapolation include differences in indoor/outdoor activity patterns, population characteristics, household characteristics that relate to exposure, and the pollution mixture. Likewise, the study of ancillary benefits involves future societies that may have dissimilar housing, populations, health care systems, and pollutant mixtures compared to the present day or the timeframe of the epidemiological research.

Air pollution exposure can be categorized as short-term (i.e., a few days) or longer term (i.e., a few months or years). Health impacts can be classified as those that take place immediately or a short time after exposure, or those that have a gradual or much-delayed response, such as cancer and neurological disease. Cohort studies of PM, which evaluate long-term exposure, generally provide higher estimates for mortality than do time-series studies, which evaluate short-term exposure 7273. Often more information is available regarding health impacts of short-term exposure because such exposure estimates are more readily available. However, the use of only acute-exposure impacts may underestimate the total mortality burden from air pollution 74. Co-benefits studies have used different approaches to address chronic and acute health impacts. Whereas one study 75 included estimates of chronic mortality, excluding acute mortality effects, another 23 incorporated acute mortality only.

While air pollution has been quantitatively linked to many health consequences, there are other health events, including several pediatric and neurological endpoints, for which concentration-response functions have not yet been developed. Some of these health responses are less severe than the more commonly studied effects. However as a counter example, recent studies elucidated the link between O₃ and mortality 51767778. Although less severe health endpoints have lower monetary valuations than more severe impacts, they often occur in larger numbers. Thus, the more grave outcomes such as death and hospital admissions are best viewed as indicators of the much broader spectrum of adverse health effects resulting from air pollution.

The public health burden of mortality associated with air pollution depends not only on the increased risk of death, but also on the length of life shortening. Several recent studies provide evidence that short-term mortality displacement of a few days or less does not account for the observed PM mortality effect estimates 7980818283. Past evaluations of air pollution's effect on life expectancy generally considered only deaths among adults above 30 years of age, but some studies 848586 suggest that infants may be among the sub-populations particularly affected by long-term PM exposure, which would indicate a much larger reduction in life expectancy. Currently, considerable uncertainty remains as to the amount of life-shortening associated with air pollution.

We identified several approaches for economic valuation of averted health consequences (step 3 of Figure 1): COI; human capital; a variety of WTP methods; and quality-adjusted life year (QALY) approaches.

The COI method totals medical and other out-of-pocket expenditures and has been used for acute and chronic health endpoints. For instance, separate models of cancer progression and respiratory disease were used to estimate medical costs from these diseases over one's lifetime 93. COI incorporates direct medical costs, such as for physicians' visits and medications, and indirect costs, including lost income from work loss days. However the approach does not capture other consequences of illness such as psychological suffering, physical pain, transportation to medical appointments, dietary restrictions, and expenditures for friends or family acting as caretakers. The approach can have a welfare theoretic basis, but does not reflect the full damage of illness, hence results usually underestimate costs and should be considered a lower bound. Some COI studies assign a medical expenditure based on primary diagnosis 94.

Early attempts to value mortality risk reductions applied the human capital approach, which estimates the "value of life" as lost productivity. This method is generally recognized as problematic and not based on modern welfare economics, where preferences for reducing death risks are not captured. Another limitation is incorporation of racial- or gender-based discrimination in wages. This method assigns value based solely on income, without regard to social value, so unpaid positions such as homemaker and lower paid positions such as social worker receive lower values. Because data are often available for superior alternatives, this approach is rarely used in health benefit studies.

WTP generates estimates of preferences for improved health that meet the theoretical requirements of neoclassical welfare economics, by aiming to measure the monetary amount persons would willingly sacrifice to avoid negative health outcomes. Complications arise in analysis and interpretation because changes in environmental quality or health often will themselves change the real income (utility) distribution of society. A valuation procedure that sums individual WTP does not capture individual preferences about changes in income distribution. Another complication is that the value of avoided health risk may differ by type of health event and age. For instance, in one study WTP to reduce cancer was about a third larger than that for a similar chronic, degenerative disease 95. VSL estimates can be adjusted based on existing health condition or age, or by the use of a value per life-year saved 96. Use of the value of a statistical life year (VSLY) is very controversial, however, because it implies that age and WTP are proportionally and inversely related, although the literature does not support this assumption. Estimates for children are very limited; however VSLs are generally higher for children 97 and the empirical literature suggests that children's values are approximately twice the value for an adult. WTP measures are theoretically superior to the "supply-side" measures of health damage because they can capture the complete value of health, including pain and suffering.

The hedonic labor market WTP approach relates wage differentials to health risk differences across occupations and industrial/commercial sectors, under the theory that in competitive labor markets, workers in risky jobs should receive wage premiums equal to the value they place on avoiding health risks 9899. Such studies can ask workers their perception of health risks to address differences between perceived and actual risk. These studies are numerous and form the foundation for most VSL estimates. However, they are problematic for application to health effects of pollution, because of less directly relevant behavioral contexts and/or the populations. In particular, reducing air pollution may lower some health risks disproportionately for older persons who are not in the labor market. These benefits, furthermore, may be more likely for people with chronic heart or lung disease and may have a delayed effect, all of which would not be captured in the labor market studies.

A small literature of consumer preference studies estimates WTP to reduce health risks from purchases or other actual consumer decisions (e.g., purchase of smoke detectors 100, driving behavior under different speed limits 101). These studies typically find lower VSLs than other approaches 101. A difficulty about these studies is statistically separating the health risk-reducing attribute from other valued attributes. A large body of literature applies a hedonic property value approach 102, which provides a revealed WTP for air pollution reductions but is dependent on housing market perceptions about pollution and links to non-health effects.

The stated preference WTP approaches, of which CV and choice experiments are most prominent, are survey methods presenting hypothetical choices (e.g., willingness to pay some amount or prefer one set of attributes over another) to recover preferences for health risk reductions. Results can be sensitive to question wording and ordering, and cognition difficulties when understanding small changes in probabilities are required. However these methods can be molded to a particular population or context. Respondents can be tested for their cognition and understanding of the survey's concepts.

Some of the best known stated preference studies for mortality examine traffic fatalities 103104 and fewer studies are available for air pollution contexts 105106107108. A CV survey found that WTP was higher when death risk reduction takes place now rather than later in life or if the individual was mentally healthy 109. Age had a relatively minor effect on VSL, and physical health status had no effect. These results are consistent with those from a study of adults in the US and Canada, which did not find strong evidence that WTP is lower for older persons or for those with chronic heart or lung conditions or cancer 110. A recent WTP study of three countries also found that VSL is not significantly lower for older populations, however persons admitted to the hospital or emergency room for CVD or respiratory causes had higher VSL 111. The first study to investigate WTP for increased life expectancy (one year in expectation) added between ages 75 to 85 years found implied VSLs to range from $70,000 to $110,000, but did not provide indication of whether respondents understood the complex scenario, and offered respondents an unrealistically large reduction in risk 106.

Two studies applied choice experiments to examine WTP to reduce risks of chronic respiratory disease 112113. Subjects chose between two cities for residence, both preferred to their present city and differing in risk of developing chronic bronchitis or respiratory disease and in one other characteristic: the probability of dying in an automobile accident or cost of living. Several studies evaluated the WTP to reduce cancer morbidity risks 114115.

Three of the first CV studies for acute health responses used bidding procedures to elicit values for respiratory-symptom days, with average estimates from $5 to $25 depending on the symptom, its severity, and whether a complex of symptoms is experienced 116117118. CV techniques have advanced since these studies, however they offer consistent ranges of WTP estimates. In one of the few European studies of this type, over 1,000 Norwegians were interviewed to ascertain WTP to avoid various ac ute health effects (e.g., one more day over their usual annual frequency). The values for avoiding symptoms are slightly smaller than those found in older US studies, but the asthmatic values are far larger 119. A survey of 832 Taiwanese investigated WTP to avoid participants' most-recent episode of acute respiratory illness 120. Statistical techniques are used to relate these values to the duration and severity of the episode and other variables.

Another approach is the averting-behavior method, which infers WTP by observing and placing values on behavior used to avoid adverse health outcomes. For instance, if someone stays indoors with the air conditioner on because of high air pollution, the added electricity costs might relate to WTP to avoid health impacts. Defensible estimates under this approach require stringent assumptions, and in practice the method is rarely used, particularly in an acute-health context.

The QALY approach attempts to account for the quality of life lost by adjusting for time "lost" from disease or death. This method is welfare-theoretic only under very restrictive assumptions, so it is difficult to conceptualize the significance of any particular QALY score. The estimates may be very insensitive for distinguishing among different severities and types of acute morbidity. See the recent Institute of Medicine report 121 for a full review of this approach as it could be applied in a regulatory, cost-benefit analysis setting.

A QALY analysis of USEPA's Heavy Duty Engine/Diesel Fuel regulations found that for situations in which mortality dominates other health outcomes, QALY and WTP methods can provide similar results 122. If morbidity and non-health consequences are predominant, results from QALY and WTP analysis may differ. Another use of QALYs investigated over 230 WTP estimates, finding that variation in WTP values is affected by QALY estimates of illness severity, illness duration, income, and age 123. There also exists literature providing QALY estimates for chronic diseases, for example for various severities of asthma 124.

Valuations of mortality risk reductions associated with environmental policies are usually the largest category of benefits, both among health responses and compared to other attributes. For instance, a USEPA analysis of the Clean Air Act estimated a value of $100 billion annually for reduced premature mortality out of $120 billion in total benefits, compared to costs of approximately $20 billion 7. European and Canadian studies similarly found that mortality risk dominates analysis of pollution reductions 125126. Next to mortality, reductions in the probability of developing a chronic respiratory disease have been estimated to be the most valued, recognizing that values for other types of diseases are sparse. Reductions in acute effects are lower valued.

Table 2 provides a sample of values typically used by practitioners of health benefits analyses from several major studies or models: the USEPA's BenMAP, which is used in Regulatory Impact Analyses of Regulations 754; the ExternE model 125, which is used by the European Union (EU) in its regulatory analyses, taken from its Clean Air For Europe (CAFÉ) Program (AEA) 127; the Air Quality Valuation Model (AQVM) for Canada 126; the Australian Bureau of Transport and Regional Economics (BTRE) assessment of transportation-related pollutants in Australia 128; and a study of the benefits of environmental improvement in New Zealand 129. Within the table, health values are converted to common, comparable currency using purchasing power parity (PPP) and constant 2000 dollars. The WTP for reducing risks of mortality and chronic morbidity is expressed, for convenience, as VSL and the value of a statistical case (VSC) of chronic disease. This term is merely shorthand for the WTP for a given risk reduction divided by that risk reduction. This relationship is useful because VSLs or VSCs can be multiplied by estimates of the "lives saved" or "chronic cases saved" to obtain benefits.

The table shows a fairly wide range of VSL values, with the highest in the US. Rank ordering of values across the other health endpoints is very similar across studies, although some different sets of health endpoints are considered and there are many blank cells (i.e., categories for which information is unknown or not incorporated) outside of the US and EU. The relatively close agreement between the US and EU likely results from reliance on a common pool of studies, results, and interpretations as well as the social cost of electricity studies in the US and the ExternE effort in Europe, which benefited from close collaboration between the participating researchers in both efforts 130131. In addition, the Canadian studies were informed by the AQVM developed by researchers active in the US social costing debate 132.

We evaluated economic valuation methods on three criteria: (i) the degree to which methods are based on preferences for such health improvements, which we took to be in agreement with welfare economics principles; (ii) the number of studies following the technique, which is an imperfect measure of degree of consensus and attractiveness of the technique to researchers; and (iii) additional major limitations, serving to capture other issues, such as data shortcomings. Based on this admittedly subjective judgment, we then rated the reliability of the different approaches from A (very reliable) to D (unreliable). The assessment is intended to provide comparison among approaches, rather than an absolute assessment of accuracy.

As a first step of the evaluation, we compared theoretical predictions and empirical results of economic valuation studies for mortality (Table 3). Under the theoretical framework, WTP should increase with the size of the risk change. The life cycle model also implies lower WTP when risk change is further in time. Persons facing higher baseline risks should have higher WTP for a given risk reduction (the "dead anyway" effect) 133. Higher incomes or wealth should relate to higher WTP. With borrowing against future earnings, the relationship between WTP and age should be an inverted U-shape according to life cycle models. Finally, these models do not make a prediction regarding health status.

These theoretical predictions are not always matched by empirical results, and Table 3 demonstrates that no simple consistent relationship exists between WTP for mortality and other factors listed, other than income. This could be due to differences in the underlying approaches used to solicit results, or indication of a more complicated system (e.g., age's impact on VSL may further depend on other factors). Our subjective evaluation of the valuation methods for mortality, chronic morbidity, and acute morbidity are provided in Tables 4, 5, and 6, respectively. No single method is fully satisfactory. Due to the array of methods available for estimating the economic impact of health and the limitations of any single approach, we recommend the application of multiple methods.

In addition to the issues of credible economic evaluation of the benefits and costs of climate change policies, a central issue in comparing these values is the discount rate applied 134. Selection of the discount rate, which accounts for differential value of costs and benefits occurring in the far future compared to those taking place in the present or near feature, can greatly alter results of cost benefit analysis, such as for climate change. In fact, a recent disagreement regarding climate change policy analysis by two leading economists centered largely on the use of a different discount rate 11134. While some aspects of benefit/cost analysis are well-suited to monetary terms, the issue of an appropriate discount rate carries ethical implications regarding the relative impacts on various populations.