Upload
trinhtu
View
216
Download
1
Embed Size (px)
Citation preview
Offshoring Health Risks: The Impact of the US Lead
Regulation on Infant Health in Mexico∗
Shinsuke Tanaka† Kensuke Teshima‡
August 2017
Abstract
This study examines the effect of tightening the environmental standard on lead in theUnited States (US) on (i) air quality in the US, (ii) trade patterns between the US andMexico and (iii) infant health in Mexico as a test of pollution haven hypothesis. We providenew evidence based on the innovative research design that allows us to exploit exogenousvariation in the regulatory stringency as well as to single out the specific substance andpollution-intensive product. Consistent with the theoretical prediction, our findings showthat improvement in air quality in the US was achieved by offshoring the pollution-intensiveproduct and infant health risks to Mexico.
Keywords: Pollution Haven Hypothesis, Environmental Regulation, Infant Health
JEL Codes: F18, Q56, Q58, I18
∗We thank Adam Storeygard, Akira Hibiki, Alan Finkelstein, Arik Levinson, Koichiro Ito, Laura Gee, LucasDavis, Marco Gonzales-Navarro, Melissa Dell, Michael Klein, Michikazu Kojima, Paulina Oliva, Sahar Parsa,Shin-Yi Chou, Shomon R. Shamsuddin, Yoichi Sugita, Yutaka Arimoto, and participants at Ashecon, AEA-ASSA, Hitotsubashi University, JETRO-IDE, Keio University, Kyoto University, Tohoku University, LACEA-TIGN, MIT, National institute for Environmental Studies in Japan, NBER Summer Institute, North AmericanSummer Meeting of the Econometric Society, and PacDev for helpful comments and discussions. We are gratefulto Omar Trejo and Paola Ugalde Araya for excellent research assistance. The usual disclaimer applies.†The Fletcher School, Tufts University. 160 Packard Ave. Medford, MA, USA. Email: [email protected]‡CIE-ITAM. Av. Santa Teresa # 930, Mexico, D.F. 10700, Mexico. Email: [email protected]
I. Introduction
Whether and the extent to which low-income countries should bear responsibility for environ-
mental protection has centered contentious debates at international policy conferences. Low-
income countries often express prevailing concerns that tightening regulatory stringency will
hinder economic growth and call for stringent environmental regulations only in developed
countries. In contrast, economists and policymakers have long feared that raising environmen-
tal standards only in rich countries will induce the pollution-intensive industries to relocate
in seeking for lower environmental compliance costs, damaging the environment and exposing
public health at risk in the recipient countries.1 Known as the pollution haven hypothesis, the
idea is solidly grounded on the Ricardo’s theory of comparative advantage —differences in the
stringency of environmental regulations create potential for trading pollution-intensive products
from countries with weak environmental regulations to countries with stringent regulations.
The empirical evidence to date, however, still remains inconclusive about this theoretical
prediction. Some recent work focuses on the inflow or outflow of foreign direct investment (FDI),
mostly among multinational firms, associated with variation in the regulatory stringency (Keller
and Levinson 2002; Eskeland and Harrison 2003; Javorcik and Wei 2003; Hanna 2010; Kellenberg
2009; and Chung (2014)).2 The important limitations with this strand of literature are that
FDI flows are not direct evidence of trading pollution-intensive products, and that production
relocation may occur outside the multinational firms. Other related literature, although not
directly testing the pollution haven hypothesis, highlights reductions in production and births
of polluting firms in counties not compliant with the Clean Air Acts in the United States (US)
(Henderson 1996; Levinson 1996; Becker and Henderson 2000; and Greenstone 2002), but they
shy away from implications for international trade or production shifts. Others, such as Davis
and Kahn (2010), focus on trade of used or waste products induced by trade liberalization but
not by variation in environmental regulations.3
In this study, we investigate and provide new evidence of the effect of tightening environ-
mental standards in an industrialized country on the international trade and health outcomes
in a recipient country. In particular, we exploit the revision of the national ambient air qual-
ity standards (NAAQS) on lead in the US that effectively raised the production costs of used
1Non-governmental organizations and media have also alerted the issue. See, for instance, Commission forEnvironmental Cooperation (2013) and New York Times (2011).
2The early work, which was predominantly based on cross-sectional settings, find little evidence to support thehypothesis. See Jaffe and Stavins (1995) for the summary. The recent work, which exploits panel settings to controlfor unobserved heterogeneity across jurisdictions or instrumental variables to address endogeneity in regulatorystringency, have found associations between environmental regulations and production. See Brunnermeier andLevinson (2004) and Copeland and Taylor (2004) for details.
3The critical assumption that this strand of literature provides inference for pollution haven hypothesis is thattrade barriers inhibit trade of pollution-intensive products in a greater extent than they do on clean products,thereby requiring comparisons of dirty and clean industries. Levinson (2008) clarifies the distinction between thisstrand of literature and the direct test of pollution haven hypothesis as follows. Consider the regression model:
Yit = α+ βEit + γTit + θEitTit +X ′itδ + νi + εit, (1)
where Yit is pollution-intensive production in a country i at time t, E measures the stringency of environmentalregulations, T measures the extent of trade barrier, such as tariffs. X is a vector of controls and ν is countryfixed effects. In this model, the direct test of pollution haven hypothesis is ∂Y/∂E = β + θT . In contrast, testingwhether trade barriers inhibit pollution-intensive production shift is ∂∂Y/∂E∂T = θ.
1
lead-acid batteries (ULAB), which constitute more than 90% of the domestic secondary produc-
tion of lead. Importantly, the policy intervention created variation in the regulatory stringency
between the US and Mexico, the leading destination of the US lead export. The innovative
research design, coupled with exogenous variation in the regulatory stringency as well as the
ability to single out the specific pollution-intensive product, allows us to shed clear light on
the impacts on the trade patterns between the two countries. Further, although the pollution
haven hypothesis is often alerted because of public health concerns in countries where pollution-
intensive production occurs, we know of no other study that has simultaneously examined the
health consequences due to the production relocation. Overall, our findings support the pollu-
tion haven hypothesis suggesting that the stringent environmental policy in a developed country
can induce offshoring of pollution-intensive production and health risks to a low-income country
with lax environmental regulations.
In order to rigorously test the pollution haven hypothesis, the empirical strategy employs
unique combinations of public and non-public datasets on the environment in the US, trade
between the US and Mexico, and health outcomes in Mexico that provide a rare level of detail
and that were not previously employed in the related fields. In addition, three salient features
allow us to perform the analysis with greater precision and offer causal inference. First, among
manufactured products, the dominant fraction of lead today is consumed by lead-acid batteries
(LAB) used in automobiles ((nearly 90 % in the US in 2010). Second, the dominant source of
lead production today is recycling of ULAB (around 90 % in the US in 2010). Third, while this
recycling process is the major source of lead emissions to the environment, it does not typically
emit a high volume of other toxic substances relative to lead. These features allow us to focus
on the particular substance and the particular product, making it possible to shed clear light
on the impact of the regulation change on the environment, trade flows, and health outcomes.
Since no study has ever documented whether the revised NAAQS constrained the lead pro-
duction in the US, our first contribution is to offer evidence on whether the revised NAAQS
effectively reduced lead concentrations in ambient air. The new NAAQS on lead was imple-
mented at the beginning of 2009, and the revised standard became ten times tighter than the
previous one, which was commensurate with the ongoing standard in Mexico. Using exact lo-
cations of the lead-emitting plants and monitoring stations across the nation, we empirically
guide our research design, which compares areas in close proximity to the emission sources and
those slightly farther away, defined by the distance lead can travel once it is emitted from the
sources, separately for the battery-recycling plants and other lead-emitting plants. Based on the
triple-difference approaches, we find that the policy made substantial reductions, nearly 50%,
in lead concentrations within 1 mile of the battery-recycling plants. The finding support that
the new policy was binding the domestic lead production, the compliance was high, and human
exposure to lead in the environment was effectively reduced.
Second, we document a trend break in the exports of ULAB from the US to Mexico. Mexico
has historically been a major destination of lead exports from the US, accounting for nearly 50%
of the total lead exports. The policy change in the US caused a marked increase in the exports,
resulting in quadrupling the total number of ULAB shipment. We also find reduced domestic
lead production in the US and increased imports of lead by the US to meet the sustained
2
demand for LAB from the automobile industry and production/exports of new LAB. Using
the confidential customs records of the Mexican exporters, we also find evidence that exports
by lead-emitting Mexican firms to the US have also increased. These results suggest that lead-
emitting production, both recycling of ULAB and other lead-emitting activities ,shifted from
the US to Mexico.
Lastly, using exact locations of lead-emitting plants and residential locations at a fine dis-
aggregated level in Mexico, we identify critical distance over which households are exposed to
pollution emitted from sources and show that birth outcomes in Mexico deteriorated in prox-
imity to battery recycling plants (i.e., within 2 mile) relative to areas slightly away (between
2-4 mile).
Our paper is also related to an emerging literature examining new channels through which
trade affects health. Atkin (2013) analyzes the consequence of trade on nutrition when the
preference of household exhibits habit formation. Pierce and Schott (2016) analyze the impact
of import competition on mortality in the US, whereas Bombardini and Li (2016) analyze the
impact of export market access on the environment and child mortality in China. We highlight
a distinct channel, which altogether help understand how trade factors can shape health.
The rest of the paper is organized as follows. Section II describes the background on the
revision of NAAQS on lead, and Section III presents the data sources and the summary statistics.
Section IV examines the effect of the new NAAQS on the environment in the US. Section V
investigates the effect on trade flows between the US and Mexico. Section VI explores the effects
on birth outcomes in Mexico. Finally, Section VII concludes.
II. The US National Ambient Air Quality Standards on Lead
Lead is a metal known to affect almost every organ and system in the human body. Children
under six years old including fetuses are considered most susceptible to the effects of lead, which
are often linked to retarded fetal growth, behavioral problems, learning disabilities, lower IQ,
and later criminal activities.4 The US had historically made substantial reductions in lead in
ambient air, achieving nearly 92 percent reductions in average airborne lead concentrations
between 1980 and 2013 (US EPA). Much of the reduction came from the efforts to permanently
phase out lead in gasoline for motor vehicles since the mid-1970s.
However, lead still exists in ambient air in the US and continues to endanger public health
and welfare. In 2010, an estimated 535,000 children aged one to five had blood lead levels high
enough to damage their health, and costing over $3 billion in medical and special education
costs (CDC). Increasing scientific evidence suggests that there is no safe level of lead in blood,
and even a lower level than originally thought has adverse effects on health and behavior (WHO
2010, Emory et al. 1999, Canfield et al. 1999, Lanphear et al. 2005, Nigg et al. 2008). In response,
the EPA heavily strengthened the national ambient air quality standards (NAAQS) on lead in
2008. The revised standard is 0.15 µg/m3, not to be exceeded by a rolling 3-month average,
4See, for example, Needleman et al. (1990), Pocock et al. (1994), Schwartz (1994), Burns et al. (1999), Torres-Sanchez et al. (1999), Nevin (2000), Dietrich et al. (2001), Bellinger and Needleman (2003), Berkowitzn et al.(2006), Hu et al. (2006), Schnaas et al. (2006), Reyes (2007), Braun et al. (2008), Cecil et al. (2008), Wrightet al. (2008), Nilsson (2009), Mirghani (2010), Rau et al. (2013), Gronqvist et al. (2014), Aizer et al. (2015), andBillings and Schnepel (2015).
3
which is 10 times tighter than the previous one, 1.5 µg/m3, set over 30 years ago in 1978.
This standard is identical for both the primary (health-based) and secondary (welfare-based)
standards. Importantly, such revision created substantial variation in the standards between
the US and Mexico.5.
A review of the lead standards was initiated in November 2004, leading to the release of the
Plan for Review of the National Ambient Air Quality Standards for Lead in February 2006, and
subsequently an advanced notice of proposed rule-making on December 5, 2007. The revision
on the NAAQS for lead was signed on May 1, 2008, and the EPA published the final rule
in Federal Register on November 12, 2008, which became effective on January 12, 2009. To
assess compliance with the revised NAAQS on lead, the EPA expanded the lead monitoring
network, requiring monitors to be placed in areas that are expected to exceed the standards,
such as those near industrial facilities that emit one ton or more of lead per year. Compliance to
the new standards is evaluated over a 3-year period. The EPA completed final designations of
attainment/nonattainment status for every county in November, 2011, and states were required
to submit State Implementation Plans for nonattainment counties to comply with the standards.
All states are required to meet the standards by January 2017.
III. Data Sources and Summary Statistics
A. The Locations of Lead Emitting Plants
We identify the geocoded locations of the lead-emitting plants in the US using the Toxic Release
Inventory (TRI). Managed by the Environmental Protection Agency (EPA), the TRI aims at
enhancing informed decision-making among communities and other stakeholders by publicly
disclosing information on toxic chemical releases and pollution prevention activities reported
by industries and the government. Plants are required to report their emissions of specified
chemicals if they employ at least ten full-time employees, operate in a certain industry, and
are engaged in manufacturing, processing, or using regulated substances more than a specified
threshold. Despite the benefits of the TRI allowing for identifying the exact locations of the
emission sources, which otherwise is not feasible, the two limitations with the TRI have been
widely concerned. First, although reporting is mandatory, information provided through the
TRI contains self-reported values from individual facilities, which are known to suffer from
substantial errors (de Marchi and Hamilton 2006; Koehler and Spengler 2007; and Bennear
2008). Second, plants do not appear in the TRI unless they satisfy the (varying) minimum
mandates for reporting. To circumvent these issues, we follow Currie et al. (105) in keeping all
facilities that have ever emitted lead and/or lead compounds at least once in the relevant period,
from 2001 to 2013, and in not relying on the information on the reported emission amount.
In total, there are close to 13,000 facilities across the country that emitted lead during this
period. On average, each facility that emitted any amount in the year emitted approximately
77,270 pounds of lead and lead compounds combined per year with standard deviation of 3.3
5The lead standard in Mexico has been 1.5 µg/m3 for three-month average since 1993 (NOM-043-SEMARNAT-1993). In Canada, Ontario sets the level of 0.5 µg/m3 for 24-hour average and 0.2 µg/m3 for 30-day average,while Quebec sets 0.1 µg/m3 for one-year average.
4
million pounds. The list of facilities that recycle ULAB in the US is found in CEC (2013). There
are in total 15 plants that are operated by 7 firms in the US.
The Mexican counterpart of the US TRI data is Registro de Emisiones y Transferencia de
Contaminantes (RETC), or Registry of Emission and Transfer of Pollutants, a registry of estab-
lishments’ reports of emissions and transfers of pollutants.6 Operated and managed by Secretar
de Medio Ambiente y Recursos Naturales (SEMARNAT), or Ministry of the Environment and
Natural Resources, plants in the specified industries that emit specified chemical substances are
obliged to register to the RETC every year. The RETC publishes the information on the exact
locations of each plant (the full and geocoded address), the sector, the amount of emissions by
chemical substance, and the emission modes. The information used in this study is also publicly
available on the website of the SEMARNAT.
B. The Ambient Air Quality
The data on ambient air quality in the US are obtained from the Air Quality System, a pub-
licly accessible database managed by the EPA. The data provide information on air pollutant
concentrations in the ambient air monitored at more than 4,000 stations across the nation.
Unfortunately, there is no such data that monitor air quality in Mexico.
C. The International Commodity Trade
We obtained information on the exports of US commodities from the US International Trade
Commission (ITC) at the monthly level between 1997 and 2014, corrected based on the commodity-
specific statistical corrections provided by the US Census Bureau, Foreign Trade Division.7
We obtained information on commodities exported from Mexico to the US using the ad-
ministrative records of the Mexican customs agency on every transaction crossing the Mexican
border from 2005 to 2011. Prior to carrying out an international transaction, Mexican exporters
and importers must fill out a customs form, on which they report the tariff classification code
of the products, their value and the quantity, and the destination (or source) country, among
other information.
D. Birth Outcomes in Mexico
Our data on birth outcomes come from two sources. The two sources have respective advantages
and disadvantages: the first dataset covers enough pre-shock years but somewhat limited (though
6As part of North American Free Trade Agreement (NAFTA) negotiations, Mexico made the commitment toestablish a registry of public environmental information on emissions and pollutant transfers, which is similar tothe TRI in the US. After a pilot program that started in 1997, the mandatory registration system was legislatedin 2001, and the registry started in full operation in 2004.
7Note that numerous corrections are reported for years around 2009 with regard to ULAB exports to Mexico.After we made all these corrections, we verified the information with the Mexican data on ULAB imports fromthe US The two numbers are closely matched except one month, May 2007, when the number from the US isimplausibly high, while the number from Mexico is in a reasonable range. Given that most of the US numbersaround this time were reported as errors and corrected, we believe this particular month also contains an error,though overlooked or unreported by ITC, and thus we adopted the Mexican number.
5
relevant) population, while the second dataset covers the universe of the births but for a shorter
period.
D.1. Hospital Discharge Data: 2005-2013
Mexican Ministry of Health makes the hospital discharge record for the hospitals under the
operation of the ministry. The Ministry of Health hospitals serves population who are not
under the Mexican social security system, and covers around one third of all the births in
Mexico. Another one third is covered by hospitals operated under the social security system,
and the last one third is covered by private hospitals. Population served by the hospitals under
the Ministry of Health hospitals are primarily in the informal sector of the economy, thus are
relatively disadvantaged than those who go to other types of hospitals. The the Ministry of
Health hospital discharge records include information on the weight of new-born babies when
mothers stayed in these hospitals.
D.2. Birth Certificate Data: 2008-2013
Second, we use the universe of birth certificates between 2008 and 2013. The birth certificate is a
preliminary identification document issued at birth, only once, by the Mexican National Health
System (NHS)8 to every newborn born alive9 within the Mexican territory since September
2007, regardless of parents’ nationality or immigration status.10 de Salud, or Army). The data
can be obtained from the National Basic Health Information System on the website of Ministry
of Health.
The issuing of the NHS birth certificate is a relatively new legal mandate in Mexico. The
only proof of birth for all births occurred prior to September 1st, 2007 is the civil registry-issued
birth certificate (acta de nacimiento). As a consequence, our data necessarily start from 2008.
While being born on Mexican soil concedes the Mexican nationality, the civil registry requires
the submission of the NHS birth certificate in order to issue its civil birth certificate (acta de
nacimiento), which is the first valid official identification document for civil purposes. Therefore,
it is mandatory for the health system personnel to issue the NHS birth certificate immediately
following births.
8The National Health System (NHS), established in 1984 as part of the General Health Act (Ley General deSalud), is the network of all listed hospitals, health facilities and care providers, including both governmental andprivate. The NHS? goal is to provide access to the healthcare services.
9The legal definition of a newborn born alive considers all human fetuses or newborns that have been expelledor extracted from the mother?s body, regardless of lengths of gestation and/or whether newborns are physicallyseparated from mothers, show voluntary or involuntary muscular movements, are breathing, or their umbilicalcord is beating at some point.
10By law, when a baby is born alive in a listed medical facility, and therefore within the NHS or at any otherlocations with assistance of a registered midwife or doctor, the professional who delivers the baby is obliged toissue the birth certificate within the next 24 hours after the birth. If under any given circumstances the birthcertificate was not issued within this timeframe, the certificate is issued according to the biological mother?sclinical record and filled by the health care provider who delivered the baby, or by the head of the correspondinghealthcare facility or her representative. If a non-listed person assists the delivery, and consequently she does notpossess any issuing capacities, the NHS birth certificate will be supplied by the corresponding listed healthcarefacility within the next year. Usually, this is the closest healthcare unit to the place of birth, but it may not bethe case and rather depend on the mother?s affiliation status to any of the current social security schemes (IMSS,ISSSTE, Pemex, Secretar
6
In addition, birth certificates are used to provide statistics of the number of births and
health of newborns and mothers. For these purposes, the data not only include demographic
information of mothers (i.e., age, residential locality, educational, occupation, and prenatal care
and pregnancy history) but also report birth outcomes such as birthweight, length of gestation
period as well as Apgar and Silverman test scores.11
IV. The Effect on the Environment in the US
We begin with investigating the effect of revised NAAQS on air quality in the US to understand
whether or not the new standards were binding the production activities in the US.
A. Graphical Evidence
If the newly revised NAAQS had any contributions to reductions in lead production in the US, we
would expect lead concentrations in ambient air near lead-emitting facilities to fall subsequent to
the policy implementation. Moreover, we should observe a larger impact on ambient air quality
closest to the lead-emitting facilities.
Figure 1 plots the lead pollution gradient of distance to the lead-emitting facility during the
pre-reform period (years 2003 through 2008) in a blue dashed line. Ambient lead concentrations
are the highest at monitoring stations in closest proximity to lead-emitting plants and fall with
distance from the plants until monitoring stations located about 2 miles away and beyond
detect almost no lead. This evidence itself does not directly indicate that lead travels about 2
miles once emitted from the facilities, because lead in ambient air can also be caused by the
atmospheric re-suspension of contaminated soil/road dust.
Figure 1 adds the pollution gradient during the post-reform period (years 2009 through 2013)
in a red solid line. The pollution concentrations are similar before and after the policy reform in
distance over 2 miles. However, there is a clear decline in pollution concentration with proximity
to lead-emitting facilities within 1 mile, and the decline is greatest for the monitors nearest to
toxic facilities. The evidence suggests that lead concentrations have declined at monitors within
1 mile from the lead-emitting plants, indicating that the areas within 1 mile from a lead-emitting
plant constitute the affected areas. The estimated distance is consistent with Currie et al. (105).
The notion that the decline in lead concentrations in ambient air within 1 mile found in
Figure 1 reflects a causal effect of the revised NAAQS would require further evidence. In par-
ticular, the timing of decline must coincide with the implementation of the new standard. An
identification threat is that lead concentrations in ambient air have been declining over time
even without the revision of NAAQS, and in such a case, we would still get similar figures as in
Figure 1, but the decline may simply reflect a preexisting time trend.
11The Apgar test is performed on a baby at 5 minutes after birth. Each letter represents five factors used toevaluate the baby?s condition; Appearance (skin color), Pulse (heart rate), Grimace response (reflexes), Activity(muscle tone), and Respiration (breathing rate and effort). The Silverman test score rates upper chest retractions,lower chest retractions, xiphoid retractions, nasal flaring, and expiratory grunting immediately after birth. Notethat both scores are scaled from 0 to 2 per factor, and thus from 0 to 10 in total, higher Apgar scores are better(a score of 8-10 is considered normal), while lower Silverman scores are better (normal babies have a score closeto 0).
7
Figure 2 shows lead concentrations in ambient air gradient over time. The trends in ambient
lead concentration levels are measured separately for air quality monitoring stations located
within 1 mile from a lead-emitting facility (a red solid line) and those between 1 to 2 miles from
a lead-emitting facility (a blue dashed line). Prior to the revision of the NAAQS in 2008, the
levels of lead concentrations stayed constant at both plants within 1 mile and between 1-2 mile
from the lead-emitting plants. An important implication is that these two areas would follow
similar trends in lead in air over time without an intervention, supporting the key assumption for
the econometric model described below. Interestingly, we observe a spike in lead concentrations
in ambient air in 2008.
Importantly, the two areas illustrate distinct trends in lead concentrations after 2009; lead
concentration levels in ambient air closest to the toxic facilities substantially fell, while lead
concentration levels in the proximate 1-2 mile areas did not decline after 2009 and have been
constant over time. Under the assumption that these two areas would have had similar trends
in ambient air quality had the revised NAAQS not been implemented, the differences in the
amount of reductions can be interpreted as the causal effect of the new EPA regulation on lead.
How much do battery recycling plants account for the reductions in lead concentrations in
ambient air within 1 mile from the lead-emitting sources, as observed in Figure 2? To illustrate
this question, we now plot lead concentrations at monitors within 1 mile from battery recycling
plants and those within 1 mile from other lead-emitting plants in Figure 3. The figure shows that
average lead concentrations near battery recycling plants exceeded the new standard prior to the
revision of the NAAQS, while those near other lead-emitting plants were already below the new
standard. While lead in ambient air remained constant in both areas up to 2008, only the one
near battery recycling plants showed substantial reductions after 2009. The figure suggests that
reductions in lead in ambient air were primarily driven by battery recycling plants, suggesting
that battery recycling plants were most affected by the new NAAQS standard on lead.
Figure 4 highlights heterogeneous effects by the revised NAAQS across battery-recycling
plants.
It shows lead concentrations in ambient air before the policy reform on the x-axis and after
on the y-axis by averaging observations at monitors within 1 mile for each battery-recycling
plants. The lead concentrations near a number of plants exceeded the new standard of 0.15
µg/m3 in the pre-reform period. While the policy achieved substantial reductions in lead in
ambient air, making most of them comply with the new standard in the post-reform period, the
policy had little effect on the plants that were compliant with the standard to begin with, and
two plants remained incompliant with the new standard.
B. Regression-adjusted Evidence
We examine the policy effect on lead in ambient air based on the difference-in-differences (DD)
model, which compares changes in lead concentrations, from before to after the policy change,
between areas in the close proximity to lead-emitting plants that are affected by the policy
effect and areas slightly farther away, far enough to not be subject to the policy effect yet
still adjacent to be comparable. Such estimation requires knowledge of how far lead can travel
once it is emitted from the sources. There is little guidance in the literature, so it is one of
8
our contributions to provide rigorously estimated evidence based on the quasi-experimental
research setting. In particular, we estimate the distance gradient of the policy effect by running
the equation of:12
Leadjt = α+ f(β, Distj , Postt) + δPostt + µt + γj + εjt, (2)
where the outcome variable is the monthly averages of lead concentrations in ambient air at
monitor j at time t. f(.) is various orders of polynomials in distance from the monitor to the
nearest lead-emitting plant, Dist interacted with the post-reform dummy, Post, which takes the
value of one for all periods after 2009. We additionally include time fixed effects (i.e., year and
month), µt, and monitor fixed effects, γj (, which are equivalent to the monitor-plant pair fixed
effects). To capture the distance gradient in a flexible manner, we include up to quintic functions
in distance. The parameter of interest is the marginal effect of the post dummy (equivalently
the vector of β and δ), which infer the differences in the outcome variable between before and
after the policy reform over various distances. The sample includes monitors up to 10 miles
away from the nearest lead-emitting plant. The standard errors are clustered at the monitor
level.
Because individual coefficients of interest are difficult to interpret, we plot the figure showing
the distance gradient effect in Figure 5 based on the quintic function.13
Figure 5 presents the consistent estimate as the graphical evidence above. The reductions
in lead concentrations were the largest in the close vicinity of the lead-emitting plants, and
the effects wane as monitors are located away from the pollution sources. The point estimates
become almost constant after 2 miles, indicating that any reductions in lead concentrations any
farther can be explained by the overall time effect.
The results above allow us to define the areas that are affected by the policy effect, (denoted
by Near hereafter), and the adjacent areas that are slightly farther away, (denoted by Adjacent
hereafter), that are not directly affected by the policy but are otherwise similar. The resulting
difference-in-differences model is:
Leadjt = α+ βNearj × Postt + µt + γj + εjt, (3)
where Near is a dummy variable taking the value of one for monitors within a certain distance
from the lead-emitting plants (i.e., 1 mile). The parameter of interest is β, which measures the
reductions in lead concentrations in ambient air between before and after the policy reform at
monitors in the neighborhood of lead-emitting plants, relative to the monitors slightly away.
The major advantage of including monitor fixed effects (or equivalently the monitor-plant
pair fixed effects) rather than plant fixed effects is that it can address the selection bias. Because
the EPA required to expand the monitoring system when revising the NAAQS on lead, the model
without monitor fixed effects is likely to underestimate the policy effects when more monitors
were built near large lead-emitting sources after the policy reform. In contrast, advantages of
12The empirical strategy employed here is similar to that found in Linden and Rockoff (2008), Muehlenbachset al. (2015), and Currie et al. (105)
13The figures based on other polynomial functions and the underlying estimated coefficients for all polynomialsare presented in the Online Appendix I
9
the model with plant fixed effects rather than monitor fixed effects are that 1) the parameters
can be estimate using all monitors that reported even only before or after the policy reform,
which can enhance statistical efficiency, and 2) the parameters can be estimated by comparing
the two adjacent areas for a given plant. We provide the results based on plant fixed-effects
model (that are essentially similar to the model with monitor fixed effects) in Online Appendix
III
Table 1 reports the estimated policy effect using the equation 3. Column (1) includes a
Near dummy in place of the monitor fixed effects, indicating that there was about 0.046 µg/m3
reduction in lead concentrations in ambient air at monitors within 1 mile from the lead-emitting
plants relative to those between 1-2 mile away. Column (2) includes monitor fixed effects to
capture fine variation in average ambient air quality across monitors, suggesting about 0.068
µg/m3 reductions in lead concentrations. The point estimate is statistically significantly different
from zero at the 1% level. Also, the point estimate is economically significant, suggesting that
lead concentrations near a toxic plant declined by 54% in the aftermath of the n ew lead
regulation.14 The comparison of areas between the close proximity to the lead-emitting plants
(i.e., within 0.5 mile) and slightly away (i.e., between 1-2 mile) in Columns (3) and (4) reveal
that the effects are larger in closer areas. Column (5) presents similar results using alternative
definitions of the treatment and control groups. We present various alternative definitions of
the comparisons in the Online Appendix Table 2. Further, we present the results with plant
fixed effects in the Online Appendix Tables 3 and 4. All these results are reassuring that the
revision of NAAQS resulted in substantial reductions in lead concentrations within 1 mile from
the lead-emitting plants.
The graphical evidence above has suggested that most of the reductions in lead occurred
near battery recycling plants among all lead-emitting plants. In order to investigate how much
reductions in lead were achieved in the neighborhood of battery recycling plants in a more
rigorous manner, we estimate the following triple differences model15;
Leadjt =α+ β1Nearj ×Batteryj × Postt+ β2Batteryj × Postt + β3Nearj × Postt + β4Nearj ×Batteryj+ µt + γj + εjt,
(4)
where the additional variable, Battery, is the dummy variable taking the value of one if the
nearest lead-emitting plant is the battery recycling plant. This model has several advantages.
First, the coefficient, β2, differences out any effects in the neighborhood of battery recycling
plants other than the direct policy effect, including any differential trends between the neigh-
14The average lead concentration level in ambient air at monitors within 1 mile from the lead-emitting plantbefore 2008 is 0.125µg/m3.
15We present the two difference-in-differences models that constitute this triple differences model in the OnlineAppendix III. In those difference-in-differences model, we additionally included the number of other lead-emittingplants interacted with the post dummy to control for the effects of the number of other lead-emitting plants incase other lead-emitting plants are located near battery-recycling plants. Such a variable can also be included inthe triple-differences model, although it is less important because the effects by other lead-emitting plants arealready controlled for. Nonetheless, the results with such a variable included provides quantitatively the sameresults.
10
borhood of battery recycling plants and the neighborhood of other lead-emitting plants. Second,
the coefficient, β3, controls for any differential trends between the close proximity to the lead-
emitting plants (both battery recycling and other lead-emitting plants) and their adjacent areas.
Lastly, the coefficient, β4, which is practically absorbed by the monitor fixed effects, captures
permanent differences in air quality between the close proximity to battery recycling plants and
their adjacent areas. Consequently, the estimated coefficient of interest, β1, can be interpreted
as the policy effect on air quality near battery recycling plants.
Table 2 presents the results based on the triple-differences model as specified by the equation
4. Column (1) uses the sample of monitors up to 2 miles away from the lead-emitting plants, in
which those within 1 mile is defined as the Near area. The estimated coefficient suggests that
the lead concentrations in ambient air near battery-recycling plants declined by 0.172 µg/m3
after the revision of NAAQS, which is 50% reductions from the pre-reform level. We test the
robustness of the preferred estimate in Column (1) to various alternative definitions of the two
areas. Column (2) expands Adjacent area up to 10 miles, Column (3) compares 0-2 miles as
Near and 2-4 as Adjacent, and Column (4) expands Adjacent up to 10 miles, all of which has
virtually no impact on the estimated effect, giving confidence that the finding is not driven by
comparisons of the specific areas.
V. The Effect on Trade and Production
A. The Effect on the US Exports of ULAB: Graphical Evidence
Figure 6 below illustrates the monthly exports of used lead-acid batteries (ULAB)16 from the
US to the two major destinations, Mexico and Canada, which together account for close to
90 percent of total US lead exports. The figure shows a clear pattern. There were few exports
before 2004. While the exports to Mexico increased since 2004, the amount of exports stayed
constant around five hundred thousand ULABs a month up to 2009. Note that there are two
months, May and June 2008, that clearly deviate from the overall trend. This is likely to reflect
the market response to the signed revision of NAAQS on lead on May 1, 2008. In contrast,
the figure illustrates a clear trend break around the beginning of 2009, after which the ULAB
exports to Mexico continued to surge over time.
B. The Effect on the US Exports of ULAB: Regression-Adjusted Evidence
While the figure above provides visual evidence that there exists a trend break of accelerated
ULAB exports to Mexico around 2009, we now conduct a formal test of identifying a potential
trend break. We take two approaches. First, without assuming a relevant control group to
construct a valid counterfactual, we rely on the time-series analysis to look for a structural
break. Without prior knowledge of when the structural break occurs, one can conduct the
Quandt likelihood ratio (QLR) test (Quandt 1960). The basic idea is to test whether, among all
possible break points, the break is strongest, in a statistical sense, in or around the beginning
of 2009. Following Andrews (1993, 2003) and Hansen (2000), we test for a structural break in
16Although ULAB is reported as a commodity itself, it consists of spent lead-acid batteries, spent primarybatteries, waste and scrap of batteries.
11
the amount of ULAB exports in month of the year, τ , by estimating a series of the following
regression:
lnYt = α+ β1τDt(τ)× Trendt + β2τDt(τ) + β3Trendt + εt, (5)
where the dependent variable is the log of number of ULAB exports, is an indicator variable
equal to one for all months after and zero otherwise and Trend is a monthly trend. We then
conduct an F test of the joint significance of β1τ and β2τ for each between 2007 and 2010 using
the sample of 2004-2014.17
Figure 7 plots these test statistics. The figure illustrates a clear pattern. QLR statistics starts
increasing since the beginning of 2009 and peaks in July 2009, while statistically significant trend
breaks at the 1 percent level are observed from May through August 2009 using the asymptotic
critical values provided by Andrews (1993, 2003).18 Evidence of a structural break around the
beginning of 2009 suggest that the revised NAAQS led to a break in the ULAB export trend.
Our second approach to test for a break point is to conduct a difference-in-differences analysis
by comparing the ULAB exports to an appropriate comparison group. We use the exports of
new lead-acid batteries (NLAB) as the control group. The exports of NLAB can serve as the
valid control group for ULAB for three major reasons. First, these two products should be
comparable not only because they are categorized with the same four-digit HTS codes but they
are also essentially the same product and differs only in the extent of usage. Second, to the extent
that the revised NAAQS concerns lead emissions, only recycling of ULAB should be affected,
while production of NLAB should not be subject to the policy impact because its production
emits little lead. Lastly, using NLAB as the comparison group addresses a concern that any
demand shock for LAB in Mexico triggered increased ULAB exports from the US, in which
case exports of both NLAB and ULAB (as a source of lead necessary to produce NLAB) would
increase. Despite the similarity and comparability in the nature of the product characteristics,
there still remains a concern that the exports of NLAB are inherently and historically distinct
from those of ULAB. We can test such differential preexisting trends by directly conducting
an event study and testing deviations in trend for each year.19 In particular, using the sample
period of 2004-2014, we estimate the model;
17Because Figure 6 illustrates another trend break around 2004, we limit the sample for this analysis after2004. The span to test for a structural break is necessarily smaller than the entire sample period because onerequires sufficient data to estimate a trend before and after a potential break point. The literature provides littleguidance over the choice of the appropriate window. Our focus between 2007 and 2010 is primarily motivatedby the trend provided in Figure 6, whereas trimming 15% from the boundaries of the sample, as suggested byAndrews (1993, 2003), has no virtual effect on the result. Based on the evidence suggested by Figure 6, we alsoreplaced the values in May and June 2008 as missing values as these two months are likely to be affected by therevised NAAQS itself, although the result is unchanged without such corrections.
18Andrews (1993, 2003) provides (asymptotic) critical values to account for the fact that there is typically alarge probability that some of the null hypothesis of no structural break is rejected when one tests a set of breakpoints simultaneously. Figure 7 presents the critical value of the QLR statistics at the 1 percent significance levelof 14.1.
19Additionally, one may be concerned that the use of NLAB may not satisfy the SUTVA assumption if domesticLAB production relies on the domestic lead production via recycling of ULAB, which in turn induces the NLABexports. However, as we discuss later, there is no evidence that domestic LAB production fell after 2009. Ifanything, the production was on an increasing trend. Evidence suggests that there is little linkage betweendomestic LAB production and lead production, discarding the channel via which the regulation indirectly affectedNLAB exports.
12
lnYimt = α+∑τ
βτ (ULABi × τmt) + β2ULABi + µt + νm + εimt, (6)
where ULAB is a dummy variable taking the value of one for ULAB and zero for NLAB
and τmt is a dummy variable taking the value of one for a month-of-year, mt, and zero for all
others. β2 captures permanent differences in the amount of exports between ULAB and NLAB.
We additionally include year fixed effects, µt , and month fixed effects, νm, to control for overall
growth of trade over time and seasonality effects, respectively. The parameter of interest is a
vector of βt, which indicate any changes in trend of ULAB exports relative to NLAB exports
compared to the comparison month, which we set at December 2008. The standard error is
clustered at the month-of-year level.
Figure 8 below plots individual coefficients for each month.20 The figure highlights a sharp
relative increase in the exports for ULAB right after 2009. Also, the figure shows that the
coefficients are zero since mid-2004, except the two months affected by signing of the revised
NAAWS, suggesting that ULAB and NLAB would have followed similar trends had NAAQS on
lead not been revised in 2009, confirming the validity of NLAB as a comparison group.
Overall, the results from both approaches mirror each other, presenting strong evidence that
substantially increasing numbers of ULAB were exported from the US to Mexico starting right
after the implementation of the new NAAQS in 2009.
C. Effects on Other Dimensions of the US Economy
In this subsection, we explore the regulation effects on other dimensions, both descriptively and
quantitatively, to understand the overall impacts of the revised NAAQS on the economy. First,
we explore lead production in the US
Figure 3 illustrates lead recovered from ULAB processed in the US ULAB has historically
been the dominant source of recoverable lead scrap, accounting for 94% of all lead produced
from secondary sources in 2009, particularly because primary lead production has fallen over
time. The amount recovered was increasing up to 2007 achieving about 14% increase in lead
production from batteries since 1996. Such trend reversed in 2008, which marks lowered demand
for lead due to recession and potentially a fire at one of the fifteen major plants.21 The production
continued to be stagnant until 2014, when substantial reductions in the amount of secondary
production was observed.
20In case one is interested in knowing the estimate of the simple difference-in-differences model of;
lnYimt = α+ β1ULAB × Postmt + β2ULABi + µt + νm + εimt,
where Post is a dummy taking the value of one for all months after January 2009, the estimated coefficientof interest, β1 is 0.665 (0.065) and statistically significant at the 1% level. The magnitude of the coefficient issubstantial, suggesting more than 60% increase in the amount ULAB exports relative to NLAB exports after2009.
21The RSR Corp. was caught by the fire in early July 2008 at one of their plants in Los Angeles, California, andwas forced to cease its operation until repairs completed in late November. Although the amount of productionlost by the closure is not clear, the smelter has the capacity to produce 10,000 metric tons per month of refinedlead under normal conditions. Suppose RSP had produced 10,000 metric tons for every month during the five ofthe closure, the total lead production in 2008 would have been greater than that of 2007 by 2,000 metric tons.Because a large majority of lead is produced by 7 companies operating 15 plants, an idiosyncratic event like thisat even one place can have substantial impacts on total lead production.
13
The trend observed in Figure 3 is consistent with Figure 4, which illustrate a great extent
of heterogeneity in the policy effects ? while greatest reductions in pollution concentrations
were observed around plants with initially high pollution levels to begin with (new compliers),
some firms were already compliant even before the policy revision (already compliers), and
others remained non-compliant (non-compliers). Note that only new compliers contribute to
the production reduction, whereas both new compliers and already compliers contribute to
increased trades, which necessarily resulted in greater effects on trade pattern than production
pattern.22
For the sake of understanding the overall trend in domestic lead production and portraying
counterfactual of what would have happened had there not been the revision of NAAQS, it
is critical to examine driving factors that constitute domestic demand for lead. To this end,
we illustrate domestic LAB production in Figure 4, domestic vehicle production in Figure 5,
and the total US exports of NLAB as characterizing demand for US produced LAB by the
rest of the world in Figure 6. The LAB industry constitutes the primary demand for domestic
lead production, as it consumes more than 90 percent of lead produced domestically (check
the number). In turn, the demand for LAB is governed by auto production, as these LABs
are primarily used as automobile starting, lighting, ignition batteries, while some are also used
as industrial-type batteries for standby power supplies for telephones and computers. Thus,
productions in these two industries, both domestically and internationally, help understand
domestic lead production.
Overall, the evidence suggests that there was robust demand for lead production from all
three elements after 2009, as economy recovered from the recession. While we observe evident
reductions in vehicle productions and NLAB exports in 2008/2009 due to recession, both made
quick recovery in 2010 and continued to increase afterwards. These trends after 2009 contrasts
with the trend of domestic lead production, as presented in Figure 3, which continued to be
stagnant. The sustained domestic production of LAB and LAB exports suggests that the sec-
ondary lead production should have continued to rise, yet it has failed to recover meaningfully
by being constrained by the NAAQS revision.
If, as evidence above suggests, demand for LAB continued to grow in the US, while do-
mestic lead production is constrained, how did the industry meet supply shortage in LAB? To
investigate this question, we plot the US imports of NLAB in Figure 7 and of lead in Figure
8. These two figures show that the amount of NLAB imports was not affected around 2009,
while the amount of lead imports increased substantially since 2009. Note that these imports do
not necessarily reflect imports from Mexico. While we do observe increased lead imports from
Mexico after 2009, they share only about 20% after 2009. Thus, these two figures indicate that
the US LAB industry increased lead imports to meet the shortage in lead to produce LAB,
which could have been domestically produced without the revised NAAQS.
We aggregate the information at the exporter-product-destination-year level. We further
aggregate all the non-US countries, therefore the destination variable is a dummy indicating
whether the destination is the US or not.
22While new compliers must reduce production to comply with the new standard and increased exports asa consequent, already compliers may also have increased exports because they were constrained in expandingdomestic production.
14
We identify lead-emitting exporters in the Mexican customs data by matching the firm’s
name, address, and a part of tax payer identification code in the RETC. The Mexican exporting
firms that are linked to the RETC dataset account for more than 70% of all the Mexican exports
each year.
D. Effect on the Exports of Mexican Lead-emitting Plants
This section presents the results of the analysis for Mexican exports of lead emitting plants. The
purpose of this analysis is to show that there is an increase in exports to US by Mexican lead
emitting plants, which is consistent with the prediction of the pollution haven hypothesis that
the tightening of the US lead regulation would increase the Mexican comparative advantage in
lead-intensive products.
LogExportsisut = βLeadi ∗ USu ∗ Postt + δtLeadi ∗ Y earDummy + µst + γut + εisut (7)
Exportsisut is exports of firm i in sector s for destination u (U.S. or non-U.S.) at yeat t.
Leadi is a dummy indicating whether firm i owns at least one of its plants register lead as
one of its emitting substances. USu is a dummy indicating whether the destination country is
the U.S. We analyze whether the exports of lead emitting plants to U.S. increased after the
regulation change. This is a triple-difference strategy, where one treatment is lead emitting
plants (as opposed to plants emitting other substances) and the other treatment is US. The
parameter of interest is β, which can be interpreted as a causal effect of the regulation on
exports. We control lead-emitting plants-year effects, sector-year effects, as well as, US-year
effects, which control, respectively, differential supply trends between lead-emitting plants and
non-lead-emitting-plants, differential sectorial trends, and differential demand trends between
the U.S. and other countries.
Table 3 shows the results. Regardless of the definition of sector, we find that the exports to
the U.S. by Mexican lead-emitting firms increased after the regulation change, relative to their
exports to the other destinations, and relative to Mexican firms not emitting lead but other
substances.
The results from the section shows that the improvements in the U.S. domestic environmen-
tal quality documented in Section IV are accompanied by increased U.S. exports of lead contents
to Mexico for recycle and production. In return, exports from lead-emitting Mexican firms to the
U.S. have also increased. These together suggest that lead-emitting production process shifted
from the U.S. to Mexico in a manner consistent with the pollution haven hypothesis.
VI. The Effect on Infant Health in Mexico
A sharp increase in US exports of the ULAB to Mexico we showed in the last section suggests
that there is also a sharp relocation in the recycling of these ULAB from the US to Mexican
battery recycling plants, resulting increases in lead emissions by Mexican recycling plants. We
show the health consequences of such relocations in this section by presenting the results of
15
the analysis for infant heath in Mexico. As we did for the analysis of US lead emission, we
use distance from the nearest authorized battery-recycling plant to define the treatment status
of each residential locality; births adjacent to a battery-recycling plant are considered as the
treatment group and births near but slight away from a battery-recycling plant as the control
group. We compiled the list of the authorized recycling plants at various point of time from the
SEMARNAT website. We limit the sample to municipalities with at least one battery recycling
plants.
For the case of Mexico, the treatment status of lead-emitting plants other than battery
recycling plants is less clear than the US case. On the one hand, they may be indirectly affected
by the change in the US lead regulation through an increase in relative cost advantage in
Mexican production of lead-intensive products. On the other hand, the Mexican production
of lead-intensive products is likely to have its own distinct trends due to factors unrelated to
the US regulation change, such as the growth of Mexican mining activities and the automobile
industry.
Considering these factors, the regression equation for the analysis of this section is the one
below:
Healthijmt = βBatteryj∗Postt+δ(NOtherLeadFirmsj)∗Postt+ρXijmt+µmt+γj+εijmt (8)
where i, j, m and t denote individual, locality, municipality and year, respectively. Healthijmt
is a dummy indicating whether a baby’s birth weight is less than 2500 gram, a commonly used
measure of low-birth weight and infant health. Batteryj , is the dummy variable taking the
value of one if the locality is within 2 miles of a battery recycling plant. NOtherLeadFirmsj is
the number of lead emitting plants that are not battery recycling plants within 2 miles of the
locality. Postt is taking the value of one if year is 2009 or after. Xijmt includes characteristics
of mothers such as age and the number of previous births and their interactions with the post
dummy. The estimated coefficient of interest, β, captures the difference in the changes in the
birth outcome between localities very near from battery recycling plants and those in the same
municipality but slightly far away from battery recycling plants, controlling for changes in
activities by other lead emitting plants and for changes in mother and locality characteristics.
β can be thus interpreted as the effect of the US lead regulation change on infant health near
Mexican battery recycling plants.
A. Hospital Discharge Data
First, we show the results of the regression for the hospital discharge sample. Columns (1)-(3) in
Table 4 show the results of the analysis 2005-2013 period. Column (1) shows that The coefficient
on Batteryj × Postt is 0.023 and statistically significant at the conventional level. The mean
pre-2009 low birth weight rate is around 10 percent. This implies that the affected locality saw
an increase of the incidence of low birth weight by 23 percent, a large effect. Columns (2) and
(3) show that the effect is robust to controlling for the presence of other lead-emitting plants
and changes associated with it, as well as mother characteristics. The patterns of the coefficients
16
suggest that the presence of other lead emitting plants is associated with improvement of the
birth outcome. This may be because the presence of lead emitting plants and its interaction
with the post dummy may be capturing local economic growth if there is a boom in economic
activities which use lead intensively.
Columns (4)-(6) show the results of the placebo analysis 2005-2008 period treating counter-
factually 2007 and 2008 as the post period. None of the coefficient on Batteryj × Postt is
statistically significant at the conventional level and is very close to zero once the presence
of other lead-emitting plants is controlled for. This suggests that the localities very close to
battery recycling plants and those slightly far away had had an identical trend prior to the lead
regulation change in the US that increased the Mexican imports from US of ULAB, providing
support for the assumption for the difference in differences analysis.
Next, we run an alternative regression where we replace Batteryj × Postt with Batteryj
interacted with each year dummy, with the omitted category being 2008, and show how the
coefficients evolve over years in Figure 9. The figure shows that the difference in the fraction of
low-birth weight births of the localities very close to battery recycling plants from those slightly
far away had been shrinking if any but rapidly expanded exactly after the US regulation change.
B. Birth Certificate Data
We now show the results of the regression for the hospital discharge sample, covering all the
births but for a shorter period starting 2008. Table 5 shows the results. Columns (1)(2) show
the results for all the births near and slightly far away battery recycling plants, Columns (3)(4)
show the results restricting to the births born at one of the Ministry of Health hospitals, and
Columns (5)(6) show the results restricting to the births born at all other hospitals. Columns
(3)(4) confirm the results in the previous section. Interestingly and importantly, the negative
effect of the increases in Mexican battery recycling due to the US regulation change exists only
for babies who are born at one of Ministry of Health hospitals, which serve primarily families
without social security access. This means that such effects are concentrated on disadvantaged
families. This may be because mothers in disadvantaged families are arguably more likely to live,
work and commute in an environment with higher exposure to outside air. Mothers in disad-
vantaged families may be less knowledgeable about the danger of lead and thus be less involved
in avoidance behavior. Regardless of the mechanism, the result has an important implication
for policy makers in designing policies to mitigate the negative health effect.
VII. Conclusion
This study examines the effect of variation in the environmental regulatory stringency on relo-
cation of pollution intensive production as a test of pollution haven hypothesis. In particular,
we exploit the tightened NAAQS on lead in the US in 2008. Our findings are three-hold. i) The
new NAAQS effectively constrained domestic lead production in the US, resulting in substantial
reductions in lead in ambient air near battery-recycling plants. ii) The battery recycling plants
increased exports of ULAB to Mexico as a way to circumvent the regulatory stringency and
lower compliance costs with weaker environmental regulation. iii) The increased lead produc-
17
tion in Mexico worsened health status among Mexican newborns near battery-recycling plants.
Overall, our findings provide support to the theoretical prediction that unbalanced stringency
in environmental standards may spur flows of pollution intensive activities to countries with lax
environmental standards.
Our study offers an important policy implication for the contemporary debate in the interna-
tional climate change conferences that simply tightening standards in developed countries only
will end up with offshoring pollution and health risks to developing countries. It is, therefore,
necessary to design policies that limit the scope of such negative externalities to low-income
countries.
18
References
Aizer, A., J. Currie, P. Simion, and P. Vivier (2015). Do low levels of blood lead reduce children’s
future test scores?
Andrews, D. W. K. (1993). Tests for parameter instability and structural change with unknown
change point. Econometrica 61 (4), 821–856.
Andrews, D. W. K. (2003). Tests for parameter instability and structural change with unknown
change point: A corrigendum. Econometrica 71 (1), 395–397.
Atkin, D. (2013). Trade, tastes and nutrition in india. American Economic Review 103 (5),
1629–1663.
Becker, R. and V. Henderson (2000). Effects of air quality regulations on polluting industries.
Journal of political Economy 108 (2), 379–421.
Bennear, L. S. (2008). What do we really know? the effect of reporting thresholds on inferences
using environmental right-to-know data. Regulation and Governance 2 (3), 293–315.
Berkowitzn, Z., P. Price-Green, F. Bove, and W. Kaye (2006). Lead exposure and birth out-
comes in five communities in shoshone county, idaho. International Journal of Hygiene and
Environmental Health 209, 123–132.
Bombardini, M. and B. L. Li (2016). Trade, pollution and mortality in china. mimeo.
Braun, J., T. Froehlich, J. Daniels, K. Dietrich, R. Hornung, P. Auinger, and B. Lanphear
(2008). Association of environmental toxicants and conduct disorder in u.s. children: Nhanes
2001-2004. Environmental Health Perspectives 116 (7), 956–962.
Brunnermeier, S. and A. Levinson (2004). Examining the evidence on environmental regulations
and industry location. Journal of Environment & Development 13 (6-41).
Burns, J., P. Baghurst, M. Sawyer, A. McMichael, and S. Tong (1999). Lifetime low-level
exposure to environmental lead and children’s emotional and behavioral development at ages
11–13 years: The port pirie cohort study. American Journal of Epidemiology 149, 740–749.
Cecil, KM, B. C. A. C. D. K. A. M. E. J., S. Wessel, I. Elangovan, R. Hornung, K. Jarvis, and
B. Lanphear (2008). Decreased brain volume in adults with childhood lead exposure. PLoS
Medicine 5 (5), e112.
Chung, S. (2014). Environmental regulation and foreign direct investment: Evidence from south
korea. Journal of Development Economics 108 (C), 222–236.
Copeland, B. and M. Taylor (2004). Trade, growth, and the environment. Journal of Economic
Literature 42, 7–71.
Currie, J., L. Davis, M. Greenstone, and R. Walker (105). Environmental health risks and
housing values: Evidence from 1,600 toxic plant openings and closings. American Economic
Review 2, 678–709.
19
Davis, L. W. and M. E. Kahn (2010). International trade in used vehicles: The environmental
consequences of nafta. American Economic Journal: Economic Policy 2 (4), 58–82.
de Marchi, S. and J. T. Hamilton (2006). Assessing the accuracy of self-reported data: An
evaluation of the toxics release inventory. Journal of Risk and Uncertainty 32 (1), 57–76.
Dietrich, K., M. Ris, P. Succop, O. Berger, and R. Bornschein (2001). Early exposure to lead
and juvenile delinquency. Neurotoxicol and Teratology 23 (6), 511–518.
Emory, E., R. Pattillo, E. Archibold, M. Bayorh, and F. Sung (1999). Neurobehavioral effects
of low-level lead exposure in human neonates. American Journal of Obstetrics and Gynecol-
ogy 181, S2–S11.
Eskeland, G. and A. Harrison (2003). Moving to greener pastures? multinationals and the
pollution haven hypothesis. Journal of Development Economics 70 (1), 1–23.
Greenstone, M. (2002). The impacts of environmental regulations on industrial activity: Ev-
idence from the 1970 and 1977 clean air act amendments and the census of manufactures.
Journal of Political Economy 110 (6), 1175–1219.
Gronqvist, H., J. Nilsson, and P. Robling (2014). Early childhood lead exposure and criminal
behavior: Lessons from the swedish phase-out of leaded gasoline? mimeo.
Hanna, R. (2010). Us environmental regulation and fdi: evidence from a panel of us-based
multinational firms. American Economic Journal: Applied Economics 2, 158–189.
Hansen, B. E. (2000). Testing for structural change in conditional models. Journal of Econo-
metrics 97 (1), 93–115.
Henderson, J. (1996). Effects of air quality regulation. American Economic Review 86 (4),
789–813.
Hu, H., M. Tellez-rojo, D. Bellinger, D. Smith, A. Ettinger, H. Lamadrid-Figueroa, J. Schwartz,
L. Schnaas, A. Mercado-Garcia, and M. Hernandez-Avila (2006). Fetal lead exposure at
each stage of pregnancy as a predictor of infant mental development. Environmental Health
Perspectives 114, 1730–1735.
Jaffe, A., P. S. P. P. and R. Stavins (1995). Environmental regulations and the competitiveness
of us manufacturing: what does the evidence tell us? Journal of Economic Literature 33 (132-
163).
Javorcik, B. and S. Wei (2003). Pollution havens and foreign direct investment: dirty secret or
popular myth? The B.E. Journal of Economic Analysis & Policy 3 (2), 1–34.
Kellenberg, D. (2009). An empirical investigation of the pollution haven effect with strategic
environment and trade policy. Journal of International Economics 78 (2), 242–255.
Keller, W. and A. Levinson (2002). Pollution abatement costs and foreign direct investment
inflows to u.s. states. Review of Economics and Statistics 84 (4), 691–703.
20
Koehler, D. A. and J. D. Spengler (2007). The toxic release inventory: Fact or fiction? a case
study of the primary aluminum industry. Journal of Environmental Management 85 (2),
296–307.
Lanphear, B., R. Hornung, J. Khoury, K. Yolton, P. Baghurst, D. Bellinger, R. Canfield, K. Di-
etrich, R. Bornschein, T. Greene, S. Rothenberg, H. Needleman, L. Schnaas, G. Wasser-
man, J. Graziano, and R. Roberts (2005). Low-level environmental lead exposure and chil-
dren’s intellectual function: An international pooled analysis. Environmental Health Perspec-
tives 113 (7), 894–899.
Levinson, A. (1996). Environmental regulations and manufacturers’ location choices: Evidence
from the census of manufactures. Journal of Public Economics 62 (1–2), 5–29.
Levinson, A. (2008). New Palgrave Dictionary of Economics, Chapter Pollution Haven Hypoth-
esis.
Linden, L. and J. E. Rockoff (2008). Estimates of the impact of crime risk on property values
from megan’s laws. American Economic Review 98 (3), 1103–1127.
Mirghani, Z. (2010). Effect of low lead exposure on gestational age, birth weight and premature
rupture of the membrane. Journal of Pakistan Medical Association 60 (12), 1027–1030.
Muehlenbachs, L., E. Spiller, and C. Timmins (2015). The housing market impacts of shale gas
development. American Economic Review 105 (12), 3633–3659.
Needleman, H., A. Schell, D. Bellinger, A. Leviton, and E. Allred (1990). The long-term effects
of exposure to low doses of lead in childhood. an 11-year follow-up report. New England
Journal of Medicine 322 (2), 83–88.
Nevin, R. (2000). How lead exposure relates to temporal changes in iq, violent crime and unwed
pregnancy. Environmental Research 89, 1–22.
Nigg, J., G. Knottnerus, M. Martel, M. Nikolas, K. Cavanagh, W. Karmaus, and M. Rap-
pley (2008). Low blood lead levels associated with clinically diagnosed attention-
deficit/hyperactivity disorder and mediated by weak cognitive control. Biological Psychi-
atry 63 (3), 325–331.
Nilsson, J. (2009). The long-term effects of early childhood lead exposure: Evidence from the
phase-out of leaded gasoline. mimeo.
Pierce, J. and P. Schott (2016). Trade liberalization and mortality: Evidence from u.s. counties.
mimeo.
Pocock, S., M. Smith, and P. Baghurst (1994). Environmental lead and children’s intelligence:
a systematic review of the epidemiological evidence. British Medical Journal 309 (6963),
1189–1197.
Quandt, R. (1960). Tests of the hypothesis that a linear regression system obeys two separate
regimes. Journal of American Statistical Association 55, 320–330.
21
Rau, T., L. Reyes, and S. Urzua (2013). he Long-term Effects of Early Lead Exposure: Evidence
from a Case of Environmental Negligence. NBER WP 1891.
Reyes, J. (2007). Environmental policy as social policy? the impact of childhood lead exposure
on crime. The B.E. Journal of Economic Analysis and Policy 7 (1).
Schnaas, L., S. Rothenberg, M. Flores, S. Martinez, C. Hernandez, E. Osorio, S. Velasco, and
E. Perroni (2006). Reduced intellectual development in children with prenatal lead exposure.
Environmental Health Perspectives 114, 791–797.
Schwartz, J. (1994). Low-level lead exposure and children’s iq: a meta- analysis and search for
a threshold. Environmental Research 65 (1), 42–55.
Torres-Sanchez, L., G. Berkowitz, L. Lopez-Carrillo, L. Torres-Arreola, C. Rios, and M. Lopez-
Cervantes (1999). Intrauterine lead exposure and preterm birth. Environmental Research
Section A81 (297-301).
Wright, J., K. Dietrich, D. Ris, R. Hornung, S. Wessel, B. Lanphear, M. Ho, and M. Rae (2008).
Association of prenatal and childhood blood lead concentrations with criminal arrests in early
adulthood. PLoS Medicine 5 (5), e101.
22
Figures
Figure 1: Air Quality over Distance
Notes: Cross-sectional variation in average lead concentration levels inambient air at monitoring stations over distance from the nearest lead-emitting facility, separately for the pre-reform period (years in and before2008) in a blue dashed line and for the post-reform period (years after2009) in a red solid line.
23
Figure 2: Air Quality Over Time
Notes: Time-series variation in average lead concentration levels in ambi-ent air at monitoring stations between 2001 and 2013, separately for themonitoring stations within 1 mile from the nearest lead-emitting facilityin a red solid line and for the monitoring stations between 1 to 2 milesfrom the nearest lead-emitting facility in a blue dashed line. The blackdashed vertical line at January 2009 indicates the timing of the new EPAlead regulation implementation.
Figure 3: Battery Recycling Plants vs. Other Lead-emitting Plants
Notes: Time-series variation in average lead concentration levels in ambi-ent air at monitoring stations between 2001 and 2013, separately for themonitoring stations within 1 mile from the nearest battery-recycling facil-ity in a red solid line and for the monitoring stations within 1 mile fromany other lead-emitting plant in a blue dashed line. The black dashedvertical line at January 2009 indicates the timing of the new EPA leadregulation implementation.
24
Figure 4: Lead Concentrations near Battery-recycling Plants
Notes: This figure presents the lead concentrations before the policychange on the x-axis and after on the y-axis. Each dot represents theaverage using monitors within 1 mile from each battery-recycling plants.The dashed lines indicate the revised NAAQS on lead, while the solid lineindicates the 45-degree line.
Figure 5: Distance gradient of the policy effect
Notes: This figure illustrates distance gradient of the policy effect on leadconcentrations in ambient air estimated based on equation 2 using thefifth-order of polynomial in distance from monitors to the nearest lead-emitting plants. The underlying estimates are presented in online ap-pendix table 1.
25
Figure 6: The US Monthly Exports of ULAB
Notes: Each dot in the figure above indicates the amount of ULABexported from the US to Mexico (in green) and Canada (in orange).The corresponding lines indicate smoothed local polynomial trends withthe bandwidth of three months. The trend for Mexico does not considerMay and June, 2008, as these two deviate from the trend due to theannouncement of the NAAQS revision.
Data source: The US ITC with corrections provided by the USCensus Bureau Foreign Trade Division
Alternative Figure to Figure 6
Notes: The figure above illustrates the monthly exports of ULAB fromthe US to Mexico (in blue) and Canada (in red).
Data source: The US ITC with corrections provided by the USCensus Bureau Foreign Trade Division
26
Figure 7: Testing for a Trend Break in ULAB Exports
Notes: The figure above presents the QLR statistics for a potentialtrend break in months between 2007 and 2010 using the sample of2004-2014. The asymptotic critical value at the 1 percent significancelevel is provided by Andrews (1993, 2003).
Figure 8: Event-study Estimates
Notes: The figure plots individual coefficients of the interactions betweenULAB and month-of-year dummies as specified in Equation 6. Thecomparison period is December 2008.
27
Figure 9: The Effects on the Incidence of Low-Birth-Weight Babies in Mexico
Notes: The figure plots individual coefficients of the interactions betweenthe dummy indicating being near from a battery recycling plant andyear dummies in a regression of a dummy indicating the low birth weighton the interactions above controlling for mother characteristics and thenumber of other lead emitting plants and their interactions with the post2009 dummy. The comparison period is 2008. The dashed lines show the95 percent confidence intervals.
28
Tables
Table 1: The Policy Effect on Air Quality near Lead-emitting Plants
(1) (2) (3) (4) (5)
Near×Post -0.0463** -0.0677*** -0.0777*** -0.0458** -0.0472***(0.0217) (0.0159) (0.0211) (0.0203) (0.0130)
N 25,779 25,779 19,149 13,836 29,638
Treat 0-1 0-1 0-0.5 0.5-1 0-2Control 1-2 1-2 1-2 1-2 2-4Monitor FE N Y Y Y YNum of monitors 439 439 319 257 512
Notes: This table presents estimates based on equation 3 using various treatment and controlgroups as specified in the respective column. All specifications include time fixed effects. Standarderrors are clustered at the monitor level.
Table 2: The Policy Effect on Air Quality near Battery-recycling Plants
(1) (2) (3) (4)
Near × Post × Battery -0.172*** -0.172** -0.157** -0.157***(0.0626) (0.0726) (0.0644) (0.0594)
N 25,779 19,149 13,836 29,638
Near 0-1 0-0.5 0.5-1 0-2Adjacent 1-2 1-2 1-2 2-4Mean 0.344 0.353 0.308 0.359Num of monitors 439 319 257 512
29
Table 3: The Policy Effect on Mexican firm-product level Exports: DDD
Sector-level Product-level(1) (2)
Lead × US × Post 0.184** 0.307**(0.069) (0.053)
N 171,561 536,727
Notes: The dependent variables are Log Exports at the Firm-HS-Chapter level(the first two digit of the HS code) and the Firm-HS6 digit level. Both specifi-cations include Leat-emitting-plants-year interaction effects, Chapter (or HS6)-year interaction effects, and US-year interaction effects. All standard errors areclustered at the firm level.
Table 4: The Policy Effect on Birth Weights of Mexican New Born Babies
(1) (2) (3) (4) (5) (6)Analysis Main Main Main Placebo Placebo PlaceboPeriod 2005-13 2005-13 2005-13 2005-08 2005-08 2005-08
Battery × Post 2009 0.023** 0.031** 0.031**(0.009) (0.008) (0.008)
N Other lead × Post 2009 -0.005** -0.005*(0.002) (0.002)
Battery × Post 2007 -0.020 -0.001 -0.001(0.017) (0.021) (0.021)
N Other lead × Post 2007 -0.007 -0.007(0.010) (0.010)
Mother Control No No Yes No No YesMunicipality × Year Yes Yes Yes Yes Yes YesN 248,327 248,327 248,327 96,839 96, 839 96, 839
Near 0-2 0-2 0-2 0-2 0-2 0-2Adjacent 2-4 2-4 2-4 2-4 2-4 2-4
Notes: The dependent variable is a dummy indicating the low birth weight(birth weight less than 2500 gram). All the specifications include municipality-year effects All standard errors are clustered at the locality level.
30
Table 5: The Policy Effect on Birth Weights of Mexican New Born Babies: Birth Certificate
(1) (2) (3) (4) (5) (6)Analysis All Ministry of Health OthersPeriod 2008-2013
Battery × Post 2009 0.002 0.013 0.013 0.031** -0.011 -0.004(0.010) (0.010) (0.014) (0.013) (0.013) (0.014)
N Other lead × Post 2009 -0.002** -0.003* 0.001(0.001) (0.001) (0.001)
Mother Control No Yes No Yes No YesMunicipality × Year Yes Yes Yes Yes Yes YesN 319,412 319,412 142,892 142,892 176, 520 176, 520
Near 0-2 0-2 0-2 0-2 0-2 0-2Adjacent 2-4 2-4 2-4 2-4 2-4 2-4
Notes: The dependent variable is a dummy indicating the low birth weight(birth weight less than 2500 gram). All the specifications include municipality-year effects All standard errors are clustered at the locality level.
31
Not for Publication
Offshoring Health Risks: The Impact of the US Lead
Regulation on Infant Health in Mexico
Online Appendix
I. Distance Gradient of the PolicyEffect
Figure 1: Distance gradient of the policy effect
Panel A: Quadratic
Panel B: Cubic
1
Panel C: Quartic
Panel D: Quintic
Notes: This figure illustrates distance gradient of the policy effect on leadconcentrations in ambient air estimated based on equation 2 using thevarious polynomial in distance, as specified in the panel, from monitors tothe nearest lead-emitting plants. The underlying estimates are presentedin online appendix Table 1.
2
Table 1: Distance Gradient of the Policy Effect
(1) (2) (3) (4)
Distance × post 0.0503*** 0.0892*** 0.141** 0.172(0.0146) (0.0291) (0.0607) (0.106)
Distance2 × post -0.00502*** -0.0211*** -0.0564* -0.0884(0.00158) (0.00816) (0.0305) (0.0786)
Distance3 × post 0.00137** 0.00838 0.0197(0.000587) (0.00513) (0.0224)
Distance4 × post -0.000408 -0.00197(0.000268) (0.00267)
Distance5 × post 0.0000723(0.000112)
Post -0.143*** -0.158*** -0.172*** -0.178***(0.0381) (0.0413) (0.0456) (0.0509)
N 30,874 30,874 30,874 30,874Monitor FE Y Y Y Y
Notes: This table presents estimates based on equation 2 using the various polynomialin distance necessary to produce Figure 1
3
II. Illustrating the Empirical Strategy
In this section, we illustrate empirical strategies employed in the main analysis and additional
robustness checks conducted in the Online Appendix below in greater detail. To begin, Figure
2 is useful in understanding the sample areas examined. Area A represents a buffer around
a battery-recycling plant (battery), defined by Near in the main analysis, and Area B is its
adjacent area. Likewise, Area C represents a buffer around other lead-emitting plant (other),
and Area D is its adjacent area. Note that Area A and B are defined not in exclusive of being C
or D. That is, Area A and B are defined simply based on the distance between a monitor and
the nearest battery-recycling plant, and there may exist other lead-emitting plants in Areas A
and/or B. The converse is also true.
With these four defined areas, we can describe that air pollution, P in each area is comprised
of:
PA = P battery +Near +Macro
PB = Macro
PC = P other +Near +Macro
PD = Macro
(9)
It is clear that a simple comparison of air quality between the two adjacent areas (i.e., Area
A vs. Area B or Area C vs. D) does not produce changes in pollution emitted because the
effect of being located near lead-emitting plants, Near, remains. For example, lead emitted
from these plants may be deposited in soil and continue to pollute ambient air for a long time
as wind blows top soil.
The main advantage of the difference-in-differences (DD) model employed in the main anal-
ysis is to compare changes in air quality, from before to after the policy change, between areas
near and slightly farther away from lead-emitting plants. In particular, using all four areas,
A+B + C +D, it estimates:
DD1 = ∆PA+C −∆PB+D (10)
where ∆ refers to the changes from before to after the policy change. Assuming that the effect
by being near lead-emitting plants (whether it is a battery-recycling plant or other lead-emitting
plant) is time-invariant, (i.e., Near is constant over time because lead deposited in soil is likely
to stay for many years), the DD estimate differences out both time trends and permanent
differences across areas, allowing to disentangle the policy effect on air quality.
We can relax some of the assumptions above by employing several variations in the DD
approach, which motivates the triple-difference (DDD) approach in the main analysis. First,
we relax the assumptions that the policy effect is homogenous between battery-recycling plants
and other lead-emitting plants. That is, ∆Pbattery 6= ∆Pother. To highlight the policy effect on
air quality in the vicinity of battery-recycling plants, we can focus only on Areas A and B and
estimate:
DD2 = ∆PA −∆PB (11)
4
DD2 can estimate the policy effect on air quality near battery-recycling plants with the condi-
tions that Area B provides counterfactual of what would have happened to air quality in Area
A had there not been the policy change. That is Near is constant over time in Area A, and
Macro is the same between Areas A and B. We can alternatively relax the first assumption
and be independent of the second assumption by estimating:
DD3 = ∆PA −∆PC (12)
This recovers the policy effect on air quality near battery-recycling plants with the condition
that any changes in air quality near lead-emitting plants other than due to battery-recycling are
homogenous between battery-recycling and other lead-emitting plants (i.e. The combination of
Near and Macro effects is similar).
DD2 and DD3 motivate the DDD approach:
DDD = (∆PA −∆PB)− (∆PC −∆PD) (13)
The main attraction is that we no longer require the assumption that Areas A and B follow
the parallel trend as required in estimating DD2, or Areas A and C follow the parallel trend as
required in estimating DD3. Instead, we can directly control for differential trends. In particular,
we control for any differential patterns in growth between Areas A and B by differencing out
the differences in growth between Areas C and D. The identification assumption is that the
second difference provides the counterfactual of the first difference had there not been the policy
reform.
Figure 2: The Sample Areas Examined
A
B
C
D
-- Battery recycling plant-- Other lead-emitting plant
Notes: This figure illustrates the sample areas examined in the analysis.See the text for detailed definitions of the four areas.
5
III. Additional Robustness Checks: The Policy Effect on Lead
in Ambient Air
In this section, we provide additional robustness checks to the main policy effect on lead con-
centrations in ambient air based on various DD and DDD models.
First, we test the robustness of the main policy effect on lead near lead-emitting plants, as
presented in Table 1 in the main text, using various using various alternative definitions of Near
and Adjacent areas. Table 2 shows that the point estimates are robust and stable in Columns
(1) through (5). The point estimate substantially falls and no longer statistically significant
when we compare the areas in 1-2 miles away and 2-4 miles away from lead-emitting plants,
suggesting that the policy effect concentrates within 1 mile.
Second, we test the robustness of the main policy effect on lead near lead-emitting plants, as
presented in Table 1 in the main text, using plant fixed effects instead of monitor fixed effects.
As described in the main text, the disadvantage with including plant fixed effects is that the
point estimate may be understated due to the fact that more monitors were required to be build
near lead-emitting sources. On the other hand, the advantages with the plant fixed effects are
that it not only enhances statistical efficiency by utilizing all available data, but it also allows
comparing the Near and Adjacent for a given plant. In particular, we run the model:
Leadjt = α+ βNearj × Postt +Nearj + µt + ηp + εjt, (14)
where ηp denotes the plant fixed effects. The standard errors are clustered at the plant level.
Table 3 presents the results, analogous to Table 1, based on Equation 14 above. As predicted,
the point estimates become slightly lower in magnitude compared with those with monitor fixed
effects, yet they remain similar in magnitude. The finding reassures that the main results are
not driven by inappropriate comparisons of monitors at different locations. We also report the
estimates analogous to the Online Appendix 2, which leads to the same conclusion.
Third, we present the policy effect on lead using various other DD models. In particular, we
estimate the DD3 using Areas A and C based on the regression:
Leadjt = α+ β1Batteryj × Postt + β2Otherj × Postt + µt + γj + εjt, (15)
where Battery is a dummy variable for the area in the vicinity of a battery-recycling plant.
Because monitors near a battery-recycling plant may also be near other lead-emitting plants,
we also include the number of other lead-emitting plants, Other, interacted with the post-
reform period dummy. The results in Table 5 show substantially greater reductions in lead
concentrations in ambient air near battery-recycling plants than those near other lead-emitting
plants, highlighting that the policy effect concentrates in the vicinity of battery-recycling plants.
The estimates are robust to the inclusion of the number of other lead-emitting plants and
alternative sample distance.
Fourth, we estimate the DD2 using Areas A and B based on the regression:
Leadjt = α+ β1Nearj × Postt + β2Otherj × Postt + µt + γj + εjt. (16)
6
The results presented in Table 6 provides consistent results as above.
Finally, the DD2 and DD3 estimates motivate the triple-difference (DDD) estimate based
on:
Leadjt =α+ β1Nearj ×Batteryj × Postt+ β2Batteryj × Postt + β3Nearj × Postt + β4Nearj ×Batteryj+ µt + γj + εjt,
(17)
Table 12 presents the DDD estimate based on Equation 17. Again, it shows that there was little
effect on NOX and PM2.5.
Overall, the findings in this section suggest that lead is the largest primary pollutant emitted
from battery-recycling plants during the process of recycling ULAB to produce lead, and the
revised NAAQS had only affected lead concentrations in the vicinity of battery-recycling plants
but not the other pollutants.
Table 2: The Policy Effect on Lead near Lead-emitting Plants: Alternative Definitions
(1) (2) (3) (4) (5) (6)
Near × post -0.0671*** -0.0674*** -0.0472*** -0.0480*** -0.0663*** 0.00459(0.0160) (0.0159) (0.0130) (0.0126) (0.0168) (0.00548)
N 29,908 30,874 29,908 30,874 22,432 11,065
Near 0-1 0-1 0-2 0-2 0-1 1-2Adjacent 1-5 1-10 2-5 2-10 2-4 2-4Monitor FE Y Y Y Y Y YNum of monitors 522 543 522 543 375 210
Notes: This table presents DD1 estimates based on equation 3 using various alternative treatment and controlgroups as specified in the respective column. All specifications include time fixed effects. Standard errors areclustered at the monitor level.
Table 3: The Policy Effect on Lead near Lead-emitting Plants using Plant Fixed Effects
(1) (2) (3) (4) (5)
Near × post -0.0463*** -0.0490** -0.0673 -0.0412* -0.0658**(0.0155) (0.0225) (0.0423) (0.0246) (0.0288)
Near 0.0912** 0.181* 0.288 0.131*** 0.0211**(0.0381) (0.103) (0.207) (0.0466) (0.0106)
N 25,779 25,779 19,149 13,836 29,638
Near 0-1 0-1 0-0.5 0.5-1 0-2Adjacent 1-2 1-2 1-2 1-2 2-4Plant FE N Y Y Y YNum of plants 332 332 255 214 392
Notes: This table presents estimates based on equation 14 using various treatment andcontrol groups as specified in the respective column. All specifications include time fixedeffects. Standard errors are clustered at the plant level.
7
Table 4: The Policy Effect on Lead near Lead-emitting Plants using Plant Fixed Effects: Alter-native Definitions
(1) (2) (3) (4) (5) (6)
Near × post -0.0583** -0.0602** -0.0658** -0.0670** -0.0836** -0.00729(0.0231) (0.0233) (0.0287) (0.0281) (0.0353) (0.0113)
Near 0.153* 0.154* 0.0211** 0.0213** 0.0305* 0.00998*(0.0912) (0.0914) (0.0106) (0.0104) (0.0159) (0.00507)
N 29,908 30,874 29,908 30,874 22,432 11,065
Treat 0-1 0-1 0-2 0-2 0-1 1-2Control 1-5 1-10 2-5 2-10 2-4 2-4Plant FE Y Y Y Y Y YNum of plants 402 419 402 419 288 183
Notes: This table presents estimates based on equation 14 using various alternative treatment andcontrol groups as specified in the respective column. All specifications include time fixed effects. Thestandard errors are clustered at the plant level.
Table 5: The Policy Effect on Lead near Battery-recycling Plants compared with near OtherLead-emitting Plants
(1) (2) (3) (4) (5)
Battery × Post -0.162** -0.154** -0.157*** -0.164** -0.147**(0.0629) (0.0688) (0.0594) (0.0730) (0.0648)
# Lead emitting plants × Post 0.00489(0.00824)
N 18,573 18,573 25,779 11,943 6,630
Sample distance 0-1 0-1 0-2 0-0.5 0.5-1Mean 0.344 0.344 0.359 0.353 0.308Num of monitors 302 302 439 182 120
Notes: This table presents DD3 estimates based on equation 15. The sample includes all monitors inthe distance, as specified in the respective column, from any lead-emitting plants (Areas A + C in theonline appendix figure 2). All specifications include monitor and time fixed effects. The standard errors areclustered at the monitor level.
8
Table 6: The Policy Effect on Lead near Battery-recycling Plants compared with Adjacent Areas
(1) (2) (3) (4) (5)
Near × Post -0.208*** -0.174** -0.210*** -0.227*** -0.0923*(0.0632) (0.0714) (0.0733) (0.0227) (0.0507)
# Lead emitting plants × Post -0.047(0.0546)
N 2,662 2,662 2,149 765 2,662
Near 0-1 0-1 0-0.5 0.5-1 0-2Adjacent 1-2 1-2 1-2 1-2 2-4Mean 0.344 0.344 0.353 0.308 0.359Num of monitors 33 33 26 14 33
Notes: This table presents estimates based on equation 16. The sample includes monitors near batteryrecycling plants only (Areas A + B in the online appendix figure 2). All specifications include monitor andtime fixed effects. The standard errors are clustered at the monitor level.
9
IV. The Policy Effect on Other Pollutants
In this section, we investigate the effects of the revised NAAQS on other pollutants in the
ambient air in the vicinity of the lead-emitting plants. Lead-emitting plants do emit various
other pollutants. As the reference of types of chemical emitted from these plants, Table 7
presents the 20 largest volumes of chemicals emitted from these lead-emitting plants, and Table
8 presents the analogous chemicals emitted by the battery-recycling plants. While lead-emitting
plants in general emit zinc compounds in the largest volume, followed by lead compounds,
battery-recycling plants primarily emit lead and lead compounds.
Table 9 presents the DD1 estimates using all lead-emitting plants. It compares areas within
1 mile from a lead-emitting plant with those 1-2 mile away from a lead-emitting plant. We find
that the policy had little impact on other major pollutants in the air. All point estimates are
statistically indistinguishable from zero and economically not significant. The point estimates
suggest that there was only about 1-2% reduction in these pollutants. The results highlight one
of the advantages in our empirical framework —while other pollutants such as PM2.5 often mask
the emission sources, we can isolate the specific pollutant, lead, to highlight the policy effect.
Before we proceed to present analogous results based on a variant of DD models, let us note
that we slightly modify the way to construct Areas A and B. While the analysis thus far defined
these areas based on the nearest plant, in the following analysis, we prioritize defining Areas A
and B over C and D. That is, a monitor is in Area A if it is located in the close vicinity of the
battery-recycling plant as defined by Near even if there exists other lead-emitting plant in a
shorter distance. This practice is necessary due to the data limitation —because monitors near
battery-recycling plants tend to report only lead and not other pollutants possibly because lead
is the primary concern. Even with this practice, there is no monitor that reports PM10 and SO2
in the post-reform period, so the DD estimate is not feasible for them, and there exists only one
monitor near the battery-recycling plant that reports NOX in both pre- and post-reform period
within 2 mile of the battery-recycling plant.
That being said, Table 10 presents the DD3 estimates using areas within 1 mile from battery-
recycling plants and other lead-emitting plants based on Equation ??. Note that because mon-
itors near battery-recycling plants report only before the policy change, the DD estimate is not
feasible for them. The estimated effect shows that NOX increased near battery-recycling plants,
while PM2.5 decreased near battery-recycling plants, although the magnitude of the latter effect
is small.
Additionally, Table 11 reports the DD2 estimates using areas in the close proximity to
battery-recycling plants and their adjacent areas. Due to the limited number of monitors away
from these plants (because monitors are in general located near the sources), we expand the
adjacent area up to 10 mile. The finding is similar to Table 10.
10
Table 7: Chemicals Released from All Lead-emitting Plants
Chemical name CAS number Total release
ZINC COMPOUNDS N982 9.69LEAD COMPOUNDS N420 6.19HYDROCHLORIC ACID (1995 AND AFTER ’ACIDAEROSOLS’ ONLY)
007647010 5.13
BARIUM COMPOUNDS N040 2.99COPPER COMPOUNDS N100 2.8MANGANESE COMPOUNDS N450 2.66ARSENIC COMPOUNDS N020 2.61NITRATE COMPOUNDS N511 1.82SULFURIC ACID (1994 AND AFTER ’ACIDAEROSOLS’ ONLY)
007664939 1.61
METHANOL 000067561 1.6AMMONIA 007664417 1.09HYDROGEN FLUORIDE 007664393 0.72CHROMIUM COMPOUNDS(EXCEPT CHROMITEORE MINED IN THE TRANSVAAL REGION)
N090 0.62
VANADIUM COMPOUNDS N770 0.59ZINC (FUME OR DUST) 007440666 0.55NICKEL COMPOUNDS N495 0.4ALUMINUM (FUME OR DUST) 007429905 0.32LEAD 007439921 0.23XYLENE (MIXED ISOMERS) 001330207 0.21TOLUENE 000108883 0.21
Notes: This table reports the largest twenty total amounts of chemicals released from 12,853 lead-emitting plants over the period of 2001-2013 The unit of total release is billion pounds.
11
Table 8: Chemicals Released from Battery Recycling Plants
Chemical name CAS number Total release
LEAD COMPOUNDS N420 100.36LEAD 007439921 30.39ANTIMONY COMPOUNDS N010 13.29ANTIMONY 007440360 8.38CHROMIUM COMPOUNDS(EXCEPT CHROMITEORE MINED IN THE TRANSVAAL REGION)
N090 4.47
ARSENIC COMPOUNDS N020 3.41ARSENIC 007440382 2.14COPPER COMPOUNDS N100 1.8BARIUM 007440393 0.99COPPER 007440508 0.89BARIUM COMPOUNDS N040 0.81ZINC (FUME OR DUST) 007440666 0.46CADMIUM 007440439 0.44SULFURIC ACID (1994 AND AFTER ’ACIDAEROSOLS’ ONLY)
007664939 0.35
TETRACHLOROETHYLENE 000127184 0.1ZINC COMPOUNDS N982 0.07NICKEL COMPOUNDS N495 0.06SELENIUM 007782492 0.03CHLORINE 007782505 0.01TOLUENE 000108883 0.01
Notes: This table reports the largest twenty total amounts of chemicals released from 15 battery-recycling plants over the period of 2001-2013 The unit of total release is million pounds.
Table 9: The Policy Effect on Other Pollutants near Lead-emitting Plants
Pollutant NOX PM2.5 PM10 SO2
(1) (2) (3) (4)
Near × Post -0.376 -0.211 0.274 -0.0808(1.184) (0.156) (0.447) (0.259)
N 31,617 107,087 99,321 62,023
Near 0-1 0-1 0-1 0-1Adjacent 1-2 1-2 1-2 1-2Mean 28.71 12.98 25.20 4.35N of monitors 338 982 1001 601
Notes: This table presents DD1 estimates based on equation 3 usingvarious other pollutants as the outcomes. All specifications includetime fixed effects. Standard errors are clustered at the monitor level.
12
Table 10: The Policy Effect on Other Pollutants near Battery-recycling Plants compared withnear Other Lead-emitting Plants
Pollutant NOX PM2.5
(1) (2)
Battery × Post 5.435*** -0.370***(1.419) (0.142)
# of other lead emitting plants × Post -1.513* -0.188***(0.873) (0.0546)
N 15,593 58,660
Sample distance 0-1 0-1Mean 11.23 11.79N of monitors 170 516
Notes: This table presents DD3 estimates based on equation 15 using vari-ous other pollutants as the outcome. The sample includes all monitors in thedistance, as specified in the respective column, from any lead-emitting plants(Areas A + C in the online appendix figure 2). All specifications include mon-itor and time fixed effects. The standard errors are clustered at the monitorlevel.
Table 11: The Policy Effect on Other Pollutants near Battery-recycling Plants compared withAdjacent Areas
Pollutant NOX PM2.5
(1) (2)
Near × Post 7.443** -0.241(3.315) (0.505)
# of other lead emitting plants × Post -2.478** -0.404***(1.151) (0.132)
N 2,487 4,238
Near 0-1 0-1Adjacent 1-10 1-10Mean 11.23 11.79N of monitors 21 39
Notes: This table presents estimates based on equation 16 using various otherpollutants as the outcome. The sample includes monitors near battery recyclingplants only (Areas A + B in the online appendix figure 2). All specificationsinclude monitor and time fixed effects. The standard errors are clustered at themonitor level.
13
Table 12: The Policy Effect on Other Pollutants near Battery-recycling Plants: DDD
Pollutant NOX PM2.5
(1) (2)
Near × Post × Battery 9.420*** 0.0186(3.415) (0.427)
# of other lead emitting plants × Post -1.383* -0.194***(0.711) (0.0532)
N 58,545 172,123
Near 0-1 0-1Adjacent 1-10 1-10Mean 11.23 11.79N of monitors 628 1634
Notes: This table presents the DDD estimates based on equation 17 using var-ious other pollutants as the outcome. The sample includes monitors near lead-emitting plants (Areas A + B + C + D in the online appendix figure 2). Allspecifications include monitor and time fixed effects. The standard errors areclustered at the monitor level.
14
V. Additional Figures on Trade
Figure 3: Lead Recovered From Old Battery Processed in the US
Notes: Unit is 1,000 metric tons.
Data source: Minerals Yearbooks by USGS (respective year)
15
Figure 4: Domestic LAB Production
Notes: This figure plots real value of shipment of LAB (in constantDecember 1984 dollar), in which the nominal values are deflated byproducer price index industry data for LAB.
Data source: Annual surveys of manufacturers.
Figure 5: Domestic Vehicle Production
Notes: The figure presents monthly basis domestic auto production.Units are thousands of units, seasonally adjusted.
Data source: US Bureau of Economic Analysis, retrieved fromFederal Reserve Economic Data, Federal Reserve Bank of St. Louis.
16
Figure 6: US Exports of NLAB
Notes: The figure illustrates the total amount of NLAB exported fromthe US to the rest of the world.
Data source: The US ITC
Figure 7: The US Imports of NLAB
Notes: The figure illustrates the total amount of NLAB imported to theUS from the rest of the world.
Data source: The US ITC.
17
Figure 8: The US Imports of Lead
Notes: The figure illustrates the total amount of lead imported to the USfrom the rest of the world. The unit is 1,000 metric tons. Lead includesunwrought lead, lead waste and scrap, lead plates, sheets, strip and foil,and other articles of lead.
Data source: The US ITC.
Figure 9: Type of the US Lead Export
Notes: The figure presents the types of lead exported from the US toMexico and Canada. The unit is in 1,000 metric tons.
Data source: The US ITC.
18
Figure 10: Share of Lead Exports by Country
Notes: The figure presents the share of total lead exported from the USto Mexico and Canada.
Data source: The US ITC.
19