Environmental disaster, pollution and infant health: Evidence from the Deepwater Horizon oil spill

Journal Pre-proof Environmental disaster, pollution and infant health: Evidence from the Deepwater Horizon oil spill Louis-Philippe Beland, Sara Oloomi PII:

S0095-0696(17)30582-X

DOI:

https://doi.org/10.1016/j.jeem.2019.102265

Reference:

YJEEM 102265

To appear in:

Journal of Environmental Economics and Management

Received Date: 26 August 2017 Revised Date:

29 March 2019

Accepted Date: 14 September 2019

Please cite this article as: Beland, L.-P., Oloomi, S., Environmental disaster, pollution and infant health: Evidence from the Deepwater Horizon oil spill, Journal of Environmental Economics and Management (2019), doi: https://doi.org/10.1016/j.jeem.2019.102265. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier Inc.

Environmental Disaster, Pollution and Infant Health: Evidence from the Deepwater Horizon Oil Spill∗ Louis-Philippe Beland

Sara Oloomi

March 2019

Abstract In 2010 the Gulf Coast experienced the largest oil spill affecting U.S. waters in history. Evaporating crude oil and dispersant chemicals can cause major health problems. This paper examines the impact of the Deepwater Horizon oil spill on air quality and infant health outcomes. Using U.S. Environmental Protection Agency (EPA) AirData, vital statistics data from National Center for Health Statistics (NCHS), and a difference-in-difference methodology, we find that the oil spill of 2010 increased concentrations of PM2.5, NO2, SO2, and CO in affected coastal counties, increased incidence of low birth weight (<2500 grams) and premature born infants (<37 weeks of gestation). Heterogeneity effects reveal more pronounced adverse infant health outcomes for black, Hispanic, less educated, unmarried, and younger mothers. Results are robust to a wide range of controls and robustness checks.

JEL Classification: I12, J13, Q53 Keywords: Infant Health, Pollution, Oil spill.

∗ We would like to thank Dr. Carl Childs at NOAA for sending us the list of coastal counties that received oil coming ashore and answering our questions.

1

Environmental Disaster, Pollution and Infant Health: Evidence from the Deepwater Horizon Oil Spill∗ March 2019

Abstract In 2010 the Gulf Coast experienced the largest oil spill affecting U.S. waters in history. Evaporating crude oil and dispersant chemicals can cause major health problems. This paper examines the impact of the Deepwater Horizon oil spill on air quality and infant health outcomes. Using U.S. Environmental Protection Agency (EPA) AirData, vital statistics data from National Center for Health Statistics (NCHS), and a difference-in-difference methodology, we find that the oil spill of 2010 increased concentrations of PM2.5, NO2, SO2, and CO in affected coastal counties, increased incidence of low birth weight (<2500 grams) and premature born infants (<37 weeks of gestation). Heterogeneity effects reveal more pronounced adverse infant health outcomes for black, Hispanic, less educated, unmarried, and younger mothers. Results are robust to a wide range of controls and robustness checks.

JEL Classification: I12, J13, Q53 Keywords: Infant Health, Pollution, Oil spill.

∗ We would like to thank Dr. Carl Childs at NOAA for sending us the list of coastal counties that received oil coming ashore and answering our questions.

1

1

Introduction

In 2010, the Gulf Coast experienced the largest oil spill affecting U.S. waters in history, with an estimated release of about 134 million gallons of oil. The 2010 oil spill in the Gulf of Mexico is an exogenous event that affected the coastal counties of Alabama, Florida, Louisiana, and Mississippi. Recent research suggests that pregnant women, infants, and children are at a higher risk of adverse health impacts due to pollution (e.g. Currie and Neidell (2005) and Janke (2014)). Using air quality data from the U.S. Environmental Protection Agency (EPA), infant health data from the National Center for Health Statistics (NCHS), and a difference-in-difference methodology, this paper aims to investigate the impact of the oil spill on air quality (PM2.5, NO2, SO2 and CO) and on infant health outcomes (birth weight, low birth weight incidence, gestation in weeks, and premature newborn incidence). We find that the 2010 oil spill and cleanup activities increased concentrations of PM2.5, NO2, SO2, and CO. We also find that the oil spill increased the incidence of low birth weight by 0.010 (<2500 grams), and the incidence of prematurely born infants by 0.009 (<37 weeks of gestation).1 These outcomes have been shown to be good measures of infant health and have long term consequences on cognitive development (e.g. Black et al. (2007), Figlio et al. (2014), and Sanders (2012)). We distinguish the impact for mothers already pregnant at the moment of the oil spill from mothers who subsequently became pregnant, and find similar impacts. Heterogeneity effects reveal more pronounced adverse infant health outcomes for mothers who are black, Hispanic, less educated, unmarried, and younger. The results have important policy implications as certain mothers are more affected by the pollution shock. This suggests that certain mothers can successfully apply measures to mitigate the negative impacts of pollution and that resources and information should be targeted toward pregnant women in the aftermath of an environmental disaster to help prevent poor infant health outcomes. Simple actions can be implemented to mitigate negative impacts of pollution. Exposure to air pollution can be reduced by staying indoors, reducing outdoor air infiltration to indoors, cleaning indoor air using air filters, and limiting physical activities, especially outdoors and near air pollution sources (e.g. Laumbach et al. (2015) and Sbihi et al. (2016)). The environmental literature emphasize the negative impact of the cleanup activities and dispersants on pollution 1

These effects are similar to the literature on the impact of air pollution on infant health outcomes (e.g. Currie et al., 2009B; Currie & Walker, 2011; and Lavaine & Neidell, 2017).

2

following the Deepwater Horizon oil spill (e.g. Afshar-Mohajer et al., (2018) and De Gouv et al. (2011)). Recent research have also shown ineffective use of dispersant after the Deephorizon oil spill (e.g. Paris et al. (2018)). Our results also have implications regarding how cleanup activities are conducted and the importance of minimizing pollution and negative health effects for infants. Our results are robust to a wide range of specifications and robustness checks such as: instrumental variable estimations, different control groups (such as propensity score matching), placebo estimations and controls, including the county’s monthly average employment and wages. Event study graphs show that the oil spill affects infant health outcomes and air quality negatively following the oil spill without any prior trends. The rest of the paper is organized as follows: Section 2 reviews the literature; Section 3 discusses the oil spill, data and descriptive analysis; Section 4 presents the methodology; Section 5 presents the results; and Section 6 concludes.

2

Literature review

Our paper is related to recent papers documenting the negative impacts of oil on pollution and health. For example, Lavaine and Neidell (2017) find that a temporary reduction in oil production due to a strike leads to a significant reduction in sulfur dioxide (SO2) concentrations and increases in birth weight and gestational age of newborns. Currie and Walker (2011) and Knittel et al. (2016) find that traffic congestion leads to air pollution and decreases infant health outcomes. More closely related, oil spills have been shown to pose a threat to human health from inhalation or dermal contact with oil and dispersant chemicals, as well as threats via seafood safety (e.g. Solomon and Janssen (2010); Reddy et al. (2012); and D’Andrea & Reddy (2014)). De Gouw et al. (2011) documents that the Deepwater Horizon oil spill and the cleanup activities generated a wide plume with high concentrations of organic aerosol (>25 micrograms per cubic meter) which was attributed to the formation of secondary organic aerosol (SOA) emitted from a wide area around the Deepwater Horizon oil spill. Their research provides convincing evidence for the formation of secondary organic aerosol (SOA), which has been shown to generate PM2.5, from less volatile hydrocarbons. Afshar-Mohajer et al. (2018) have also shown that the dispersant chemicals used to cleanup oil spills can transform crude oil into a toxic and dangerous cloud able to travel several

3

miles and affect human lungs. McGowan et al. (2017) documented that several cleanup workers exposed to dispersants suffered lung irritation, nausea, vomiting, and rashes. These suggest that cleanup activities are important factors leading to increases in pollution.2 Our paper also joins a growing body of literature that studies the negative impacts of pollution on health outcomes.3 Recent papers document the negative impact of pollution on infant health outcomes in the United States. Chay and Greenstone (2003) use variations in air pollution reductions across sites due to the recession of 1981-1982 and find that a one percent reduction in air pollution leads to a 0.35 percent decline in infant mortality. Currie et al. (2013) find a negative impact for water contamination and Currie and Neidell (2005) for air quality on infant health outcomes (see also Currie & Schmieder (2009A) and Currie et al. (2009B)). Deschenes et al. (2009) find that exposure to hot ambient temperatures during the second and third trimesters negatively affects birth weight. Other papers have shown the adverse impact of pollution on human health leads to avoidance activities to reduce the negative impacts of pollution. Some optimizing individuals compensate for increases in pollution by reducing their exposure. Zivin et al. (2011), for example, find that following water contaminations there is an increase in bottled water sales (see also Moretti and Neidell (2011) and Janke (2014)). Prenatal exposure to pollution has also been found to have long term impacts. Figlio et al. (2014) study the relationship between birth weight and cognitive development. They find that the effects of infant health on cognitive development are important and consistent through the school career. Moreover, they find that the effect is invariant to school quality and similar across different family backgrounds. They conclude that infant health outcomes are an important predictor of school outcomes. Black et al. (2007), using twins’ data from Norway, find that birth weight is an important predictor of both short-term and long-run outcomes. In particular, they find that low birth weight negatively affects educational attainment and earnings. Figlio et al. (2014) find that prenatal exposure to environmental toxins reduces later developmental outcomes of children by reducing test scores, increasing the likelihood of repeating a grade, and getting suspended from school (see also Almond et al. (2011A) and Sanders (2012)). 2

See also Middlebrook et al. (2012). The literature of pollution on health and economic outcomes is vast. Zivin and Neidell (2013) offer a good survey of the literature. 3

4

This paper contributes to the literature by studying the impact of one of the largest oil spills in U.S. history, an exogenous shock to the Gulf Coast, on air quality and infant health outcomes. We analyze the impact on mothers already pregnant during the oil spill and perform our analysis by trimesters. We also use the oil spill, an exogenous event that affected the coastal counties in the Gulf region, as an instrumental variable to estimate the impact of air pollution on infant health. We also contribute to the literature by distinguishing the impact by characteristics of the mothers: race, education, marital status, and age. Our results suggest that certain mothers can successfully apply measures to mitigate negative impacts of pollution and that resources and information should be targeted toward pregnant women in the aftermath of an environmental disaster to help prevent poor infant health outcomes. Exposure to air pollution can be reduced by staying indoors, reducing outdoor air infiltration to indoors, cleaning indoor air using air filters, and limiting physical activities, especially outdoors and near air pollution sources (e.g. Laumbach et al. (2015) and Sbihi et al. (2016)). Our results also have implications regarding how cleanup activities are conducted.

3

Deepwater Horizon Oil Spill Description, Data Sources, and Descriptive Statistics

3.1

Deepwater Horizon Oil spill

The Deepwater Horizon oil spill (also called Gulf of Mexico oil spill or BP oil spill) is recognized as one of the worst oil spills in U.S. history and one of the worst environmental disasters.4 The Deepwater Horizon rig was situated in the Gulf of Mexico, approximately 50 miles off the coast of Louisiana. On April 20, 2010, a final cement seal on an oil well failed, causing an explosion; eleven of the 126 workers aboard the rig were killed. The rig sank on the morning of April 22, the pipe leaked, and oil began to discharge into the Gulf. The well was capped on July 15, 2010, 87 days later. The U.S. government estimates that about 134 million gallons of oil were released into the Gulf of Mexico over that 87 day period. By early June 2010, oil had washed up on Louisiana’s coast and along the Mississippi, Florida, and Alabama coastlines. Efforts were put in place shortly after the oil spill to control the spread of 4

For more details on the oil spill, see the National Commission on the BP Deepwater Horizon Oil Spill, (2011), Reddy et al. (2012), Bishop et al. (2017), Michael et al. (2013), and English et al. (2018).

5

oil to beaches and other coastal ecosystems using floating booms to contain surface oil and chemical oil dispersants to break it down underwater. In December 2012, several miles of coastline remained subject to evaluation and/or cleanup operations. Figure 1 presents a map of affected coastal counties using the National Oceanic and Atmospheric Administration’s (NOAA) list of coastal counties that received oil coming ashore as our treatment group, which we refer as narrow coastal counties.5 The oil spill, along with the response and cleanup activities, caused extensive damage to marine and wildlife habitats, as well as the fishing and tourism industries. British Petroleum (BP) agreed to pay nearly 62 billion dollars to settle liabilities related to the oil spill.

3.2

Data sources and descriptive statistics

The main dataset used in this paper is the National Vital Statistics data from the National Center for Health Statistics (NCHS). Vital Statistics from NCHS provides birth data of all the births registered in the United States. We use data from 2006 to 2012. We have access to more than 28 million birth observations over the period and we have county identifiers.6 Of those 28 million, more than 9 million are in southern states.7 We use this dataset to investigate the impact of the oil spill on infant health outcomes including premature births (<37 weeks of gestation), low birth weight (<2500 grams), gestation in weeks, and birth weight in grams. The mean gestation in our sample is 38.46 weeks and the mean birth weight is 3229.77 grams. The incidence of premature and low birth weight are 13.30% and 9.24%, respectively. This dataset also enables us to control for the age, education, race, marital status, and other risk factors of the pregnancy including plurality, place of the birth, prenatal visits, sex of the infant, and characteristics of fathers. We also use this dataset to investigate the impact of the oil spill by identifying characteristics of the mother. In Table 1 we present descriptive statistics for the characteristics of mothers and fathers over our sample of years (2006 to 2012) for our treatment and control groups. Our treatment group is the narrow coastal counties that received oil coming ashore as described in Figure 1, , which we refer 5

A list of these counties are presented in Figure 1. It also distinguish between counties that received a high level versus low trace amount. In the robustness section, we also use broader definition of treatment group in Table A.1 such as all coastal counties of Alabama, Florida, Louisiana, and Mississippi. 6 We made a request and were granted access to the restricted data from NCHS for this project. These restricted data give access to county identifiers. 7 We consider the following list of U.S. states as southern states: Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas and Virginia. We have investigated different control groups in our robustness checks, including propensity score matching (see the heterogeneity and robustness section).

6

as narrow coastal counties. Our comparison group is the other infants born in other southern counties in the same period who were unaffected by the oil spill. Table 1 shows that the treatment and control groups are similar, but our treatment group has a higher non-white population and a higher fraction of mothers with a college degree, for example.8 As an alternative control group, we use propensity score matching and Table 1 also presents descriptive statistics for our one-to-one propensity score matching control group. Table 1 shows that our propensity score matching control group is very similar to our treatment group based on observables. Figure 2 shows that there are no prior trends before the oil spill for air quality (PM2.5 and SO2) and infant health (incidence of premature babies and low birth weights) and that the oil spill significantly increases the incidence of premature, low birth weight babies and pollution in coastal counties. We also present event study graphs for mothers’ characteristics before and after the oil spill in Figure A.2 in the appendix. Figure A.2 presents several event study graphs investigating potential compositional changes for characteristics of mothers (education, race of mother, age of mother, and marital status). Figure A.2 suggests that our results are not driven by compositional changes in mother’s characteristics, with no prior trends. We use pollution data from the U.S. Environmental Protection Agency (EPA) AirData.9 We have information on daily average concentrations for the following major pollutants: nitrogen dioxide (NO2), particulate matter (PM2.5), sulfur dioxide (SO2), and carbon monoxide (CO). Nitrogen dioxide (NO2) is 1-hour daily data measured in .01 parts per million (ppm). Particulate matter (PM2.5) is 24-hour daily data measured in micrograms per cubic meter (µg/m3). The mean emissions for NO2 and PM2.5 in our sample are 89.90 and 11.44, respectively. Sulfur dioxide (SO2) is 1-hour daily data measured in .01 parts per million (ppm). Carbon monoxide (CO) is 1-hour daily data measured in parts per million (ppm). The mean emissions for SO2 and CO in our sample are 22.06 and 0.37, respectively. The pollutants are regulated by the Clean Air Act and are targeted by the EPA for their negative impacts on health, the environment, and properties. Of those pollutants, particulates have the strongest impact on health and can lead to or exacerbate respiratory problems, especially for people with asthma (e.g. Gent et al., 2003). NO2 and SO2 also contribute to the formation of particulates. As discussed in the literature section, other papers have documented the 8 9

Several of the differences are statistically significant due to our large sample size. See www.epa.gov/air/urbanair/ for more details.

7

negative impact of the Deepwater Horizon oil spill and the cleanup activities on air pollution and discuss mechanisms, including the formation of SOA (e.g Afshar-Mohajer et al. (2018), De Gouv et al. (2011) and McGowan et al. (2017)). Figure A.1 in the appendix presents the location of monitoring stations for our pollutants (PM2.5, NO2, SO2, and CO). Our results are based on 264 monitoring stations for PM2.5, 108 for NO2, 146 for SO2 and 84 for CO in the southern United States. Our treatment group is coastal counties of Alabama, Florida, Louisiana, and Mississippi that received oil coming ashore. We refer to these coastal counties as narrow coastal counties. We use the other monitoring stations in other southern states as our control group.10

4

Methodology

We estimate the impact of the Deepwater Horizon oil spill on infant health outcomes using a difference-in-difference strategy. We estimate the following: HealthOutcomeict = β0 + β1 OSCict + β2 OSCict ∗ Af terApril2010 (1) +β3 Af terApril2010 + β4 Xict + βM Y + βC + γt + γt2 + ict HealthOutcomeict represents the outcome variable for infant i at county c and month-year t. We consider the following infant health outcomes: gestation in weeks, premature birth (<37 weeks of gestation), birth weight (in grams) and low birth weight (<2500 grams). OSCict is a dummy variable that takes a value of one if the mother lives in an oil spill county (OSC) and zero otherwise. We use the National Oceanic and Atmospheric Administration’s (NOAA) list of coastal counties that received oil coming ashore as our treatment group. As discussed in the data section, we refer to these counties as narrow coastal counties.11 Af terApril2010 is a variable indicating that the oil spill occurred at time t. It takes a value of one for the months and years after April 2010 (after the oil 10

We follow a similar process than Currie and Neidell (2005) and Currie et al. (2009B). Each monitoring station provides daily data on pollutant levels. Some counties have multiple monitoring stations. Therefore, we averaged the monthly pollutants data across multiple stations within a county to calculate average pollution for each county. Some counties do not have a monitoring station for some of the pollutants; hence, we assigned the values of the closest monitoring station as follows: first, we calculate the centroid of each county with the missing monitoring station. We then calculate the distance between the centroid of each county with closest monitoring stations in coastal counties as defined in Figure A.1 using geographical measures of the monitoring sites (latitude and longitude). Finally, we identify the closest monitoring station that has the shortest distance. 11 A list of these counties are presented in Figure 1. It also distinguish between counties that received high level versus low trace amount. Table A.1 also presents broader definition of treatment group.

8

spill) and takes a value of zero for the period prior to April 2010. We use data from 2006 to 2012.12 OSCct ∗ Af terApril2010 is the interaction term identifying the treatment group for whether the county c is affected by the oil spill after April 2010. β2 is the coefficient of interest which captures the impact of the oil spill on infant health outcomes. Our comparison group is the other infants born in other southern counties in the same period who were unaffected by the oil spill.13 Xict is a vector of characteristics of mothers and fathers with information on race, age, marital status, education, and risk factors of the pregnancy. βM Y and βC represent fixed effects for month by year and county, respectively. γt and γt2 are linear and quadratic time trends. Standard errors are clustered at the county level. The sample is limited to mothers 20 to 45 years old. In the heterogeneity and robustness sections, we investigate whether results are robust to several different control groups (all U.S., all U.S. except California and New York and propensity score matching), robustness checks, placebo tests, and present event study graphs. We also present instrumental variable estimates using the oil spill as the instrument for air pollution. We also investigate whether results vary by mothers’ characteristics and by trimesters for mothers already pregnant at the moment of the oil spill. We also estimate the impact of the Deepwater Horizon oil spill on air quality using a differencein-difference strategy. We estimate the following: Airpolluantsct = β0 + β1 OSCct + β2 OSCct ∗ Af terApril2010 (2) +β3 Af terApril2010 + βM Y + βC + βS + γt + γt2 + ct Airpolluantsct measures the air quality for the air monitoring stations in coastal counties (parishes) c of Alabama, Florida, Louisiana, and Mississippi at month-year t. We use information on average concentrations for four major pollutants: PM2.5, NO2, SO2, and CO. OSCct is a dummy variable that takes a value of one for narrow coastal counties of Alabama, Florida, Louisiana, and Mississippi that received oil coming ashore.14 Af terApril2010 , OSCct ∗Af terApril2010 , β2 , βM Y , 12

In December 2012, several miles of coastline remained subject to evaluation and/or cleanup operations. We consider the following list of U.S. states: Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas, and Virginia. 14 Each monitoring station provides daily data on pollutants levels. Some counties have multiple monitoring stations. Therefore, we averaged the monthly pollutants data across multiple stations within a county to calculate average pollution for each county. Some counties do not have a monitoring station for some of the pollutants; hence, we assigned the values of the closest monitoring station as follow: first, we calculate the centroid of each county with the missing monitoring station. We then calculate the distance between the centroid of each county with closest 13

9

βC , γt and γt2 are defined as above. βS represents fixed effects for monitoring stations.

5

Results

5.1

Main Results

We first look at the impact of the oil spill and cleanup activities on air pollution in coastal counties, using air monitoring stations. Table 2 presents results for outcomes of the majors pollutants PM2.5, NO2, SO2, and CO. Table 2 shows that the concentrations of PM2.5, NO2, SO2, and CO increased significantly in monitoring stations in coastal counties of Alabama, Florida, Louisiana, and Mississippi and therefore the air quality is significantly negatively affected.15 Using the mean emission in our sample and calculating a percentage increase, also presented in Table 2, we find that emissions for PM2.5, NO2, SO2, and CO increased by 12.03%, 3.89%, 17.78% and 10.35%, respectively. These results are consistent with other papers in the environmental sciences that found an increase in air pollution following the Deepwater Horizon oil spill and clean up activities (e.g Afshar-Mohajer et al. (2018) and De Gouv et al. (2011)). We next investigate the impact of the oil spill on infant health outcomes. Table 3 shows the impact of the oil spill for several infant health outcomes. Column (1) of Table 3 investigates whether the oil spill has an impact on the incidence of having a premature baby. Column (2) investigates the impact of the oil spill on the incidence of low birth weight (<2500 grams). Columns (3) and (4) study the impact of the oil spill on gestation (in weeks) and birth weight (in grams). Table 2 shows that the oil spill significantly increases the incidence of premature birth (by 0.009) and the incidence of low birth weight (by 0.010).16 We also find that the oil spill significantly decreases gestation and birth weight. In sum, Table 3 shows that the Deepwater Horizon Oil spill has significant negative impacts on infant health outcomes. monitoring stations in coastal counties as defined in Figures A.1 using geographical measures of the monitoring sites (latitude and longitude). Finally, we identify the closest monitoring station that has the shortest distance. 15 Table A.7 in the appendix uses the number of hours per month that pollutants exceeded the standard. Table A.7 finds that the oil spill has no significant impact on the number of hours per month that pollutants exceeded the standard. CO never exceeds the standard. 16 These outcomes have been shown to be good measures of infant health and the literature documents long term benefits of positive infant health outcomes (e.g. Black et al. (2007), Figlio et al. (2014), and Sanders (2012)). These results are similar to estimates presented in the literature (e.g. Currie et al., 2009B; Currie & Walker, 2011) or slightly below (Lavaine & Neidell, 2017).

10

5.2

Heterogeneity & Robustness

We next investigate several heterogeneity and robustness checks. Table 4 presents heterogeneity estimates by individual characteristics of mothers including race (white, black and Hispanic), age groups (age is less than 25, age 26 to 35 and above 35), educational attainment (college graduate, some college, and high school or less), and marital status (married or unmarried). Table 4 shows higher adverse infant health outcomes for black, Hispanic, less educated, unmarried, and younger mothers. For example, Table 4 shows that the oil spill increases the incidence of premature babies for high school diploma or less (by 0.013), black (by 0.017), and Hispanic (by 0.019) mothers. This suggests important policy implications as certain mothers are more affected by the oil spill. Table 4 suggests that certain mothers can successfully apply measures to mitigate negative impacts of pollution.17 These results suggest that resources and information should be targeted toward pregnant women in the aftermath of an environmental disaster to help prevent poor infant health outcomes. Exposure to air pollution can be reduced by staying indoors, reducing outdoor air infiltration to indoors, cleaning indoor air using air filters, and limiting physical activities, especially outdoors and near air pollution sources (e.g. Laumbach et al. (2015) and Sbihi et al. (2016)). Table 5 studies the impact of the oil spill separately for women already pregnant at the moment of the oil spill and women who became pregnant subsequently. Table 5 shows that both women already pregnant and women who subsequently became pregnant are affected. Infant health outcomes are affected for both groups and the difference is not statistically different.18 Table 6 studies the impact of the oil spill on infant health outcomes by trimesters for women already pregnant at the moment of the oil spill, using interaction terms. It shows that the negative impacts of the oil spill are present in all three trimesters for women already pregnant. Table 7 presents results for several specifications. Panels A-C present results for alternate control 17

These results might be due to those groups having a lower income, which might prevent them from applying measures to mitigate negative exposure from pollution. The role of income on infant health as been discussed in several papers such as Currie (2009), Case et al. (2002), and Hoynes et al. (2015). Income is not available in the NCHS data. Tables A.4 and A.5 in the appendix add additional controls for the county’s monthly average employment and wages and find similar impact on infant health. 18 Table A.6 in the appendix investigates by characteristics whether the oil spill affects the propensity to become a mother. Table A.6 suggests that women in our sample did not alter their childbearing decision due to the oil spill. We also present event study graphs for mother’s characteristics before and after the oil spill in Figures A.2. These figures present several event study graphs investigating potential compositional changes for characteristics of mothers (education, race of mother, age of mother, marital status). Similarly to Table A.6, Figure A.2 suggests that our results are not driven by compositional changes in mother’s characteristics, with no prior trends.

11

groups and Panel D for standard errors clustered at the state level. Table 7, Panel A investigates whether results are similar if we use all U.S. births as our control groups and Panel B presents results for all births but removes births in the states of New York and California from our control group - two big states less likely to be comparable to our treatment group. Table 7, Panel C presents propensity score matching.19 Table 7, Panel D presents results using our main control group for standard errors clustered at the state level and results are qualitatively the same. Table 7 shows, once again, that the oil spill leads to a decrease in infant health outcomes. In Table 8, we implement instrumental variables estimations using the oil spill as the instrument to predict PM2.5 level of air pollution. The results of the impact of pollution on infant health, using the oil spill as the instrument, are presented in Table 8 and show, once again, that the oil spill leads to a decrease in infant health outcomes with results qualitatively the same as our main Table 3.20 We next implement placebo tests in Tables 9 and 10, which are generated by turning on the oil spill dummy in different years before the oil spill incident. This placebo intervention should have no significant impact on air quality and infant health outcomes. If there is a positive significant relationship, then there are correlations between the trend and the oil spill. Tables 9 and 10 show that placebo treatments do not produce a significant effect on air quality and infant health outcomes. We take this as further evidence that prior trends are not generating these results; this makes us confident in our main results of Tables 2 and 3. Figure 2, Panel A shows the event study graphs for the effect of the oil spill on incidence of premature and Panel B for low birth weight babies. It shows that the oil spill significantly increases the incidence of premature and low birth weight babies in affected coastal counties, without discernible prior trends. Figure 2, Panel C performs a similar exercise for air quality for PM2.5 and Panel D for SO2; it shows that the oil spill and clean up activities affect air quality negatively without any prior trends.21 19 We implement matching to pre-process the data, following the guidelines of Ferraro and Miranda (2017). Oneto-one propensity score matching is performed on mothers’ characteristics as well as the risk factors of the pregnancy, using mothers in southern states not impacted by the oil spill. Table 1 shows descriptive statistics for the treatment and propensity score control group. Table 1 shows that our propensity score matching control group is very similar to our treatment group based on observables. 20 In Panel A of Table 8, we first assign each county to the concentration of PM2.5 of the closest monitoring station. If there is more than one monitoring station in a county, we use the average of monitoring stations in the county. In Panel B, we only use counties with an air monitor as the spatial interpolation of PM2.5 can lead to bias. Bento et al. (2015) find that areas near monitors tend to be lower income, more non-white. Results are similar in Panel A and B of Table 8. 21 Graphs for other outcomes are qualitatively similar.

12

We next investigate the spatial dispersion of the results in several ways. Table A.1.A in the appendix investigates different definitions of the treatment group. We define the treatment group in three alternative ways: narrow coastal counties that received oil coming ashore (most exposed to the pollution effects of the oil spill and our main treatment group), coastal counties, and coastal counties and close counties (to consider possible spillover). In Table A.1.A, we find a significantly more pronounced impact on infant health outcomes for narrower definition compared to the broader definition of treatment group.22 Table A.1.B separates our treatment into counties with high level of oil and counties with small traces of oil. Table A.1.B finds a negative impact of the oil spill on infant health in both cases but the effect is significantly higher in counties with high traces. Table A.2.A investigates the heterogeneity of the impact for the different coastal states separately using interaction terms (for Alabama, Florida, Louisiana, and Mississippi). Tables A.2.A shows that some states experience a significantly more pronounced effect than others for certain outcomes.23 Table A.2.B present a placebo treatment group, using coastal counties of Texas and coastal counties in the Florida Peninsula as a treatment group, which did not receive oil coming ashore. Table A.2.B finds no significant impact of the oil spill using this placebo treatment group. This provides additional evidence that the results on infant health are due to the oil spill and not to other possible confounding changes in the states.24 Overall, results are robust to alternative specifications and robustness checks. These numerous robustness checks provide confidence that the oil spill increased air pollution and had negative impacts on infant health outcomes.

5.3

Discussion of other potential mechanisms

We next discuss other potential mechanisms. We next investigate the impact of changes to the labor market due to the oil spill. Aldy (2014) finds that the local labor market is affected after 22

We use the National Oceanic and Atmospheric Administration’s (NOAA) list of narrow coastal counties that received oil coming ashore as our main treatment group. However, it is possible that there is some spillover to neighboring counties and we investigate this in Table A.1. NOAA list of coastal counties are defined in Figure 1. 23 These states have a larger fraction of mothers most affected by oil spill in narrow coastal counties: unmarried, lower level of education and minority group. 24 Table A.3 presents the effect of the oil spill on PM2.5 by state and shows an increase in PM2.5 emissions in all states after the oil spill. We focus on PM2.5 as it is the pollutant with the most monitoring stations over the period and monitoring stations are present in all states. It is also one of the pollutants that has the most severe impact on health. De Gouw et al. (2011) documents that the Deepwater Horizon oil spill and the cleanup activities generated secondary organic aerosol (SOA), which has been shown to generate PM2.5.

13

the oil spill with some counties/parishes positively affected and others negatively.25 To account for changes in the labor market, we first add controls for employment and wage to our specification and re-estimate Table 3.26 Table A.4 in the appendix shows that we find qualitatively similar impacts to Table 3. Table A.5 separately studies the impact of the oil spill on counties with positive and negative employment effects. We find an impact on infant health outcomes in both groups. Tables A.4 and A.5 provide suggestive evidence that the negative impact on infant health is transmitted through the effect on pollution and not the labor market.27 Another potential channel is increased stress due to the oil spill. Stress is one factor that has been shown to affect infant health (e.g Black et al. (2016)). We find in Table 5 that both women already pregnant and women who subsequently became pregnant are affected by the oil spill. It is less likely that women that became subsequently pregnant after the oil spill experience high level stress from the oil spill. The Deepwater Horizon oil spill might also have an impact on water pollution and seafood contamination. We cannot test these potential effects directly with our data but our coefficients on the impact of air pollution on infant health are similar to the literature on the impact of air pollution on infant health (e.g. Currie et al., 2009B; Currie & Walker, 2011; and Lavaine & Neidell, 2017).28

6

Conclusion

This paper examines the impact of the oil spill of 2010 in the Gulf of Mexico on air pollution and health outcomes of newborns. Using air data from the EPA, Vital Statistics data from the National Center for Health Statistics, and a difference-in-difference methodology, we find that the oil spill of 2010 and the cleanup activities increased the concentration of SO2, NO2, CO, and PM2.5 in coastal counties and has a significant negative impact on infant health outcomes. Particularly, we find an 25

Aldy (2014) finds that Louisiana coastal parishes, and oil-intensive parishes in particular, experienced a net increase in employment and wages. In contrast, he finds that several Florida counties experienced a decline in employment. 26 We use labor data from monthly county/parish-level data (quarterly data for wages) from the Quarterly Census of Employment and Wages (QCEW), similarly to Aldy (2014). 27 Tables A.4 and A.5 are not fully addressing potential differential economic effects of the oil spill. It is possible that there is intra-county heterogeneity in economic consequences of the oil spill and that lower income families are significantly more affected. This could have repercussions on the analysis of the heterogeneity results. 28 Currie et al. (2009B) find that an increase in air pollution leads to an increase in low birthweight incidence by 0.008. Currie and Walker (2011) find that a decrease in air pollution due a congestion policy (EZPAS) decrease low birth weight by 0.009. Similarly, Lavaine and Neidell (2017) find that a decrease in air pollution due to an oil refinery strikes, decrease low birth weight incidence by 0.011.

14

increase in the incidence of low birth weight by 0.010 (<2500 grams) and incidence of premature birth by 0.009 (<37 weeks of gestation).29 Heterogeneity effects reveal higher adverse infant health outcomes for black, Hispanic, less educated, unmarried, and younger mothers. Results are robust to a wide range of controls and robustness checks. This is important as the literature documents the long term benefits of positive infant health outcomes (e.g. Black et al. (2007), Figlio et al. (2014), and Sanders (2012)). According to Butler and Behrman (2007), the economic burden associated with preterm birth in the United States is at least $66,591 in 2018 dollars per infant born before term. Therefore, using our estimates, we can approximate the related economic costs due to the increase in premature babies to be more than 300 million dollars. The associated economic costs are important and the results have important policy implications as certain mothers are more affected by the pollution shock. Our results suggest that certain mothers can successfully apply measures to mitigate negative impacts of pollution. Resources and information should be targeted toward pregnant women in the aftermath of an environmental disaster to help prevent poor infant health outcomes. Simple actions can be implemented to mitigate negative impact of pollution. Exposure to air pollution can be reduced by staying indoors, reducing outdoor air infiltration to indoors, cleaning indoor air using air filters, and limiting physical activities, especially outdoors and near air pollution sources (e.g. Laumbach et al. (2015) and Sbihi et al. (2016)). These information can be transmitted during prenatal visits. The environmental literature emphasize the negative impact of the cleanup activities and dispersants on air pollution following the Deepwater Horizon oil spill (e.g. Afshar-Mohajer et al., (2018) and De Gouv et al. (2011)). Recent research have shown ineffective use of dispersant after the Deephorizon oil spill (e.g. Paris et al. (2018)). Our results also have implications regarding how cleanup activities are conducted and the importance of minimizing pollution and negative health effects for infants. More research is needed to understand fully the long-term impacts of dispersants on the coastal regions.

References Afshar-Mohajer, N et al.. (2018). A laboratory study of particulate and gaseous emissions from crude oil and crude oil-dispersant contaminated seawater due to breaking waves. Atmospheric Environment, 179, 29

These effects are similar to the literature on the impact of air pollution on infant health outcomes (e.g. Currie et al., 2009B ; Currie & Walker, 2011; and Lavaine & Neidell, 2017).

15

177-186. Aldy, Joseph E. “The Labor Market Impacts of the 2010 Deepwater Horizon Oil Spill and Offshore Oil Drilling Moratorium,” NBER working paper, (2014). Almond, Douglas, and Janet Currie, ”Killing me softly: The fetal origins hypothesis.” The Journal of Economic Perspectives 25.3 (2011A): 153-172. Bento, A., Freedman, M., & Lang, C. (2015). Who benefits from environmental regulation? evidence from the clean air act amendments. Review of Economics and Statistics, 97(3), 610-622. Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes, ”From the Cradle to the Labor Market? The Effect of Birth Weight on Adult Outcomes” Quarterly Journal of Economics 122.1 (2007). Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes. ”Does grief transfer across generations? bereavements during pregnancy and child outcomes.” American Economic Journal: Applied Economics 8.1 (2016): 193-223. Bishop, Richard C., et al. ”Putting a value on injuries to natural assets: The BP oil spill.” Science, 356.6335 (2017): 253-254. Butler, Adrienne Stith, and Richard E. Behrman, eds. Preterm birth: causes, consequences, and prevention. National Academies Press, 2007. Case, Anne, Darren Lubotsky, and Christina Paxson. 2002. “Economic Status and Health in Childhood:The Origins of the Gradient.” American Economic Review 92 (5) 1308-34. Chay, Kenneth Y., and Michael Greenstone. ”The impact of air pollution on infant mortality: evidence from geographic variation in pollution shocks induced by a recession.” The Quarterly Journal of Economics, 118.3 (2003): 1121-1167. Currie, Janet. 2009. “Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood,and Human Capital Development.” Journal of Economic Literature 47 (1): 87–122. Currie, Janet and Johannes F Schmieder, ”Fetal Exposures to Toxic Releases and Infant Health,” American Economic Review, 2009A, 99 (2), 177-83. Currie, Janet, Matthew Neidell, and Johannes F Schmieder, ”Air pollution and infant health: Lessons from New Jersey,” Journal of Health Economics, 2009B, 28 (3), 688-703. Currie, Janet, and Matthew Neidell. ”Air pollution and infant health: what can we learn from California’s recent experience?.” The Quarterly Journal of Economics, 120.3 (2005): 1003-1030. Currie, J., & Walker, R. Traffic congestion and infant health: Evidence from E-ZPass. American Economic Journal: Applied Economics, (2011). 3(1), 65-90.

16

D’Andrea, Mark A., and G. Kesava Reddy. ”Health risks associated with crude oil spill exposure.” The American Journal of Medicine, 127.9 (2014): 886-e9. Deschenes, O., Greenstone, M., & Guryan, J.. ”Climate change and birth weight”. The American Economic Review, 99(2), (2009), 211-217. De Gouw, J. A., et al. (2011). Organic aerosol formation downwind from the Deepwater Horizon oil spill. Science, 331(6022), 1295-1299. English, E., et al.. (2018). Estimating the Value of Lost Recreation Days from the Deepwater Horizon Oil Spill, Journal of Environmental Economics and Management, 91 (2018) 26e45 Ferraro, Paul J., and Juan Jos´ e Miranda. ”Panel data designs and estimators as substitutes for randomized controlled trials in the evaluation of public programs.” Journal of the Association of Environmental and Resource Economists 4.1 (2017): 281-317. Figlio, David,Guryan, J., Karbownik, K., & Roth, J, ”The effects of poor neonatal health on children’s cognitive development.” The American Economic Review, 104.12 (2014): 3921-3955. Gent JF et al.. Association of Low-Level Ozone and Fine Particles With Respiratory Symptoms in Children With Asthma. JAMA. 2003;290(14):1859-1867. Hoynes, Hilary, Doug Miller, and David Simon. ”Income, the earned income tax credit, and infant health.” American Economic Journal: Economic Policy 7.1 (2015): 172-211 Janke, Katharina. ”Air pollution, avoidance behaviour and children’s respiratory health: Evidence from England.” Journal of Health Economics 38 (2014): 23-42. Knittel, Christopher R., Douglas L. Miller, and Nicholas J. Sanders. ”Caution, drivers! Children present: Traffic, pollution, and infant health.” Review of Economics and Statistics, 98.2 (2016): 350-366. Laumbach, Robert, Qingyu Meng, and Howard Kipen. ”What can individuals do to reduce personal health risks from air pollution?.” Journal of Thoracic Disease 7.1 (2015): 96. Lavaine, Emmanuelle, and Matthew J. Neidell, ”Energy production and health externalities: Evidence from oil refinery strikes in France.” Journal of the Association of Environmental and Resource Economists, 4.2 (2017): 447-477. Michel, Jacqueline et al. ”Extent and degree of shoreline oiling: Deepwater Horizon oil spill, Gulf of Mexico, USA.” PloS one 8, no. 6 (2013): e65087. Moretti, Enrico, and Matthew Neidell, ”Pollution, health, and avoidance behavior evidence from the ports of Los Angeles.” Journal of Human Resources, 46.1 (2011): 154-175. Middlebrook, Ann M., et al. ”Air quality implications of the Deepwater Horizon oil spill.” Proceedings

17

of the National Academy of Sciences 109.50 (2012): 20280-20285. McGowan, Craig J., et al. ”Respiratory, dermal, and eye irritation symptoms associated with Corexit EC9527A/EC9500A following the Deepwater Horizon oil spill: findings from the Gulf study.” Environmental health perspectives 125.9 (2017). Paris, Claire B., et al. ”BP Gulf Science Data Reveals Ineffectual Subsea Dispersant Injection for the Macondo Blowout.” Frontiers in Marine Science, 5 (2018): 389. Sanders, Nicholas J. ”What doesn’t kill you makes you weaker prenatal pollution exposure and educational outcomes.” Journal of Human Resources, 47.3 (2012): 826-850. Sbihi, Hind, et al. ”Perinatal air pollution exposure and development of asthma from birth to age 10 years.” European Respiratory Journal (2016): ERJ-00746. Solomon, Gina M., and Sarah Janssen. ”Health effects of the Gulf oil spill.” JAMA 304.10 (2010): 1118-1119. National Commission on the BP Deepwater Horizon Oil Spill, Deep water: the Gulf oil disaster and the future of offshore drilling, Perseus Distribution Digital, 2011. Reddy, C. M. et al. ”Composition and fate of gas and oil released to the water column during the Deepwater Horizon oil spill.” PNAS, 109.50 (2012): 20229-20234. Zivin, Joshua Graff, and Matthew Neidell, ”Environment, health, and human capital.” Journal of Economic Literature, 51.3 (2013): 689-730. Zivin, Joshua Graff, Matthew Neidell, and Wolfram Schlenker. ”Water quality violations and avoidance behavior: Evidence from bottled water consumption.”The American Economic Review 101.3 (2011): 448-453.

18

Table 1.A: Summary Statistics Treatment

Main Control

Diff.

PSM Control

Diff.

0.1846

0.1765

-0.0080

0.1840

-0.0006

[0.388]

[0.381]

(0.001)

[0.388]

0.5957 [0.491] 0.2940 [0.456]

0.5640 [0.496] 0.2607 [0.439]

-0.0315 (0.001) -0.0334 (0.001)

0.5959 [0.491] 0.2949 [0.456]

(0.002) 0.0002 (0.002) 0.0009

0.0738 [0.261] 26.6687 [5.784] 0.1034 [0.305]

0.1284 [0.335] 26.9859 [6.055] 0.1093 [0.312]

0.0547 (0.001) 0.3163 (0.018) 0.0059 (0.001)

0.0736 [0.261] 26.6550 [5.774] 0.1037 [0.305]

-0.0001 (0.001) -0.0136 (0.024) 0.0003

0.8191 [0.385]

0.7958 [0.403]

-0.0232 (0.001)

0.8193 [0.385]

0.0001

Mother characteristics Mother has college degree Mother is white Mother is black Mother is Hispanic Mother’s age Mother is < 20 years old

Mother is 20 to 35 years old

Mother is over 35 years old

(0.002)

(0.001)

(0.002)

0.0775 0.0949 0.0174 0.0770 -0.0005 [0.267] [0.293] (0.001) [0.267] (0.001) Note: This table shows summary statistics (mean and standard deviation). PSM control refers to characteristics from our propensity score matching. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Source: National Center for Health Statistics (NCHS).

19

Table 1.B: Summary Statistics (continued) Treatment

Main Control

Diff.

PSM Control

Diff.

0.0851 [0.279]

0.0919 [0.290]

0.0064 (0.001)

0.0844 [0.329]

-0.0007

Mother is married

0.5084

0.5753 [0.494] 0.5110

0.0670 (0.0014)

0.5087 [0.500]

Infant is male

[0.500] 0.5103 [0.500]

[0.500]

0.0007 (0.0015)

0.5103 [0.500]

0.0003 (0.002) -0.0001

0.4055

0.4038

[0.491]

[0.491]

-0.0147 (0.0014)

0.4057 [0.491]

0.3136 [0.464] 0.1648 [0.371]

0.3139 [0.464] 0.1632 [0.370]

0.0004 (0.0014)

0.3139 [0.464]

-0.0015 (0.0011)

0.1644 [0.371]

0.1156 [0.320] 0.0348 [0.184]

0.1058 [0.308] 0.0339 [0.181]

-0.0099 (0.0009) -0.0010 (0.0005)

0.1155 [0.320] 0.0348 [0.182]

-0.0001 (0.0013) -0.0001

0.0317

0.0303 [0.171]

0.0317 [0.175]

-0.0000

[0.175]

-0.0014 (0.0005)

0.5095 [0.500] 0.2187 [0.413] 0.0694

0.4817 [0.500] 0.1791 [0.383] 0.1110

-0.0278 (0.0015) -0.0396 (0.0011)

0.5091 [0.500] 0.2192 [0.414]

-0.0005 (0.002) 0.0005

Risk factors of pregnancy Fewer than 4 prenatal visits

First in birth order Second in birth order Third in birth order Fourth or higher in birth order Multiple Births

(0.001)

(0.002) 0.0001 (0.002) 0.0003 (0.002) -0.0004 (0.002)

(0.000)

Father characteristics Father is < 20 years old Father is white Father is black Father is Hispanic

0.0417 0.0691 [0.254] [0.314] (0.0009) [0.254] Note: This table shows summary statistics (mean and standard deviation). PSM control refers to characteristics from our propensity score matching. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Source: National Center for Health Statistics (NCHS).

20

(0.001)

(0.002) -0.0003 (0.001)

Table 2: Impact of the oil spill on air pollutants

Oil spill

(1) PM2.5

(2) NO2

(3) SO2

(6) CO

1.3767*** (0.339)

3.5003*** (0.281)

3.9229** (1.900)

0.0383** (0.019)

Mean in sample 11.44 89.90 22.06 0.3699 % change 12.03% 3.89% 17.78% 10.35% Note: This table shows the impact of the oil spill shock in April 20, 2010 on air pollutants. All the regressions include monitor, month by year and county fixed effects and state linear and quadratic time trends. The percentage change is calculated by dividing the regression coefficient from the mean in sample. The results are based on 264 monitoring stations for PM2.5, 108 for NO2, 146 for SO2 and 84 for CO in the southern United States. Time period is 2006 to 2012. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: US Environmental Protection Agency (EPA) AirData

21

Table 3: Impact of the oil spill on infant health outcomes

Oil Spill

(1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

0.0094*** (0.003)

0.0095*** (0.002)

-0.1034*** (0.031)

-9.3493** (4.745)

Note: This table shows the main results. LBW means low birth weight. Control variables consist of mother characteristics including mother’s age, mother’s education, mother’s race, whether the mother is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

22

Table 4: Heterogeneity effect of the oil spill on infant health outcomes for all women (1) all

(2) College degree

(3) Some college

(4) High School Or less

(5) Age over 35

(6) Age 26-35

Panel A Premature (<37 weeks)

0.0094*** (0.003)

0.0015 (0.003)

0.0050 (0.004)

0.0131** (0.005)

0.0042 (0.006)

0.0081** (0.003)

LBW (<2500 grams)

0.0095*** (0.002)

-0.0035 (0.004)

0.0070** (0.003)

0.0169*** (0.004)

-0.0006 (0.005)

0.0063*** (0.002)

Gestation (weeks)

-0.1034*** (0.031)

-0.0672 (0.044)

-0.0466 (0.042)

-0.1406*** (0.051)

0.0982 (0.140)

-0.0992*** (0.028)

BW (grams)

-9.3492** (4.745)

1.3554 (9.836)

-17.8310*** (3.814)

-16.8458*** (7.428)

-4.0834 (9.272)

-10.0549* (5.629)

(8) White

(9) Black

(10) Hispanic

(11) Married

(12) Unmarried

premature (<37 weeks)

(7) Age below 25 0.0119*** (0.003)

0.0054 (0.004)

0.0168*** (0.003)

0.0186*** (0.005)

0.0040 (0.004)

0.0142*** (0.003)

LBW (<2500 grams)

0.0145*** (0.002)

-0.0033 (0.004)

0.0184*** (0.005)

0.0212*** (0.004)

-0.0031 (0.004)

0.0180*** (0.003)

Gestation (weeks)

-0.1457*** (0.038)

-0.0559 (0.035)

-0.2293*** (0.065)

-0.1498*** (0.040)

-0.0575* (0.033)

-0.1714*** (0.043)

BW (grams)

-11.5069** (5.180)

-6.8417 (4.912)

-20.0028** (10.187)

-25.0283** (11.458)

-4.0244 (9.487)

-15.8396** (7.264)

Panel B

Note: This table shows subgroup analysis for the impact of the oil spill on infant health outcomes. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

23

Table 5: Impact of the oil spill on infant health outcomes by pregnancy status at time of oil spill (1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

Oil spill * pregnant

0.0088*** (0.003)

0.0099*** (0.003)

-0.1340*** (0.028)

-13.7980*** (5.042)

Oil spill * not pregnant

0.0097*** (0.003)

0.0093*** (0.002)

-0.0911** (0.039)

-9.4772* (5.451)

Note: This table shows the results by pregnancy status at the moment of the oil spill, using interaction terms. Control variables consist of mother characteristics including mother’s age, mother’s education, mother’s race, whether the mother is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male, as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

24

Table 6: The effect of the oil spill on infant health outcomes at different trimesters (1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

Oil spill * 3rd Trimester

0.0109*** (0.004)

0.0071*** (0.002)

-0.1308*** (0.038)

-5.4833 (6.025)

Oil spill * 2nd Trimester

0.0089** (0.004)

0.0090*** (0.003)

-0.1269*** (0.032)

-17.7161*** (5.700)

Oil spill * 1st Trimester

0.0066** (0.003)

0.0138*** (0.004)

-0.1462*** (0.037)

-20.1291*** (7.122)

Note: This table shows the results for the impact of the oil spill on infant health outcomes for different trimesters for mothers already pregnant, using interaction terms. Time spans have been limited to include women who have been exposed to the oil spill shock at first, second, or third trimesters of pregnancy. Control variables consist of mother characteristics including the mother’s age, education, race, whether she is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male, as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

25

Table 7: Impact of the oil spill on infant health outcomes using different specifications (1) (2) (3) (4) Premature LBW Gestation BW (<37 weeks) (<2500 grams) (weeks) (grams) Panel A–all U.S. Oil Spill

0.0098*** (0.003)

0.0085*** (0.002)

-0.0804*** (0.020)

-5.9449** (2.706)

0.0091*** (0.003)

0.0086*** (0.002)

-0.0844*** (0.020)

-6.7165** (2.716)

0.0135*** (0.005)

0.0090** (0.004)

-0.0717* (0.042)

-9.1513* (5.151)

0.0094** (0.003)

0.0095*** (0.002)

-0.1034** (0.044)

-9.3493* (5.221)

Panel B – all U.S. excluding NY and CA Oil Spill

Panel C – Propensity score matching Oil Spill

Panel D – Clustering state level Oil Spill

Note: This table shows the main results for different control groups for Panels A-C and different clustering in Panel D. Control variables consist of mother characteristics including mother’s age, education, race, whether she is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, whether the child is male, as well as father’s age group and race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. The time period is 2006 to 2012. The sample of Panel A is derived from 28,437,956 births in the U.S. over the period. One-to-one propensity score matching is done on mother characteristics: race, age, marital status and education, as well as the risk factors of the pregnancy. Standard errors are clustered at the county level for Panels A-C and at the state level for Panel D. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS). 26

Table 8: Instrumental Variable estimations - Impact of the oil spill on infant health outcomes (1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

0.0129*** (0.004)

0.0090*** (0.002)

-0.1036*** (0.027)

-13.9355*** (4.831)

0.0100*** (0.003)

0.0071*** (0.002)

-0.0805*** (0.020)

-11.1067*** (3.734)

Panel A – interpolation PM2.5

Panel B – no interpolation PM2.5

Note: This table shows the results using instrumental variables, using the oil spill to predict PM2.5 level of air pollution. Control variables consist of mother characteristics including mother’s age, mother’s education, mother’s race, whether the mother is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. In Panel A of Table 8, we first assign each county to the concentration of PM2.5 of the closest monitoring station. If there is more than one monitoring station in a county, we use the average of monitoring stations in the county. In Panel B, we only use counties with air monitor as spatial interpolation of PM2.5 can lead to bias. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Sources: National Center for Health Statistics (NCHS) and US Environmental Protection Agency (EPA) AirData

27

Table 9: Placebo test for the impact of the oil spill on pollutants (1) PM2.5

(2) NO2

(3) SO2

(4) CO

Oil spill * April 2008

-0.5909 (0. 497)

-0.4071 (0.298)

0.3520 (2.895)

-0.0244 (0.017)

Oil spill * April 2009

-0.3416 (0.467)

0.1215 (0.308)

2.9656 (2.424)

-0.0128 (0.016)

Note: This table shows the results for the placebo effect of oil spill on April 2008, and April 2009 on air pollution. All the regressions include monitor, month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: US Environmental Protection Agency (EPA) AirData

28

Table 10: Placebo test for the impact of the oil spill on infant health outcomes (1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

Oil spill * April 2008

0.0038 (0.004)

-0.0016 (0.003)

0.0588 (0.052)

1.7137 (5.146)

Oil spill * April 2009

0.0005 (0.003)

0.0006 (0.002)

0.0265 (0.036)

-2.7931 (3.999)

Note: This table shows the results for the placebo effect of oil spill on April 2008, and April 2009 on infant health outcomes. Control variables consist of mother characteristics including mother’s age, education, race, whether she is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, whether the child is male, as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

29

Figure 1 - Map

States Florida

Counties/ Parishes NOAA affected counties: Bay*, Escambia, Franklin*, Gulf*, Okaloosa, Walton*.

Alabama

NOAA Coastal counties: Bay, Calhoun, Charlotte, Citrus, Collier, De Soto, Dixie, Escambia, Franklin, Gadsden, Gilchrist, Glades, Gulf, Hardee, Hernando, Hillsborough, Holmes, Jackson, Jefferson, Lafayette, Lee, Leon, Levy, Liberty, Madison, Manatee, Marion, Monroe, Okaloosa, Pasco, Pinellas, Polk, Santa Rosa, Sarasota, Sumter, Suwannee, Taylor, Wakulla, Walton, Washington. NOAA affected counties: Baldwin, Mobile.

Mississippi

NOAA Coastal counties: Baldwin, Clarke, Covington, Escambia, Geneva, Mobile, Monroe, Washington. NOAA affected counties: Hancock, Harrison, Jackson.

Louisiana

NOAA Coastal counties: Amite, George, Hancock, Harrison, Jackson, Lamar, Marion, Pearl River, Pike, Stone, Walthall, Wilkinson. NOAA affected counties: Iberia, Jefferson, Lafourche, Orleans, Plaquemines, St. Bernard, St. Tammany*, Terrebonne, Vermilion*.

NOAA Coastal counties: Acadia, Ascension, Assumption, Avoyelles, Beauregard, Calcasieu, Cameron, East Baton Rouge, East Feliciana, Evangeline, Iberia, Iberville, Jefferson, Jefferson Davis, Lafayette, Lafourche, Livingston, Orleans, Plaquemines, Pointe Coupee, Rapides, Sabine, St. Bernard, St. Charles, St. Helena, St. James, St. John the Baptist, St. Landry, St. Martin, St. Mary, St. Tammany, Tangipahoa, Terrebonne, Vermilion, Vernon, Washington, West Baton Rouge, West Feliciana. Note: * Refers to counties with light or moderate trace only among NOAA affected counties. Source: National Oceanic and Atmospheric Administration’s (NOAA). 30

Figure 2 – Event study graphs Panel A

Impact on premature incidence

0.05 0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 -0.05 -5

-4

-3

-2

-1

0

1

2

3

4

5

Exposure (quarters) Panel B 0.05

Impact on lbw incidence

0.04 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 -0.05 -5

-4

-3

-2

-1

0

1

Exposure (quarters) 31

2

3

4

5

Panel C

Impact on PM25 level

12 8 4 0 -4 -8 -12 -5

-4

-3

-2

-1

0

1

2

3

4

5

Exposure (quarters) Panel D 18

Impact on SO2 level

14 10 6 2 -2 -6 -10 -14 -18 -5

-4

-3

-2

-1

0

1

2

3

4

5

Exposure (quarters) Note: Figure 2 presents event study graphs. Figure 2, Panel A shows the event study graph for the effect of the oil spill on incidence of premature and Panel B for incidence of low birth weight babies (right). Figure 2, Panel C shows the event study graph for the effect for level of PM2.5 and Panel D for level of SO2. Exposure is defined as the quarters before and after the oil spill. Source: National Center for Health Statistics (NCHS) and US Environmental Protection Agency (EPA) AirData

32

Appendix

Table A.1.A: Impact of oil spill on infant health outcomes for women using different treatment groups (1) Premature (<37 weeks) Narrow coastal counties (main results)

Coastal counties

Coastal counties + close counties

0.0094*** (0.003)

0.0073*** (0.003)

0.0043* (0.002)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

0.0095*** (0.002)

-0.1034*** (0.031)

-9.3493** (4.745)

-0.058615** (0.024)

-8.8389*** (2.849)

-0.0299 (0.021)

-7.8184*** (2.775)

0.0081*** (0.001)

0.0043*** (0.002)

Note: This table shows the results for the impact of oil spill on infant health outcomes using different definitions of the treatment group as defined in Figure 1. Control variables consist of mother characteristics including mother’s age, education, race, whether she is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, whether the child is male, as well as father’s age group and race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

33

Table A.1.B: Impact of the oil spill on infant health outcomes by oil intensity (1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

Panel AOil spill – High

0.0095*** (0.003)

0.0105*** (0.002)

-0.0980*** (0.033)

-11.6008* (5.927)

Panel BOil spill – Low or moderate trace

0.0081** (0.002)

0.0079*** (0.001)

-0.0994* (0.060)

-3.8568 (3.664)

Note: This table shows the results by impact of the oil spill for high and low and moderate trace. Control variables consist of mother characteristics including mother’s age, mother’s education, mother’s race, whether the mother is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male, as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS) and Quarterly Census of Employment and Wages (QCEW).

34

Table A.2.A: The effect of the oil spill on infant health outcomes for different coastal counties (CC) (1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

Oil spill * CC of Alabama

0.0182*** (0.005)

0.0134*** (0.002)

-0.1026*** (0.024)

-14.6960** (6.524)

Oil spill * CC of Florida

0.0164*** (0.004)

0.0109*** (0.004)

-0.2137*** (0.038)

-23.6277*** (6.477)

Oil spill * CC of Louisiana

0.0111*** (0.003)

0.0071*** (0.002)

-0.0391* (0.020)

-1.9043 (19.217)

Oil spill * CC of Mississippi

0.0014 (0.003)

0.0094** (0.004)

-0.1135*** (0.039)

-20.7949*** (2.931)

Note: This table investigates potential heterogeneity of the effect by states, using interaction terms. Control variables consist of mother characteristics including mother’s age, education, race, whether she is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, whether the child is male, as well as father’s age group and race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

35

Table A.2.B: Impact of the oil spill on infant health outcomes

Oil Spill Placebo treatment

(1) Premature (<37 weeks)

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

0.0055 (0.006)

0.0025 (0.003)

-0.0699 (0.050)

-5.0351 (8.935)

Note: This table shows the results using placebo treatment of coastal counties of Florida and Texas as treatment. LBW means low birth weight. Control variables consist of mother characteristics including mother’s age, mother’s education, mother’s race, whether the mother is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

36

Table A.3: The effect of the oil spill on PM2.5 by state, using interaction terms

PM2.5

(1) Alabama

(2) Louisiana

(3) Florida

1.3878*** (0.267)

0.5190*** (0.117)

0.4197*** (0.109)

(5) Mississippi 0.8090*** (0.135)

Note: This table investigates potential heterogeneity of the effect on PM2.5 by states, using interaction terms. This table investigates the impact of the oil spill on narrow counties by state. All the regressions include monitor, month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. * p < 0.10, ** p < 0.05, *** p < .01 Source: US Environmental Protection Agency (EPA) AirData

37

Table A.4: Impact of the oil spill on infant health outcomes, controlling for labor market effects (1) Premature (<37 weeks) Oil Spill

0.0101*** (0.003)

(2) LBW (<2500 grams) 0.0095*** (0.002)

(3) Gestation (weeks)

(4) BW (grams)

-0.1021*** (0.031)

-10.5763** (5.158)

Note: This table replicates the main results by controlling for change in labor market in coastal counties due to the oil spill. To account for changes in the labor market, we add controls for average employment rate and wage by counties to our specification and reestimate Table 3. Control variables consist of mother characteristics including mother’s age, mother’s education, mother’s race, whether the mother is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS) and Quarterly Census of Employment and Wages (QCEW).

38

Table A.5: Impact of the oil spill on infant health outcomes by labor market impact (1) Premature (<37 weeks) Panel AOil spill - Positive labor effect

Panel BOil spill - Negative labor effect

(2) LBW (<2500 grams)

(3) Gestation (weeks)

(4) BW (grams)

0.0106*** (0.004)

0.0075*** (0.002)

-0.0759*** (0.021)

-9.4729** (4.157)

0.0085** (0.004)

0.0124*** (0.002)

-0.1245** (0.047)

-16.2220*** (6.033)

Note: This table shows the results by impact of the oil spill on the labor market (employment and wage): positive versus negative labor market effect. Control variables consist of mother characteristics including mother’s age, mother’s education, mother’s race, whether the mother is married, and risk factors of the pregnancy including birth order, an indicator for whether it is a multiple birth, and whether the child is male, as well as father’s age group and father’s race. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS) and Quarterly Census of Employment and Wages (QCEW).

39

Table A.6: The effect of the oil spill on characteristics of mothers (1) College degree

(2) Some college

(3) High School Or less

(4) Age over 35

(5) Age 26-35

0.0270 (0.022)

0.0216 (0.016)

0.0463 (0.035)

0.0009 (0.001)

-0.0011 (0.002)

(6) Age below 25

(7) White

(8) Black

(9) Hispanic

(10) Married

(11) Unmarried

0.0002 (0.002)

-0.0033 (0.003)

0.0006 (0.002)

0.0036 (0.002)

-0.0023 (0.005)

0.0002 (0.002)

Panel A Propensity to become pregnant Panel B

Propensity to become pregnant

Note: This table shows the results for the effect of the oil spill on characteristics of new mothers. All the regressions include month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. The sample is derived from 9,324,839 births in the southern United States over the period. Standard errors are clustered at the county level. * p < 0.10, ** p < 0.05, *** p < .01 Source: National Center for Health Statistics (NCHS).

40

Table A.7: Number of hours per month that pollutant exceed the standard

Oil spill

(1) PM2.5

(2) NO2

(3) SO2

(6) CO

0.0544 (0.464)

0.0007 (0.001)

-0.6511 (0.584)

-

Note: This table shows the impact of the oil spill shock in April 20, 2010 on number of hours per month the pollutant exceeded the standard. All the regressions include monitor, month by year and county fixed effects and state linear and quadratic time trends. Time period is 2006 to 2012. Standard errors are clustered at the county level. CO is never exceeding the standard. * p < 0.10, ** p < 0.05, *** p < .01 Source: US Environmental Protection Agency (EPA) AirData

41

Figures A.1 – Map of monitoring station in the United-States Panel A - PM2.5

Panel B - NO2

42

Panel C - SO2

Panel D - CO

Note: Figure A.1 presents map of monitoring stations by pollutants. Source: US Environmental Protection Agency (EPA) AirData

43

Figures A.2 – Characteristics of mothers

Impact on having college degree

Panel A 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -5

-4

-3

-2

-1

0

1

2

3

4

5

2

3

4

5

Exposure (quarters) Panel B

Impact on mother white

0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -5

-4

-3

-2

-1

0

1

Exposure (quarters)

44

Panel C

Impact on mother's being hispanic

0.1 0.08 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 -0.1 -5

-4

-3

-2

-1

0

1

2

3

4

5

2

3

4

5

Exposure (quarters) Panel D

Impact on mother's being black

0.1 0.08 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 -0.1 -5

-4

-3

-2

-1

0

1

Exposure (quarters)

45

Panel E 4

Impact on mother's age

3 2 1 0 -1 -2 -3 -4 -5

-4

-3

-2

-1

0

1

2

3

4

5

2

3

4

5

Exposure (quarters) Panel F

Impact on mother not married

0.1 0.08 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 -0.1 -5

-4

-3

-2

-1

0

1

Exposure (quarters)

Note: Figure A.2 presents event study graphs for characteristics of mothers. Source: National Center for Health Statistics (NCHS) 46