Incorporating environmental equity into risk assessment: A case study of power plant air pollution control strategies Jonathan Levy, Sc.D. Assistant Professor of Environmental Health and Risk Assessment, Harvard School of Public Health The David Bradford Seminars in Science, Technology and Environmental Policy, Princeton University April 17, 2006 Risk assessment – basic definition (NRC, 1983) Risk Assessment Hazard Identification Exposure Assessment Dose-Response Assessment Risk Characterization Environmental justice – basic definitions • A societal goal, defined as the provision of adequate protection from environmental toxicants for all people, regardless of age, ethnicity, gender, health status, social class, or race (Sexton and Anderson, 1993). • The fair treatment and meaningful involvement of all people regardless of race, color, national origin, or income with respect to the development, implementation, and enforcement of laws, regulations, and policies. (U.S. EPA, 1998). Environmental justice Egalitarian Risk assessment/ benefit-cost analysis Utilitarian Process-oriented Outcome-oriented Focused on high-risk subpopulations Concerned with proximity Community-driven Focused on total population Concerned with exposure/risk Analyst-driven PM From Chestnut et al., 2006 Central health estimates (primary + secondary PM, annual) Mortality Hospital admissions Emergency room visits Asthma attacks Brayton Point Current Benefits impacts 79 55 76 53 Salem Harbor Current Benefits impacts 33 23 32 22 1,000 700 420 300 5,300 3,700 2,200 1,600 Levy et al., 2002 Questions asked… • Are populations near the plant “disproportionately” affected by the plant emissions? • Would emission control reduce “environmental injustice”? Benefits from NOx and SO2 controls at Salem and Brayton (mg/m3 of PM2.5, annual avg) 0.16 45 0.15 0.14 44 0.13 0.12 0.11 43 0.1 0.09 0.08 42 0.07 0.06 0.05 41 0.04 0.03 40 -75 0.02 -74 -73 -72 -71 -70 -69 -68 -67 0.01 0 Levy et al., 2002 Analytical challenge • Risk analysts have developed simple, meaningful indicators that can capture the magnitude of the benefits of pollution control from a source or set of sources – QALYs, deaths, hospitalizations, etc. • Is there a simple, meaningful indicator that can capture the distribution of the benefits of pollution control from a source or set of sources, in a way that informs EJ concerns? “Equity” = Distribution of Health Benefits “Efficiency” = Magnitude of Health Benefits Our approach 1. Clarify terminology 2. Develop inequality indicators that are meaningful in a pollution control context 3. Evaluate whether the premise behind our indicators is supported by environmental justice or risk assessment practitioners 4. Apply indicators to a case study of national power plant control strategies to determine information value 5. Extend model to local-scale pollution control decision where small-scale demographics may be influential Key points on terminology • Important communication gaps between risk assessment and environmental justice related in part to loose terminology – EJ: Equality = equal access/participation (process) – RA: Equality = equal outcomes • Moving to equity (or justice) requires determination of those inequalities that are deemed unjust and unfair (avoidable? undeserved? remediable?), which is well beyond domain of quantitative analysis • We focus here on equality of outcomes, considering subpopulations of concern from EJ perspective Levy et al., 2006 Developing indicators • Numerous income inequality studies developed axiomatic approach to select indicators • We modify the standard list of axioms and propose additional axioms relevant to health benefits analysis Standard axioms Analytic tractability Computable in standard applications Appropriateness Reflects values of decision makers Anonymity/impartiality Not dependent on characteristics of affected individuals Increase when income transferred from poor to rich (decrease for transfer from rich to poor) No change for uniform proportional increases Pigou-Dalton transfer principle Scale invariance Normalization Follows defined range Subgroup decomposability Principle of population Segmented into constituent parts (additive separable) Invariance to replication of population Scale invariance • In economics, supported for the case of changing income to different currencies – For risk assessment, parallel argument for concentration measures • For real changes in income/risk, it is less clear – Argument for increased inequality: Absolute gaps have increased, new assets have not been distributed equitably – Argument for decreased inequality: Diminishing marginal utilities of income/risk • We do not require scale invariance in this context (but would not reject a scale-invariant measure) Anonymity • Runs counter to basic premise of environmental justice (concern with sociodemographic factors and comparisons between groups) • Understanding geographic/demographic patterns of health risks may facilitate the development of pollution control strategies • We reject anonymity (and prefer indicators where relevant individual characteristics can be incorporated) Additional axioms (1) • The analyst must not impose a value judgment about the relative importance of transfers at different percentiles of the risk distribution Additional axioms (2) • The welfare measure must be as close to a measure of health risk as possible. If quantifying risk is impossible or there is no differential susceptibility, then exposure should be evaluated. If quantifying exposure is impossible or there is no differential exposure, then concentrations in relevant media should be evaluated. Additional axioms (3) • The inequality indicator should not be applied without consideration of the baseline distribution of risk. Additional axioms (4) • The inequality indicator should be estimated for the geographic scope and resolution that are used for the health benefits analysis, but the sensitivity of the findings to scope and resolution should be evaluated. In particular, an inequality indicator should be estimated with the finest geographic resolution possible, given available data and analytical capabilities. Additional axioms (5) • When efficiency-equality tradeoffs are important for policy decisions, the inequality indicator should be derived for multiple competing policy alternatives. If this is not possible, qualitative interpretations are most appropriate. Some candidate indicators • Gini coefficient • Variance of logarithms • Atkinson index – Note: We evaluated 19 indicators, but present a subset to illustrate key issues The Gini coefficient Cumulative risk 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 Cumulative population 0.8 1 Modifying Gini for pollution control Cumulative risk 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 Cumulative population Original Lorenz curve Post-control Lorenz curve 1 Equation for Gini G 1 2 n x x i i j j 2m Average absolute difference between all pairs of individuals, normalized by dividing by twice the mean Evaluating Gini • Widely used and satisfies many basic criteria, but… – Not subgroup decomposable unless subgroups strictly ordered by income – Most sensitive to transfers in middle of distribution – Structured on rank of incomes rather than absolute value, yielding somewhat arbitrary weights • Gini may not be interpretable in many applications, but could be considered for sensitivity analyses Evaluating variance of logs • Theoretically appealing, esp. for lognormal data • However… – Violates principle of transfers • Marginal transfers from high-risk to low-risk increase variance of logs if high value greater than e times geometric mean of the distribution – Implicitly attaches more weight to transfers at the low end than at the high end of the distribution – Only subgroup decomposable if geometric means replace arithmetic means in subgroup data • Not applicable to health benefits analysis… Atkinson index 1 n xi 1 i 1 x n 1e 1 1e • Member of generalized entropy family (derived specifically to be decomposable) • Fulfills transfer principle • Societal preferences about inequality incorporated through e – Higher e = more weight on transfers at low end Conclusions about indicators • Atkinson index, if applied appropriately, best addresses needs of inequality assessment in health benefits analysis • Could be supplemented by other indices for sensitivity analyses: – Gini: Alternate viewpoint about inequality (comparisons to all those better off vs. average, additive vs. weighted additive formulation) – Theil: Alternate statistical formulations from generalized entropy family Power plant case study • What would happen if we used cap-andtrade programs to reduce emissions from power plants nationally, rather than mandatory controls for all plants? – Would this result in an “environmental injustice”? – What do optimal reductions given a national emissions cap look like, considering efficiency and equity? EPA Announces Landmark Clean Air Interstate Rule (March 10, 2005) “CAIR will permanently cap emissions of sulfur dioxide (SO2) and nitrogen oxides (NOx) in the eastern United States… Under CAIR, states will achieve the required emissions reductions using one of two options for compliance: 1) require power plants to participate in an EPA-administered interstate cap and trade system that caps emissions in two stages, or 2) meet an individual state air emission limits through measures of the state's choosing.” Criticism… • The data released by environmentalists this week show about 50 percent of the 1,041 coal-fired electric generating units expected to be in operation in 2020 would not be equipped with the best pollution equipment on the market to reduce sulfur dioxide and nitrogen oxide emissions [under CAIR] • The distribution of emission controls has become a controversial topic for the Bush administration since it linked its air pollution policies to a market-friendly, cap-and-trade system. EPA officials maintain the administration's approach is the most cost-effective for the electric utility industry while guaranteeing the decline of air pollution because of obligatory emission caps on power plants. Greenwire February 24, 2006 Extracted from EGRID, NEI for 425 plants PM S-R matrix, county resolution C-R function from ACS, county mortality Model validation for seven power plants in GA (Levy et al., 2003) S-R/ CALPUFF, primary PM S-R/ CALPUFF, SO2/sulfate S-R/ CALPUFF, NOx/nitrate Bowen 0.8 1.1 0.4 Hammond 0.9 1.1 0.4 Harllee Branch 0.9 1.1 0.4 Jack McDonough 0.6 1.0 0.4 Scherer 0.9 1.0 0.4 Wansley 0.9 1.2 0.4 Yates 0.9 1.1 0.4 Emission reductions • Developed logical (or illogical) approaches by which 75% reductions could be achieved, to span efficiency/equity space – 75% reductions from all plants – Reductions to meet target emission rates in lb/MMBTU (for those above target) – Elimination of plants with highest/lowest health benefit per unit emissions of SO2/NOx/primary PM – Elimination of plants in counties with highest/lowest background PM2.5 concentrations – Elimination of highest/lowest emitters of SO2/NOx/primary PM • Supplemented with random emission control scenarios How do we capture equity? • Given policy context/nature of debate, primarily concerned about spatial equity • Multiple ways we might incorporate “baseline” (Axiom 1) – For concentrations: Total PM2.5, power plantrelated PM2.5 – For health risk: Total mortality, PM2.5-related mortality, power plant PM2.5-related mortality • Multiple inequality indicators • Consideration of concentrations vs. health risks 0.002 Equity benefits (decrease in inequality indicator) 0.0018 0.0016 Indicator: Atkinson index, e = 0.25 Outcome: Mortality Baseline: PM-related mortality High background PM High SO2 iF 0.0014 0.0012 High PM iF Reduction to target/MMBTU High SO2 emissions High NOx emissions High PM emissions 0.001 75% reduction for all Low PM iF Low NOx iF 0.0008 Low NOx emissions 0.0006 High NOx iF Low SO2 iF 0.0004 0.0002 0 10000 Low SO2 emissions Low PM emissions Low background PM 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 Sulfate/SO2 iF Reductions in concentrations (High SO2 iF) Baseline concentrations Post-control baseline concentrations (High SO2 iF) Does the choice of indicator matter? 0.002 Equity benefits (decrease in inequality indicator) 0.0018 0.0016 Indicator: Atkinson index, e = 0.25 Outcome: Mortality Baseline: PM-related mortality High background PM High SO2 iF 0.0014 0.0012 High PM iF Reduction to target/MMBTU High SO2 emissions High NOx emissions High PM emissions 0.001 75% reduction for all Low PM iF Low NOx iF 0.0008 Low NOx emissions 0.0006 High NOx iF Low SO2 iF 0.0004 0.0002 0 10000 Low SO2 emissions Low PM emissions Low background PM 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.007 Equity benefits (decrease in inequality indicator) 0.006 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: PM-related mortality High background PM 0.005 High SO2 iF High PM iF 0.004 High SO2 emissions Reduction to target/MMBTU High PM emissions High NOx emissions 75% reduction for all Low NOx iF Low PM iF Low NOx emissions 0.003 0.002 High NOx iF Low SO2 iF Low SO2 emissions Low PM emissions 0.001 Low background PM 0 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.045 Equity benefits (decrease in inequality indicator) 0.04 0.035 Indicator: Atkinson index, e = 3.0 Outcome: Mortality Baseline: PM-related mortality High background PM High PM iF 0.03 High SO2 iF 0.025 Low NOx iF High SO2 emissions Reduction to target/MMBTU High PM emissions 0.02 75% reduction for all High NOx emissions 0.015 Low SO2 iF Low NOx emissions 0.01 Low PM iF 0.005 Low SO2 emissionsHigh NOx iF Low PM emissions Low background PM 0 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.007 High background PM Equity benefits (decrease in inequality indicator) 0.006 Indicator: Theil Outcome: Mortality Baseline: PM-related mortality High SO2 iF 0.005 High PM iF Reduction to target/MMBTU High SO2 emissions High NOx emissions High PM emissions 0.004 75% reduction for all Low PM iF 0.003 Low NOx iF Low NOx emissions High NOx iF 0.002 Low SO2 iF Low SO2 emissions Low PM emissions 0.001 Low background PM 0 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.014 Equity benefits (decrease in inequality indicator) 0.012 Indicator: Gini Outcome: Mortality Baseline: PM-related mortality High background PM High SO2 iF 0.01 0.008 Reduction to target/MMBTU High SO2 emissions High NOx emissions High PM emissions High PM iF 75% reduction for all Low PM iF 0.006 Low NOx iF Low NOx emissions High NOx iF 0.004 Low SO2 iF Low SO2 emissions Low PM emissions 0.002 Low background PM 0 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.01 Equity benefits (decrease in inequality indicator) 0.009 0.008 Indicator: MLD Outcome: Mortality Baseline: PM-related mortality High background PM High SO2 iF 0.007 High PM iF 0.006 High SO2 emissions Reduction to target/MMBTU High PM emissions High NOx emissions 0.005 75% reduction for all Low NOx iF 0.004 Low PM iF 0.003 Low SO2 iF High NOx iF Low SO2 emissions Low PM emissions 0.002 Low background PM 0.001 0 10000 Low NOx emissions 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 Does the choice of baseline matter (and do we even need to consider the baseline)? 0.007 Equity benefits (decrease in inequality indicator) 0.006 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: PM-related mortality High background PM 0.005 High SO2 iF High PM iF 0.004 High SO2 emissions Reduction to target/MMBTU High PM emissions High NOx emissions 75% reduction for all Low NOx iF Low PM iF Low NOx emissions 0.003 0.002 High NOx iF Low SO2 iF Low SO2 emissions Low PM emissions 0.001 Low background PM 0 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.00024 Equity benefits (decrease in inequality indicator) 0.00023 0.00022 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: All-cause mortality High background PM 0.00021 High SO2 iF High PM iF High SO2 emissions Reduction to target/MMBTU 0.0002 High PM emissions High NOx iF NOx emissions LowHigh PM iF 75% reduction all emissions Low for NOx Low NOx iF 0.00019 0.00018 Low SO2 iF Low PM emissions Low SO2 emissions Low background PM 0.00017 0.00016 0.00015 0.00014 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.2 0.1 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: Power plant PM-related mortality High background PM High SO2 iF Reduction to target/MMBTU High SO2 emissions Equity benefits (decrease in inequality indicator) High PM iF High PM emissions 75% reduction for all High NOx emissions Low NOx iF Low NOx emissions 0 Low SO2 iF -0.1 10000 -0.2 -0.3 -0.4 Low PM iF 11000 12000 13000 Low SO2 emissions Low PM emissions Low background PM High NOx iF -0.5 Health benefits (Deaths/year) 14000 15000 Health benefits (Deaths/year) 10000 0.1 0.15 Inequality indicator 0.2 0.25 11000 12000 13000 14000 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: None Low background PM Low Low PM SO2emissions emissions Low PM iF Low SO2 iF High LowNOx NOxiFemissions 75% reduction for all High NOx emissions 0.3 0.35 High PM emissions Reduction to target/MMBTU High SO2 emissions Low NOx iF High SO2 iF High PM iF 0.4 0.45 High background PM 0.5 15000 Reductions in concentrations Does it matter whether we use mortality or concentrations in the equity calculation? 0.007 Equity benefits (decrease in inequality indicator) 0.006 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: PM-related mortality High background PM 0.005 High SO2 iF High PM iF 0.004 High SO2 emissions Reduction to target/MMBTU High PM emissions High NOx emissions 75% reduction for all Low NOx iF Low PM iF Low NOx emissions 0.003 0.002 High NOx iF Low SO2 iF Low SO2 emissions Low PM emissions 0.001 Low background PM 0 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0 Equity benefits (decrease in inequality indicator) -0.0005 -0.001 High background PM Indicator: Atkinson index, e = 0.75 Outcome: Concentrations Baseline: Background PM2.5 High SO2 iF -0.0015 Reduction to target/MMBTU High NOx emissions High SO2 emissions High PM iF High PM emissions 75% reduction for all -0.002 Low NOx iF Low PM iF -0.0025 Low NOx emissions Low SO2 iF Low SO2 emissions Low PM emissions High NOx iF -0.003 -0.0035 Low background PM -0.004 -0.0045 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 What happens if we optimize on subsets of pollutants (or single pollutants)? 0.007 Equity benefits (decrease in inequality indicator) 0.006 0.005 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: PM-related mortality Pollutant: PM2.5, NOx, SO2 High background PM High SO2 iF High PM iF 0.004 High SO2 emissions Reduction to target/MMBTU High PM emissions High NOx emissions 75% reduction for all Low NOx iF Low PM iF Low NOx emissions 0.003 0.002 High NOx iF Low SO2 iF Low SO2 emissions Low PM emissions 0.001 Low background PM 0 10000 11000 12000 13000 Health benefits (Deaths/year) 14000 15000 0.006 Equity benefits (decrease in inequality indicator) 0.005 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: PM-related mortality Pollutant: SO2 only High background PM High SO2 iF High PM iF 0.004 High SO2 emissions Reduction to target/MMBTU High PM emissions High NOx emissions Low NOx iF 75% reduction for all 0.003 Low PM iF 0.002 Low SO2 iF Low NOx emissions High NOx iF Low LowSO2 PM emissions Low background PM 0.001 0 8000 9000 10000 Health benefits (deaths/year) 11000 12000 Equity benefits (decrease in inequality indicator) 0.0008 0.0006 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: PM-related mortality Pollutant: NOx only High background PM High PM iF High SO2 iF 0.0004 Low NOx iF High SO2 emissions Reduction to target/MMBTU High PM emissions for all High75% NOxreduction emissions 0.0002 Low NOx emissions Low SO2 iF Low PM emissions Low PM iF SO2 Low emissions High NOx iF Low background PM 0 500 600 700 800 Health benefits (deaths/year) 900 1000 0.0016 Equity benefits (decrease in inequality indicator) 0.0014 0.0012 Indicator: Atkinson index, e = 0.75 Outcome: Mortality Baseline: PM-related mortality Pollutant: PM only High background PM High SO2 iF 0.001 High PM iF High SO2 emissions 0.0008 Low NOx iF Reduction to target/MMBTU High PM emissions High NOx emissions 75% reduction for all 0.0006 Low NOx emissions Low PM iF High NOx iF SO2 iF Low 0.0004 Low PM emissions Low SO2 emissions 0.0002 0 1400 Low background PM 1500 1600 1700 1800 1900 Health benefits (deaths/year) 2000 2100 2200 What can we conclude? (I) • The emissions reductions are substantial enough that the efficiency differences among scenarios are not huge in relative terms (but may be important in absolute terms) • For power plants and PM, strong concordance between the more efficient and more equitable strategies, implying limited tradeoffs What can we conclude? (II) • Our conclusions are robust across numerous indicators and formulations, but clearly show the importance of properly accounting for baseline/background conditions What’s missing? • Linkage with economics of power plant control – Plausibility of control options, economic efficiency/equity considerations • Equity other than spatial equity – No subgroup decomposability • Important effect modifiers (no variability in CR), differential susceptibility • Consideration of local (vs. national) perspective Concluding thoughts • Axiomatic approach allowed us to develop inequality indicators that are meaningful and can be used in parallel with standard indicators of efficiency • Power plant case demonstrates the viability of the approach and the potential for formally optimizing on efficiency and equity • Ongoing analyses with smaller spatial scales, socioeconomic equity concerns, effect modification/differential baseline disease rates will allow for further refinement Acknowledgments • • • • • • • Susan Chemerynski Jessica Tuchmann Julia Forgie Sue Greco Andrew Wilson Len Zwack National Science Foundation (SES-0324746) References • • • Levy JI, Spengler JD. Modeling the benefits of power plant emission controls in Massachusetts. J Air Waste Manage Assoc 52: 5-18 (2002). Levy JI, Wilson AM, Evans JS, Spengler JD. Estimation of primary and secondary particulate matter intake fractions for power plants in Georgia. Environ Sci Technol 37: 5528-5536 (2003). Levy JI, Chemerynski SM, Tuchmann JT. Incorporating concepts of inequality and inequity into health benefits analysis. International Journal for Equity in Health 5:2 (2006).