Air pollution requires multipollutant analysis: the case of Santiago, Chile.

In many ways, air pollution control presents itself as a textbook example for cost-benefit analysis. Air quality certainly is a public good in the sense that it can be enjoyed by one without excluding the other, so there certainly is a role for government in weighing trade-offs and guiding resource

Analytical Approach: The Compensation Criterion and the Samuelson Condition

In the public finance tradition, we simplify the analysis by invoking the "compensation criterion": Assuming that costless transfers are available to the planner, there is no need to let the consideration of how control costs are distributed influence the control strategy. The optimal pollution control program then minimizes the costs of a given air quality target and will furthermore be characterized by

(1) [Mathematical Expression Omitted]

for all consumers, g, h, and all goods, j, k.(1)

In the numerator on the right-hand side is the marginal cost of pollution abatement for consumer g in sector j per unit of emission reduction. The denominator is the marginal cost of good k. Thus, the right-hand-side expression is the marginal rate of transformation between emission reductions and production of good k. For example, if j = k = private car transport, then the fraction expresses the rate at which resources can be transferred from car services to reduced emissions, as when manpower is transferred from car production to producing catalytic converters. Optimum also requires that right-hand side be uniform across consumers g and sectors j, k, so that the marginal costs of emission reductions are equalized across the economy. On the left-hand side is the marginal rate of substitution between the public good - pollution reductions - and private goods, aggregated across consumers. The optimality condition (1) is the Samuelson condition for optimal supply of pure public goods: The sum across consumers in willingness to pay for public goods shall equal the marginal cost of provision. This condition is not incentive compatible without explicit coordination. We can think of a city government implementing such coordination and of our analysis as an input into its policy process.

In this analysis, a limited set of four discretely described pollution abatement strategies is considered. Thus, optimality is described by the subset of these strategies maximizing net benefits rather than the continuous formulation in condition (1).

The Air Pollution Problem in Santiago

Santiago has severe air pollution problems. The city has grown to a population of 4.8 million and is located between mountain ranges with conditions contributing to thermal inversions, especially in the winter months of May through August (see table 1; see also Ostro et al., World [TABULAR DATA FOR TABLE 1 OMITTED] Bank). Pollution concentrations typically exceed both local standards and World Health Organization (WHO) guidelines, frequently by as much as 100%.

Module 1: Emission Inventory and the Control Scenario

Although this paper will concentrate on the health benefits of air pollution control in Santiago, the study context is one of four modules: (a) emission inventory and control strategy, (b) dispersion and exposure model, (c) health effects estimation and valuation, and (d) conclusion. The modules serve the purpose that they can be replaced individually as the analytic basis improves or as policy scenarios change.

With a base case established by the updated emission inventory, a "package" of control measures was devised with the aim of quantifying its implications for annual emissions of the main air pollutants: small particles ([PM.sub.10]), ozone, N[O.sub.x], and S[O.sub.2]. Lead was excluded from the analysis because a phase-out was in process anyway, and carbon monoxide was excluded because health effects quantification is not yet feasible (Ostro 1993). The perspective was supposed to be strictly local, so no global implications were considered. Important criteria for the control measures were that they should have an important impact on emissions, that the technical basis for assessing their costs and effects on emissions should be reasonably well developed in advance, and that they should not be seen as a priori clearly rejected or attractive. From a somewhat longer list of strategies, the following set of mutually and separately feasible actions were selected as the control scenario: (a) emission standards for gasoline vehicles, (b) emission standards for trucks, (c) natural gas for buses, and (d) conversion of wood-burning industrial sources to distillate fuel oil (Weaver et al., World Bank).

Module 2: Dispersion and Exposure Model

A dispersion model of the Santiago airshed had been developed previously for other purposes (Ulriksen, Ferdandez, and Munoz) and was used to estimate ambient pollutant concentrations across the city. The model has the following form:

(2) [Mathematical Expression Omitted]

where [M.sub.ij] is the mass of pollution in cell (i, j), [Q.sub.ij] is the rate of emissions of the pollutant in the cell, and [Mathematical Expression Omitted] is the net flow of pollution from the cell through its P walls. The model simulates meteorological conditions and the functioning of the airshed. For our purposes, its key service is to estimate the concentration reductions and exposure reductions, paying attention to the location of people, the location of emissions, and atmospheric conditions.

[TABULAR DATA FOR TABLE 3 OMITTED]

Table 1 shows the results from the dispersion and exposure model for the control scenario. The ambient air pollution concentrations are averages across the city and are weighted by population densities. Apart from the total reductions in air pollution concentrations, the table also shows improvement above the assumed pollution standard, which is the part of the air quality improvement that occurs in locations in noncompliance with the ambient standard.

Module 3: Estimation and Valuation of Health Effects

In peer-reviewed scientific literature, dose response functions have now been estimated for linking health problems to several pollutant measures (e.g., ozone, N[O.sub.x], and various measures of particles such as [PM.sub.10]). However, although the studies cover a number of health end points (e.g., premature mortality, respiratory hospital admissions, asthma attacks, eye irritation, and chronic bronchitis) and have used different data sets and methodologies, they can hardly be claimed to give comprehensive coverage. The literature was started with studies of mortality during smog periods in London winters, and studies from cities of industrialized countries still comprise the bulk of the literature. "Borrowing" dose response functions and adjusting results is therefore necessary for cost-benefit analysis.

Transferring Dose Response Functions from the Literature

In this study, dose response functions are based on a literature review (Ostro 1994) that also provides a detailed discussion of the approach. Because a dose response function from the literature will depend on the particular conditions of the study in question (e.g., population characteristics, climate, and period), one should be aware that there are many potential sources of uncertainty. In a conference at the World Bank on this issue, many important questions were raised (see Falk et al., McMichael et al.), including curvature in functional form and the need to correct for differences in pollution levels and population characteristics. Also, important problems relate to how effects measured in one time perspective are converted to results applied to annual averages, as was done in this study.

With particular reference to estimates for premature deaths, the use of studies based on short-term time-series variation (which dominate the literature) raises two types of problems. The first problem is that deaths may be measured as premature, even though some of them may be premature only by a few days or weeks. This, the so-called harvesting effect, would indicate that estimated effects are biased upward if longer prematurity is assumed. The second problem is that such studies may fail to capture chronic effects (e.g., carcinogenic effects) so as to underestimate total effects on mortality (see McMichael et al., Ostro 1996).

Estimation of Dose Response from Santiago Time-Series Data

For premature mortality, we conducted separate research to estimate dose response relationships for Santiago (Ostro et al.). Table 2 displays results showing that dose response relationships in Santiago are much like those found in other short-term time-series studies in the literature (notice that one important adjustment, for age composition, is partly made when coefficients are reported as percentage change in mortality).

Part B of table 1 shows the dose response relationships applied (including the one estimated for Santiago) and the estimated health effects for the four ambient air pollutants when implementing the control scenario. For [PM.sub.10], the estimate of 0.47 premature deaths per microgram/[m.sup.3] (per hundred thousand citizens) is found by multiplying a 0.11% increase (table 2) with the average daily mortality (total, except accidents) in Santiago of 55 and then convening to annual and per hundred thousand. To combine all morbidity effects into one category, we have used work-day equivalents as a common denominator. This involved using directly the days (or hours) lost to illness and the average daily wage ($9.55) to convert treatment costs into work-day equivalents (see Sanchez). We used the same methodology for premature mortality (choosing a human capital approach rather than willingness to pay because of limited analytical resources; see below)(2). These work-day equivalents would provide suitable units of measurement for cost-effectiveness analysis, leaving freedom of interpretation for their value in general as well as for the relative value of illness versus premature mortality (see World Bank).

Using the average wage to value these losses, total health benefits for the control scenario are $103 million per year, of which 67% is due to [PM.sub.10] reductions (part C of table 1). Only 8% of total benefits are due to mortality, a fraction that would likely be higher if value estimation were by willingness to pay, as the downward bias in the human capital approach is likely greater for mortality than for morbidity.

Module 4: Discussion and Comparison of Costs and Benefits

Part C of table 1 shows the central estimates for the benefits of the control strategy, assigned to the different emitted pollutants. The resulting values, in benefits per ton, are useful as reference points when these or other strategies are subjected to evaluation.

In table 3, benefits and costs are compared by program component, taking account of the "joint product" of each strategy. The fact that benefits and costs can be compared first at this stage is a good demonstration of the need to conduct analysis in a multipollutant framework: Because control strategies invariably will produce emission reductions of various pollutants as joint products, not even cost-effectiveness analysis can be conducted without a value-based priority between emitted pollutants. This in turn must be based on a value-based comparison of morbidity and mortality effects, as demonstrated here. The table demonstrates that the central estimates for benefits - even when incorporating only modestly assessed health benefits - exceed costs for each individual component strategy in the control scenario.

Table 3. Evaluation of Program Components

                        Benefits     Costs
                        (million    (million     B/C
                         US $)       US $)      Ratio

Fixed sources             26          11         2.4
Gasoline vehicles         33          14         2.4
Buses                     37          30         1.2
Trucks                     8           4         1.8

Control strategy         103          60         1.7

We also report briefly on sensitivity analysis for the quantity part of benefit estimation (part C of table 1). The reported figures in the table represent the extreme case produced for morbidity and mortality, respectively, for each pollutant in percentage adjustments. To produce high-case estimates, we did the following variations. First, we removed the assumption that health benefits accrue only to air quality improvements above a certain threshold (table 1). This increased the benefits of S[O.sub.2] reductions in particular because so many of those improvements were in areas of compliance. It had only a minor effect on total benefits (only 4%), as the ozone benefits were only marginally affected, and [PM.sub.10] benefits were not affected at all, as all improvements were in noncompliance areas. It should be noted that there is no true scientific base for assuming a threshold effect (see Ostro 1994, p. 6), so these higher benefit adjustments are indeed quite plausible. Another high-case scenario was produced by using the higher end of the confidence intervals (95%) in the original studies or in some cases by using coefficients from alternative studies. When these intervals have been combined for various morbidity effects, a value-weighted average has been applied. Finally, a high case reflects an effort to include - for [PM.sub.10] - the chronic effects that likely are not captured in the shortterm time-series studies. The 100% upward adjustment for premature mortality of [PM.sub.10] is inspired by the coefficient of a long-term prospective study in the United States (Pope et al.). The risk ratio of that study translates to a 4.5% increase in total mortality per 10 micrograms [PM.sub.10] (i.e., four times the 1.1% in our short-term Santiago analysis; table 2). We have subjectively adjusted it to a 2.2% increase in mortality per 10 micrograms [PM.sub.10]. This is based on a view stated by one of the authors of that study (Arden Pope, at the previously mentioned World Bank workshop, and personal communication) that a substantial adjustment may be needed to make it quantitatively comparable because of differences in pollution measures. A low case was constructed by using the lower end of the confidence intervals.

A brief summary of this sensitivity analysis is that the low case is sufficiently low to make the control strategy marginally attractive, even though two subcomponents will remain very unattractive. In sum, however, the control strategy appears to be an attractive proposal, yielding more than it costs in mere health benefits.

Importantly, it would have been conceptually more correct to base the valuation of these health effects on willingness-to-pay estimates. Likely, this would have given higher health benefits, and the need to go in this direction is therefore stronger as one turns toward additional, more expensive options providing pollution reductions (for a review of such estimates, see Viscusi).

It may be interesting to contrast our findings with recent findings from the United States that some environmental and other regulations now come at a price, making them highly questionable on the basis of cost-benefit analysis (see Arrow et al., Krupnick and Portney). The Krupnick and Portney study is directly relevant because it deals with health effects of ozone reductions. They concluded that a $10 billion program for U.S. cities in noncompliance would yield a mere $1 billion in health benefits.

When compared to U.S. cities, Santiago is characterized not only by lower personal incomes (and thus lower valuation of health benefits) but also by a smaller, "tighter" airshed (leading to high concentration reductions for each ton reduced) and by cheap control options that have not yet been exhausted. It appears that the cheaper controls and the greater response in the airshed are significant enough to make pollution control pay, even with modestly assessed benefits in a poorer economy.

The findings, interpretations, and conclusions expressed in this paper are those of the author and do not necessarily represent views of the World Bank, its executive directors, or the countries they represent. The author acknowledges study contributions (see references) from P. Ulriksen (emission inventory and dispersion model), C. Weaver (emission inventory and control strategy), B. Ostro. J. Sanchez, and C. Aranda (health effect estimation and valuation) and thanks J. Dixon and J. von Amsberg, colleagues at the World Bank, and Juan Escudero of Santiago's Pollution Control Agency.

1 See Eskeland for more details on this framework: u, utility, is defined over consumption, x, and damages, d; the function d translates emissions, e, into concentrations and health effects; e defines emissions as function of consumption, x, and pollution abatement, a; c(a, x) = F describes production feasibility.

2 For premature mortality, the procedure involved making assumptions about prematurity (see Sanchez). On average, air-pollution related deaths (189 in the central scenario) were premature by 12.9 net present years, or 45,000 USD.

References

Arrow, K. J., M. L. Cropper, G. C. Eads, et al. "Is There a Role for Benefit-Cost Analysis in Environmental, Health, and Safety Regulation?" Science 272 (12 April 1996).

Dockery, D. W., C. A. Pope, and Xiping Xu. "An Association between Air Pollution and Mortality in Six U.S. Cities." New England J Medicine 329 (December 1993): 1753-59.

Eskeland, G. S. "A Presumptive Pigovian Tax: Complementing Regulation to Mimic and Emissions Fee." World Bank Econ. Rev. 8 (September 1994): 373-94.

Falk, H., D. Flanders, A. Haddix, et al. "Review of Urban Air Pollution Health Impact Methodology." Paper prepared for workshop at World Bank, Centers for Disease Control and Prevention, Atlanta GA, 1995.

Krupnick, A. J., and P. R. Portney. "Controlling Urban Air Pollution: A Benefit-Cost Assessment." Science 252 (1991): 522-28.

McMichael, A. J., et al. "Review of Methods Proposed, and Used, for Estimating the Population Health Risks of Exposure to Urban Air Pollution." Report prepared for World Bank, Latin American Department, Washington DC, 1995.

Ostro, B. "Estimating the Health and Economic Effects of Air Pollution Control Strategies in Santiago, Chile." Mimeograph, background paper for World Bank, Washington DC, 1993.

-----. "Estimating the Health Effects of Air Pollution: A Methodology with an Application to Jakarta." Policy Research Working Paper 1301, World Bank, Washington DC, 1994.

-----. "A Methodology for Estimating Air Pollution Health Effects." Paper presentation. World Health Organization, Geneva, April 1996.

Ostro, B., J. M. Sanchez, C. Aranda, and G. S. Eskeland. "Air Pollution and Mortality: Results from a Study of Santiago, Chile." J. Exposure Analysis and Environ. Epidemiology 6(1996): 97-114. Also published as Policy Research Working Paper 1453, World Bank, Washington DC.

Pope, C. A., III, M. J. Than, N. M. Namboodiri, et al. "Particulate Air Pollution as a Predictor of Mortality in a Prospective Study of U.S. Adults." Amer. J. Respiratory. and Critical Care Medicine 151 (March 1995): 669-74.

Sanchez, J. M. "Unit Cost Estimates of Health Outcomes Associated with Atmospheric Pollution in Santiago." Background paper for World Bank, mimeograph, Washington DC, 1994.

Ulriksen, P., M. Ferdandez, and R. Munoz. "Simulacion de los efectos de estrategias de control de emisiones sobre las concentraciones de contaminantes atmosfericos en Santiago, mediante un modelo simple de dispersion de contaminates." Mimeograph, World Bank, Washington DC, 1994.

Viscusi, W. K. "The Value of Risks to Life and Health." J. Econ. Literature 31(December 1993): 1912-46.

Weaver, C., and M. Turner. "Cost and Emissions Benefits of Selected Air Pollution Control Measures for Santiago Chile." Background paper for the World Bank, mimeograph, Washington DC, 1993.

World Bank. "Chile: Managing Environmental Problems: Economic Analysis of Selected Issues," chap. 3: The net benefits of an air pollution control scenario for Santiago. Report No. 13061-CH, World Bank, Washington DC, 1994.

Gunnar S. Eskeland is with the Policy Research department of the World Bank.