Environmental Inequality and
the Black-White Test Score Gap
Anna Aizer
Janet Currie
Peter Simon
Patrick Vivier
Racial Disparities in Test Scores
Large literature examining causes of the gap
• Reviewed by Jencks and Phillips, eds (2008 and 2011) and Magnuson
and Waldfogel, eds (2008)
• Consider the following factors:
•
•
•
•
•
•
•
Family income
School segregation
Family structure
Parenting practices
Quality of educational inputs
School segregation (Reber, 2010)
Neighborhood segregation (Card and Rothstein, 2007)
• Even considering all the above factors, substantial gaps remain
New sources to consider?
• Jencks and Phillips (2008) recommend looking at other, less
traditional sources, to explain the persistence of the gap
• Chay, Guryan and Mazumder (2009) alternative explanation: relative
improvement in infant health of African Americans for cohorts born in
the American south between 1963 and 1971.
New Source: Environmental Inequality
• Large literature documenting that African American and low income
families’ greater exposure to environmental toxins (Brown, 1995;
Brulle and Pellow, 2006; Ringquist, 2005; Mohai and Saha, 2006)
• Cause of the disparities:
• Intentional siting of hazards in low income/minority neighborhoods
• Housing prices reflect environmental quality (Chay and Greenstone, 2005;
Currie et al, 2014)
• Disparities in regulatory response to an existing hazard
Environmental Justice (EJ)
• Inequity in the regulatory response to pollution by race/class led to
new federal regulation in 1994:
• “Preventing discrimination in the development, implementation and
enforcement of environmental laws, regulations and policies” (EPA)
• Research examining the extent to which the regulatory response has
become less discriminatory is mixed
• Sites in areas with a high share of black families or poor families still less likely
to receive superfund designation (O’Neal, 2007)
• Conditional on superfund designation, no longer appears to be any disparity
in the duration of clean up (Burda and Harding, 2014)
The Questions:
• Can environmental inequality explain any of the racial gap in test
scores?
• African-Americans disproportionately exposed to pollutants
• Conditional on exposure, may have fewer resources to counter negative
effects (eg, nutrition)
• Can environmental regulation reduce disparities in test scores?
Focus on a Specific Pollutant: Lead
• Long epidemiological literature linking lead exposure with diminished
cognition
• Lead has no biological value and is toxic. Mimics calcium in the body,
interfering with renal, endocrine, cardiovascular and nervous systems.
• In the brain, lead damages the prefrontal cerebral cortex, hippocampus and
cerebellum (Finkelstein et al, 1998)
• Nutrition affects rates of absorption
• Children particularly sensitive (Lidsky and Schneider, 2003)
• Greater exposure (inhalation and ingestion)
• Conditional on exposure, lead more likely to be absorbed by children
• Conditional on absorption, lead more likely to affect a developing nervous
system
Historical Sources of Lead
• Gasoline: in the 1970s, a gallon of gasoline contained 4.0 grams of
lead
• Household Paint: contained high lead content (10-40% for some
colors) prior to world war II
• Post war, manufacturers drastically reduced the amount of lead contained in
paint.
Regulatory Response
• 1970, US Surgeon General issued his first formal statement on lead
poisoning, naming it a national health problem.
• Since then, 2 major regulatory changes:
• Eliminating lead from gas: 1985 0.5 grams/gallon, 1986 0.1 grams/gallon,
1996 zero
• Eliminating lead from household paint: completely de-leaded in 1978
• Lead in the air has declined from 2.5 ug/m3 in 1980 to less than 0.5
by 2012 (EPA)
• Nationally, preschool age children with lead levels in excess of 10 has
declined from 8.6% in 1998 to 1.3% in 2004.
Is Lead Still a Problem?
• Lead remains a hazard due to residual lead found in the air, dust and soil.
• 4.5 million households in the US are still exposed to high levels of lead and
half a million preschool aged children have elevated blood lead levels (CDC,
2011)
• Disadvantaged children disproportionately affected (1999-2002 NHANES)
• Lead levels 50% higher for African American Children
• Lead levels 30% higher for low income children
• Elimination of elevated blood levels a goal in Healthy People 2020, the ten
year national objectives for improving the health of all Americans
• More recent focus on the danger of low lead levels.
• CDC “lead levels of concern” which have declined over time from 60, to 10 to 5 and
now they no longer use the term – any level above zero is a concern.
Greater Exposure of African-Americans to
Lead
• Nationally, African Americans more likely to live in old (pre 1978)
housing
• 73% of African Americans vs. 63% of whites
• Racial disparities greater than income disparities
• Racial differences greater when we limit to within region/area
Black
White
Hispanic
<=100% FPL
>=200% FPL
RI Share in Housing Built
Pre 1978
Pre 1945
0.83
0.52
0.74
0.37
0.83
0.53
0.81
0.43
0.76
0.4
African American Families Concentrated in
Historic, Urban Centers
Providence
Rest of RI
Share housing built pre 1978
Share housing built pre 1945
0.81
0.49
0.68
0.27
Share black population living in:
Share white population living in:
0.86
0.51
0.14
0.49
Share population <100% FPL living in:
Share population >100% FPL lining in:
0.77
0.55
0.23
0.45
Share population <100% FPL & black living in:
Share population <100% FPL & white living in:
0.89
0.6
0.11
0.4
Calculations from US census IPUMS data for 1990, 2000 and 2010
Sample includes all children <=10 years old
Structure of This Paper:
• Document disparities in lead levels by race and income
• Examine a policy aimed at reducing lead levels among RI children
• Disproportionate declines in the lead levels of African American & the poor
• Link declining disparities in lead levels with declining disparities in
test scores
• Establish a causal relationship between lead levels and future cognitive
achievement
• Identification issues
Data
• Data on child blood lead levels (BLLs) in first 72 months of life (RIDOH)
linked with third through eight grade test scores (RIDE)
•
•
•
•
On average, 4.7 (median=4) BLLs per child
Date of test, location of child, method, level
NECAP test scores for reading and math in grades 3-8, free-lunch status, IEP
Linked with vital statistics data: maternal race, ethnicity, education, marital
status, birth weight, gender, month prenatal care initiated, birth order
• Covers birth cohorts 1997-2004 in the state.
• In RI, 80% of all children screened at least once by 36 months
• In RI, 15-20% in private school as of 2010
• Final sample of 57,000 linked children
Part I: Documenting Disparities in Lead Levels
& Trends
Appendix Table 1: Within Tract Correlations Between Lead Levels and Child and Family Characteristics
Avg lead level
Married at Birth
Maternal age at birth
Mother African-American
Maternal education in years
Mother has at least one risk factor
Always free/reduced lunch
male
Birth order
Birth weight/100
Gestation in Weeks
Month prenatal care initiated
Observations
R-squared
-0.244
[0.0214]
-0.0280
[0.00174]
0.328
[0.0325]
-0.0526
[0.00371]
0.172
[0.0223]
0.359
[0.0267]
0.169
[0.0160]
0.207
[0.00870]
-0.0147
[0.00180]
0.0389
[0.00510]
0.0865
[0.00730]
53,994
0.206
Part II: Why the disproportionate decline in
lead levels for some groups?
• Policy by the DOH and the HRC to reduce lead levels in RI children
• DOH: Required all landlords of homes in which a child had an elevated lead level
to obtain a “Lead Free” certificate.
• State AG to file suit against non-compliant landlords
• HRC: Required all landlords to obtain “Lead Safe” certificates to rent their homes
•
•
•
•
Provided training and information to landlords about lead abatement
Low-interest loans for abatement: section 8, family day care
No resources for enforcement, only the threat of litigation
Target neighborhoods with historically high lead burden – contracted with community based
organizations in the 4 “core cities”, the oldest cities in the state with greatest share of
disadvantaged children
• 35% of children in the core cities poor vs. 9.4% rest of the state
• 36% of children in the core cities African American vs. 2% rest of state
• 73% of African American children and 46% of the poor live in the core cities
• In 1997, 331 certificates issued; by 2010, 41,000 had been issued.
Did Certificates Work?
Leadi= α1certificates/old homesnt + α2 Xic + α3Xim +
α4 Xntn + γn + γt + µi
• Xic : BW, birth order, gestation, gender, year of birth, free lunch status
• Xim: Maternal characteristics: race, ethnicity, years of schooling, marital
status at birth, month prenatal care initiated, maternal age at birth
• Xntn : share poor, median income, share African American, share Hispanic,
share housing build pre-war, share housing built post 1978, ln(population)
• Year of birth and tract FE
Table 1: First Stage - Predicting a Child's Average Lead Level
DOH certificates at birth/old housing units
(1)
(2)
-0.0878
-0.184
[0.0152]
[0.0285]
(DOH certificates at birth/old housing units) Squared
(3)
(4)
0.000502
[0.000125]
HRC certificates at birth/old housing units
-0.113
[0.0147]
Loans at birth/old housing units
-0.00707
[0.00280]
Observations
57,303
57,303
57,303
57,303
R-squared
0.204
0.204
0.205
0.204
F Statistic (excluded instruments)
33.18
17.19
59.00
12.25
Table 3: Certificates and Lead Levels: Heterogeneity by Child Characteristic
DOH certificates at birth/old housing units
(DOH Certificates at birth/old housing)*Black
(1)
(2)
(3)
(4)
-0.0785
[0.0156]
-0.0637
[0.0232]
-0.0372
[0.0252]
-0.129
[0.0394]
-0.0836
[0.0174]
(DOH Certificates at birth/old housing)*Free Lunch
-0.0630
[0.0250]
(DOH Certificates at birth/old housing)*Maternal Education
0.00347
[0.00308]
(DOH Certificates at birth/old housing)*Male
-0.00817
[0.0162]
Observations
57,303
57,303
57,303
57,303
R-squared
0.205
0.204
0.204
0.204
Can lead certificates explain disproportionate
decline in lead among black children?
• More certificates in neighborhoods with a larger share black
• Within neighborhoods, black children’s lead levels disproportionately
affected by certificate availability
• Of the 2.3 point decline in average lead levels among African
Americans for 1998-2004 birth cohorts, 21% is explained by the rise
in certificates
Part III: Does Lead Exposure Reduce Cognitive
Achievement?
• Review the existing work
• Discuss identification issues & strategy
• Present IV estimates based on certificates as an IV
• Examine impact of lead exposure on
•
•
•
•
Reading and math scores
Distributional impacts
Test scores over time (3rd through 8th grade)
Heterogeneous effects by child characteristic
Existing Research on Lead and Child
Outcomes
• Long epidemiological literature linking lead and cognitive and noncognitive outcomes
• Known for centuries that high levels of lead are associated with worse
cognitive outcomes
• More recent focus on low levels of lead
• Individual-level analysis:
• linked preschool lead levels with test scores and behavioral outcomes including ADHD
and hyperactivity (Chardramouli et al, 2009; Canfield et al, 2003; Nigg et al, 2010, and
Wasserman, 1997).
• McLaine et al (2013) links preschool lead levels with kindergarten reading readiness
among Providence children.
Two Challenges to Estimating
• Endogneity of lead exposure: omitted variable bias/confounding
• Bellinger (2008) “It has long been recognized that due consideration must be
given to the possibility that any observed association between lead exposure
and neurodevelopment is, in part, an artifact of residual
confounding…Indeed, adjusting for SES and related covariates can result in
reductions of 50% of more in the magnitude of lead’s regression coefficient.”
Measurement Error in Blood Lead Levels (BLLs)
1. Individual blood lead levels (BLLs) measured with considerable error
• “The ratio of imprecision to measurement value, particularly at BLLs
<10ug/dL, is relatively high.” (CDC, 2007)
• The CDC allows measurement error up to 4 ug/dl or 10%, whichever is higher.
2. Infrequent testing & short half life of lead in blood
• We have on average 4 measures for each child for the first 72 months of life
• Lead half life in blood is 36 days
• Remains in tissue much longer
Cohort Level Analysis – identification more
straight forward, but ecological fallacy
• Lead and Cognition
• Ferrie, Rolf and Troesken (2012) – variation from lead piping, examining
impact on intelligence tests of WWII enlistees.
• Effects only for low SES children.
• Lead and Crime
• Masters et al(1998); Reyes (2007); Nevin (2000 and 2007); Mielke and Zahran
(2012); Nilsson (2014) and Reyes (2014)
• Dramatic declines in lead levels (from the de-leading of gasoline) linked with
the dramatic declines in crime rates
Lead, SES and Cognitive Outcomes
IV Strategy
• Use certificates in one’s tract at birth to identify impact of lead on
future cognitive achievement
• Exclusion restriction: neighborhoods that see more growth in the
number of certificates not otherwise changing in ways that would
independently affect child cognitive outcomes
Table 4: Lead and Reading Test Scores (Avg)
Avg lead level
(1)
(2)
(3)
(4)
(5)
(6)
OLS
Tract FE
Tract FE
Tract FE IV
Mom FE
Mom FE IV
-1.029
-0.274
-1.931
-0.105
-3.197
[0.0188]
[0.0190]
[0.837]
[0.0413]
[1.609]
57,303
57,310
57,303
Randomly Drawn Lead Level
-0.0792
[0.0142]
Observations
60,582
57,310
57,310
R-squared
0.048
0.138
0.136
0.062
Table 5: Lead Levels and Third Grade Reading Proficiency
(1)
(2)
Substantially below
Proficient <30
(3)
(4)
Proficient >=40
(5)
(6)
with Distinction >56
Panel A: Reading
Tract FE
IV
Tract FE
IV
Tract FE
IV
Avg lead level
0.00541
0.0790
-0.00858
-0.176
-0.00281
0.0205
[0.000537]
[0.0261]
[0.000795]
[0.0450]
[0.000729]
[0.0308]
Observations
53,978
53,972
53,978
53,972
53,978
53,972
R-squared
0.033
Average of dependent variable
0.067
21%
0.058
71%
10%
Table 5: Lead Levels and Third Grade Math Proficiency
(1)
(2)
Substantially below
Proficient <30
(3)
(4)
Proficient >=40
(5)
(6)
with Distinction >56
Panel B: Math Proficiency
Tract FE
IV
Tract FE
IV
Tract FE
IV
Avg lead level
0.00484
0.0644
-0.00479
-0.0316
-0.00176
0.00381
[0.000619]
[0.0279]
[0.000844]
[0.0355]
[0.000591]
[0.0246]
Observations
54,035
54,029
54,035
54,029
54,035
54,029
R-squared
0.032
Average of dependent variable
0.066
12%
0.034
60%
14%
Table 6: Lead Levels and Reading Proficiency Over Time
(1)
(2)
Third Grade
(3)
(4)
Eighth Grade
FE
FE IV
FE
FE IV
-0.00792
-0.0710
-0.00782
-0.120
[0.00135]
[0.0626]
[0.00111]
[0.0648]
Observations
16,511
16,511
17,850
17,850
R-squared
0.069
Avg lead level
0.070
2 sets of robustness checks
• Alternative measures of lead
• Alternative sets of Instruments
Table 7: Different Measures of Lead and Reading Test Scores (Avg)
Average level - censor at 25
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
FE
IV
FE
IV
FE
IV
FE
IV
-0.278
-1.925
-0.252
-1.038
-0.140
-1.285
-0.309
-3.631
[0.0192] [0.833]
Lead Area Under the Curve (AUC)
[0.0147] [0.432]
Max lead level
[0.0107] [0.573]
Geometric Mean of Lead
[0.0207] [1.990]
Observations
57,310
R-squared
0.138
57,303
57,309
0.140
57,302
57,310
0.138
57,303
56,278
0.140
56,271
Table 8: Second Stage Estimates of Lead and Average Reading Scores from Alternative IVs
(1)
Avg lead level
Observations
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
-1.825 -1.736 -1.931 -1.812 -1.939 -1.717 -1.277 -1.770 -1.892 -1.807
[0.775] [0.752] [0.837] [0.816] [0.679] [0.622] [0.553] [0.594] [0.666] [0.611]
57,303 57,303 57,303 57,303 57,303 57,303 57,303 57,303 57,303 57,303
First Stage Instruments:
DOH Certificates at birth/1000*Original (1997) Lead Level
DOH Certificates at age 5/1000*Original (1997) Lead Level
DOH certificates at birth/old housing units
DOH certificates at age 5/old housing units
HRC certificates at birth/1000*Original (1997) Lead Level
HRC certificates at birth/Old housing units
HRC certificates at age 5/Old housing units
Total certificates at birth/1000*Original (1997) Lead Level
Total certificates at birth/old housing units
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Additional robustness exercises:
• Including more controls:
• race*year FE
• free lunch* year FE
• Lead level in 1997 (in tract)*year
• Propensity score estimates
• Limit to tracts with predicted high numbers of certificates (by 2004)
Summary (thus far)
• Racial disparities in lead levels and test scores
• Lead has a negative and causal impact of cognitive achievement that
does not fade over time and is not heterogeneous
• Policy in 1997 disproportionately lowered the lead levels of African
American children
• The declines did not simply reflect declines more generally by income
• Important spatial component
• African American children concentrated in urban areas with higher initial lead
levels that received more “treatment” over this time period
• Did the narrowing of the racial gap in lead levels result in a reduction
in the racial gap in test scores?
Trends in Lead Over Time
Two exercises:
• Calculate what share of the reduction in the black-white test score
gap was due to the reduction in lead levels based on the causal
estimate of the impact of lead on educational outcomes
• Note: assume the impact of lead on test scores not vary by race
• Examine directly how the gap in lead levels relates to the gap in test
scores
Trends in disparities in test scores
• Racial gap in test scores fell from 9.7 for those born in 1998 to 6.3 for
those born in 2004 (from 70% to 45% of a standard deviation).
• For lead, the gap fell from 2.2 to 0.9 over this same period
• Based on the range of causal estimates, the decline in the gap in lead
levels explains 37-76% of the decline in the test score gap.
• Income gap in test scores and lead both fell,but by smaller amounts:
• Income test score gap fell from 9.3 to 8.4 (67% to 60% of a std deviation)
• Income lead level gap fell from 1.83 to 0.99
Table 9: Disparities in Lead and Disparities in Test Scores - Tract and County Level
(1)
(2)
(3)
(4)
White-Black Reading Scores
Tract Level Disparities
County Level Disparities
OLS
IV
OLS
IV
-0.746
-2.101
-1.422
-1.321
[0.127]
[1.487]
[0.822]
[2.179]
Observations
1,736
1,695
40
40
R-squared
0.146
0.077
0.853
0.853
White-Black Lead Scores
Each observation is a tract (county) year. Also included are county (tract) and year of birth fixed effects.
All regressions weighted by tract (county) population
Note that the White-Black lead score difference is negative, generally, since white lead levels are lower than black lead levels.
Conclusions
• RI implemented a policy that targeted and disproportionately reduced
the lead levels of African American and low income children
• The resulting declines in racial disparities in lead exposure can explain
a substantial share of the recent decline in racial test score disparities.
• Eliminating the black-white test score gap single most effective way to
reduce racial economic inequality (Jencks and Phillips, 2011).
• If we continue to target environmental regulation at children at
greatest risk, then declines in disparities in future economic outcomes
may follow.