Intergenerational health consequences of in utero exposure to maternal stress: Evidence from the 1980 Kwangju uprising

Social Science & Medicine 119 (2014) 284e291

Contents lists available at ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

Intergenerational health consequences of in utero exposure to maternal stress: Evidence from the 1980 Kwangju uprising* Chulhee Lee Department of Economics, Seoul National University, Seoul, South Korea

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 2 July 2014

The evidence that demonstrates the negative effects of maternal psychological stress during pregnancy on a wide variety of offspring outcomes is growing. Animal studies suggest that negative influences of maternal stress during pregnancy persist across multiple generations, but the direct evidence to confirm that the effect is present among human populations is scarce. This study draws evidence on the intergenerational influences of maternal stress from the Kwangju uprising (May 18e27, 1980), arguably the bloodiest incident that occurred in South Korea since the end of the Korean War in 1953. The results of difference-in-difference estimations suggest that in utero exposure to the Kwangju uprising significantly diminished the offspring birth weight and length of gestation, and increased the risks of low birth weight and preterm birth. Exposure to stress during the second trimester of pregnancy exerted the strongest negative effect on grandchildren's birth outcomes. © 2014 Elsevier Ltd. All rights reserved.

Keywords: Stress in pregnancy Birth outcomes Low birth weight Preterm birth Intergenerational effect Kwangju uprising

1. Introduction The strong and persistent effects of in utero exposure to adverse environment on health and socioeconomic outcomes in older adults are widely accepted. In recent years, a growing number of studies have offered semi-experimental evidence on the consequences of exogenously generated shocks to fetal health, including disease, famine, pollution, and economic recession. Growing evidence demonstrates that maternal psychological stress during pregnancy, a more subtle form of health shock, negatively affects a variety of offspring outcomes. Animal studies suggest that negative influences of maternal stress during pregnancy persist across multiple generations, but limited direct evidence confirms that the negative influences are present among human populations. The current study draws evidence on the intergenerational influences of maternal stress from the Kwangju uprising (KU, hereafter), arguably the bloodiest incident that occurred in South Korea

* This study benefited from the criticisms and suggestions of Sok Chul Hong and Alain Gagnon, and the participants of the 2013 Asia-Pacific Economic and Business History Association Conference, IUSSP International Population Conference and Korean Economic Association Conference, and seminars at Korea University and Seoul National University. I would also like to thank Tommy Bengtsson and two anonymous referees for their helpful comments. This research was supported by the National Research Foundation of Korea Grant funded by the Korean Government (NRF-2013S1A5A2A01018232). The author takes the responsibility for any remaining errors. E-mail address: [email protected].

http://dx.doi.org/10.1016/j.socscimed.2014.07.001 0277-9536/© 2014 Elsevier Ltd. All rights reserved.

since the end of the Korean War in 1953. We hypothesize that the maternal stress in pregnancy from the KU negatively affected the fetal health of the children, and that the impact of the adverse early-life shock even transmitted to the health of the next generation. To investigate this hypothesis, we compare several measures of birth outcomes between the children of women exposed to the KU and the rest of newborn babies. The difference-in-difference estimation method is applied. The results suggest that in utero exposure to the KU adversely affected offspring birth outcomes. The impact of exposure to maternal stress differs by stage of pregnancy when the shock is received. Exposure to stress during the second trimester of pregnancy exerted the strongest negative effect on grandchildren's birth weights and the risks of low birth weight and preterm birth. As for the length of gestation, the second and third trimesters were equally critical. The intergenerational effects of the exposure to stress in pregnancy were more strongly associated with boys than with girls. 2. Maternal stress and offspring outcomes A voluminous body of research has established that in utero exposure to adverse environment has strong and persistent effects on health and socioeconomic outcomes in older adults. Since David J. Barker and his colleagues developed and popularized the fetal origins hypothesis (Barker, 1992, 1994), various non-experimental studies on the long-term effects of low birth weight have

C. Lee / Social Science & Medicine 119 (2014) 284e291

supported this thesis (Currie and Hyson, 1999; Behrman and Rosenzweig, 2004; Black et al., 2007; Currie and Moretti, 2007). In recent years, a growing number of studies have offered semiexperimental evidence on the consequences of exogenously generated shocks to fetal health, including hunger, disease, and pollution. Such traumatic events include the 1918 Pandemic Influenza (Almond, 2006; Almond and Mazumder, 2005), the Dutch Famine (Neugebauer, Hoek, and Susser, 1999; Roseboom, Meulen, Ravelli et al., 2001; Bleker et al., 2005), the Chinese Famine (St. Clair et al., 2005; Luo et al., 2006; Chen and Zhou, 2007; Meng and Qian, 2009), and the Chernobyl disaster (Almond et al., 2009). Evidence also shows that maternal psychological stress during pregnancy, a more subtle form of negative health shock, negatively affects a wide variety of offspring outcomes. A possible mechanism by which maternal stress affects the health of the next generation is hormonal change. Human studies suggest that maternal stress during pregnancy, associated with raised plasma levels of corticotrophin releasing hormone (CRH) and cortisol, may increase the likelihood of preterm birth, developmental delays, and behavioral abnormalities in the children (Weinstock, 2001). Aizer et al. (2009) reports that poor mothers have elevated levels of the stress hormone cortisol, and the increased cortisol in pregnancy negatively affects the cognition, health, and educational attainment of their children. An emerging research by economists and other social scientists provides evidence pertaining to the effects of maternal stress on offspring health and economic outcomes based on semiexperimental episodes. The present paper is closely related to the studies that examined the effects of exposure to violent events such as the September 11, 2001 terrorist attacks (Lauderdale, 2006; Eccleston, 2011), landmine explosion in Columbia (Camacho, 2008), the Korean War (Lee, 2014a), the 2005 Chilean earthquake (Torche, 2011), and natural disasters in the U.S. (Simeonova, 2011; Currie and Rossin-Slater, 2012). Animal studies suggest that the negative influences of maternal stress during pregnancy persist across multiple generations (Stewart et al.,1975,1980; Pinto and Shetty,1995; Sheriff et al., 2010). The presence of the multi-generational effect of maternal stress in human populations is less clear, although recent research suggests that early-life exposure to famine has negative effects on the health and socioeconomic status of the next generation (Fung and Ha, 2010; Kim and Fleisher, 2010). Epidemiological studies have reported that parental and offspring birth weights are positively related (Emanuel et al., 1992; Klebanoff et al., 1997). Currie and Moretti (2007) found that if a mother was born with low birth weight, her child is significantly more likely to have low birth weight. If these findings are combined with the result demonstrating that maternal stress in pregnancy reduces offspring birth weights, we might expect to find the multi-generational effects of maternal stress. A possible mechanism behind the intergenerational transmission of birth weight is maternal growth, especially the uterine and ovarian size. Exposure of the fetus to a negative shock in utero may permanently alter physiology in adulthood. Such physiological changes may result in an adverse intrauterine environment for the offspring of the mother, leading to physiological changes in the next generation. Evidence shows that maternal size is associated with the size of the offspring (Hackman et al., 1983; Emanuel, 1997). Another possible explanation is that a stimulus or insult received during critical periods of growth and development may permanently alter the programming of physiology by epigenetic modification (Rakyan et al., 2001). 3. The Kwangju uprising in May 1980 The Kwangju uprising took place on May 18, 1980 in the southwestern city of Kwangju (with a population of 730,000 at the

285

time) in the course of political crisis that followed the assassination of President Park in October 1979. Student demonstrations demanding democratic reform turned to mass protests by citizens who were angered by the brutal acts of violence conducted by paratroopers (special unit trained for guerrilla warfare), and the incident soon escalated into a bloody warfare between armed civilian militias (called the Citizen Army) and martial law troops. An estimated 500 civilians were killed and over 3000 were wounded in the violence that engulfed the city (Shin and Hwang, 2003). Testimonials frequently refer to ongoing medical problems, financial difficulties, and general hardship of the victims (Byun and Lewis, 2000). A more systematic study by a medical anthropologist/nurse found that 92% of Kwangju victims developed post-traumatic stress disorders; they were much more likely to suffer from physical ailments, anxiety, and depression than nonvictims in Kwangju (Byun, 1996). Even for the citizens who avoided any direct damage, the incident was extremely stressful. How the traumatic experiences during the KU affected the lives of the vast majority of the citizens who were not visibly influenced by the violence during the event was not systematically documented. No attempts have been made to understand the impact of the tragic event on offspring outcomes. The KU provides a unique opportunity for identifying the impact of maternal psychological stress on pregnancy. First, the tragic incident of May 1980 was a completely unanticipated shock that was unavoidably sustained by all citizens of Kwangju. The aggressive move of the new military regime to suppress the democratization movement was largely unknown to the public until it actually happened. Once the crackdown started, the citizens could not keep away from the terror by moving out of the city because all routes leading into and out of Kwangju were blocked by government troops. Second, the major type of negative shocks to fetal health caused by the KU is likely to be maternal psychological stress. Although the incident was highly violent, the vast majority of the citizens and almost all pregnant women were physically unaffected. The primary targets of the brutal assault of martial law troops were young males. A number of public buildings were burnt or destroyed, but the majority of residential and commercial buildings were untouched. Provisions of foods and public utilities were uninterrupted. Finally, the duration of potential exposure to maternal stress was brief: only 10 days between May 18 and May 27 in 1980. Shortly after the Citizen Army rebellion was crushed by the government troops at the provincial office on May 27, the uprising ended abruptly. On the surface, the city resumed its ordinary life only in a few days. A few politicians, activists, and students were arrested and jailed on the accusation of instigating the uprising. The agony and anger of the Kwangju citizens continued as they were often treated as “mobsters” or “communist spies.” Nevertheless, the direct exposure to intense violence lasted about 10 days. 4. Data and methods We hypothesize that maternal stress in pregnancy negatively affected the fetal health of their children, and that the impact of the adverse early-life shock even transmitted to the health of the next generation. In the balance of the paper, the women who experienced the KU while in pregnancy, their children who were exposed to the KU in utero, and their grandchildren will be denoted, respectively, G1, G2, and G3. The first step of the research is to select G2. We assume that individuals whose mother resided in Kwangju in May 1980 and who were born between June 1980 and February 1981 were exposed to the violence in utero. We do not regard persons born during and shortly after the incident (May 18e31) as

286

C. Lee / Social Science & Medicine 119 (2014) 284e291

“treated,” because their births could be negatively influenced by the chaotic conditions of hospitals in Kwangju during the uprising as well as maternal stress. To study the intergenerational transmission of maternal stress caused by the violence during the KU, the following difference-indifference model is employed in the regression analyses:

extremely small number of the treated fathers included in the 2000 and 2002 vital statistics. Only 48 children who were born in Kwangju and whose fathers were born between June 1980 and February 1981 are included in the data. The number of treated fathers is so small because very few men in Korea marry before the age of 21.

yi ¼ a þ b1 Ii þ b2 ðIi  Si Þ þ b3 Si þ g1 YOBi þ g2 YOB2i þ g3 YOB3

5. Results

þ dXi þ εi ; (1) where y denotes birth outcomes, I is the dummy variable that identifies a mother (G2) as in utero between June 1980 and February 1981, S is the dummy variable that indicates whether the child (G3) was born in the city of Kwangju, YOB stands for the mother's year of birth, and X represents other characteristics of the child and parents that potentially influence birth outcomes, including year of birth, sex, multiple births, and parents' education and occupation. The data used in this study are micro files of the 2000 and 2002 Vital Statistics for birth in Korea. These data contain the universe of the digitalized birth registration records for all children born in the given years, which provide information on the characteristics of the parents and the newborn babies. The Korean legal system requires all citizens to be registered with a local government office within one month after birth. The documents required for the registration include birth certificates that are recorded and issued by hospitals. Information on the characteristics of parents and babies is drawn largely from birth certificates. The variables offered by the data include the child's sex, place of birth, date of birth, length of gestation (in week), birth weight, and age, education, and occupation of the parents. We can also determine if the birth weight is lower than 2.5 kg (referred to as low birth weight, or LBW) and if the length of gestation is less than 37 weeks (referred to as preterm birth). The micro files of the data are available from 1991, but the exact birth dates of the parents are reported only in the 2000 and 2002 micro data. The data for the other years only provide the ages of the parents at the time of registration. For the years except 2000 and 2002, thus, accurately identifying the year and month of birth is difficult. Only the birth records of children born in 2000 or 2002 are utilized in this study due to this data feature. Several drawbacks of the data used in this study limit the scope and quality of the study. First, the actual birthplace of G2 cannot be identified from the data, and we cannot but use the birthplace of G3 as a proxy for the place of residence of G1 or the birthplace of G2 20 years earlier. If geographic migrations are considered, some of the young women who gave birth in Kwangju in the early 2000s were born elsewhere. This type of measurement error results in downward bias in the estimated effect of exposure to the KU. This potential bias problem arising from geographic mobility is discussed in Section 6. Second, G2 who were included in the 2000 and 2002 birth records were aged 20e22 years at the time of delivery. Given that the average age of marriage and first birth in the early 2000s was the late 20s, available data allow us to examine the birth outcomes of children born to a selected group of mothers who gave birth at a relatively young age. Possible bias issues related to this drawback are addressed in Section 6. Finally, the effects of early-life conditions may be transmitted across generations through both maternal and paternal lines (Emanuel et al., 1992; Klebanoff et al., 1997; Fung and Ha, 2010). However, examining the intergenerational consequences of the fathers' prenatal exposure to the KU is difficult because of the

If the impact of maternal stress caused by the KU transmitted across two generations, G3 might reveal worse birth outcomes than babies whose grandmothers were not influenced by the incident. To verify this possibility, we compare some measures of birth outcomes, namely, average birth weight, probability of LBW, length of gestation, and probability of preterm birth, between G3 and the rest of the children. Ordinary least squares regressions for birth weight and gestation period are conducted according to the difference-in-difference model of Equation (1). In all statistical tests, standard errors are adjusted for clustering at the levels of child's province of birth and mother's year of birth. If the exposure of a pregnant woman to the Kwangju uprising negatively affected the birth outcomes of their grandchildren, then the coefficient of the interaction between the dummy variable for birth in Kwangju (B_Kwangju) and that for having a mother born between June 1980 and February 1981 (B_Jun80-Feb81) should be negative. The regression results for birth weight are reported in Table 1. In the baseline model (Column 1), polynomial terms of the year of birth of the mother (YOB, YOB2, and YOB3) are included to consider long-term changes in birth weight across birth cohorts. The variables of key determinants of birth weight, such as year of birth (year 2002, omitted category is year 2000), sex of the child (male), multiple birth (twins or triplets), and parity order (second, third or higher), are also included. In the second model (Column 2), the variables of the educational attainment and occupation of the mother are added. Schooling is classified into four categories: (1) middle school or lower, (2) high school, (3) college or higher, and (4) education unknown. Occupation is classified into eight categories: (1) professional (2) clerical job, (3) service and sales, (4) farming, (5) skilled, (6) unskilled, (7) not working, and (8) occupation unknown. In the final model (Column 3), the variables of the socioeconomic status of the father and mother are included. The prenatal exposure of the mother to the KU reduced the offspring birth weight. The coefficient for the interaction term (B_Kwangju*B_Jun80-Feb81) is negative and significant. The magnitude of the coefficient suggests that the average birth weight of Kwangju grandchildren is lower than that of non-Kwangju grandchildren by about 56 g, controlling for the year of birth of the mother and some key birth characteristics. Adding parental socioeconomic characteristics does not significantly change the result. The year of birth of the mother is negatively related to birth weight. Given that the analysis is based on the 2000 and 2002 birth records, it perhaps captures the effect of early-age delivery. The effects of birth characteristics and parental socioeconomic status (SES) show largely anticipated signs. Boys are heavier at birth than girls. Multiple births reduce birth weight, and higher-parity children are heavier at birth. Parental education is positively related to birth weight. Table 2 reports the regression results for the other three birth outcomes based on the full model with controls for parental SES. The in utero exposure of the mother to the KU reduced the gestation period of the child by about 2.3 days. G3 were about 2.4% more likely to have LBW than the rest, and the difference is statistically significant. Similarly, G3 were at a higher risk of preterm birth. As

C. Lee / Social Science & Medicine 119 (2014) 284e291

287

Table 1 Mother's in utero exposure to the Kwangju uprising and offspring birth weight (in grams). Mean

(1)

(2)

vy/vx Intercept Mother's place & timing of birth B_KWANGJU B_KWANGJU*B_Jun80-Feb81 B_Jun80-Feb81 Mother YOB Mother YOB2 Mother YOB3 Characteristics of birth Year 2002 Male Twins Triplets Birth parity: second Birth parity: third or higher Mother's SES Mother middle school or less Mother high school Mother college or higher Mother's education unknown Mother professional job Mother clerical job Mother service/sales Mother farming Mother skilled job Mother unskilled job Mother not working Mother occupation unknown Father's SES Father middle school or less Father high school Father college or higher Father's education unknown Father professional job Father clerical job Father service/sales Father farming Father skilled job Father unskilled job Father not working Father occupation unknown R-square F-value P-value

(3)

P-value

vy/vx

P-value

vy/vx

P-value

4704.42

<0.0001

4479.98

<0.0001

4526.38

<0.0001

0.03259 0.00021 0.00607 71.6 5144.7 370,645.7

5.20 55.82 15.17 105.28 2.06 0.01

0.0059 <0.0001 0.0075 0.0125 0.0007 <0.0001

3.63 57.56 16.19 88.38 1.71 0.01

0.0475 <0.0001 0.0045 0.0300 0.0034 0.0003

2.68 57.27 15.89 87.68 1.66 0.01

0.1440 <0.0001 0.0053 0.0279 0.0036 0.0004

0.43653 0.52389 0.01793 0.00017 0.41977 0.09843

14.63 101.26 834.89 1232.68 14.87 72.86

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

13.22 101.21 835.36 1230.06 16.80 78.24

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

13.43 101.19 835.87 1229.99 16.52 78.60

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

42.68 NI 10.12 196.21 3.92 11.86 7.67 10.93 7.18 14.82 NI 32.36

<0.0001

28.02 NI 1.93 169.67 5.65 9.43 6.94 1.79 10.23 9.81 NI 28.07

<0.0001

0.03731 0.55457 0.40549 0.00263 0.04831 0.06670 0.01336 0.00594 0.00265 0.00188 0.84368 0.00628

<0.0001 <0.0001 0.0374 <0.0001 0.0330 0.0971 0.3489 0.1377 <0.0001

20.44 NI 12.21 97.55 0.45 NI 2.55 12.36 5.72 7.79 7.04 7.19

0.04728 0.44835 0.49977 0.00460 0.17894 0.34924 0.13535 0.03093 0.08688 0.06344 0.06096 0.00823

0.0835 <0.0001 0.0025 <0.0001 0.0514 0.7849 0.1872 0.3285 <0.0001 <0.0001 <0.0001 <0.0001 0.6666 0.0778 <0.0001 0.0016 <0.0001 0.0009 0.2912

0.0764 0.0772

0.0777

3343.32

2478.93

<0.0001

<0.0001

5659.43 <0.0001

Note: The number of observation is 1,124,848. The dependent mean is 3262.142 g. NI denotes the omitted category (control group). Standard errors are adjusted for clustering at the levels of child's province of birth and mother's year of birth. Source: Micro data on the Korean Vital Statistics (Birth) for 2000 and 2002.

with birth weight, adding parental SES does not change the magnitude and statistical significance of the coefficient for the interaction term. The impact of exposure to maternal stress could differ by stage of pregnancy when the shock is received. To verify this possibility, G3 are classified into three groups, depending on which trimester of pregnancy their mothers experienced the event. For example, mothers born between December 1980 and February 1981 were in the first trimester when the KU occurred. Similarly, those born between September 1980 and November 1980 and between June 1980 and August 1980 were exposed to the KU when they were in the second and third trimesters, respectively. In the new regression equations, three dummy variables of the month or year of birth of the mothers corresponding to the three groups (B_Dec80-Feb81, B_Sep80-Nov80, and B_Jun80-Aug80) and their interactions with the dummy variable for birth in Kwangju (B_Kwangju) are included, replacing B_Jun80-Feb81 and

B_Kwangju*B_Jun80-Feb81. Other control variables of the year of birth of the mother, birth characteristics, and parental SES are included. Table 3 shows how the in utero exposure of the mother to the KU in each trimester affects each of the four birth outcomes. Column 1 shows exposure during the second trimester exerted the strongest negative effect on offspring birth weight, diminishing birth weight by 103 g. Maternal exposure to stress during the first and final trimesters also reduces the birth weight of G3, but the effects are less significant and smaller in magnitude (39 and 23 g, respectively). Similarly, maternal exposure to stress during the second trimester of pregnancy is particularly harmful for offspring risks of LBW and preterm birth (Columns 3 and 4). Whether the intergenerational consequences of the in utero exposure of the mother to the KU differ by offspring gender is also examined by separate regression analyses for boys and girls. The results suggest that the overall effects of the in utero exposure of the mother to the

288

C. Lee / Social Science & Medicine 119 (2014) 284e291

Table 2 Mother's in utero exposure to the Kwangju uprising and birth outcomes.

B_KWANGJU B_KWANGJU* B_Jun80-Feb81 B_Jun80-Feb81 Mother YOB, YOB2,YOB3 Characteristics of birth Mother's SES Father's SES R-square F-value P-value

(1) Gestation

(2) LBW

vy/vx

vy/vx

P-value

0.081 <0.0001 0.0002 0.328 0.0097 0.0236 0.022 Yes Yes Yes Yes 0.0757 702.38 <0.0001

0.2983 0.0017 Yes Yes Yes Yes 0.1073 542.44 <0.0001

(3) Preterm birth P-value

vy/vx

P-value

0.8075 0.0031 <0.0001 0.0152

0.0006 0.0218

0.5252 0.0003 Yes

0.8890

Yes Yes Yes 0.0787 463.21 <0.0001

Note: The number of observation is 1,124,848. The dependent mean is 39.175 for (1), 0.0388 for (2), and 0.0406 for (3). The same independent variables used in the regressions presented in Table 1 are included. Standard errors are adjusted for clustering at the levels of child's province of birth and mother's year of birth. Source: Micro data on the Korean Vital Statistics (Birth) for 2000 and 2002.

Kwangju uprising are stronger for boys than for girls (Lee, 2014b, Table 9). 6. Discussion 6.1. Robustness Several robustness checks are performed (See Table 9 in Lee, 2014b for detailed results). First, to compare mothers (G2) with similar year of birth, the children of women born between 1975 and 1985 are included in regressions, excluding the children of mothers who are much older or younger than the cohort who experienced the Kwangju uprising in utero. The results are practically the same as the baseline regression based on the full sample. Second, only babies born in the six metropolitan cities (i.e., Seoul, Busan, Incheon, Daejeon, Daegu, and Kwangju) are included in regression analyses to make comparisons across mothers living under similar conditions (e.g., similar medical facilities and ecological environment). For all four measures of birth outcomes, regression results based on the urban sample are remarkably similar to the results from the whole sample. Third, children born in counties adjacent to Kwangju are included in the treated group along with those born in Kwangju to consider the potential spillover effects of the uprising. Although major clashes between martial law troops and citizens occurred in Kwangju, neighboring areas were affected by the uprising to some extent. Regression results show that the prenatal exposure of the mother to the KU during the second trimester of pregnancy had significantly negative effects on birth outcomes. However, the overall effects are weaker than those of the baseline results, in which only the children born in Kwangju are regarded as treated. Fourth, children (G3) born anywhere in Chonnam Province are regarded as treated by the Kwangju incident. Kwangju and Chonnam residents probably share similar political, social, and cultural backgrounds, although they differ in terms of the extent of urbanization. In particular, the entire region supported Former President Kim Dae Jung and most actively protested against the military regime that emerged in 1980. The purpose of the final robustness check is to determine if the negative effect captures the influences of exposure to the violence during the KU or the effects of the political oppression of the military regime during the period that were felt by a larger population in the region. Regression results do not support this conjecture. The negative effect of maternal

exposure to the KU is likely to reflect the negative intergenerational effects of stress caused by the first-hand experience of the violent event. Fifth, a birth month fixed-effect model is estimated with dummy variables on the mother's month of birth added. Seasonal variations in birth outcomes are found even in rich countries (Chodick et al., 2009). Individuals exposed to the KU in each trimester of pregnancy were born in a particular season; thus, seasonal influences might be compounded with the effect of exposure to maternal stress. However, the regression results reject this hypothesis. Finally, the independent variables of interest are redefined by excluding the mothers born in February 1981 from the treated group. The females who were in the very early stage of pregnancy during and shortly after the uprising (those born in February 1981) could be selected because the probabilities of conception and miscarriage can be affected by the incidence. In such cases, the intergenerational health consequences of exposure to the KU during the first trimester of pregnancy can be biased. The regression results suggest that the selection bias, if any, is not significant.

6.2. Potential bias problems A critical weakness of the data used in this study is that the birthplace of the parent is not reported. The place of residence of women in the sample at the time of giving birth must be an inaccurate proxy for their birthplace because of geographic mobility. The direction and magnitude of potential bias arising from the problem depend on the extent and pattern of migrations from and to Kwangju. Statistics computed from the 2% sample of the 2000 census suggest considerably high geographic mobility of female birth cohorts born around 1980 (those aged 18e22 years in 2000) from their birth to 2000. Nearly half of the women born in Kwangju had remained in Kwangju or adjacent counties by 2000, and about an equal proportion had moved out of the Chonnam Province. Of the Kwangju female citizens aged 18e22 years in 2000, 20% were Kwangju natives, another 20% came from adjacent counties, 40% from other Chonnam areas, and the remaining 20% from other provinces. The considerably high geographic mobility must lead to a misidentification of treated persons. That is, G3 include both the children of women actually exposed to the KU while in utero and the children of those who were born elsewhere and migrated to Kwangju. As noted above, any negative effects of the KU were likely limited to those who were directly exposed to the incident. If this type of bias is considered, we suspect that the actual negative effects of in utero exposure to the KU on offspring birth outcomes might be even more serious than suggested by the results of this study. Another possible concern related to geographic mobility is that the results of this study might be attributed to the consequences of selective migration. As difference-in-difference estimation is employed and the socioeconomic characteristics of parents (i.e., education and occupation) are controlled for, the so-called “selective migration story” is plausible if the following condition holds: the selection of Kwangju residents around 2000 out of all women born in Kwangju was discontinuously negative for the cohort born between June 1980 and February 1981 in terms of unobservable personal characteristics. Although the validity of the selective migration story is difficult to examine directly, some circumstantial evidence can be offered based on the micro-sample of the 2000 census. Using a sample of women born in Kwangju aged 18e22 years in 2000, the following difference-in-difference regressions were conducted to determine

C. Lee / Social Science & Medicine 119 (2014) 284e291

289

Table 3 Mother's in utero exposure to the Kwangju uprising and offspring birth weight by stage of pregnancy. (1) Birth weight

B_KWANGJU B_KWANGJU*B_Dec80-Feb81 B_KWANGJU*B_Sep80-Nov80 B_KWANGJU*B_Jun80-Aug80 B_Dec80-Feb81 B_Sep80-Nov80 B_Jun80-Aug80

(2) Gestation

(3) LBW

(4) Preterm birth

vy/vx

P-value

vy/vx

P-value

vy/vx

P-value

vy/vx

P-value

2.68 36.80 104.65 27.11 29.37 5.48 13.79

0.1440 0.0227 <0.0001 0.0141 0.0037 0.5798 0.2110

0.081 0.090 0.539 0.510 0.013 0.066 0.012

<0.0001 0.1084 <0.0001 <0.0001 0.7028 0.0216 0.7622

0.0002 0.0171 0.0403 0.0125 0.0039 0.0049 0.0032

0.8077 0.0075 <0.0001 0.0002 0.4583 0.0814 0.3443

0.0031 0.0188 0.0462 0.0153 0.0061 0.0049 0.0001

0.0006 0.1637 <0.0001 <0.0001 0.1875 0.1598 0.9749

Note: The number of observation is 1,124,848. The dependent mean is 3262.142 for (1), 39.175 for (2), 0.0388 for (3), and 0.0406 for (4). The same independent variables used in the regressions presented in the column 3 of Table 1 are included. Standard errors are adjusted for clustering at the levels of child's province of birth and mother's year of birth. Source: Micro data on the Korean Vital Statistics (Birth) for 2000 and 2002.

if educational attainment as of 2000 was significantly lower for Kwangju citizens born in 1980.

Ei ¼ a þ b1 Ii þ b2 ðIi  Ki Þ þ b3 Ki þ g1 AGEi þ g2 AGE2i þ εi ;

that of February-born mothers who gave birth in other places. This result rejects the conjecture that fewer Kwangju natives were born around February 1981.

(2)

where E denotes educational attainment (dummy variables for no high school diploma and for entering college), I denotes the dummy variable for women born in 1980 (a proxy variable for being in utero during the Kwangju uprising), K is the dummy variable for whether the woman lived in Kwangju in 2000, and AGE stands for the age of the woman in 2000. The results of regressions suggest outmigrants from Kwangju were positively selected, but migrant selections were not significantly different between the 1980 birth cohort and neighboring cohorts (Lee, 2014b, Table 12). Selective marriage is another possible source of bias. As only the 2000 and 2002 Vital Statistics on birth provide information on the dates of birth of parents, the empirical analyses above were restricted to babies born to relatively young mothers. Mothers exposed to the KU while in utero were in their early 20s at the time of giving birth. If only selected women married early and selections in marriage among young women were discontinuously negative for those who were in utero during the KU, the results of this study regarding the intergenerational impact of prenatal exposure to stress could be explained by selective marriage. To examine if this selective marriage hypothesis is plausible, difference-in-difference regressions were performed employing similar methods used for those conducted regarding selective migration. The results show that married women are less educated, but selections in marriage are not significantly different between the 1980 birth cohort and neighboring cohorts (Lee, 2014b, Table 13). Overall, circumstantial evidence suggests that the negative intergenerational effects of maternal stress in pregnancy are not attributable to bias problems related to selective migration or marriage. Finally, potential bias problem resulting from selections in earlylife deaths is a matter of possible concern in interpreting the results of this study. If individuals who were in utero during the KU were positively selected in terms of characteristics correlated with offspring birth outcomes, the effect of exposure to maternal stress in pregnancy estimated in this study is subject to a downward bias. Furthermore, if the risk of stress-caused miscarriage is particularly high in the early stage of pregnancy, the effect of exposure to the uprising during the first trimester of pregnancy could be more seriously underestimated. It is difficult to obtain direct evidence regarding selections in live birth such as incidence of miscarriage. However, circumstantial evidence suggests that this type of bias is not serious. For instance, the percentage of G2 who were born in February is relatively high compared with that of G2 born on other months or compared with

6.3. Significance of the effects in comparative perspectives The intergenerational effects of the in utero exposure to the KU are considerably large in magnitude, judging from comparisons with those of maternal stress during pregnancy on the birth outcomes found in previous studies. The largest estimates of the effects of prenatal exposure to the 2005 Chilean earthquake suggest that those exposed in the first trimester suffer a 51 g decline in birth weight and a 1.7% rise in the probability of low birth weight (Torche, 2011). Similarly, the effects of in utero exposure to floods in the second trimester of pregnancy were estimated as the strongest consequence of early-life exposure to natural disasters in the U.S.; such an exposure can reduce birth weight by 18 g and gestation age by approximately 2 days (Simeonova, 2011). Cohorts exposed to the September 11, 2001 terrorist attack during the first or second trimester weighed 12 ge14 g less and had gestation length 1 day to 1.5 days shorter on average (Eccleston, 2011). In addition, the Supplemental Nutrition Program for Women, Infants, and Children (WIC), the largest-scale nutrition supplementation program in the U.S., caused a 63 g increase in birth weight among enrolled mothers (Bitler and Currie, 2005). Prenatal exposure to maternal stress caused by the KU has emerged as a powerful predictor of birth outcomes compared with other personal and parental characteristics that are considered major determinants of fetal health. The effect of exposure to the KU in the second trimester on birth weight (a decline of about 103 g) is similar in magnitude to the estimated difference in birth weight between boys and girls (approximately 101 g) and is larger than the difference between the first and third or higher-parity children (approximately 79 g). The effects of prenatal exposure to maternal stress have considerably greater magnitude than those of parental education, which is widely acknowledged as an important determinant of child health. The children of mothers with less than high school education weigh approximately 28 ge43 g less than those of mothers with high school education (Table 1). The difference between the children of college graduates and high school graduates is small (2 ge10 g). Parental occupation has even weaker effects than the aforementioned factors. Previous studies on the effects of low birth weight suggested that the deteriorated birth outcomes observed among the treated children would exert long-term health and socioeconomic consequences. The estimates of Black et al. (2007) offer particularly useful basis for evaluating how significant those effect would be, because their study considers a wide range of short-term and longterm effects. In addition, their study successfully reduces the

290

C. Lee / Social Science & Medicine 119 (2014) 284e291

potential upward bias problem that might arise from unobservable family and genetic factors through twin fixed-effect estimations. Exposure to the Kwangju uprising in the second trimester diminished birth weight by 103 g (Table 1), approximately 3.2% of the average birth weight of the sample (3262 g). If the magnitudes of the effects of birth weight estimated by Black et al. (2007: pp. 423e425) are applied, a decline in birth weight of this magnitude is expected to bring the following consequences: a 1.3 rise in firstyear deaths per 1000 births, a 0.18 cm decrease in height, a 0.04 decrease in body mass index (BMI), a 0.02 decrease in IQ score (measured on a scale from one to nine), a 0.3 percentage point decrease in the probability of high school completion, a 0.32% decrease in full-time earnings, and a 0.5% increase in the birth weight of the first child for mothers. These anticipated effects are not large in magnitude. However, they are not negligible, considering that they capture the remaining effects of the health shock transmitted across multiple generations. 6.4. Limitations Despite the efforts exerted in this study to consider the potential sources of bias, the estimated results on the intergenerational health effects of in utero exposure to the KU may not be accurate. The series of empirical tests performed above in this section are too crude to preclude bias problems arising from selections in birth, migration, and marriage. The results of the present study may not be generalized for the entire population exposed to the KU because only the children born to relatively young mothers are considered in the analyses. The most fundamental weakness of the present work is that the treated group is inaccurately identified because of various drawbacks of the available data. As explained in detail, whether or not the mothers of children born in Kwangju are Kwangju native themselves is difficult to identify because their place of birth is unknown. Even when the mothers spent the prenatal period in Kwangju in May 1980, we do not know whether or not they were actually influenced by the violent event. Although the KU engulfed the entire city, some citizens may have been untouched by the event. Despite these shortcomings, the present study still concludes that prenatal exposure to the maternal stress caused by the KU adversely affected offspring birth outcomes. Although not decisive, the available evidence suggests that the treated group is not negatively selected. Considering the features of the identification strategy employed in this study, we may interpret the estimated effect as the average between the treated and non-treated groups. Thus, the true effect for the treated population is probably larger than that suggested in this paper. 7. Summary and implications This study investigated how the stress experienced by women during the KU in South Korea affected the birth outcomes of their grandchildren. For this purpose, such measures as birth weight, length of gestation, and risks of low birth weight (<2.5 kg) and preterm birth (<37 weeks) were compared between the children of mothers exposed to the KU while in utero and the children of untreated mothers through the difference-in-difference method. The prenatal exposure of the mother to the violent incident is estimated to reduce offspring birth weight and length of gestation by 57 g and 2.3 days, respectively, and increase the probability of LBW by 2.4%. Similarly, the probabilities of LBW and preterm birth of treated children were higher by, respectively, 2.4% and 1.5%. The impact of exposure to maternal stress varied with the stage of pregnancy when stress was received. The in utero exposure of the

mother to stress during the second trimester exerted the strongest adverse effect on offspring birth outcomes, reducing birth weight and length of gestation by 103 g and 3.8 days, respectively, and increasing the probability of LBW and preterm birth by 4.0% and 4.6%, respectively. The offspring length of gestation is also significantly diminished by the prenatal exposure of the mother to stress during the third trimester. The intergenerational effects of exposure to stress in pregnancy were stronger for boys than for girls. The results are robust to changes in sample and specification. It is unlikely that the negative intergenerational effects of maternal stress in pregnancy are caused by bias problems related to selective migration or marriage. Regarding the health consequences of exposure to stress, a traumatic experience can have a lasting impact on the life of a person even if he or she is physically unaffected. These invisible scars are often ignored when assessing the severity of a disastrous event. In the case of the KU, only about 4000 dead, missing, injured, and detained people are regarded as its victims, and eligible to receive compensation for their losses. In reality, however, a much larger fraction of Kwangju citizens of the time must have been victimized by the violence. As noted above, preterm birth and LBW are related to poor health and socioeconomic performance over the life cycle. Thus, maternal stress in pregnancy not only harms the well-being of mother and child, but also incurs economic cost. The finding that the adverse effects of maternal stress in pregnancy transmit across generations suggests that, if intergenerational consequences are considered, the economic cost of a traumatic event or stressful circumstance could be larger than suggested by an estimate obtained through short-term consideration. This leads to another policy implication: the long-run potential benefits of efforts to prevent children or pregnant women from being exposed to stressful events or environments are greater than conventionally believed. References Aizer, A., Stroud, L., Buka, S., 2009. Maternal Stress and Child Well-being: Evidence from Siblings. Brown University Working Paper. Almond, D., 2006. Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 population. J. Polit. Econ. 114 (4), 672e712. Almond, D., Edlund, L., Palme, M., 2009. Chernobyl's subclinical legacy: prenatal exposure to radioactive fallout and school outcomes in Sweden. Q. J. Econ. 124 (4), 1729e1772. Almond, D., Mazumder, B., 2005. The 1918 influenza pandemic and subsequent health outcomes: an analysis of SIPP data. Am. Econ. Rev. Pap. Proc. 95 (2), 258e262. Barker, D.J.P., 1992. Fetal and Infant Origins of Adult Disease. British Medical Journal of Publishing Group, London. Barker, D.J.P., 1994. Mothers, Babies, and Disease in Later Life. British Medical Journal of Publishing Group, London. Behrman, J.R., Rosenzweig, M.R., 2004. Returns to birth weight. Rev. Econ. Stat. 86 (2), 586e601. Bitler, M., Currie, J., 2005. Does wic work? the effect of wic on pregnancy and birth outcomes. J. Policy Anal. Manage. 24 (1), 73e91. Black, S.E., Devereux, P.J., Salvanes, K.G., 2007. “From the cradle to the labor market? the effect of birth weight on adult outcomes. Q. J. Econ. 122 (1), 409e439. Bleker, O.P., Roseboom, T.J., Ravelli, A.C.J., van Montfans, G.A., Osmond, C., Barker, D.J.P., 2005. Cardiovascular disease in survivors of the Dutch famine. In: Hornstra, G., Uauy, R., Yang, X. (Eds.), The impact of maternal malnutrition on the offspring: nestle nutrition workshop series pediatric program, vol. 55. Karger, Basel, Switzerland, pp. 183e195. Byun, J., 1996. 15-year aftermath-victim-syndrome of the 5e18 Kwangju civil uprising. Korean Cult. Anthropol. 29 (2), 221e259 (in Korean). Byun, J., Lewis, L.S., 2000. The Kwangju Uprising After 20 Years; the Unhealed Wounds of the Victims. Dahaes, Seoul. Camacho, A., 2008. Stress and birth weight: evidence from terrorist attack. Am. Econ. Rev. Pap. Proc. 98 (2), 511e515. Chen, Y., Zhou, L.A., 2007. The long-term health and economic consequences of 1959e1961 famine in China. J. Health Econ. 26 (4), 659e681. Chodick, G., Flash, S., Deoich, Y., Shalev, V., 2009. Seasonality in birth weight: review of global patterns and potential causes. Hum. Biol. 81 (4), 463e477.

C. Lee / Social Science & Medicine 119 (2014) 284e291 Currie, J., Hyson, R., 1999. Is the impact of shock cushioned by socioeconomic status? The case of low birth weight. Am. Econ. Rev. 89 (2), 245e250. Currie, J., Moretti, E., 2007. “Biology as destiny? Short- and long-run determinants of intergenerational transmission of birth weight. J. Labor Econ. 25 (2), 231e264. Currie, J., Rossin-Slater, M., 2012. Weathering the Storm: Hurricanes and Birth Outcomes. NBER Working Paper No. 18070. Eccleston, M., 2011. In Utero Exposure to Maternal Stress: Effects of 9/11 on Birth and Early Schooling Outcomes in New York City. Harvard University Working Paper. Emanuel, I., 1997. Invited commentary: an assessment of maternal intergenerational factors in pregnancy outcomes. Am. J. Epidemiol. 146, 820e825. Emanuel, I., Filakti, H., Alberman, E., Evans, S.J., 1992. Intergenerational studies of human birth weight from the 1958 birth cohort: evidence for a multigenerational effect. Br. J. Obstet. Gynecol. 99, 67e74. Fung, W., Ha, W., 2010. Intergenerational Effects of the 1959e61 China Famine. Working Paper. Hackman, E., Emanuel, I., van Belle, G., Daling, J., 1983. Maternal birth weight and subsequent pregnancy outcome. J. Am. Med. Assoc. 250, 2016e2019. Klebanoff, M.K., Schulsinger, C., Mednick, B.R., Secher, N.J., 1997. Preterm and smallfor-generation-age birth across generations. Am. J. Obst. Gynecol. 176 (3), 521e526. Kim, S., Fleisher, B.M., 2010. The China Leap Forward Famine: The Lasting Impact of Mother's Fetal Malnutrition on their Offspring. Ohio State University Working Paper. Lauderdale, D.S., 2006. Birth outcomes for Arabic-named women in California before and after September 11. Demography 43 (1), 185e201. Lee, C., 2014a. In-utero exposure to the Korean War and its long-term effects on economic and health outcomes. J. Health Econ. 33 (1), 76e93. Lee, C., 2014b. Intergenerational Health Consequences of in Utero Exposure to Maternal Stress; Evidence from the 1980 Kwangju Uprising. Working Paper, Seoul National University. Luo, Z., Mu, R., Zhang, X., 2006. Famine and overweight in China. Rev. Agric. Econ. 28 (3), 296e304.

291

Meng, X., Qian, N., 2009. The Long-term Consequences of Famine on Survivors: Evidence from a Unique Natural Experiment Using China's Great Famine. NBER Working Paper No. 14917. Neuzgebauer, R., Hoek, H.W., Susser, E., 1999. Prenatal exposure to wartime famine and development of antisocial personality disorder in adulthood. J. Am. Med. Assoc. 281 (5), 455e462. Pinto, M., Shetty, P., 1995. Influence of exercise-induced maternal stress on fetal outcomes in wistar rats: intergenerational effects. Br. J. Nutr. 73, 645e653. Rakyan, V.K., Preis, J., Morgan, H.D., Whitelaw, E., 2001. The marks, mechanisms and memory of epigenetic states in mammals. Biomed. J. 356, 1e10. Rosebloom, T.J., Meulen, J.H.P., Osmond, C., Barker, D.J.P., Ravelli, A.C.J., Bleker, O.P., 2001. Effects of prenatal exposure to the Dutch famine on adult disease in later life: an overview. Twin. Res. 4 (5), 293e298. Sheriff, M.J., Krebs, C.J., Boonstra, R., 2010. The ghosts of predators past: population cycles and the roles of maternal programming under fluctuating predation risk. Ecology 91 (10), 2983e2994. Shin, G.W., Hwang, K.M., 2003. Contentious Kwangju: The May 18 Uprising in Korea's Past and Present. Rowman & Littlefield Publishers, Lanham, MD. Simeonova, E., 2011. Out of sight, out of mind? Natural disasters and pregnancy outcomes in the USA. CESifo Econ. Stud. 57 (3), 403e431. Stewart, R., Preece, R., Sheppard, H.G., 1975. Twelve generations of marginal protein deficiency. Br. J. Nutr. 33, 233e253. Stewart, R.J., Sheppard, H., Preece, R., Waterlow, J., 1980. The effect of rehabilitation at different stages of development of rats marginally malnourished for ten to twelve generations. Br. J. Nutr. 43, 403e412. St Clair, D., Xu, M., Wang, P., Yu, Y., Fang, Y., Zhang, F., Zheng, X., Gu, N., Feng, G., Sham, P., He, L., 2005. Rates of adult schizophrenia following prenatal exposure to the Chinese famines of 1959e1961. J. Am. Med. Assoc. 294 (5), 557e562. Torche, F., 2011. The effect of maternal stress on birth outcomes: exploiting a natural experiment. Demography 48, 1473e1491. Weinstock, M., 2001. Alterations induced by gestational stress in brain morphology and behaviour of the offspring. Prog. Neurobiol. 65, 427e451.