thIn Utero Exposure to Maternal Stress: Effects of the September 11 Terrorist Attacks in New York City on Birth and Early Schooling Outcomes Melissa Eccleston October 11, 2011 PRELIMINARY DRAFT Abstract growing body of primarily observational research finds that exposure to maternal psychological stress while in utero may have substantial effects on physical and mental health, as well as cognitive ability. This paper estimates the causal effect of exposure to the stress of the September 11th, 2001 terrorist attacks on the cohort in utero that day using birth data from the National Vital Statistics System and New York City public school student records. The analysis finds that cohorts exposed during first or second trimester in New York City were 1-1.5 days younger, weighed 8-19 grams less, and had a 0.1% lower five-minute Apgar score at birth. Male and female newborns were affected similarly. Births in the rest of the United States were not affected. Initial educational attainment in New York City also suffered: at the age of six, boys were 7-9% more likely to be in special education and 15-18% more likely to be in kindergarten rather than first grade, with no effect on girls. The analysis accounts for alternative causal channels, namely air pollution and economic recession. Effects estimated outside the area of air pollution are very similar to estimates within it. Outcomes for a cohort exposed in utero to a period of high unemployment following September 11, but not to the terrorist attacks themselves, are not adversely impacted. The results suggest that psychological stress is an important channel through which adverse conditions experienced by pregnant women negatively impact the early life outcomes of in utero cohorts. JEL Codes: I14, I15, J13, O15. thA I. Introduction Chronic psychological stress contributes to a broad range of adverse health and cognitive outcomes. A growing body of research finds that stress may have substantial impacts even before birth. In animal experiments, maternal exposure to temporary psychological stressors during pregnancy results in offspring with impaired physical health and diminished cognitive, emotional, and behavioral abilities (Kaiser and Sachser 2005). However, our understanding of the nature and magnitude of effects for humans remains limited due to the difficulties of causal estimation. Omitted variables bias, in particular the direct effect of stressors on outcomes, makes the results of observational studies difficult to interpret. Studies that do utilize exogenous variation are limited by the small number of relevant outcomes measured. The terrorist attacks of September 11 th, 2001 created a substantial and widespread psychological shock, providing an opportunity to estimate the causal effects of stress exposure in utero on early life outcomes. The analysis focuses on New York City, where the destruction and psychological impact were greatest. Using Vital Statistics birth records from 1995 through 2004 and New York City public school student records from the 2003 through 2009 school years, I estimate the effects of in utero exposure to the attacks on maternal health in pregnancy, health at birth, and early schooling performance. In line with the human and animal literature, I define the cohort first or second trimester thin utero on September 11, and therefore conceived in the spring or summer of 2001, as exposed. The unexpected nature of the attacks and the limited duration of the psychological impact allow causal estimation by comparing outcomes of the exposed cohort to those of cohorts conceived 1 prior to the spring of 2001 and to cohorts conceived after the attacks.1 thPregnant women exposed to September 11 during first or second trimester were 2-5% more likely to suffer from a medical risk and 2% more likely to experience a complication during labor. Newborns were 1-1.5 days younger, weighed 8-19 grams less, and had a 0.1% lower five-minute Apgar score at birth.2 Male and female newborns were affected similarly. Analysis of national birth records finds neither effects in geographies proximate to the attacks nor a national impact. Diminished health at birth has long-term effects on health and socioeconomic status, mediated in part by childhood health and educational attainment (Case, Fertig and Paxson 2005; Oreopoulous et al. 2008). Additionally, results from the animal experimental literature suggest that in utero exposure may have cognitive, emotional, or behavioral effects which are not manifested or measured at birth. I find evidence of such effects for boys at the onset of their education. At the age of six, boys were 7-9% more likely to be in special education and 15-18% more likely to be in kindergarten rather than first grade, with no effect on girls. thFeatures of September 11’s economic and environmental consequences in New York City allow for assessment of their impacts on in utero cohorts. Air pollution was highly geographically concentrated in lower Manhattan and western Brooklyn, and estimation excluding these areas finds effects of similar magnitudes and significance. The economic thdownturn thcohort attributed to September 11 was prolonged, and examination of the outcomes of a conceived well after September 11 but exposed to poor economic conditions in utero finds no effects. 1 Cohorts conceived prior to the spring of 2001 were either third trimester in utero or already born on September 112th. Five-minute Apgar score is a “summary measure of the infant’s condition” rating five aspects of newborn’s health five minutes after birth (NCHS Technical Appendix 2004). 2 These results shed light on a difficult to measure mechanism through which a range of adverse shocks in utero affect individual outcomes.3 Psychological stress may be a particularly important channel in explaining the significant impacts of in utero exposure to relatively mild adversities, such as severe storms (King and LaPlante 2005; Simeonova 2009), temporary income shocks (Burlando 2010), and cold weather (Stoecker 2011).4 In addition, the estimates contribute to our understanding of the long-term effects and full costs of the attacks of thSeptember 11. The remainder of the paper is organized as follows. Section II describes the primary theffects of September 11 in New York City and their implications for the cohort in utero. Section III describes the data sources and construction. Section IV presents the main results, followed by investigations of the potential for selective migration, heterogeneous effects, and alternative causal channels. Section V concludes. II. Background II.A. September 11 th, 2001 in New York City great loss of human life and physical thThe terrorist attacks of September 11 caused property. The unprecedented scale and method of attack, vividly captured in film and photos, elicited shock, sadness, and other feelings associated with trauma in onlookers around the world. The most dramatic destruction and over 90% of the casualties occurred in New York City, which experienced substantial environmental, economic, and psychological fallout over the ensuing months and years. 3 See Almond and Currie (2011a; 2011b) for overviews of the long-run effects of in utero exposures and early life shocks. 4 With the exception of Stoecker (2011), these papers find that adverse exposures in first and/or second trimester affect outcomes. 3 Environmental effects September 11th created “an acute environmental disaster” in parts of New York City (Landrigan et al. 2004). The collision of the airliners with the towers and the towers’ subsequent collapse generated thousands of tons of airborne pollutants, including glass fiber, cement dust, and a range of chemical compounds (Clark et al. 2001). Much of the particulate matter settled in the vicinity immediately surrounding the site of the World Trade Center, leading to elevated levels of respiratory illness among rescue workers and nearby residents (Farfel et al. 2008). A large dust cloud also formed, spreading the airborne pollution across lower Manhattan, New York Harbor, and western Brooklyn in the days following the attack before dispersing (Clark et al. 2001). The effects of prenatal exposure to environmental toxins have been well documented. Air pollution has been linked to infant mortality (Currie and Neidell 2005), lead exposure to low birth weight (Reyes 2005), and radioactive fallout to diminished cognitive ability (Almond, thEdlund and Palme 2009). After September 11, two studies of pregnant women who were near the World Trade Center towers during the attacks found a higher incidence of small-forgestational age births compared to other births at the same hospital (Landrigan et al. 2004).5 thBased on satellite imagery from existing monitors and samples collected after September 11, the Environmental Protection Agency concluded that levels of particulate matter in the air only exceeded thresholds for health problems in the immediate vicinity of the World Trade Center in the weeks following the attack (Lorber et al. 2007). I will use this spatial concentration to assess if the effects of exposure could have been caused by pollution rather than stress. th percentile of weight for gestational age. One study Small-for-gestational age is defined as being below the 10 found an effect on gestational age, but neither found an effect on birth weight. A limitation of these studies is that the control group was likely affected as well. 5 4 Economic effects Employment in New York City had been slowly declining since the beginning of 2001, and dropped off sharply in the months after the attacks. Approximately 92,000 (or 2% of) private sector jobs were lost in the city from September 2001 to March 2002, and 50-75% of these losses were estimated to be a direct consequence of the attacks (Bram, Orr and Rapaport 2002). After peaking in the second quarter of 2002, the unemployment rate declined modestly, but then rose again in early 2003 before steadily declining, aided in part by $10 billion in federal economic redevelopment assistance (Makinen 2002; U.S. Department of Labor 2011). Poor economic conditions while in utero have been found to affect a range of outcomes. In the extreme, prenatal famine exposure has been shown to increase the rate of schizophrenia in adulthood, a link thought to be caused by a lack of micronutrients causing fetal genetic mutations (Neugebauer, Hoek and Susser 1999 ). Common economic cycles have also been shown to matter. Van den Berg, Lindeboom and Portrait (2006) find that cohorts in utero during recessions in the Netherlands had higher mortality rates than their counterparts in utero immediately before recessions began. Several channels could account for this effect. Reduced income could, for example, push a pregnant woman to reduce her consumption of healthcare or significantly increase her work effort. In contrast, Dehejia and Lleras-Muney (2004) find that birth outcomes during times of high unemployment improve due to healthier behaviors and, for some subgroups, positive selection into birth. I will assess the effect of in utero exposure to the economic thslowdown by examining the outcomes of a cohort conceived after September 11, when the unemployment rate was high and rising, but the psychological shock had faded for most. 5 Psychological effects The purpose of terrorism is “to intimidate a watching popular audience by harming only a few”, and the dramatic violence of the attacks affected the mental health and behavior of millions with no direct connection to the victims (Crenshaw 2000). A national phone survey conducted ththree to five days after September 11 (a weekend) found that 61% of respondents living within 100 miles of New York City, and 42% living farther away, were suffering from at least one substantial stress symptom such as repeated disturbing thoughts or dreams about the attacks, or feeling very upset when reminded of the attacks (Schuster et al. 2001).6 Additional evidence suggests that the psychological effects were both concentrated in New York City and limited in duration. In the three month period following the attacks, emergency room visits for mental or behavioral disorders rose 10% and the number of psychiatric medication prescriptions rose 18% among Medicaid enrollees resident in the City relative to enrollees in the rest of New York State (DiMaggio, Galea and Richardson 2007; DiMaggio, Galea and Madrid 2007). Self-reports from phone surveys suggest that these rates had declined back to long-run averages by four to five months after the attacks (Boscarino et al. 2004). More broadly, one month after the attacks, the rates of probable post-traumatic stress disorder (PTSD) and depression among New York City residents were estimated to be over two times higher than the national average, which itself was unaffected by the attacks (Galea et al. 2002; Schlenger et al. 2002).7 These rates had returned to national averages within five months 6 There are limited data with which to compare these results, as few studies report “the prevalence of trauma-related symptoms of stress in people who do not necessarily meet criteria for a psychiatric disorder” (Schuster et al. 2001). A study of the 1980-81 Epidemiological Catchment Area survey provides evidence consistent with a point-in-time trauma-related stress symptom prevalence rate among the general population of roughly half that found in Schuster et al. (Helzer, Robins and McEvoy 1987). 7 The surveys in these studies were conducted through random-digit telephone dialing for Manhattan residents below 110th street. 6 (Galea et al. 2003). 8 Behaviors were also affected. Surveys found increased alcohol, tobacco, and marijuana use among New York City residents who were prior consumers in the months after thSeptember 11 (Vlahov et al. 2004). Becker and Rubinstein (2008) find evidence of avoidance of commercial flights: national passenger volumes, relative to cargo, declined 10 percentage thpoints after Septembe r 11, and attack-rel ated routes were particularl y affected.9 While there is no clear theory to guide the investigation of in utero impacts across thgeographies, the evidence from September 11 suggests that both physical and social proximity are associated with larger psychological effects.10 In particular, the surrounding suburbs of New Jersey and New York State, as well as the Washington, D.C. metropolitan area, may have been affected.11 National and area-specifi c effects will be investigated in Section IV.C. II.B. Prenatal effects of psychological stress According to the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition), psychosocial stressors include difficult or negative events and situations related to one’s family, friends, work, or community. Stressors prompt cognitive and emotional reactions, and activate a physiological stress response. This physiological response is remarkably consistent across imminent external stressors as well as thoughts or feelings regarding past or potential stressors (Sapolsky 2004). A woman’s physiological stress response is thought to affect pregnancy through three interrelated biological channels. A primary component of the body’s complex stress response is Among rescue workers and those directly exposed to the attack, rates of PTSD have actually risen over time (Thorpe 2009). Becker and Rubinstein (2003) find that the NYC-London route suffered a substantial decline in passenger volume compared to the Chicago-Paris route. 10 I define social proximity as the likelihood of knowing someone affected by the attacks. The third crash site, near Shanksville, Pennsylvania, is less likely to have been affected, given that nearby residents were unlikely to have known the victims on the flight, no one on the ground was injured or killed, and the area was not the intended target. 1198 7 the production of hormones, including corticotropin-releasing hormone (CRH) and cortisol. CRH plays important roles in fetal maturation and in the process of parturition (Wadhwa et al. 2001). Observational medical studies consistently find strong correlations between maternal psychosocial stressors, CRH levels in early to middle pregnancy, and risk of preterm birth and delivery complications (Paarlberg et al. 1995; Dunkel-Schetter et al. 2000). These studies utilize women’s self-assessments of stress or self-reported incidents of stressful events. For example, Da Costa et al. (2000) find that women who report facing stressful incidents were more likely to have a complication in labor or delivery, and the newborns of women who were dissatisfied with their social support weighed less. In addition to regulating fetal maturation, hormones produced in response to stress play an epigenetic role in fetal development of nervous, endocrine, and immune systems, thereby affecting future cognitive, behavioral, and emotional abilities, as well as health (Meaney and Seckl 2004; Van den Bergh et al. 2005). Excess exposure to cortisol has been correlated with impaired development of the brain and spinal cord, as well as diminished mental and motor skills of infants (Yu et al. 2004; Huizink et al. 2003). Several observational studies have also found negative correlations between women’s blood cortisol or self-assessed level of stress during pregnancy and the emotional and mental health of their children through middle school age (de Weerth and Buitelaar 2005; Malaspina et al. 2008). Effects on health outcomes are just beginning to be studied in humans; for example, Li et al. (2010) find that in utero exposure to maternal stress increases the risk of being overweight starting at age 10. Thirdly, chronic stress has a well-established correlation with disease and deterioration of bodily systems. The theory of allostatic load posits that this is a causal relationship, by which frequent stimulation of the body’s stress response causes harm over time (McEwen and Stellar 8 1993). Some of the conditions correlated with chronic stress are known to directly affect pregnancy; for example, hypertension increases the risk of labor complications and preterm birth (Warren, Gurewitsch and Goland 1995). In addition to the biological channels, maternal stress may also affect pregnancy through behavior. Patterns of eating, sleeping, and drug use can all be altered by psychological stress, and directly affect the health of a pregnancy. While the relationships between and relative importance of these channels continues to be studied, the current evidence suggests that stress is most strongly correlated with negative outcomes when experienced in the first two trimesters of pregnancy (Wadhwa 1998; Glynn et al. 2001). This may be driven by the range of vulnerable developmental processes that occur early in pregnancy; in addition, women may become less reactive to stressors in the third trimester as a protective mechanism (Kammerer et al. 2002). However, because these observational studies lack sources of exogenous variation in stress, the concern remains that measures of stress are correlated with other determinants of fetal health, including stressors that directly affect health and relevant maternal characteristics, such as the ability to cope with difficulties. In medical research, randomized experiments with animals ranging from rats to primates have provided causal estimates of the effects of temporary stress on pregnancies. Exposure of pregnant females to a range of stressors, including unstable social groupings, unexpected stimuli, or physical discomfort, results in offspring with lower birth weight, impaired cognitive ability, or diminished behavioral skills (Kaiser and Sachser 2005). The form and severity of impairment, as well as differences by offspring gender, vary with the species, the type and severity of stressor imposed, and the timing of exposure. For humans, plausibly exogenous shocks to stress are rare, but a small number have been used to identify causal impacts of stressful events on birth outcomes. Hansen, Lou and Olsen 9 (2000) examine the effects of a deadly diagnosis or death of a loved one during first trimester of pregnancy in a sample of Danish women, finding a marked increase in the rate of cranial neural crest malformations.12 With an approach similar to this paper, Camacho (2008) utilizes the unexpected timing of terrorist landmine explosions in Colombia and finds that cohorts in utero first trimester when a landmine exploded in the municipality of their birth weighed approximately 9 grams, or 0.3%, less than average. However, birth weight is the only outcome Camacho is able to examine, leaving unanswered the potential for effects on the health of a pregnancy, other measures of health at birth, and later life outcomes. While the existing literature does not include the use of exogenous shocks to measure effects beyond birth, Aizer, Stroud and Buka (2009) reduce the scope for bias from unobserved maternal characteristics by utilizing variation in blood cortisol across pregnancies for siblings. They find that childhood health, IQ, and ultimate educational attainment are negatively impacted by high maternal blood cortisol during pregnancy.13 Recent medical findings suggest that female fetuses may be less susceptible or better able to adapt to changes in their uterine environment, and therefore potentially less likely to be impacted by maternal stress (Stark, Wright and Clifton 2009; Charil et al. 2010). However, evidence of gender differences in susceptibility to the effects of stress is mixed. After negative shocks such as economic recessions and natural disasters, which cause both psychological and physical stress, the sex ratio at birth is often found to shift towards females (Catalano et al. 2006). Some studies, including Aizer, Stroud and Buka (2009), find no differences across gender in the magnitude and type of effects. 12 Cranial neural crest are precursor cells responsible for the formation of most structures of the head and face. These cells are known to be sensitive to a range of environmental signals, and resulting defects include cleft lip and cleft palate. 13 Aizer, Stroud and Buka (2009) find weak evidence for effects on birth outcomes, which they believe may be due to the way in which their sample was constructed (see pgs. 18 and 24). 10 thII.C. Psychological effects of September 11 on pregnancy outcomes thA handful of studies have examined September 11’s psychological effects on a narrow range of pregnancy outcomes, and most have found negative impacts. In New York City and surrounding New York State suburbs, Eskenazi et al. (2007) find an increase in the rate of very low birth weight (<1500 grams) births for pregnancies that were first or second trimester on thSeptember 11.14 Catalano et al. (2005; 2006) find that the sex ratio at birth dipped in January 2002 in both California and New York City, which the authors suggest is due to increased rates of spontaneous abortion of male fetuses caused by the psychological trauma of the attacks. thThe spike in incidents of discrimination and hate crimes following September 11 was itself a source of stress for the targeted population. Lauderdale (2006) estimates the effect of increased discrimination on the birth outcomes of Arabic women in California. She finds that births to women with distinctly Arabic names were two times more likely to be low birth weight in the six calendar months after the attacks. Results from other studies of births after September 11th vary, potentially because they have focused on populations less likely to have been affected. In a sample of Dutch infants, Smits et al. (2006) find a 48-gram reduction in birth weight for ththose in utero on September 11, but Rich-Edwards et al. (2005) find that a sample of women in thBoston who were in their first trimester of pregnancy on September 11 were less likely to deliver preterm (Rich-Edwards et al. 2005). Endara et al. (2009) find no effect on birth outcomes for a sample of active-duty military families residing across the U.S. By utilizing the full range of relevant outcomes available in the Vital Statistics birth data, I provide evidence of the effects of psychological stress on health at birth and one potential causal channel, maternal health in pregnancy. In addition, the evidence from early schooling 14 Eskenazi et al. (2007) use data from 1996 through 2002 and therefore have a limited post-September 11th sample. The authors also consider the rate of pre-term birth, finding no evidence of impact. Effects on birth weight in the suburbs are weaker than those in New York City. 11 records provides the first estimate of effects beyond birth using plausibly exogenous variation in stress. III. Data III.A. Pregnancy outcomes Data on pregnancy outcomes come from the National Center for Health Statistics’ Vital Statistics public use data files.15 The birth data files contain the information recorded on every birth certificate issued in the United States. I combine this data with the fetal death data files, which contain information recorded on all reported fetal deaths. I limit the sample to live births and fetal deaths occurring to women whose residence is listed as one of the five counties, equivalent to the five boroughs, of New York City. I pool the data between 1995 and 2004, resulting in almost 1.2 million observations.16 The data report month and year of birth, as well as estimated weeks of gestation at birth. Using this information, I determine whether a pregnancy was in progress, and if so, in which thtrimester, on September 11, 2001. I first assign all births in a given month to a birthdate equal to the midpoint of that month. I subtract from this assigned birthdate the gestation estimate to arrive at a conception date. thI then count the number of weeks from the conception date to September 11, 2001. t , 2001 were 0-13 weeks in utero on thBirths conceived between June 18 and September h 10 thSeptember , and therefore defined as exposed during first trimester. Births 11 conceived thbetween March th and June 17, 2001 were 14-27 weeks in utero and therefore defined as 12 exposed during second trimester.17 15 These files are available online at http://www.cdc.gov/nchs/data_access/Vitalstatsonline.htm The sample ends in 2004 because public Vital Statistics do not report geographic information in subsequent years. There may be error in birth certificate measures of gestation length (Wier 2007). I present an alternate version of specification (1), ignoring gestation estimates entirely, in Appendix Table I. Results are very similar. 1716 12 III.A.i. Following the medical literature, I identify six outcomes that may be affected by maternal stress. 1) For live births, I consider two indicators of the health of a pregnancy. 18a. Any maternal medical risk factors: An indicator variable for the presence of one or more factor. These factors are a set of diseases and health conditions that a pregnant woman might enter pregnancy with or develop over its course, and that can negatively impact a pregnancy’s outcome. Factors that have been shown to be induced by stress are hypertension, eclampsia, and genital herpes, although other factors may be related to stress (Higgins et al. 2002; Chida and Mao 2009). b. Any complication of labor or delivery: An indicator variable for the presence of one or more complication. These complications include placental abnormalities and precipitously fast or excessively prolonged labor. Complications of labor and delivery are more likely in women with high levels of stress and in women with certain medical risks, including eclampsia (NCHS Technical Appendix 2004). 2) For live births, I consider three indicators of the health of a newborn. a. Birth weight: Weight immediately after birth, measured in grams. b. Gestational age: Estimated in weeks from the date of mother’s last menstrual period (when unknown, a clinical estimate is used). c. Five minute Apgar score (scaled from 0 to 10): A “summary measure of the 19infant’s condition” rating five aspects of newborn’s health (NCHS Technical Appendix 2004). 203) Fetal Death: An indicator variable for a pregnancy resulting in a fetal death rather than a live birth. While New York requires the reporting of fetal deaths at all stages of pregnancy, the majority of states do not require reporting under 20 weeks gestation. Fetal death is consistently underreported and gestation length measured with error. Underreporting and errors are most severe at the earliest gestational lengths (Goldhaber 1989). I therefore consider only fetal deaths occurring at 20 weeks gestation or later, and utilize only month and year of delivery data. 18 Appendix Table II presents results for stress-related and other medical risk factors separately. The five aspects are heart rate, respiratory effort, muscle tone, reflex irritability, and color. Each aspect is scored from 0 to 2, and the scores are totaled across aspects, with a score of 10 considered optimal. The score is named after its inventor, Dr. Virginia Apgar. 20 Including fetal deaths at all gestation lengths yields very similar results. 19 13 III.A.ii. Characteristics of the birth and mother recorded on birth certificates serve as control variables.21 1) Gender of the newborn. 2) Indicator of plural birth (twins or higher order). 3) Age: Mother’s age at time of delivery. 4) Race: Mother’s self-identified affiliation, for which I construct five indicator variables (Black, White, Hispanic, Asian, and Other). 5) Marital status: An indicator of whether or not mother is married at time of delivery. 6) Years of schooling: Mother’s educational attainment, for which I construct four indicator variables (0-8 years, 8-12 years, 12-16 years, 16 or more years). 7) Parity: Number of live births mother has had prior to the current one. 8) Mother’s borough of residence. III.B. Schooling outcomes Data on schooling comes from the administrative records of the New York City Department of Education for the 2003/4 through 2009/10 academic years.22 In order to thseparately identify the effect of exposure to September 11 from school year and grade effects, I limit the sample in each academic year to children who turn six in the calendar year in which that academic year begins. This allows me to observe children born before and after the exposed cohort, and to use grade level as an outcome measure, because New York City allows children to stenter first grade if their sixth birthday occurs by December 31 of the calendar year in which the 21 Other recorded characteristics, such as tobacco and alcohol use, may be partially determined by the effects of stress and are omitted. Tobacco and alcohol use, on both extensive and intensive margins, were examined as behavioral outcomes of September 11 22th exposure, and no results were found. Based on comparisons of enrollment numbers with population counts in the American Community Survey, approximately 65-80% of elementary school-aged children (K-8 grades) in New York City are in public school. Selective sorting out of public school by unaffected children in the exposed cohort would create bias in the direction of the results. 14 academic year begins. The administrative records do not include data on student’s place of birth or parents’ former places of residence. I exclude all children identified as English language learners, as their mothers are relatively less likely to have been residing in New York City when pregnant.23 With this exclusion, the sample consists of slightly more than 410K observations of six year old students across academic years. The data report exact date of birth, which I use to define the affected cohort as children thborn March th to June th, 2002 (27-40 weeks later) and December 11, 2001 to March 19th, 20 19 2002 (13-26 weeks later). If gestation for all children had been approximately 40 weeks, these definitions would correspond directly to first and second trimester in utero exposure. If gestation is shortened by exposure, using date of birth to identify exposure will incorrectly identify some children as having been exposed one trimester later than they actually were. These errors may attenuate the estimated effect of exposure in first trimester.24 III.B.i. Although the cohort exposed to September 11th in utero is still very young, I am able to construct five measures of children’s educational performance.25 1) Attendance: A count of the number of days a student is recorded absent from school. At the age of six, absence may be a plausible indicator for child health (rather than an indicator of skipping school). 2) Behavior: An indicator equal to one if a student has one or more behavioral incidents during a school year. Incidents are recorded and punished on a scale of one to four. For children aged six, commonly reported incidents include “being insubordinate” and “fighting/engaging in physically aggressive behavior”. 3) Special education: An indicator equal to one if a student is enrolled in a special 23 Over 50% of English Language Learners are immigrants, and a significant additional share is believed to have parents who are immigrants (NYC Department of Education 2007). 24 Because these errors may assign some first trimester exposures to second trimester, and some second trimester exposures to third trimester (and therefore non-exposure), their net effect on the estimate of exposure in second trimester is indeterminate. 25 Public school students in New York City begin taking state standardized tests in grade three. 15 education program. Enrollment is highly correlated with both cognitive impairments and behavioral issues (NYCLU 2011). 4) Held back: An indicator equal to one if a student is in kindergarten rather than first grade in the calendar year in which he or she turns six. 5) Average grade per course: For kindergartners through fifth graders, courses are graded on a scale of one to four. This data is available only for a 10% subsample, and is not representative of the full sample. III.B.ii. Students’ demographic and residential information is used to construct control variables.26 1) Student’s gender 2) Student’s race: grouped into five categories (White, Black, Hispanic, Asian, and Other) 3) Student’s current home zip code IV. Econometric Specifications and Results IV.A. Pregnancy outcomes Live births th This section presents evidence of the effect of in utero exposure to September 11 for births to women residing in New York City. I begin by comparing mean outcomes of live births across years, grouped by estimated time of conception in order to account for seasonality in birth thcharacteristics. As detailed in Section III.A., pregnancies in first trimester on September 11 are thconceived approximately 0-13 weeks prior, between June th and September 10, and 18 pregnancies in second trimester are conceived approximately 14-27 weeks prior, between March 12t th and June 17. h 26 Student’s free lunch status is also recorded, but is missing for nearly one quarter of the sample. Therefore I do not include it as a covariate but do examine it as an outcome when assessing selection in Section IV.D. 16 Figure I displays the mean values of the five outcomes of interest for live births conceived during these two windows annually from 1995 to 2003.27 While there is substantial annual variation, pregnancies exposed to the attacks in 2001 deviate negatively from trends across the five outcomes. As a falsification exercise, Figure II displays the same mean outcomes for births thconceived between December th and March 11, which are approximately third trimester in 27 , and for which studies predict limited effects of exposure to stres t h Similar annual variation is apparent, but the cohort’s outcomes are in line with neighbori u t years. e r Regression analysis of live births exploits the limited nature of the exposure, and mea o ththe extent to which outcomes for children in utero on September 11 differ from the outc o children born immediately before and after: n S e p t e m b e r 1 1 (1) � � = � 1 2 * � ( � 𝑛� � � � 2� ) + + + +� � � � � � are the five measures of the health of a pregnancy � birth is estimated to be first or second trimester in utero on September 11thand of a newborn graphed in Figure I. The binary variables inutero1 and inutero2 indicate if a , 2001. Xiof covariates: newborn’s gender; indicator of plural birth; mother’s parity, age, race, marital status, and educational attainment. d and γ are month and year of birth indicators, capturing both within-year seasonality and trends across years in birth outcomes. The parameters � 1 and � 2therefore identified from differences in outcomes between births occurring in the same year, are 2001 or 2002, which were and were not exposed in utero. 27 Conceptions in 2004 are omitted because the sample ends in 2004 and thus does not include those conceived in 2004 and born in 2005. 17 The outcome variables of interest * � ( � 𝑛� � � � 1� ) � +� � � � � 𝑦� is a vector Table I, Panel A presents the results of this specification. The effects are similar for the first and second trimester cohorts, ranging from 0.1% to 5% at sample means.28 Pregnant women thexposed to September 11 in their first two trimesters were 2-5% more likely to have a medical risk factor and were about 2% more likely to suffer from a complication of labor.29 Their newborns were 1-1.5 days younger, 8-19 grams lighter, and had a 0.01 point lower Apgar score. The magnitude of effects is similar to other estimates of the effect of stress (Camacho 2008). However, compared to direct harmful exposures, the effects are relatively small. For example, in the sample, the newborns of mothers who smoke are 240 grams lighter, 4 days younger, and have a 0.08 point lower Apgar score. Table I, Panels B and C present the results of this specification fully interacted with gender to examine potential differences in the magnitude of effects between males and females. The majority of the coefficient estimates are statistically equivalent for males and females, suggesting that effects of exposure are similar across genders.30 Because the timing of birth is endogenous, Table II presents an alternative specification utilizing month and year of conception indicators, rather than month and year of birth indicators. The magnitude and statistical significance of the estimates are very similar to those in Table I, Panel A. Selection into live birth thIn addition to negatively affecting the outcomes of live births, exposure to September 11 may have increased the risk of fetal death. I examine the rate of fetal deaths relative to live 28 Significant differences exist between trimester effects for three of the outcomes. Medical risks increase more for inutero2 exposure, gestation length decreases more for inutero1 exposure, and birth weight decreases more for inutero2 exposure. 29 Specifications using separate outcome indicators for stress-related and other medical risk factors are presented in Appendix Table II. Effects are significant only for stress-related factors. 30 Of the 10 coefficient comparisons, two are significantly different across gender. Pregnancies carrying females exposed in inutero2 have a larger increase in medical risks than those carrying males. Males exposed inutero1 have a larger reduction in Apgar score than females. 18 birt hs, for the sam ple as a who le and by gen der, mot ivat ed by the evid ence of a shift ed sex ratio at birt h in Cat alan o et al. (20 06). Bec aus e of the limi ted acc urac y of feta l deat h data ,I use onl y the mo st basi c info rma tion to esti mat e the effe ct of exp osu re – the mo nth and yea r in whi ch the fe tu s w as de li ve re d fo r th e si x m on th s fo ll o wi ng th e att ac ks : (2) � � 𝑎 h� = * � � 1 + ( � + 𝑏 � � 𝑛 � � 2 0 0 1 4 7 � ) + � * � ( � � 𝑟 � thwindows 18 a � ) * + � ) * � ( � 𝑏 ( � 𝑛 � � � 𝑟 𝐹 � 6 2 0 0 1 � 2 0 0 2 � ) Results of this specification, presented in Appendix Table III, show no effect on the fetal thdeath 11. rate in the months following September However, it is possible that systematic underreporting of miscarriages could mask a real effect. I therefore use specification (2) with male gender as the outcome in the sample of live births. Results, which show no significant 31,32effect on the sex ratio at birth, are presented in Appendix Table IV. In addition, while the nature of the attacks strictly limits the scope for selection into (June pregnancy, there may have been migration following the attacks that would affect the sample. Table III presents means by year for live births conceived in the first and second trimester to September th th and March 12 to June 17th). These means do not 10 present cl ea r pa tte rn , bu tI wi ll ex pl or e th is is su e fu rt he r in Se cti on I V. D. 31 The samp le used in Catal ano et al. (200 6) ends in June 2002 . Omit ting birth s after June 2002 prod uces resul ts in line with Catal ano et al. Usin g male sex as the outc ome in speci ficati on (1) with only year and mont h of birth as cova riates also yield s no signi fican t effec ts. 32 Relat edly, predi ction s in the popu lar press of a “bab y boo m” follo wing Sept embe r 11th find no supp ort in the data. In fact, the coho rt born 9-12 mont hs later (July -Sept embe r 2002 ) was one of the small est JulySept embe r coho rts born from 1995 throu gh 2004 . Ruth er (201 0) finds evide nce of incre ased fertil ity rates two to three years later, but it is not clear that this incre ase is relat ed to the attac ks. 19 IV.B. Schooling outcomes The preceding results show that exposure to September 11th in New York City negatively impacted the health at birth of the cohorts that were exposed first or second trimester in utero. While I am unable to track individual children, I can examine the schooling outcomes of the cohort that was exposed in utero. Diminished health at birth may lead to diminished childhood outcomes, and there may be additional negative effects of in utero stress exposure not manifested or measured at birth. = � 1 * � ( 𝑏� � � 𝑎20� � 𝐽� � 18 The outcome variables of interest � )+� + + + � � � � � * � ( 𝑏� � � � � 11𝑡� 𝑎19� ) + � 2 � ( 3 Boys ) who were born 3-9 months after September 11 � � underlying health and personality variation. 33 are the four schooling outcomes defined in Section I I gender, race, and home zip code. Table IV, Panel A includes the indicators of child’s I presents the results of this specification . for the full sample of six-year-old public school children. thThe outcomes are largely unaffectedXby in utero exposure to September 11. However, when d u r i n g i fully interacted with gender in Table IV, Panels B and C, there is clearly an effect among boys. th (and therefore exposed approximately f i rst or cant effects for absences secon and behavioral incidents suggests that the impact of exposure, at least at age six, may be d trimes ter) primarily cognitive. However, it may be that these outcomes do not adequately capture were 33 7-9% more likely to be Aizer, Stroud and Buka (2009) do find childhood health effects of mother’s cortisol during pregnancy with data from doctor’s visits. in 20 specia l educa tion and 15-18 % more likely to be in kinder garten rather than first grade at age six. The effect s of expos ure in first and second trimest er are statisti cally equival ent. The lack of signifi � In contrast to boys, there are no significant effects for girls across the four outcomes.34 However, I cannot conclude from identifying effects solely for boys that there are no effects for girls. In the sample, 2.6 times as many boys are in special education, and boys are held back 1.7 times more often. It may therefore be that more boys than girls are at the margin of the cognitive and behavioral outcomes measured in the data. Appendix Table V displays the estimated coefficients using average grade as the outcome in the limited available sample. While there is no discernible impact of exposure, this may be due to the small sample. When considering these results, it is worth noting that the majority of the unaffected children (74%) were exposed to the shock as young children, having been between birth and age thfour on September 11. Children with direct experience of the attacks or whose mothers’ experienced the attacks displayed elevated rates of behavioral problems in the years after thSeptember 11 (Chemtob et al. 2010). To the extent that children across the city were affected, I may be underestimating the true effect of in utero exposure to maternal stress.35,36 IV.C. Broader geographical analysis thAlthough the effects of September 11 were most substantial for the residents of New York City, there may have also been impacts on geographically and socially proximate populations. In particular, the six suburban counties of New York City home to the highest number of attack victims and the largest share of lower Manhattan workers outside of New York City, as well as the Washington D.C. metropolitan area, are natural candidates.37,38 34 Tests of the equality of coefficients across genders are easily rejected for the special education and held back outcomes. 35th found no significant effects of this Specifications including an indicator for children who were one year old or younger (the cohort most likely to be breastfeeding and thus exposed to maternal stress hormones) on September 11 The majority of research into the effects of September 11 36indicator. The coefficients on the main indicators were unchanged. th on children focused on those of school age. See, for example, Hoven 2005. 37 Five of these counties (Nassau, Westchester, Suffolk, Hudson, and Bergen) are home to a total of 18% of lower Manhattan workers. In each, less than 2% of the county labor force works in lower Manhattan. Each county, 21 To investigate this possibility on birth outcomes, I expand the sample to include all births occurring in the United States, and add geographic indicators and their interactions with the indicators of in utero timing to specification (1): � * ) + * � ( � 𝑛� * � ( � 𝑛� � � � 2� ) � � � +� � � � 1 5 = � 1 * � ( � 𝑛� ) ++ � � +� � +� � 4) + � � �* � ( � 𝑛� � � � 1� 𝑚 � =2� � � � � � 18+ � � � =6� � * � � 𝑦� (4) � � 39 t o � and � 6to � 8 � and � 5 Only three of the five birth outcomes are comparable across states over time due to 1 revisions of birth certificate content implemented in some states in 2003. In particular, the list of maternal medical risks and the list of complications of labor and delivery were altered such that it is not possible to compare them across revised and unrevised data. Therefore, the outcomes � are gestational age, birth weight, and Apgar score.The three � � indicators capture fixed differences between New York City, its suburbs, the Washington D.C. area, and the rest of the nation. The month and year of birth indicators flexibly thcapture national time trends. Estimates of � exposure to September 11 measure the national effect of in utero . Because the psychological impact of the attacks was felt broadly, it is plausible that national birth outcomes were affected. Estimates of � 24measure the extent to which outcomes of exposed in utero cohorts in the three specified areas including the sixth (Monmouth) lost at least 80 residents in the attacks. Approximately 70% of lower Manhattan workers live in one of the five New York City boroughs (US Census Bureau 2008). 38 The area is defined as the Washington Metropolitan Statistical Area. Victims at the Pentagon lived throughout the MSA. Specifications using only Arlington County (where the Pentagon is located) or only the counties that lost the highest number of residents (Fairfax, Prince William, and Prince Georges – each lost at least 20 residents) do not alter the results. 39 California and Texas did not record the five-minute Apgar score on birth certificates during the sample period, reducing the sample size for this outcome by 22%. 22 differ from the outcomes of those exposed in utero in the rest of the nation. Table V reports the results of this specification. Across outcomes, only � 2 and � 6, the estimated effect of in utero exposure in New York City, are consistently negative and statistically significant. While these results suggest that the effects were not widespread or substantial enough to be detected in broad geographies outside New York City, there are likely several groups across the country that were significantly affected but not identifiable in standard data sets, including those who knew someone killed in the attacks and, as Lauderdale (2006) finds in California, those subject to increased discrimination. IV.D. Selection IV.D.i. Birth outcomes thAs the attacks of September 11 were entirely unforeseen, and no other major event followed closely in timing, there is a limited role for selection into pregnancy.40 However, thmigration in response to September 11 may have been selective. To generate the estimates of exposure found in the previous sections, negative selection through migration would have had to thoccur particularly among women pregnant on September 11. For example, if higher socioeconomic status residents were more likely to move out the city after the attacks, and if pregnant women were more likely to move than other women, the resident population that gave birth in the months following September 11th may have had worse outcomes due to lower average socioeconomic status, rather than to the effect of exposure to the attacks. To investigate the possibility of selective migration, I treat covariates indicative of maternal socioeconomic status as outcome variables: 40 with an economic recession, the implications of which are explored in section IV.F. 23 September 11th did coincide outcome variables are mother’s race, marital status, and educational attainment. Results in Table VI, Panel A suggest that selection into New York City residence was indeed occurring among pregnant women. Those in their first or second trimester of pregnancy on thSeptember 11 were approximately 3-5% less likely to be White, 2% less likely to be married, and 4% less likely to have 16 or more years of education. To informally assess the importance of selection on the observed covariates, I present specification (1) without the vector of covariates � � in Appendix Table VI. The estimated effects are very similar to the results with covariates. In addition, the lower share of married mothers may directly reflect the increase in thdivorce rates observed in the year following September 11 (Cohan, Cole and Schoen 2009). However, there may be selection on unmeasured characteristics as well. To formally assess the potential bias, Altonji, Elder and Taber (2005) develop a method that utilizes the amount of selection on observables as a guide to selection on unobservables. However, the method is not informative in this case due to the low explanatory power of the few measured covariates. Across specifications, R-squared values range from 0.01 to 0.11, and the variance of residuals ranges from seven to 47 times greater than the variance of the index of covariates.41 Because the covariates are correlated with indicators of in utero exposure (as shown in Table VI) and explain relatively little of the variation in outcomes, if unobserved characteristics are similarly correlated with exposure, bias from these unobservables could account for the estimated effects. This possibility is tempered by the fact that bias from unobservables in the estimated 41 The index is constructed as the sum of every explanatory variable (other than in utero indicators) multiplied by its coefficient estimate. 24 The � ( 5 ) � �� = � 1 * � ( � 𝑛� � � � 1� ) * � ( � 𝑛� � � � 2� ) + + � +� 2 � 𝑚� 𝑦� + � � effects of exposure is related to outcome variance across conception windows (the dimension of exposure) and not to variance within windows. Depending on the outcome measure, variance within a conception window is five to 100 times greater than variance across windows. Although the timing of conception accounts for relatively little variance in outcomes, there may have been a unique period of selective migration out of New York City in response to September 11th. While conclusive assessments are not possible, qualitatively, two countervailing types of unobserved selection seem plausible. Pregnant women who were economically able to move may have been more likely to leave the city, shifting the distribution of unobservable maternal characteristics such that birth outcomes worsened, therefore biasing estimates towards finding theffects of September 11 exposure. On the other hand, pregnant women who were particularly affected by the attacks may have been more likely to migrate, removing the more negatively impacted pregnancies from the sample and attenuating the estimated effects. Unfortunately, data regarding either mechanism is not available. IV.D.ii. Schooling outcomes For schooling outcomes, there are two additional channels for potential selection: migration out of the city after birth and private school attendance. Each could plausibly lead to theither over or underestimates of the effect of September 11 in utero exposure. If families with the most severely affected children were more likely to move out of the city, or to enroll their children in private school, the main results would be underestimates of in utero exposure’s effects on schooling outcomes. However, if such families were more likely to stay in the city or to utilize public schools, the results would be overestimates. To investigate these possibilities, I follow the same initial approach of the previous 25 se cti on , an d tr ea t co va ri at es in di ca ti ve of a ch il d’ s so ci oe co no m ic st at us as ou tc o m e va ri ab le s: 1 * � ( � � � 𝑛 � � 11� � � � � 19� ) + � 2 � � � + � + � As with the birth data, students’ characteristics suggest some selection into New York City 4 2 residence in Table VI, Panels B and C. Students are approximately 4% less likely to be White, and 3% more likely to be Black. The similarity of these estimates to those at birth suggests that selective migration between birth and school, as well as selective public school attendance, was limited, at least for race. To informally assess the importance of selection on the observed covariates, I present specification (3) without the vector of covariates � � in Appendix Table VII. The estimated effects are again very similar to the results with covariates. The same challenge of limited covariates makes the Altonji, Elder and Taber (2005) method uninformative in the schooling data. Given the robustness of the estimates to the inclusion of covariates, as well as the potential for either positive or negative selection on unobservables, migration does not appear to fully account for the effects of in utero exposure found in Sections IV.A. and IV.B. IV.E. Heterogeneity of effects Because the experience of stress is personalized, there may have been variation in its severity across subgroups within New York City. Baseline levels of stress, access to coping resources, and the ability to compensate for bad shocks could all affect the extent to which Students are eligible for free or reduced price meals according to federal guidelines. For the 2009-10 school year, household income at 185% of the poverty line qualifies students for reduced price meals; at 130% qualifies students for free meals (http://www.fns.usda.gov/cnd/governance/notices/iegs/iegs.htm ). Approximately 63% of the total sample is listed as eligible (22% have no eligibility status listed). 42 26 ( 6 ) The � � � � = outcome variables are student’s race and free lunch � eligibility. * � ( � � � 𝑛 𝑎20� � 𝐽𝑢 19� ) + � 𝑚� � thexposure to September 11 translated to experienced stress, poor birth outcomes, and eventually poor schooling outcomes. Suggestive evidence finds that Manhattan residents with prior stressors in their lives, lower household income, and lower levels of education were all more likely to be suffering from attack-related PTSD or depression in the weeks after the attacks (Galea et al. 2002). To explore the possibility that socioeconomic status and the impact of the attacks may be negatively correlated, I fully interact specifications (2) and (3) with indicators of status. In the birth data, I use an indicator for maternal education being 16 years or more. Table VII displays the results. The negative effects of exposure on birth outcomes appear to be largely mitigated for highly educated mothers. For these mothers, first trimester exposure had a more substantial impact on their rate of labor complications, but effects on gestation length are about half those of less educated mothers, while effects on medical risks, birth weight, and Apgar score on indistinguishable from zero.43 In the schooling data, I approximate a student’s socioeconomic background with the mean household income of his or her home zip code and then divide the sample into income quartiles. Table VIII reports these results for the two affected schooling outcomes, special education and held back status. While many of the coefficients are insignificant for boys, a pattern is readily discernible: the most affected income quartile appears to be the second. These results are in line with a growing literature which finds that children in families of lower socioeconomic status suffer larger effects of adverse health shocks, possibly because their parents take fewer compensatory actions or because they are more likely to face additional negative shocks (Currie and Hyson 1999; Currie and Stabile 2003; Almond, Edlund and Palme 43 Specifications by maternal race find attenuated (but still significant) effects for White mothers compared to Black and Hispanic mothers. 27 2009). Heterogeneity of effects may also be explained by selective migration. For example, if thall pregnant women were adversely affected by September 11, but only those of high socioeconomic status could respond by moving, larger or more significant negative impacts would be measured for lower socioeconomic status groups. IV.F. Channels: Environmental or Economic instead of Stress? The environmental impact of the World Trade Center attack was substantial but highly localized: the dust cloud did not move north beyond Canal Street, and after traveling over western Brooklyn, dispersed (Landrigan et al. 2004). Among Medicaid enrollees, only residents of areas where the dust cloud passed over were more likely to receive medical care for asthma in the weeks after the attacks (Wagner 2005). To assess the extent to which the results are capturing the effects of air pollution, I would like to exclude the areas affected by the dust cloud from my analysis. This is a conservative approach, because people in the most environmentally damaged geographies may also have suffered the most psychologically. The birth data is only identified at the borough level, so I divide the five boroughs into two groups: environmentally affected (Manhattan, Brooklyn) and unaffected (Staten Island, Queens, the Bronx). Table IX presents the results of specification (1) fully interacted with an indicator for residence in one of the three environmentally unaffected boroughs, Staten Island, Queens, and the Bronx. The two environmentally affected boroughs, Manhattan and Brooklyn, are the omitted group. Equality of effects across the two borough groups cannot be rejected for medical risks, labor complications, birth weight, and Apgar score. For gestation length, equality of effects is rejected for both inutero1 and inutero2, however the estimates for the environmentally unaffected boroughs are still significantly different from zero. Overall, the estimates are somewhat smaller for Staten Island, Queens, and the Bronx compared to Manhattan 28 and Brooklyn, suggesting that part of the measured effects may be due to air pollution.44 In the schooling data, records identify children’s current place of residence at the zip code level. Due to mobility, current residence is a measure of residence in utero with error. I construct one indicator for residence in a Manhattan zip code south of Canal Street and a second indicator for residence in a western Brooklyn zip code.45 Table X presents the results of specification (3) fully interacted with the two indicators. The effects of in utero exposure appear to be the same for children living in and out of the environmentally affected area. However, the error in this measure of residence may obscure real differences, particularly if mobility after thSeptember 11 was selective such that more affected women moved. Across both data sets, the limited impact of location within New York City on the estimates suggests that air pollution is not the sole driver of the estimated effects. thIn addition to its environmental consequences, September 11 had a substantial negative impact on the New York City economy. It is plausible that the effects I am measuring may be due to economic deprivation rather than psychological stress. Pregnant women who lost their jobs or whose partners lost their jobs might have struggled to meet their needs for food, shelter, or medical care, which may have adversely impacted their pregnancy. While I cannot separately identify the economic effect from the effect of stress, I can look for evidence of its magnitude and direction. If this were a primary channel through which in utero exposure affected outcomes, I would expect that pregnancies which began well after thSeptember 11, when the psychological trauma had begun to fade but the economic effects persisted, would also be negatively affected. I check this in both the birth and schooling data by defining an affected cohort as the children who were born August to October 2002. These 44 In addition, specifications with the maternal medical risk factor most likely to indicate air pollution, acute or chronic lung disease (including asthma and tuberculosis), as an outcome find no effects of September 1145th exposure. Western Brooklyn is defined as the zip codes adjacent to Manhattan by way of the harbor (Wagner 2005). 29 thchildren were all conceived at least six weeks after September 11, 2001, but were in utero during much of the same period of economic slowdown as the exposed cohort. I compare their outcomes against cohorts born in the same months from 1995 through 2004, including a smooth quadratic year of birth trend: ( 7 ) = * � ( 𝑏 � 𝑛 𝑢� � � � � 2002� ) + + * � � 𝑎� � � 𝑖 + * � 𝑒𝑟 � � � 𝑟 + � � � 1 � � � � 2 � h� � 3 h� 2 � � Results are presented in Table XI, Panel A for the birth outcomes and Panels B and C for the schooling outcomes. There are no significantly negative effects for these later cohorts, suggesting that the economic impact of the attack is not a primary driver of the results.46 V. Conclusion thPregnancies exposed to the September 11, 2001 attacks in New York City during first or second trimester were 2-5% more likely to suffer from a medical risk and 2% more likely to experience a complication during labor. Newborns were 1-1.5 days younger, weighed 8-19 grams less, and had a 0.1% lower five-minute Apgar score at birth. At the age of six, boys were 7-9% more likely to be in special education and 15-18% more likely to be in kindergarten rather than first grade, with no effect on girls. These results do not appear to be driven by environmental or economic conditions, but rather provide evidence that psychological stress is an important channel through which adverse events experienced by pregnant women negatively impact the early life outcomes of in utero cohorts. Moreover, because early life outcomes are good predictors of later health and socioeconomic status, the results suggest that there may be substantial long-term effects of the 46 Note that the comparison group includes part of the cohort that was third trimester in utero on September 11 th. Excluding this group does not affect the results. 30 thSeptember th11 11 attacks on the exposed cohort. To the extent that the stress induced by September is similar to the stress created by more common experiences, the results provide support to the hypothesis that unmeasured psychosocial stressors explain part of the differences observed in birth outcomes and schooling performance across socioeconomic groups. 31 References Aizer, Anna, Laura Stroud, and Stephen Buka, “Maternal Stress and Child Well-Being: Evidence from Siblings,” manuscript, Brown University, 2011. Almond, Douglas, “Is the 1918 Influenza Pandemic Over? Long-Term Effects of In Utero Influenza Exposure in the Post-1940 U.S. Population,” Journal of Political Economy, 114 (2006), 672–71. Almond, Douglas and Janet Currie, “Human Capital Development Before Age Five,” NBER Working Paper No. 15827, 2011a. Almond, Douglas and Janet, “Killing Me Softly: The Fetal Origins Hypothesis,” Journal of Economic Perspectives, 25 (2011), 153-172. Almond, Douglas and Bhashkar Mazumder, “The Effects of Maternal Fasting During Ramadan on Birth and Adult Outcomes,” NBER Working Paper No. 14428, 2008. Almond, Douglas, Lena Edlund, and Marten Palme, “Chernobyl’s Subclinical Legacy: Prenatal Exposure to Radioactive Fallout and School Outcomes in Sweden,” Quarterly Journal of Economics, 124 (2009), 1729-1772. Altonji, Joseph, Todd Elder, and Christopher Taber, “Selection on Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools,” Journal of Political Economy, 113 (2005), 151-184. American Psychiatric Association, Diagnostic and Statistical Manual of Mental Disorders (Revised Fourth Edition). Barker, David, “In Utero Programming of Chronic Disease,” Clinical Science, 95 (1998a), 115– 128. Becker, Gary S. and Yona Rubinstein, “Fear and the Response to Terrorism: An Economic Analysis,” manuscript, The University of Chicago, 2004. Behrman, Jere and Mark R Rosenzweig, “Returns to Birthweight,” Review of Economics and Statistics, 86 (2004), 637-640. Boscarino, Joseph, Sandro Galea, Richard Adams, Jennifer Ahern, Heidi Resnick, and David Vlahov, “Mental Health Service and Medication Use in New York City after the September 11, 2001, Terrorist Attack,” Psychiatric Services, 55 (2004), 274-283. Brackbill, Robert M., James L. Hadler, Laura DiGrande, Christine C. Ekenga, Mark R. Farfel, Stephen Friedman, Sharon E. Perlman, Steven D. Stellman, Deborah J. Walker, David Wu, Shengchao Yu, Lorna E. Thorpe, “Asthma and Posttraumatic Stress Symptoms 5 to 6 Years Following Exposure to the World Trade Center Terrorist Attack,” Journal of the American 32 Medical Association, 302 (2009), 502-516. Bram, Jason, James Orr, and Carol Rapaport, “Measuring the Effects of the September 11 Attack on New York City,” Federal Reserve Bank of New York Economic Policy Review, November 2002. Bruckner, Tim A., Ralph Catalano, and Jennifer Ahern, “Male Fetal Loss in the US Following the Terrorist Attacks of September 11, 2001,” BMC Public Health, 10 (2010), 273-278. Camacho, Adriana, “Stress and Birth Weight: Evidence from Terrorist Attacks,” American Economic Review: Papers & Proceedings, 98 (2008), 511-515. Case, Anne, Angela Fertig, and Christina Paxson, “The Lasting Impact of Childhood Health and Circumstance,” Journal of Health Economics, 24 (2005), 365–389. Catalano, Ralph, Tim Bruckner, A.R. Marks, and Brenda Eskenazi, “Exogenous Shocks to the Human Sex Ratio: the Case of September 11, 2001 in New York City,” Human Reproduction, 21 (2006), 3127-3131. Catalano, Ralph, Tim Bruckner, Jeff Gould, Brenda Eskenazi, and Elizabeth Anderson, “Sex Ratios in California Following the Terrorist Attacks of September 11, 2001,” Human Reproduction, 20 (2005), 1221-1227. Chemtob, Claude M., Yoko Nomura, Khushmand Rajendran, Rachel Yehuda, Deena Schwartz, and Robert Abramovitz, “Impact of Maternal Posttraumatic Stress Disorder and Depression Following Exposure to the September 11 Attacks on Preschool Children’s Behavior,” Child Development, 81 (2010), 1129-1141. Chida, Yoichi and Xin Mao, “Does Psychosocial Stress Predict Symptomatic Herpes Simplex Virus Recurrence? A Meta-analytic Investigation on Prospective Studies,” Brain, Behavior, and Immunity, 23 (2009), 917-925. Clark, Roger, Robert Green, Gregg Swayze, Greg Meeker, Steve Sutley, Todd Hoefen, K. Eric Livo, Geoff Plumlee, Betina Pavri, Chuck Sarture, Steve Wilson, Phil Hageman, Paul Lamothe, J. Sam Vance, Joe Boardman, Isabelle Brownfield, Carol Gent, Laurie Morath, Joseph Taggart, Peter Theodorakos, and Monique Adams, “Environmental Studies of the World Trade Center area after the September 11, 2001 attack,” United States Geological Survey Open-File Report 010429, November 2001. Coe, Christopher L., Marian Kramer, Czeh Boldizsar, Elizabeth Gould, Alison J. Reeves, Clemens Kirschbaum, and Eberhard Fuchs, “Prenatal Stress Diminishes Neurogenesis in the Dentate Gyrus of Juvenile Rhesus Monkeys,” Biological Psychiatry, 54 (2003), 1025-1034. Cohan, Catherine, Steve Cole, and Robert Schoen, “Divorce Following the September 11 Terrorist Attacks,” Journal of Social and Personal Relationships, 26 (2009), 512-530. 33 Crenshaw, Martha, “The Psychology of Terrorism: An Agenda for the 21 st Century,” Political Psychology, 21 (2000), 405-420. Cunha, Flavio, and James J. Heckman, “The Technolgy of Skill Formation,” American Economic Review, 97 (2007), 31–47. Currie, Janet, “Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood, and Human Capital Development,” Journal of Economic Literature, 47 (2009), 87-122. Currie, Janet and Rosemary Hyson, “Is the Impact of Shocks Cushioned by Socioeconomic Status? The Case of Low Birth Weight,” American Economic Review, 89 (1999), 245–250. Currie, Janet and Mark Stabile, “Socioeconomic Status and Child Health: Why Is the Relationship Stronger for Older Children?” American Economic Review, 93 (2003), 1813–1823. Currie, Janet and Matthew Neidell, “Air Pollution and Infant Health: What Can We Learn from California’s Recent Experience?” Quarterly Journal of Economics, 120 (2005), 1003-1030. Da Costa, D., M. Dritsa, J. Larouche, and W. Brender, “Psychosocial Predictors of Labor/Delivery Complications and Infant Birth Weight: A Prospective Multivariate Study,” Journal of Psychosomatic Obstetrics and Gynecology, 21 (2000), 137-148. Dehejia, Rajeev and Adriana Lleras-Muney, “Booms, Busts, and Babies’ Health,” Quarterly Journal of Economics, 119 (2004), 1091-1130. De Weerth, Carolina and Jan K. Buitelaar, “Physiological Stress Reactivity in Human Pregnancy – A Review,” Neuroscience and Biobehavioral Reviews, 29 (2005), 295-312. DiMaggio, Charles, Sandro Galea, and Paula Madrid, “Population Psychiatric Medication Prescription Rates Following a Terrorist Attack,” Prehospital and Disaster Medicine, 22 (2007), 479-484. DiMaggio, Charles, Sandro Galea, and Lynne Richardson, “Emergency Department Visits for Behavioral and Mental Health Care After a Terrorist Attack,” Annals of Emergency Medicine, 50 (2007), 327-334. Dunkel-Schetter, C., R. Gurung, M. Lobel, P.D. Wadhwa, “Stress Processes in Pregnancy and Birth: Psychological, Biological, and Sociocultural Influences.” In: A. Baum, T. Revenson, and T. Singer, editors, Handbook of Health Psychology, 2000, 495-518. Endara, Skye M., Margaret A.K. Ryan, Carter J. Sevick, Ava Marie S. Conlin, Caroline A. Macera, and Tyler C. Smith, “Does Acute Maternal Stress in Pregnancy Affect Infant Health Outcomes? Examination of a Large Cohort of Infants Born after the Terrorist Attacks of September 11, 2001,” BMC Public Health, 9 (2009). Eskenazi, Brenda, Amy R Marks, Ralph Catalano, Tim Bruckner, and Paolo G Toniolo, “Low 34 Birthweight in New York City and Upstate New York Following the Events of September 11 th ,” Human Reproduction, 22 (2007), 3013-3020. Farfel, Mark, Laura DiGrande, Robert Brackbill, Angela Prann, James Cone, Stephen Friedman, Deborah J. Walker, Grant Pezeshki, Pauline Thomas, Sandro Galea, David Williamson, Thomas R. Frieden, and Lorna Thorpe, “An Overview of 9/11 Experiences and Respiratory and Mental Health Conditions among World Trade Center Health Registry Enrollees,” Journal of Urban Health, 85 (2008), 880-909. Galea, Sandro, Jennifer Ahern, Heidi Resnick, Dean Kilpatrick, Michael Bucuvalas, Joel Gold, and David Vlahov, “Psychological Sequelae of the September 11 Terrorist Attacks in New York City,” The New England Journal of Medicine, 346 (2002), 982-987. Galea, Sandro, David Vlahov, Heidi Resnick, Jennifer Ahern, Ezra Susser, Joel Gold, Michael Bucuvalas, and Dean Kilpatrick, “Trends of Probable Post-Traumatic Stress Disorder in New York City after the September 11 Terrorist Attacks,” American Journal of Epidemiology, 158 (2003), 514-524. Glynn, L., P.D. Wadhwa, C. Dunkel-Schetter, and C.A. Sandman, “When Stress Happens Matters: The Effects of Earthquake Timing on Stress Responsivity in Pregnancy,” American Journal of Obstetric Gynecology, 184 (2001), 637-642. Goldhaber, Marilyn, “Fetal Death Ratios in a Prospective Study Compared to State Fetal Death Certificate Reporting,” American Journal of Public Health, 79 (1989), 1268-1270. Hansen, Dorthe, Hans Lou, and Jorn Olsen, “Serious Life Events and Congenital Malformations,” The Lancet, 356 (2000), 875-880. Helzer, John, Lee Robins, and Larry McEvoy, “Post-Traumatic Stress Disorder in the General Population,” The New England Journal of Medicine, 317 (1987), 1630-1634. Higgins, J.R., J.J. Walshe, R.M. Conroy, and M.R.N. Darling, “The Relation between Maternal Work, Ambulatory Blood Pressure, and Pregnancy Hypertension,” Journal of Epidemiology and Community Health, 56 (2002), 389-393. Hoven, Christina W., Cristiane S. Duarte, Christopher P. Lucas, Ping Wu, Donald J. Mandell, Renee D. Goodwin, Michael Cohen, Victor Balaban, Bradley A. Woodruff, Fan Bin, George J. Musa, Lori Mei, Pamela A. Cantor, Lawrence Aber, Patricia Cohen, and Ezra Susser, “Psychopathology Among New York City Public School Children 6 Months After September 11,” Archives of General Psychiatry, 62 (2005), 545-551. Huizink, Anja, Pascale Robles de Medina, Eduard Mulder, Gerard Visser, and Jan Buitelaar, “Stress During Pregnancy is Associated with Developmental Outcome in Infancy” Journal of Child Psychology and Psychiatry, 44 (2003), 810-818. Kaiser, Sylvia and Norbert Sachser, “The Effects of Prenatal Social Stress on Behaviour: 35 Mechanisms and Function,” Neuroscience and Biobehavioral Reviews, 29 (2005), 283-294. Kammerer, Martin, Diana Adams, Brida von Castleberg, and Vivette Glover, “Pregnant Women Become Insensitive to Cold Stress,” BMC Pregnancy and Childbirth, 2 (2002), 8. King, Suzanne and David P. LaPlante, “The Effects of Prenatal Maternal Stress on Children’s Cognitive Development: Project Ice Storm,” Stress, 8 (2005), 35-45. Landrigan, Philip J., Joel Forman, Maida Galvez, Brooke Newman, Stephanie M. Engel, and Claude Chemtob, “Impact of September 11 World Trade Center Disaster on Children and Pregnant Women,” Mount Sinai Journal of Medicine, 75 (2008), 129-134. Landrigan, Philip J., Paul J. Lioy, George Thurston, Gertrud Berkowitz, L.C. Chen, Steven N. Chillrud, Stephen H. Gavett, Panos G. Georgopoulos, Alison S. Geyh, Stephen Levin, Frederica Perera, Stephen M. Rappaport, Christopher Small, and the NIEHS World Trade Center Working Group, “Health and Environmental Consequences of the World Trade Center Disaster,” Environmental Health Perspectives, 112 (2004), 731-739. Lauderdale, Diane S., “Birth Outcomes for Arabic-Named Women in California before and after September 11,” Demography, 43 (2006), 185-201. Li, Jiong, Jorn Olsen, Mogens Vestergaard, Carsten Obel, Jennifer Baker, and Thorkild Sorensen, “Prenatal Stress Exposure Related to Maternal Bereavement and Risk of Childhood Overweight,” PLoS ONE, 5 (2010). Lorber, Matthew, Herman Gibb, Lester Grant, Joseph Pinto, Joachim Pleil, and David Cleverly, “Assessment of Inhalation Exposures and Potential Health Risks to the General Population that Resulted from the Collapse of the World Trade Center Towers,” Risk Analysis, 27 (2007), 12031221. Makinen, Gail, “The Economic Effects of 9/11: A Retrospective Assessment,” Congressional Research Service: Report for Congress, September 2002. Malaspina, D., C. Corcoran, K.R. Kleinhaus, M.C. Perrin, S. Fennig, D. Nahon, Y. Friedlander, and S. Harlap, “Acute Maternal Stress in Pregnancy and Schizophrenia in Offspring: A Cohort Prospective Study,” BMC Psychiatry, 8 (2008), 1-9. McEwen, Bruce and Eliot Stellar, “Stress and the Individual,” Archives of Internal Medicine, 153 (1993), 2093-2101. Meaney, M.J. and J.R. Seckl, “Glucocorticoid Programming,” Annals of the New York Academy of Sciences, 1032 (2004), 63-84. National Center for Health Statistics (NCHS) Technical Appendix. Vital Statistics of the United States: 1999, Vol. I, Natality. Available on the internet at http://wonder.cdc.gov/wonder/sci_data/natal/detail/type_txt/natal99/Techapp99.pdf 36 thNational Center for Health Statistics (NCHS) Technical Appendix. Vital Statistics of the United States: 2004, Vol. I, Natality. Available on the internet at http://wonder.cdc.gov/wonder/sci_data/natal/detail/type_txt/natal04/TechAppendix04.pdf Neugebauer, Richard, Hans Wijbrand Hoek, and Ezra Susser, “Prenatal Exposure to Wartime Famine and Development of Antisocial Personality Disorder in Early Adulthood,” Journal of the American Medical Association, 282 (1999), 455–462. New York City Department of Education, Office of English Language Learners. New York City’s English Language Learners: Demographics and Performance (2007). New York Civil Liberties Union (NYCLU). Education Interrupted: The Growing Use of Suspensions in New York City’s Public Schools (2011). Oreopoulos, Philip, Mark Stabile, Randy Walld, and Leslie L. Roos, “Short, Medium, and Long term Consequences of Poor Infant Health,” The Journal of Human Resources, 43 (2008), 88-138. Paarlberg, K.M., A.J. Vingerhotes, J. Passchier, G.A. Dekker, H.P. Van Geijn, “Psychosocial Factors and Pregnancy Outcome: A Review with Emphasis on Methodological Issues,” Journal of Psychosomatic Research, 39 (1995), 563-595. Reyes, Jessica Wolpaw, “The Impact of Prenatal Lead Exposure on Health,” manuscript, Amherst College, 2005. Rich-Edwards, J.W., K.P. Kleinman, E.F. Strong, E. Oken, and M.W. Gillman, “Preterm Delivery in Boston before and after September 11th, 2001,” Epidemiology, 16 (2005), 323-327. Ruther, Matt, “The Fertility Response to September 11: Evidence from the Five Boroughs,” poster, University of Pennsylvania, 2010. Sapolsky, Robert M. Why Zebras Don’t Get Ulcers (New York, NY: Henry Holt and Company, 2004). Schlenger, William E., Juesta M. Caddell, Lori Ebert, B. Kathleen Jordan, Kathryn M. Rourke, David Wilson, Lisa Thalji, J. Michael Dennis, John A. Fairbank, and Richard A. Kulka, “Psychological Reactions to Terrorist Attacks,” Journal of the American Medical Association, 288 (2002), 581-588. Schuster, Mark A., Bradley D. Stein, Lisa Jaycox, Rebecca L. Collins, Grant N. Marshall, Marc N. Elliott, Annie J. Zhou, David E. Kanouse, Janina L. Morrison, Sandra H. Berry, “A National Survey of Stress Reactions After the September 11, 2011 Terrorist Attacks,” The New England Journal of Medicine, 345 (2001), 1507-1512. Simeonova, Emilia, “Out of Sight, Out of Mind? The Impact of Natural Disasters on Pregnancy Outcomes,” CESifo Economic Studies 57 (2011), 403-431. 37 Smits, Luc, Lydia Krabbendam, Rob de Bie, Gerard Essed, and Jim van Os, “Lower birth weight of Dutch neonates who were in utero at the time of the 9/11 attacks,” Journal of Psychosomatic Research, 61 (2006), 715-717. Stark, Michael J., Ian M.R. Wright, and Vicki L. Clifton, “Sex-specific Alterations in Placental 11Beta Hydroxysteroid Dehydrogenase 2 Activity and Early Postnatal Clinical Course Following Antenatal Betamethasone,” American Journal of Physiology, 297 (2009), R510-R514. Stoecker, Charles, “Chill Out, Mom: Extreme Cold Induced Maternal Stress in utero and Later Outcomes,” manuscript, UC Davis Department of Economics, 2011. United States Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Health Statistics (NCHS), Division of Vital Statistics, Natality public-use data available on the internet at http://www.cdc.gov/nchs/data_access/Vitalstatsonline.htm United States Department of Labor, Bureau of Labor Statistics, Local Area Unemployment Statistics available on the internet at http://www.bls.gov/lau/data.htm Van den Berg, Gerard J., Maarten Lindeboom, and France Portrait, “Economic Conditions Early in Life and Individual Mortality,” American Economic Review, 96 (2006), 290–302. Van den Bergh, Bea R.H., Eduard J.H. Mulder, Maarten Mennes, and Vivette Glover, “Antenatal maternal anxiety and stress and the neurobehavioural development of the fetus and child: links and possible mechanisms. A Review,” Neuroscience and Biobehavioral Reviews, 29 (2005), 237-258. Vlahov, David, Sandro Galea, Jennifer Ahern, Heidi Resnick, and Dean Kilpatrick, “Sustained Increased Consumption of Cigarettes, Alcohol, and Marijuana Among Manhattan Residents after September 11, 2001,” American Journal of Public Health, 94 (2004), 253-254. Wadhwa, P.D., Jennifer F. Culhane, Virginia Rauh, and Shirish S. Barve, “Stress and Preterm Birth: Neuroendocrine, Immune/Inflammatory, and Vascular Mechanisms,” Maternal and Child Health Journal, 5 (2001), 119-125. Wadhwa, P.D., “Prenatal Stress and Life-span Development.” In: H Friedman, editor, Encyclopedia of Mental Health, 1998, 265-280. Wagner, Victoria, Marleen Radigan, Patrick Roohan, Joseph Anarella, and Foster Gesten, “Asthma in Medicaid Managed Care Enrollees Residing in New York City: Results from a PostWorld Trade Center Disaster Survey,” Journal of Urban Health: Bulleting of the New York Academy of Medicine, 82 (2005), 76-89. Warren, W.B., E.D. Gurewitsch, and R.S. Goland, “Corticotropin-releasing Hormone and Pituitary-Adrenal Hormones in Pregnancies Complicated by Chronic Hypertension,” American Journal of Obstetric Gynecology, 172 (1995), 661-666. 38 Yu, In Tag, Lee Sang-Hun, Lee Yong-Sung, and Hyeon Son, “Differential Effects of Corticosterone and Dexamethasone on Hippocampal Neurogenesis In Vitro.” Biochemical and Biophysical Research Communications, 317 (2004), 484-490. 39 F i g u r e 1 Mean birth outcomes for New York City resident women by estimated time of conception Share of mothers with 1+ labor complications Share of mothers with 1+ medical risks 1995 1996 1997 1998 1999 2000 2001 2002 2003 18 - Sep 10 Conceived March 12 - June 17 1995 1996 1997 1998 1999 2000 2001 2002 2003 18 - Sep 10 Conceived March 12 - June 17 Year Conceived June Year Conceived June Mean birth weight (in grams) Mean gestation length (in weeks) 1995 1996 1997 1998 1999 2000 2001 2002 2003 - Sep 10 Conceived March 12 - June 17 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year Conceived June 18 - Sep 10 Conceived March 12 - June 17 Mean five-minute Apgar score (scale 1-10) Year Conceived June 18 1 arch 12 - June 17 9 9 5 estimated to have been Notes. The figures display mean outcome values of births conceived either March 12-June 17 or June 18-September10 annually from1 1995 through 2003. The sample includes all live births to women residing in one 9 of the five boroughs of New York City from 1995 through 2004. 9 6 1 9 9 7 1 9 9 8 1 9 9 9 2 0 0 0 2 0 0 1 2 0 0 2 2 0 0 3 Y e a r C o n c e i v e d J u n e 1 8 S e p 1 0 C o n c e i v e d .38 .26 .27 M F i g u r e 2 Birth outcomes for cohort third trimester in utero on September 11 (conceived December 27 to March 11) Share of mothers with 1+ medical risks Mean five- 1995 1996 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year Mean gestation length (in weeks) 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year Share of mothers with 1+ labor complications 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year 8.94 8.96 8.98 9 9.02 9.04 .36 .37 3245 3250 3255 3260 3265 38.8 38.85 38.9 38.95 Notes. The figures display mean outcome values of births estimated to have been conceived December 27-March 11 annually from 1995 through 2003. The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004. .34 .35 .2 .22 .24 .26 .28 Mean birth weight (in grams) 1995 1996 199 Table I Outcomes for births occurring to New York City women 1995-2004 Medical risk Labor complication indicator Gestation Birthweight Apgar indicator Panel A: All births sample mean 0.248 3,253 8.995 0.360 38.8 inutero1 0.0056* 0.0072** -0.215*** -8.01* -0.012** [0.0032] [0.0035] [0.019] [4.23] [0.0051] inutero2 0.012*** 0.0057* -0.163*** -18.70*** -0.014*** [0.0029] [0.0033] [0.017] [3.90] [0.0047] 2N 1,198,265 1,198,265 1,198,265 1,198,265 1,198,265 R0.02 0.01 0.07 0.11 0.02 Panel B: Males sample mean 0.248 0.367 38.7 3,307 8.989 inutero1 0.0024 0.0022 -0.231*** -11.495* -0.023*** [0.0044] [0.0049] [0.027] [6.030] [0.0072] inutero2 0.0056 0.0034 -0.156*** -13.145** -0.015** [0.0041] [0.0046] [0.025] [5.557] [0.0066] 2N 614,024 R0.02 0.02 9.002 614,024 614,024 614,024 614,024 0.01 0.06 0.10 Panel C: Females sample mean 0.249 0.353 38.8 3,196 inutero1 0.0090** 0.012** -0.198*** -4.546 -0.00085 [0.0045] [0.0050] [0.027] [5.927] [0.0071] inutero2 0.018*** 0.0082* -0.171*** -24.490*** -0.012* [0.0042] [0.0046] [0.025] [5.449] [0.0066] 2N 584,241 R and � 0.01 2 584,241 584,241 584,241 in specification (1). Panels B and C report estimates of � 1 and � 0.02 0.07 0.11 0.02 584,241 2 *Significant at 10%; **significant at 5%; ***significant at 1%. Notes. Panel A reports estimates of in specification (1) run separately by child’s gender. In all panels, the sample includes all liv New York City from 1995 through 2004. Regressions include mother’s parity, age (and age squ � 1 (5 categories – Black, White, Hispanic, Asian, Other) and marital status, as well as indicators of year of birth, and month of birth. Indicators of missing values are also included for covariates. Table II Outcomes for births occurring to New York City women 1995-2004 Medical risk Labor complication indicator Gestation Birthweight Apgar indicator 20.02 sample mean 0.248 0.360 38.8 3,253 8.995 inutero1 0.0072** 0.0083** -0.219*** -11.931*** -0.0067 [0.0033] [0.0036] [0.019] [4.346] [0.0052] inutero2 0.012*** 0.0078** -0.115*** -14.316*** -0.011** [0.0031] [0.0035] [0.019] [4.175] [0.0050] N 1,198,265 1,198,265 1,198,265 1,198,265 1,198,265 R1 Notes. The table reports estimates of � and � *Significant at 10%; **significant at 5%; ***significant at 1%. 0.02 0.07 0.11 residing in one of the five education (4 levels), race conception, and month of Table III: Mean values of characteristics across cohorts of births occurring to New York City women (by time of conception) Mother's race Mother's education White Black Hispanic Mother maried>12 years 16+ years Live births Sex ratio First trimester on Septemer 11 (Conceived June 18 - September 10) 1995 27.1% 27.5% 33.2% 47.1% 36.6% 19.3% 28,442 0.517 1996 27.0% 27.8% 32.4% 55.8% 39.5% 20.1% 28,014 0.512 1997 27.5% 27.5% 33.0% 55.0% 39.3% 20.6% 28,500 0.516 1998 27.0% 27.6% 32.8% 53.9% 40.8% 22.3% 27,765 0.510 1999 27.0% 27.0% 33.2% 54.8% 43.0% 23.2% 27,952 0.512 2000 27.8% 26.7% 33.5% 54.7% 43.4% 23.6% 28,089 0.519 2001 27.3% 26.0% 33.7% 55.3% 43.8% 23.8% 28,119 0.513 2002 28.9% 24.7% 32.6% 57.2% 46.5% 26.7% 28,106 0.516 2003 29.3% 23.9% 33.1% 56.2% 43.4% 26.7% 28,152 0.514 Second trimester on September 11 (Conceived March 12 - June 17) 1995 25.3% 29.5% 33.3% 46.6% 34.8% 17.9% 35,011 0.512 1996 25.2% 29.2% 32.9% 52.8% 37.7% 18.5% 33,390 0.509 1997 26.8% 29.0% 32.4% 53.5% 38.7% 20.0% 33,839 0.511 1998 25.7% 28.4% 33.8% 53.0% 40.2% 20.7% 33,048 0.512 1999 26.0% 28.0% 33.4% 53.0% 40.6% 21.4% 34,078 0.515 2000 26.8% 27.8% 32.7% 53.4% 41.7% 22.0% 33,307 0.512 2001 26.5% 27.2% 33.6% 53.7% 42.8% 22.8% 33,073 0.509 2002 27.5% 25.5% 33.1% 55.0% 44.5% 25.3% 33,253 0.512 2003 27.6% 25.9% 33.0% 54.6% 42.8% 25.2% 33,654 0.510 Notes. The table reports mean values for all live births estimated to have been conceived in one of the two specified windows to women residing in one of the five boroughs of New York City from 1995 through 2004. Table IV Outcomes for New York City public school children 2003/4-2009/10 Days absent Behavior Special Held back from school incident education indicator indicator indicator Panel A: all students sample mean 11 0.03 0.08 0.03 born Mar 20 to June 18, 2002 -0.054 0.0026 0.0047* 0.0026 [0.106] [0.0016] [0.0027] [0.0017] born Dec 11 to Mar 19, 2002 0.085 4.7x10-5 0.0037 0.0028* [0.101] [0.0015] [0.0026] [0.0016] 2N 410,297 410,297 mean 11 0.04 0.12 0.04 410,297 410,297 R0.15 0.03 0.03 0.02 Panel B: Boys sample born Mar 20 to June 18, 2002 -0.034 0.0049* 0.011** 0.0055** [0.153] [0.0028] [0.0045] [0.0026] born Dec 11 to Mar 19, 2002 0.037 0.0020 0.0083* 0.0066*** [0.145] [0.0026] [0.0043] [0.0025] 2N 210,519 0.03 210,519 210,519 210,519 Panel C: Girls sample mean 11 0.01 R0.16 0.04 0.03 0.02 0.02 born Mar 20 to June 18, 2002 -0.074 0.00029 -0.0014 -0.00029 [0.148] [0.0014] [0.0030] [0.0021] 1 born Dec 11 to Mar 19, 2002 0.140 -0.0021 -0.0011 -0.00083 [0.140] [0.0014] [0.0028] [0.0020] Notes. Panel A reports estimates of � and � 199,778 199,778 199,778 212 and � 2 in specification (3). Panels B and C report estimates of � R0.15 0.01 0.01 N 199,778 0.02 in specification (3) run separately by child’s gender. In all panels, the sample includes all six-yearold public school students in New York City who are not English Language Learners from the 2003 through 2009 school years. All regressions include indicators of child’s gender (Panel A only), race (5 categories – Black, White, Hispanic, Asian, Other), home zip code, year of birth, and month of birth. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. Outcomes for births occurring to women resident in the United States 1995-2004 Table V Gestation Birthweight Apgar sample mean 38.8 3,311 8.915 inutero1 -0.035* 1.62 0.0033 [0.019] [4.45] [0.0041] inutero1*NYC -0.18*** -8.81*** -0.016*** [0.0034] [0.78] [0.0057] inutero1*NYsuburbs -0.0052 4.83 -0.013* [0.024] [5.46] [0.0069] inutero1*DC 0.045* -2.33 -0.017** [0.026] [6.02] [0.0076] inutero2 -0.024 -3.00 0.00025 [0.018] [4.10] [0.0010] inutero2*NYC -0.14*** -17.16*** -0.015*** [0.0031] [0.72] [0.0052] inutero2*NYsuburbs 0.0019 1.55 -0.0093 [0.022] [5.12] [0.0065] inutero2*DC 0.011 5.75 0.0010 [0.024] [5.57] [0.0071] 38,672,662 38,672,662 29,980,978 R1 Notes. The table reports estimates of � to � 0.01 0.05 0.018 in specification (4). The sample includes all live births to 2N women residing in the United States from 1995 through 2004, except births in CA and TX for the third column, as these states did not report Apgar score during the sample period. Regressions include mother’s parity, age (and age squared), and indicators of mother’s education (4 levels), race (5 categories – Black, White, Hispanic, Asian, Other), marital status, and indicators for residence in New York City, New York suburbs, or the DC area, as well as indicators of child’s gender, plurality (twins or higher), year of birth, and month of birth. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. Panel A: Characteristics of New York City women giving birth 1995-2004 Mother's race white Mother's race black Mother's race hispanic Mother married Table VI Selection i Data sample mean 0.27 0.27 0.33 inutero1 -0.0144*** 0.0046 0.0067* -0.0086** -0.009 [0.0033] [0.0033 0.0053* 0.0053* -0.0116*** -0.0079*** [0.0030] [0.003 1,198,265 1,198,265 1,198,265 1,198,2 R20.001 0.001 0.0001 Characteristics of New York City public school child White Black Hispanic Free Lunch Eligible Panel B: Boys sa 0.33 0.80 born Mar 20 to June 18, 2002 -0.0045 0.0056 0.000 [0.0057] born Dec 11 to Mar 19, 2002 -0.0092** 0.010** 0.0 [0.0054] 2N 210,519 0.20 0.33 210,519 0.26 0.80 210,519 1 Panel C: Girls sa born Mar 20 to June 18, 2002 -0.0017 0.0074 -0.008 [0.0058] born Dec 11 to Mar 19, 2002 -0.0028 0.0052 0.0022 1 Notes. Panel A reports estimates of � and � in specification (5). The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004. Panels B and C report estimates of � 12 and � 22N 199,778 199,778 199,778 157,320 R0.35 0.39 0.20 0.26 in specification (6) run separately by child’s gender. The sample includes all six-year-old public school students in New York City who are not English Language Learners from the 2003 through 2009 school years. In all panels, regressions include month of birth and year of birth indicators. *Significant at 10%; **significant at 5%; ***significant at 1%. Table VII Outcomes M for births occurring to New York City e women 1995-2004 d Labor complication indicator Gestation Birthweight i Apgar c a sample mean 0.248 0.360 38.8 3,253 8.995 l inutero1 0.0075** 0.0037 -0.241*** -10.727** -0.018*** [0.0036] [0.0040] [0.022] [4.843] [0.0058] r i inutero1*mother 16+ years of education -0.010 0.013 0.122*** 15.683 0.027** s [0.0075] [0.0083] [0.045] [9.985] [0.012] k inutero2 0.014*** 0.0050 -0.180*** -20.627*** -0.015*** [0.0033] [0.0037] [0.020] [4.430] i [0.0053] n d inutero2*mother 16+ years of education -0.013* 0.0026 0.088** 14.456 0.0078 i [0.0070] [0.0078] [0.042] [9.324] [0.011] c 2N a1,198,265 1,198,265 1,198,265 1,198,265 1,198,265 R0.02 0.02 0.07 0.11 0.02 t o r Notes. The table reports estimates from specification (1) fully interacted with an indicator of mother’s education 16+ years. The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004. Regressions include mother’s parity, age (and age squared), and indicators of mother’s race (5 categories – Black, White, Hispanic, Asian, Other) and marital status, as well as indicators of child’s gender, plurality (twins or higher), year of birth, and month of birth, and all interactions. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. Table VIII (Part 1) Heterogeneity: Outcomes for New York City public school children 2003/4-2009/10 Special education indicator Held back indicator Boys: born Mar 20 to June 18, 2002 0.011 0.0081 [0.0088] [0.0067] born Mar 20 to June 18, 2002*quartile2 0.0029*** 0.021*** [0.00091] [0.0068] born Mar 20 to June 18, 2002*quartile 3 0.0010 0.014* [0.013] [0.0066] born Mar 20 to June 18, 2002*quartile4 -0.0049 -0.0050 [0.013] [0.0052] born Dec 11 to Mar 19, 2002 0.0071 0.0074 [0.0083] [0.0051] born Dec 11 to Mar 19, 2002*quartile2 0.007** 0.011* [0.0026] [0.0056] born Dec 11 to Mar 19, 2002*quartile3 0.015 0.011 [0.012] [0.0071] born Dec 11 to Mar 19, 2002*quartile4 -0.012 -0.0061 [0.012] [0.0072] 2N 210,519 210,519 R0.02 0.03 Table VIII (Part 2) Heterogeneity: Outcomes for New York City public school children 2003/4-2009/10 Special education indicator Held back indicator Girls: born Mar 20 to June 18, 2002 -0.0054 -0.0033 [0.0059] [0.0042] born Mar 20 to June 18, 2002*quartile2 0.0065 0.0027 [0.0083] [0.0058] born Mar 20 to June 18, 2002*quartile 3 0.0078 0.0084 [0.0085] [0.0059] born Mar 20 to June 18, 2002*quartile4 0.0021 0.0017 [0.0084] [0.0059] born Dec 11 to Mar 19, 2002 0.0008 -0.0062 [0.0055] [0.0039] born Dec 11 to Mar 19, 2002*quartile2 0.0061 0.0063 [0.0079] [0.0055] born Dec 11 to Mar 19, 2002*quartile3 -0.0026 0.011* [0.0079] [0.0056] born Dec 11 to Mar 19, 2002*quartile4 -0.0096 0.0045 R[0.0079] [0.0056] N 199,778 199,778 20.01 0.02 Notes. The table reports estimates from specification (3) run separately by child’s gender and fully interacted with indicators of quartiles of child’s home zip code mean income. The sample includes all six-year-old public school students in New York City who are not English Language Learners from the 2003 through 2009 school years. Regressions include indicators of child’s race (5 categories – Black, White, Hispanic, Asian, Other), home zip code, year of birth, and month of birth, and all interactions. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. Table IX Outcomes for births occurring to New York City women 1995-2004 Medical risk Labor complication indicator Gestation Birthweight Apgar indicator sample mean 0.248 0.360 38.8 3,253 8.995 inutero1 0.0057 0.0095* -0.260*** -11.815** -0.015** [0.0045] [0.0050] [0.027] [5.987] [0.0071] inutero2 0.0057 0.0081* -0.192*** -22.184*** -0.014** [0.0041] [0.0046] [0.025] [5.500] [0.0066] 2N 1,198,265 0.02 1,198,265 0.07 1,198,265 0.11 1,198,265 0.02 1,198,265 R0.02 Notes. The table reports estimates from specification (1) fully interacted with an indicator of mother’s borough of residence being Staten Island, the Bronx, or Queens (omitted boroughs are Manhattan and Brooklyn). The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004. Regressions include mother’s parity, age (and age squared), mother’s education (4 levels), and indicators of mother’s race (5 categories – Black, White, Hispanic, Asian, Other) and marital status, as well as indicators of child’s gender, plurality (twins or higher), year of birth, and month of birth, and all interactions. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. inutero2* Staten/Bronx/Queens residence inutero1* Staten/Bronx/Queens residence 0.011* -0.0054 0.061* 6.918 0.0022 [0.0058] [0.0065] [0.035] [7.809] [0.0093] -0.00031 -0.0056 0.091** 5.818 0.0053 [0.0063] [0.0071] [0.038] [8.487] [0.010] Table X Outcomes for New York City public school children 2003/4-2009/10 Special education indicatorHeld back indicator Boys: sample mean 0.12 0.04 born Mar 20 to June 18, 2002 0.010** 0.0051* [0.0046] [0.0027] born Mar 20 to June 18, 2002 * SouthCanal/Western Brooklyn 0.0048 0.0071 [0.019] [0.011] born Dec 11 to Mar 19, 2002 0.032* 0.0069*** [0.018] [0.0025] born D e c 11 to Ma r 19, 2002* SouthCa nal/We s te r n Br ooklyn 0.0063 -0.0071 [0.0047] [0.010] 2N 210,519 Girls: sample mean 0.04 210,519 R0.02 0.03 0.02 born Mar 20 to June 18, 2002 -0.0018 -0.000076 [0.0031] [0.00216] born Mar 20 to June 18, 2002 * SouthCanal/Western Brooklyn 0.0072 -0.0014 [0.012] [0.0085] born Dec 11 to Mar 19, 2002 -0.0012 -0.0013 [0.0029] [0.0020] born D e c 11 to Ma r 19, 2002* SouthCa nal/We s te r n Br ooklyn 0.0011 0.0073 [0.012] [0.0082] 2N 199,778 199,778 R0.01 0.02 Notes. The table reports estimates from specification (3) run separately by child’s gender and fully interacted with an indicator of child’s current home zip code (at the time of the school record) being located south of Canal St. or in Western Brooklyn. The sample includes all six-year-old public school students in New York City who are not English Language Learners from the 2003 through 2009 school years. Regressions include indicators of child’s race (5 categories – Black, White, Hispanic, Asian, Other), home zip code, year of birth, and month of birth, and all interactions. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. Table XI Economic effects of September 11th Panel A: Outcomes for births occurring to New York City women 1995-2004 Medical risk Labor complication indicator Gestation Birthweight Apgar indicator sample mean 0.250 0.357 38.8 3,252 8.994 born Aug 1 to Oct 31, 2002 -0.018*** -0.0042 0.026 -4.494 0.0069 [0.0027] [0.0030] [0.016] [3.676] [0.0043] N 308,306 308,306 308,306 308,306 308,306 R20.02 0.02 0.06 0.11 0.02 Panels B and C: Outcomes for New York City public school children 2003/4-2009/10 Days absent Behavior Special Held back from school incident education indicator indicator indicator Panel B: Boys sample mean 11 0.04 0.13 0.04 born Aug 1 to Oct 31, 2002 0.145 0.050*** -0.0049 -0.0031 [0.470] [0.003] [0.0055] [0.0028] 2N 53,323 53,323 53,323 R0.15 0.03 0.02 Panel C: Girls sample mean 11 53,323 0.02 0.01 0.05 0.02 born Aug 1 to Oct 31, 2002 -0.258 0.0034 -0.0011 1.1x10-6 [0.172] [0.0061] [0.0044] [0.0029] N 50,441 50,441 50,441 50,441 R20.15 0.01 0.01 0.01 Notes. Panel A reports estimates from specification (7). The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004 occurring between August and October. Regressions include mother’s parity, age (and age squared), and indicators of mother’s education (4 levels), race (5 categories – Black, White, Hispanic, Asian, Other) and marital status, as well as indicators of child’s gender, plurality (twins or higher), and a quadratic trend for year of birth. Indicators of missing values are also included for covariates. Panels B and C report estimates from specification (7) run separately by child’s gender. The sample includes all six-year-old public school students in New York City whose birthdates are between August and October, and who are not English Language Learners, from the 2003 through 2009 school years. Regressions include indicators of child’s race (5 categories – Black, White, Hispanic, Asian, Other), home zip code, year of birth, and a quadratic trend for year of birth. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. (A-1) � � = � 1 *� (� � � 𝑛� � � � � 𝑢� � e d i c a l r i s k i n d i c a t o r 20.02 L a b o r APPENDIX ) + � 2* � ( � � � 𝑛𝐽� � � 𝑎� h � ) + � � � +� 𝑚� + � � � � + � � Appendix Table I Outcomes for births occurring to New York City Mwomen 1995-2004 indicator Gestation Birthweight Apgar sample mean 0.248 0.360 38.8 3,253 8.995 c o bornApriltoJune 2002 0.0070** 0.0092** -0.018* -6.136** -0.0091* [0.0032] [0.0036 m [0.010] [2.602] [0.0051] p bornJantoMarch2002 0.010*** 0.011*** -0.027** -9.876*** -0.015*** [0.0032] [0.0 l [0.011] [2.593] [0.0052] i c a t i o n 0.01 0.07 0.11 0.02 in specification (A-1). The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004. Regressions include mother’s parity, age (and age squared), and indicators of mother’s education (4 levels), race (5 categories – black, white, hispanic, asian, other) and marital status, as well as indicators of child’s gender, plurality (twins or higher), year of birth, and month of birth. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. Notes. The table reports estimates of � 1,198,265 1,198,265 R 1 and � N 1,198,265 2 1,198,265 1,198,265 Appendix Table II Outcomes for births occurring to New York City women 1995-2004 Medical risk indicator: Medical risk stress related indicator: other sample mean 0.048 Notes. The table reports estimates of a1 and a2 inutero1 0.0050** 0.00082 [0.0020] [0.0029] inutero2 0.0086*** 0.0020 [0.0019] [0.0027] 2N 1,198,265 1,198,265 0.02 in specification (1) with maternal medical risks as the outcomes. In the first column, an indicator for having at least one stress-related medical risk (hypertension, eclampsia, or genital herpes) is the outcome. In the second column, an indicator for having at least one other medical risk (anemia, cardiac disease, diabetes, lung disease, hydramnios, hemoglobinopathy, incompetent cervix, previous infant >4000 grams or preterm, renal disease, Rh sensitization, uterine bleeding, or non-specified risk factor) is the outcome. The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004. Regressions include mother’s parity, age (and age squared), and indicators of mother’s education (4 levels), race (5 categories – Black, White, Hispanic, Asian, Other) and marital status, as well as indicators of child’s gender (Panel A only), plurality (twins or higher), year of birth, and month of birth. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. 0.231 1 Ap Ci Notes. The table reports estimates of � - � sample Born S Born O Born N Born D Born J Born F Born M 1,21 2N reported covariate *Signific Appendix Table City women 19 Born Sep 2001 0.00 Born Oct 2001 0.00 Born Nov 2001 0.00 Born Dec 2001 0.00 Born Jan 2002 -0.00 Born Feb 2002 0.00 Born March 2002 -0 27N 1,198,265 includes all live births to *Significant at 10%; **s 1 Notes. The table reports estimates of � - � Appendix Table V Average grades for New York City public school children 2003/4-2009/10 Average grade Boys Girls sample mean 2.85 3.05 born Mar 20 to June 18, 2002 -0.024 -0.018 [0.022] [0.020] born Dec 11 to Mar 19, 2002 -0.039* -0.0041 [0.021] [0.019] Notes. The table reports estimates of � and � 2N 24,566 24,259 R0.08 0.08 in specification (3) run separately by child’s gender. The sample is a non-representative convenience sample of 6-year-old public school students in New York City who are not English Language Learners from the 2003 through 2009 school 21 years. Regression includes indicators of child’s race (5 categories – black, white, Hispanic, Asian, other), home zip code, year of birth, and month of birth. Indicators of missing values are also included for covariates. *Significant at 10%; **significant at 5%; ***significant at 1%. Outcomes for births occurring to New York City women 1995-2004 Appendix Table VI Medical risk Labor complication indicator Gestation Birthweight Apgar indicator 20.002 0.0003 0.001 0.0003 0.0032 in specification (1) with only indicators of year of birth and month of birth included as covariates. The sample includes all live births to women residing in one of the five boroughs of New York City from 1995 through 2004. *Significant at 10%; **significant at 5%; ***significant at 1%. sample mean 0.248 0.360 38.8 3,253 8.995 inutero1 0.0062* 0.0075** -0.220*** -9.832** -0.013*** [0.0032] [0.0035] [0.020] [4.491] [0.0051] inutero2 0.013*** 0.0063* -0.179*** -22.678*** -0.015*** [0.0029] [0.0033] [0.018] [4.134] [0.0047] N 1,198,265 1,198,265 1,198,265 1,198,265 1,198,265 R1 Notes. The table reports estimates of � and � Appendix Table VII Outcomes for New York City public school children 2003/4-2009/10 Behavior Special Held back Days absent incident education indicator from school indicator indicator Boys: sample mean 11 0.04 0.12 0.04 born Mar 20 to June 18, 2002 -0.010 0.0052* 0.011** 0.0055** [0.158] [0.0028] [0.0044] [0.0026] born Dec 11 to Mar 19, 2002 0.132 0.0032 0.0094** 0.0065*** [0.149] [0.0026] [0.0042] [0.0025] 2N 210,519 0.02 210,519 210,519 210,519 R0.10 0.01 0.003 Girls: sample mean 11 0.01 0.04 0.02 born Mar 20 to June 18, 2002 -0.052 0.00042 -0.0013 -0.00017 [0.152] [0.0014] [0.0030] [0.0021] born Dec 11 to Mar 19, 2002 0.164 -0.0019 -0.00069 -0.00081 [0.144] [0.0014] [0.0028] [0.0020] Notes. The table reports estimates of � and � 2N 199,778 199,778 199,778 199,778 R0.10 0.001 0.01 in specification (3) run separately by child’s gender with only indicators of year of birth and month of birth included as covariates. The sample includes all 6-year-old public school students in New York City who are not English Language Learners from the 2003 through 2009 school years. *Significant at 10%; **significant at 5%; ***significant at 1%. 21 0.003