Incorporating environmental equity into risk assessment: A case study of power plant air pollution control strategies Jonathan Levy, Sc.D. Assistant Professor.

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)

RiskAssessment

HazardIdentification

ExposureAssessment

Dose-ResponseAssessment

RiskCharacterization

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 Risk assessment/

benefit-cost analysis

Egalitarian Utilitarian

Process-oriented Outcome-oriented

Focused on high-risk subpopulations

Focused on total population

Concerned with proximity

Concerned with exposure/risk

Community-driven Analyst-driven

From Chestnut et al., 2006

PM

Central health estimates (primary + secondary PM, annual)

Brayton Point Salem Harbor Current

impacts Benefits Current

impacts Benefits

Mortality 79 55 33 23 Hospital admissions

76 53 32 22

Emergency room visits

1,000 700 420 300

Asthma attacks 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 (g/m3 of PM2.5, annual avg)

-75 -74 -73 -72 -71 -70 -69 -68 -6740

41

42

43

44

45

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

0.11

0.12

0.13

0.14

0.15

0.16

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?

“Efficiency” = Magnitude of Health Benefits

“Equity” = Distribution of Health Benefits

Our approach

1. Clarify terminology2. Develop inequality indicators that are

meaningful in a pollution control context3. 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 axiomsAnalytic tractability Computable in standard applications

Appropriateness Reflects values of decision makers

Anonymity/impartiality Not dependent on characteristics of affected individuals

Pigou-Dalton transfer principle

Increase when income transferred from poor to rich (decrease for transfer from rich to poor)

Scale invariance No change for uniform proportional increases

Normalization Follows defined range

Subgroup decomposability

Segmented into constituent parts (additive separable)

Principle of population 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


• 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.


• The inequality indicator should not be applied without consideration of the baseline distribution of risk.


• 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.


• 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

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Cumulative population

Cu

mu

lati

ve

ris

k

Modifying Gini for pollution control

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Cumulative population

Cu

mu

lati

ve

ris

k

Original Lorenz curve Post-control Lorenz curve

Equation for Gini

2

12

i jji xx

nG

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

• Member of generalized entropy family (derived specifically to be decomposable)• Fulfills transfer principle• Societal preferences about inequality incorporated through

– Higher = more weight on transfers at low end

1

1

1

11

1n

i

i

x

x

n

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-and-trade 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.

GreenwireFebruary 24, 2006

PM

Extracted from EGRID, NEI for 425 plants

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 plant-

related 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

Reduction to target/MMBTU

Low SO2 iF

Low PM iFLow NOx iF

Low SO2 emissionsLow PM emissions

Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions

High PM emissionsHigh NOx emissions

High background PM

75% reduction for all

0

0.0002

0.0004

0.0006

0.0008

0.001

0.0012

0.0014

0.0016

0.0018

0.002

10000 11000 12000 13000 14000 15000

Health benefits (Deaths/year)

Eq

uit

y b

ene

fits

(d

ecr

ease

in i

ne

qu

alit

y in

dic

ato

r)

Indicator: Atkinson index, = 0.25Outcome: MortalityBaseline: PM-related mortality

Sulfate/SO2 iF

Reductions in concentrations (High SO2 iF)

Baseline concentrations

Post-control baseline concentrations (High SO2 iF)

Does the choice of indicator matter?


Low SO2 iF

Low PM iFLow NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions


High background PM


0

0.0002

0.0004

0.0006

0.0008

0.001

0.0012

0.0014

0.0016

0.0018

0.002

10000 11000 12000 13000 14000 15000


Eq

uit

y b

ene

fits

(d

ecr

ease

in i

ne

qu

alit

y in

dic

ato

r)



Low SO2 iF

Low PM iFLow NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions


High background PM


0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)



Low SO2 iF

Low PM iF

Low NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions

High PM emissions

High NOx emissions

High background PM


0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 3.0 Outcome: MortalityBaseline: PM-related mortality


Low SO2 iF

Low PM iF

Low NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions


High background PM


0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Theil Outcome: MortalityBaseline: PM-related mortality


Low SO2 iF

Low PM iF

Low NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissionsHigh PM emissions

High NOx emissions

High background PM


0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Gini Outcome: MortalityBaseline: PM-related mortality


Low SO2 iF

Low PM iF

Low NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions


High background PM


0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

0.01

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: MLD Outcome: MortalityBaseline: PM-related mortality

Does the choice of baseline matter (and do we even need to

consider the baseline)?


Low SO2 iF

Low PM iFLow NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions


High background PM


0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)



Low SO2 iF

Low PM iFLow NOx iF


Low NOx emissions

Low background PM

High SO2 iFHigh PM iF

High NOx iF

High SO2 emissions

High PM emissions

High NOx emissions

High background PM


0.00014

0.00015

0.00016

0.00017

0.00018

0.00019

0.0002

0.00021

0.00022

0.00023

0.00024

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 0.75Outcome: MortalityBaseline: All-cause mortality


Low SO2 iF

Low PM iF

Low NOx iF

Low SO2 emissions

Low PM emissions

Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions

High PM emissions

High NOx emissions

High background PM


-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 0.75Outcome: MortalityBaseline: Power plant

PM-related mortality


Low SO2 iF

Low PM iF

Low NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions

High PM emissions

High NOx emissions

High background PM


0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

10000 11000 12000 13000 14000 15000


Ineq

ual

ity

ind

icat

or

Indicator: Atkinson index, = 0.75Outcome: MortalityBaseline: None

Reductions in concentrations

Does it matter whether we use mortality or concentrations in the

equity calculation?


Low SO2 iF

Low PM iFLow NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions


High background PM


0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)



Low SO2 iF

Low PM iFLow NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions

High PM emissions

High NOx emissions

High background PM


-0.0045

-0.004

-0.0035

-0.003

-0.0025

-0.002

-0.0015

-0.001

-0.0005

0

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 0.75Outcome: ConcentrationsBaseline: Background PM2.5

What happens if we optimize on subsets of pollutants (or single

pollutants)?


Low SO2 iF

Low PM iFLow NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions


High background PM


0

0.001

0.002

0.003

0.004

0.005

0.006

0.007

10000 11000 12000 13000 14000 15000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 0.75Outcome: MortalityBaseline: PM-related mortalityPollutant: PM2.5, NOx, SO2


Low SO2 iF

Low PM iF

Low NOx iF


Low NOx emissions

Low background PM

High SO2 iF

High PM iF

High NOx iF

High SO2 emissions

High PM emissions

High NOx emissions

High background PM


0

0.001

0.002

0.003

0.004

0.005

0.006

8000 9000 10000 11000 12000

Health benefits (deaths/year)

Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 0.75Outcome: MortalityBaseline: PM-related mortalityPollutant: SO2 only


Low SO2 iF Low PM iF

Low NOx iF


Low NOx emissions

Low background PM


High NOx iF

High SO2 emissionsHigh PM emissions

High NOx emissions

High background PM


0

0.0002

0.0004

0.0006

0.0008

500 600 700 800 900 1000


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 0.75Outcome: MortalityBaseline: PM-related mortalityPollutant: NOx only


Low SO2 iFLow PM iF

Low NOx iF


Low NOx emissions

Low background PM


High NOx iF

High SO2 emissions


High background PM


0

0.0002

0.0004

0.0006

0.0008

0.001

0.0012

0.0014

0.0016

1400 1500 1600 1700 1800 1900 2000 2100 2200


Eq

uit

y b

enef

its

(dec

reas

e in

ineq

ual

ity

ind

icat

or)

Indicator: Atkinson index, = 0.75Outcome: MortalityBaseline: PM-related mortalityPollutant: PM only

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 C-R), 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).

Incorporating environmental equity into risk assessment: A case study of power plant air pollution control strategies Jonathan Levy, Sc.D. Assistant Professor.

Documents

pm slide

influential slide

driven slide

risk assessment practitioners

environmental equity

benefits of pollution

environmental policy

environmental toxicants