new impacts of natural disasters on children’s health: evidence … · 2020. 2. 25. · 1...
TRANSCRIPT
Persistence of Natural Disasters on Children’s Health:
Evidence from the Great Kanto Earthquake of 1923
Kota Ogasawara∗
March 6, 2021
Abstract
This study uses a catastrophic earthquake in 1923 to analyze the long-term ef-fects of a one-off disaster on children’s health. I find that fetal exposure to Japan’sGreat Kanto Earthquake had stunting effects on girls in the devastated area. Dis-aster relief spending helped remediate stunting among boys by late primary schoolages, whereas it did not ameliorate girls’ stunting, suggesting a biased remediationmechanism before birth and compensating investment after birth. While the ma-ternal mental stress via strong vibrations played a role in the adverse health effects,the maternal nutritional stress via physical disruption also enhanced those effects.
Keywords: child growth; child stunting; Great Kanto Earthquake; long-run effect;
natural disaster;
JEL Codes: I18; I19; N35;
∗Department of Industrial Engineering, School of Engineering, Tokyo Institute of Technology, 2-12-1,Ookayama, Meguro-ku, Tokyo 152-8552, Japan (E-mail: [email protected]).I would also like to thank Dan Bogart, Neil Cummins, Fabian Drixler, Bernard Harris, Janet Hunter,Volha Lazuka, Kazushige Matsuda, Yukitoshi Matsushita, Stephen Morgan, Eric Schneider, AnthonyWray, and the participants at the CSG seminar (Chiba), EHES conference (Tubingen), Stunting confer-ence (LSE), AED seminar (Kyoto), KIER workshop (Kyoto), WEHC (MIT), EHS conference (Belfast),and Tokyo Tech seminar for their helpful comments on the paper. The work was supported by the TokyoInstitute of Technology (grant in 2015; 2019). I wish to thank Motoya Taura and Tatsuki Inoue for theiroutstanding research assistance. There are no conflicts of interest to declare. All errors are my own.
1
arX
iv:1
805.
0814
8v13
[st
at.A
P] 1
Mar
202
1
1 Introduction
Natural disasters have significant and lasting impacts on economies. In Japan, the Great
Kanto Earthquake of 1923, an extremely huge quake with a moment magnitude scale of
7.9, was an unprecedented crisis, leaving 156,000 people killed, injured, or missing. The
massive earthquake of 1923 had positive long-term impacts on industries through the so-
called creative destruction mechanism, such as the technological upgrade of machinery and
the selection into efficient firms (Okazaki, Okubo, and Strobl 2019).1 However, a growing
body of the literature indicates that fetal exposure to disasters can impede normal human
development and thus lead to negative consequences on later-life health and socioeconomic
outcomes (Vellore 2018; Rosales-Rueda and Triyana 2019; Karbownik and Wray 2019).2
The weight of evidence from this literature implies that little is known about the adverse
long-term developmental effects of Japan’s 1923 earthquake.
To bridge this gap in the body of knowledge, the present study investigates the im-
pacts of fetal exposure to the Great Kanto Earthquake on children’s health. For this
purpose, I select Chiba prefecture, where the physical disruption was primarily caused
by strong vibrations, and establish a series of datasets on children’s health using a set
of physical examination records.3 Specifically, I construct school-level multidimensional
panel datasets on the height and weight both of primary school children aged 6–11 years
born between 1914 and 1929. Since my datasets have three-dimensional panel structures,
namely a data cube with school-year-age dimensions, I can control for the time-varying
unobserved factors of each school in contrast to cross-sectional and two-dimensional panel
data analyses (Balazsi, Matyas, and Wansbeek 2018). To better identify the impacts of
the earthquake, I further exploit the geospatial variation in the physical devastation and
then interact the variation with the children who potentially experienced the earthquake
in utero.
I find that fetal exposure to the Great Kanto Earthquake negatively affected the
growth of children. Primary school girls aged 9–11 exposed in utero in the area extremely
1See also Pereira (2009) for the positive long-term economic impacts of the Lisbon Earthquake of1755.
2See Almond and Currie (2011), Currie and Vogl (2013), and Prinz et al. (2018) for comprehensivereviews in the related literature.
3While Tokyo and Kanagawa prefectures were also greatly impacted by the earthquake hit, Chibaprefecture is considered to be the most appropriate research area. Tokyo experienced vibrations and anenormous fire at the same time, whereas Kanagawa suffered both vibrations and an enormous tsunami(Hunter 2014). These incidents complicate the identification because separating their effects is diffi-cult. Moreover, physical examination records for both affected prefectures are unavailable. However,fortunately, a set of records for Chiba prefecture are scattered but still remaining.
2
affected by the earthquake with the maximum seismic intensity scale were roughly one
cm shorter than those in surrounding cohorts. Given this result, I attempt to distinguish
different pathways of adverse health effects (i.e., mental and nutritional stress). First, I
test whether disaster relief supplies can mitigate the adverse health effects on children by
nourishing a pathway in the physically devastated area in which the strongest vibration
brought about unavoidable mental stress. Second, I employ the regional heterogeneity in
seismic activity and market disruption to test which stress is more likely to be plausible
in the area with little physical damage. From these exercises, I find suggestive evidence
that while mental stress plays an important role in the adverse health effects, nutritional
stress could be another pathway that enhances the adverse health effects on the exposed
children in the physically devastated area.
This study contributes to the literature in the following three ways. First, it adds
evidence on the long-term health impacts of fetal earthquake exposure. Previous studies
have documented the adverse effects of fetal earthquake exposure on pregnancy outcomes
such as low birth weight (Glynn et al. 2001; Torche 2011; Kim, Carruthers, and Harris
2017). However, limited evidence is available on the lasting effects of fetal earthquake ex-
posure on later-life health.4 This study bridges this gap by examining the adverse effects
of fetal earthquake exposure on the development and diseases of the juvenile population.
It is the first to show the long-term human costs of the earthquake of 1923, given that
previous studies have predominantly analyzed its impacts on industries and market func-
tions (Imaizumi, Ito, and Okazaki 2016; Hunter and Ogasawara 2019; Okazaki, Okubo,
and Strobl 2019).
Second, by providing evidence on the ameliorating effects of disaster relief on child
stunting, this study contributes to the literature on the optimal timing of child invest-
ment (Heckman 2012). Empirical evidence on how the complementarities in parental
investments respond to early-life shocks is at best mixed (Almond and Mazumder 2013).
Within the limited volume of evidence, the recent study by Vellore (2018) finds that New
Deal-related spending can ameliorate the adverse long-term effects of Dust Bowl exposure
on human capital formation. The consistent evidence found in the present study implies
that prenatal adverse effects can be mitigated by the end of primary school. My results
also shed light on the heterogeneity in remediation effects with respect to types of disaster
4One exception is Caruso and Miller (2015), who test the link between fetal exposure to the AncashEarthquake of 1970 and self-reported disability status in adulthood, showing little evidence on such alink in modern Peru.
3
relief. However, I find a clear gender imbalance in the compensating effects of the disaster
relief, which may be driven by the prenatal remediation mechanism and postnatal-biased
resource allocation regulated by the institutional context.
Third, this study enriches our understanding of the mechanisms behind the adverse
health effects of fetal health shocks. While a growing body of evidence documents the
long-term effects of early-life health shocks on human capital, identifying the pathways
of those effects has remained a challenge (Currie and Vogl 2013). I provide suggestive
evidence not only on the role of the maternal mental stress underlying the adverse health
effects of fetal earthquake exposure, but also on the enhancing role of maternal nutritional
stress in the physically disrupted area. This is in line with the call for future research
by Prinz et al. (2018), who suggest that mental health issues are becoming increasingly
important for understanding labor market outcomes in adulthood.
2 Background
2.1 Great Kanto Earthquake
The Great Kanto Earthquake hit the southern area of the Kanto district including the
seven prefectures shown in Figure 1: Tokyo, Kanagawa, Chiba, Saitama, Shizuoka, Ya-
manashi, and Ibaraki. Although both the physical and the human damage were con-
centrated on Tokyo and Kanagawa prefectures, Chiba prefecture was also considerably
affected not only by the main shock but also by the aftershocks.5 Roughly one in ten
households in Chiba were damaged by the earthquake.6 Figure 2a shows the spatial distri-
bution of the percentage distribution of affected households.7 The affected municipalities
were concentrated on the western coast (uchi-bo), especially in the counties of Awa and
Kimitsu, because this area includes the seismic fault plane named the Kamogawa-teichi
5While Tokyo was the largest prefecture with approximately four million inhabitants in 1922, Kana-gawa and Chiba were middle-sized prefectures with 1.36 and 1.34 million inhabitants, respectively (Statis-tics Bureau of the Cabinet 1924b, p. 347). The shares of the agricultural, industrial, and commercialsectors in Chiba at that time were 70%, 10%, and 10%, respectively (Statistics Bureau of the Cabinet1924a, pp. 26–27).
6At that time, most houses in Japan were typically built with wood. However, even the concretebuildings of the municipal offices were destroyed in highly impacted areas (Division of Social Affairs,Chiba Prefecture 1933a, pp. 88–96). Tokyo City Office (1925 p. 161) reported that 86.5% of householdsin Kanagawa were affected and almost half of those in Tokyo. This included being burnt, destroyed, orwater-damaged by the earthquake and subsequent fires and tsunami.
7The spatial distribution of victims is similar but shows more regionally smaller distribution patterns.Thus, I prefer to use the percentage distribution of affected households in my empirical analysis (seeFigure A.2 in Online Appendix A).
4
fault zone (Takemura 2003).8 Overall, 53% of municipalities suffered physical disruption.
In affected areas, roads and railways were largely destroyed and the production of
newspapers, postal services, and telegraph and telephone services also completely stopped.
Topographical changes that made traveling difficult were observed in many municipalities.
Although railway services had largely restarted by the end of September 1923, passengers
still had to walk between heavily damaged sections.9 Further, roughly nine in ten post
offices including telegraph and telephone stations were damaged.10 Wells, the main water
source for people at that time, became contaminated by sand and salt, causing difficulties
in obtaining drinking water. Physical disruption was also observed in the agricultural and
industrial sectors.11 Because approximately 14,000 hectares of arable land were affected,
including damage to reservoirs, embankments, and irrigation equipment as well as the
upheaval or depression of ground surfaces, agricultural households struggled to sell their
products because trade partners were mainly in Kanagawa and Tokyo. The industrial
sector was affected not only by the physical damage but also by the crisis in the financial
system, as approximately 77% of banks in Tokyo were burnt down by the fire.
2.2 Possible Paths
Earthquakes lead to adverse health shocks to a fetus through two main paths. The first
path is maternal mental stress in pregnant women (Hibino et al. 2009). Prenatal maternal
stress, especially posttraumatic stress disorder, increases the risk of adverse pregnancy
outcomes (Yonkers et al. 2014). Glynn et al. (2001), for instance, investigate 40 pregnant
women who experienced an earthquake of a magnitude of 6.8 that occurred in California in
1994 during pregnancy or shortly after, finding that maternal stress experienced in early
pregnancy is associated with a shorter gestational period. Torche (2011) also investigates
the influence of acute stress exposure to the large Chilean earthquake of 2005 on birth
weight using birth registry data. She shows that maternal stress results in a decline
in birth weight and an increase in the proportion of low birth weight deliveries. Kim,
Carruthers, and Harris (2017) also provide evidence that psychological maternal stress
8Indeed, pictures taken in the aftermath of the earthquake show the unimaginable scale of devastationin Awa (Online Appendix B).
9Railway transportation was well developed by the early 20th century in Japan (Tang 2014; 2017).10See Division of Social Affairs, Chiba Prefecture (1933a, p. 151) and Chiba Prefecture (1925c, vol. 5,
p. 152).11Descriptions of the agricultural sectors are taken from the Division of Social Affairs, Chiba Prefecture
(1933a, pp. 126–130, p. 133, 136, pp. 141–143) and Chiba Prefecture (1924, p. 11, 103). The number ofcollapsed banks was obtained from the Bank of Japan (1986, p. 48).
5
from exposure to the Northridge earthquake of 1994 increased the likelihood of low birth
weight. Such low birth weight due to reduced gestational age and intrauterine growth
restriction can have adverse effects on the development of children (Victora et al. 2008;
Datta Gupta, Deding, and Lausten 2013).12
The second path is maternal nutritional deprivation, which can be driven by several
channels (Barker 1992; 1998). The first direct channel is either external injury or a
shortage of food because of shocks to transportation (Division of Social Affairs, Chiba
Prefecture 1933b). The second channel is indirect pecuniary shocks (Banerjee et al. 2010;
Bozzoli and Quintana-Domeque 2014). Declines in household income were caused not
only by shocks to agricultural production and/or sales but also by the destruction of
banks as described. Despite the short-run impacts, the third indirect channel is the
increased price of food and daily commodities that could decrease the relative wealth
of households (Hunter and Ogasawara 2019).13 The fourth channel is declining sanitary
conditions. Infection can significantly reduce fetus nutrition via inflammation, high fever,
lost appetite, vomiting, and compilations (Metzger et al. 1982; Tomkins et al. 1994).14
3 Empirical Analysis
3.1 Data
To estimate the impacts of fetal earthquake exposure on children’s health, I first assem-
bled school-level datasets of height and weight, given that child stunting is the best overall
indicator of the well-being of children (de Onis and Branca 2016). Height is the main
measurement of the overall health outcome of children, as this measure reflects accu-
mulated nutritional status and is associated with cognitive ability and long-term adult
health and socioeconomic outcomes (Fogel 1994; Case and Paxson 2008; Currie and Vogl
2013).15 Weight is used as a secondary measurement of children’s health, even though it
12See Stein et al. (2014) for a comprehensive summary of the association between parental disordersand offspring outcomes. See also Au Yeung et al. (2016) for a discussion on the association between lowbirth weight and the risk of disease.
13Although the government intervened to stabilize markets by issuing the Emergency Requisitioningand Antiprofiteering Ordinance within a week of the earthquake to deal with panic buying and rapidlyrising commodities prices, the retail prices of food and other necessities still increased after the earthquake(Division of Social Affairs, Chiba Prefecture, 1933a, pp. 293–296).
14A fifth possible channel would be the work burden caused by the reconstruction of cities and towns.However, this kind of heavy labor was more likely to be conducted by men than women.
15I do not use the height-for-age z-score of modern WHO standards to control for the age effects becausethe pubertal growth spurt of children in the early 20th century could have occurred at slightly older agesthan in modern healthy children. This would lead to a distorted height-for-age profile for my sampled
6
might be sensitive to instantaneous effects.16 In light of these biological outcomes, Japan
is a good context for my study because of its comprehensive school physical examination
records. Since these physical examinations had to be conducted in April of each year
for all schools under the Gakuseiseito shintaikensa kitei (Official Regulations for School
Physical Examination) from 1897, most schoolchildren undertook one.
I collected an available set of the annual statistical reports named Gakko seitojido
shintaikensa tokei (Statistics of Physical Examinations for Students; SPES) compiled
by Chiba prefecture to construct primary school-year-age-level multidimensional panel
datasets. The primary school dataset includes children aged 6–11 measured between 1925
and 1935 (i.e., born between 1914 and 1929).17 My dataset on the primary school students
from all 434 schools covers approximately 95% of the juvenile population aged 6–11 in
Chiba at that time. Thus, the target population in the analyses using the dataset can be
regarded as almost the entire child population in the same age range in the prefecture.
I systematically divide my primary school samples into two groups to investigate the
heterogeneous stunting effects; 6–8 (early primary) and 9–11 (late primary). Panel A of
Table 1 reports the summary statistics on the biological outcomes.18 The average height
and weight of both exposed and unexposed cohorts seem to be very similar, implying the
importance of observing potential regional heterogeneity in the impacts of the earthquake.
In fact, Figure B.1 in Online Appendix B, which illustrates average height and weight by
cohort, gender, and area, with 95% confidence intervals, suggests that the stunting effects
may be concentrated in the physically devastated areas.
I used data on the physical damage from the official report for the Great Kanto Earth-
quake named Taisho shinsaishi (History of the Taisho Earthquake; HTE) published by
the Social Welfare Bureau of the Cabinet in 1926. Since the HTE surveyed all damaged
children (see Schneider 2019).16Moreover, the body mass index is not used herein because it is designed to capture the degree of
adult obesity. Child growth disturbs the measurement of obesity at different ages. This issue makes itdifficult to identify whether the observed child stunting comes from fetal shocks or just the timing ofchild growth. See Schneider (2019) for a detailed explanation of this mechanism.
17Since the academic term runs from April to March in Japan, children in the first grade of primaryschool are aged 6 and 7 and those in the final grade are aged 11 and 12. To ensure the consistency ofthe data structure, however, I refer to the range of ages in my primary students sample as 6–11 yearsthroughout this paper. The systematic data on the secondary schools are unfortunately unavailable. TheSPES indicates that the number of secondary schools in the physically devastated areas was indeed toosmall to conduct any statistical inferences.
18I excluded some observations because of outliers, missing values, and consolidations of the municipal-ities. Consequently, 99% of my school-year panels are balanced. I confirm that trimming my sample tokeep a balanced panel structure produces results virtually identical to those presented in the main resultsin Table 3 (not reported). This means that the particular subsets of my school panels do not disturb mymain results.
7
households in Chiba by November 15, 1923, it provides a complete picture of the degree of
physical disruption at the municipality level (Division of Social Affairs, Chiba Prefecture,
1933a, pp. 412–420).19 I calculated the physical disruption rate (PDR) as number of
affected (collapsed and semi-collapsed) households per 100 households for each munici-
pality, as shown in Figure 2a.20 However, the distribution of the PDR is highly skewed, as
in the case of the effect of the Chernobyl radioactive fallout in Sweden (Almond, Edlund,
and Palme 2009). Hence, to relax the functional form assumption, I begin by systemati-
cally dividing municipalities into three categories according to the Japan Meteorological
Agency seismic intensity scale (JMA-SIS): JMA-SIS of 5, 6, and 7.21 Table 2 summarizes
the classification. The mean PDRs reported in this table suggest that the physical dis-
ruption was concentrated in the JMA-SIS7 area with a mean rate of approximately 80%.
A large number of affected municipalities were in the Awa and Kimitsu counties. While
the municipalities in the SIS6 area were moderately affected (approximately 23%), the
municipalities in the JMA-SIS5 area were hardly physically damaged by the earthquake
(approximately 0.4%). This distribution of damage suggests that the negative effects of
the earthquake were obvious in the JMA-SIS7 area. Figure 2b illustrates these regions.
3.2 Identification Strategy
I use the quasi-experimental estimation strategy that matches the exogenous shocks due to
fetal earthquake exposure with the corresponding birth cohorts. If the physical disruption
enhanced the earthquake stress placed on the fetus, a higher degree of disruption may
be associated with stronger stunting effects. I thus include the product terms between
19Since more than two months had passed after the earthquake, which occurred on September 1,the number of affected households cumulatively documented in the report should be accurate. Oneobservation (Nakagawa village in Kimitsu county) showed an unrealistically large value in the number ofaffected households. This was considered to be a misprint and was replaced with the figure based on thesurvey conducted by the prefecture on October 3, 1923 reported in the Division of Social Affairs, ChibaPrefecture (1933b, pp. 2–3).
20The number of households in each municipality is based on the 1920 Population Census. Accordingto the 1925 Population Census, the number of households increased from 259,026 in 1920 to 270,796 in1925 (Statistics Bureau of the Cabinet 1926). Assuming this increasing trend, total households around1923 might have increased by roughly 5,900, accounting for 17 households more per municipality from1920. This figure accounts for a 2% increase per municipality based on 1920 values. Thus, the data fromthe 1920 Population Census are considered to be plausible to use as a proxy for the number of householdsin municipalities around the time of the earthquake hitting.
21To calculate the JMA-SIS, I first predicted the collapse rate using the prediction equation proposedby Moroi and Takemura (1999) based on data on collapsed wooden houses from the Hyogoken-NanbuEarthquake of 1995: Collapse Rate = −1.61 + 0.46 × PDR + 0.0051 × PDR2, where the PDR is the physicaldisruption rate defined earlier. Then, the predicted collapse rate is converted into seven categories of theJMA-SIS based on Takemura and Moroi (2002): the rate greater than 30% is defined as 7 (shindo nana),greater than 1% and less than 30% is defined as 6 (shindo roku), greater than 0.1% and less than 1% isdefined as upper 5 (shindo go).
8
the 1923 and 1924 birth cohort indicator variables and indicator variable for the JMA-
SIS7 area to allow the effects of earthquake stress to vary across areas.22 My baseline
specification is given as follows:
ysta = α+ β0I (YOB=1923)ta + β1I (YOB=1923)ta × SIS7 gs + β2I (YOB=1924)ta
+β3I (YOB=1924)ta × SIS7 gs + x′gcsta
γ + µsa + λt + esta(1)
where s indexes schools, t indexes survey years, and a indexes ages.23 The variable y is
either height or weight, I(·) is an indicator variable that equals one for children born in
1923 or 1924, SIS7 is an indicator variable for the JMA-SIS7 area, x is a vector of the
county birth-year-level control variables, µ is a school age-specific fixed effect, λ is a year
fixed effect, and e is a random error term.24
Since the earthquake hit on September 1, 1923, the physical and human loss mostly
occurred in that month.25 This timing suggests that children born between September
1923 and July 1924 experienced the earthquake in utero. This means that the 1923 birth
cohort includes those children impacted by the earthquake in utero because children born
in 1923 in my sample were individuals born between April 1923 and March 1924 as the
academic year starts in April and ends in March in Japan.26 In this vein, the 1924 birth
cohort also includes children born between April and July 1924 who experienced the
earthquake at early gestation. Thus, I also included the 1924 birth cohort variables in
equation (1). Since the proportion of affected children in the 1924 birth cohort was smaller
than that in the 1923 birth cohort, however, I expect that the adverse health effects should
be clearer in the 1923 birth cohort than in the 1924 birth cohort. Considering the feature
of exposure, I expect the estimated coefficients on the affected cohort indicators I(·) and
their area interaction terms I (·) × SIS7 to be negative.27
22I confirm that the effects of fetal earthquake exposure in the JMA-SIS6 area are identical to those inthe JMA-SIS5 area in the statistical sense. Online Appendix C.2.1 presents and discusses these results.
23The group variables gs and gcs indicate municipalities and counties, respectively. Although I use thenotation ta to indicates the year of birth for the sake of simplicity, it is identical to t− a. For instance,the county birth-year-level controls can be expressed as xgc
staor xgc
s,t−a.24Although earthquake severity is measured at the municipal level, I use data on the biological outcomes
at the school-year-age level to control for the unobserved factors of each school using the bilateral-specificfixed effect and exploit the full information around the JMA-SIS7 area. Despite this design, I find thatthe main results in Table 3 are mostly unchanged if I use the municipal-year-age-level datasets (notreported).
25While the share of extrinsic deaths in total deaths is normally around 1%, this share dramaticallyincreased to 27% in that month (Online Appendix A).
26The average pregnancy term was nine to ten months at that time in Japan (Tokyo City Office, 1926).Hereafter, I simply refer to the 1923 birth cohort instead of the 1923 academic year birth cohort forsimplicity.
27Drixler (2016) highlights the imprecision of birth data in prewar Japan. In light of this study, one
9
The timing and spatial distribution of the physical disruption must be exogenous to
improve the identification. While the timing of the earthquake was obviously unexpected,
the distribution of the disruption used was more likely to be dominated by the distribution
of the fault plane found after the earthquake, suggesting that the distribution is plausibly
exogenous.28 I also allow the unobserved factor of school to vary over age by introducing
a fixed effect (µ) in each school-age cell in all the specifications (Davis 2002).29 The ad-
vantage of this approach is that the school-specific unobservable preference on improving
children’s health can be controlled for.30 I further use the county birth-year-level vari-
ables at baseline, namely the fetal death rate,31 rice yield, coverage of doctors, and school
enrollment rate of the parental generation.32 These variables are included to control for
the potential mortality selection effects, wealth levels related to agricultural productivity,
accessibility to medical care around birth, and changes in parental characteristics across
cohorts, respectively (Bozzoli, Deaton, and Quintana-Domeque 2009; Brown and Thomas
2018).
To assess the potential spatial and school cohort-specific correlations, I intend to clus-
ter the standard errors at the 13-county level. For the statistical inference, however, I
adopt the wild cluster bootstrap-t method to deal with the issue of the small number of
must be careful about the potential age heaping in April. For example, children born in March 1923 mighthave been disadvantaged in their development, perhaps during their early primary school ages, comparedwith those born in April 1922. This implies that parents might have had an incentive to register theirchildren born in March 1923 as born in April 1923. However, I confirm that there was no such systematicage heaping in April using vital statistics (Statistics Bureau of the Cabinet 1925). In addition, while the1923 birth cohort includes children born after the earthquake, I confirm that postnatal exposure had nostatistically significant stunting effects on the affected birth cohorts (see the “Robustness” section).
28I also confirm that the distribution of the physical disruption did not depend on that of soil compactionmeasured as the spectral intensity values observed in 2011, suggesting that my key measure of earthquakestress does not depend on the potential agricultural productivity that might be correlated with children’sbiological standards. The map of soil compaction is based on boring data from 50,000 monitoring pointsand thus the spectral intensity values are measured in 250-m meshes. This means that the map couldshow nearly the same distribution of soil compaction in the early 20th century, even though the data areinvestigated in 2011. See Online Appendix B for the details.
29Since I introduce the fixed effect for each school-age cell, the identification depends on the withinvariation over the measured year. This means that the increasing trends in the growth of child heightshould be similar over school-age cells. I confirm that the growth patterns of the sampled children overthe measured years are similar (Online Appendix C.1).
30In addition, my approach can efficiently control for unobservable instantaneous shocks using the yearfixed effect (λ). Another advantage of including the year fixed effect is that the unobservable trends inpotential wealth and public health can be captured.
31The fetal death rate is the number of still births per 1,000 births. Data on fetal deaths as well as livebirths between 1913–1930 are taken from the SRCP and Vital Statistics for Municipalities (Online Ap-pendix B). I linearly interpolated the missing values for 1922–1924 and 1926–1929 using the values of1921, 1925, and 1930 given that the time-series plots of the rates in Chiba prefecture at that time werenearly linear (Figure C.2a in Online Appendix C.2.5).
32The rice yield is the volume of rice yield per hectare; coverage of doctors is the number of doctorsper 100 people; and the school enrollment rate of the parental generation is the 17-year lagged primaryschool enrollment rate from the year of birth. See Online Appendix B for the summary statistics anddata sources of these control variables.
10
clusters in the cluster-robust variance estimator (Cameron et al. 2008).33 The regression
equation is estimated separately for boys and girls for each developmental stage to inves-
tigate both gender differences in the effects and the potential catch-up growth against the
shocks (Steckel and Ziebarth 2016).34 The age bins are systematically divided into 6–8
(early primary) and 9–11 (late primary) because the growth pattern of children during
primary school ages were almost linear at that time.35
3.3 Main Results
Table 3 presents the results from my baseline specification denoted in equation (1). Panel
A lists the estimates of β for height. Columns (1)–(2) present the results for the boys,
whereas columns (3)–(4) present the results for the girls. In column (1), primary school
boys aged 6–8 born in 1923 are found to be approximately 0.21 cm shorter than the
surrounding cohorts. Column (2) shows a similar but slightly smaller stunting effects on
the primary school boys aged 9–11. These results imply that the stunting effects of fetal
earthquake exposure are likely to diminish in later years of primary schooling because of
the catch-up growth during primary school age (Frankenberg et al. 2017). The estimated
coefficients on the area interaction terms are statistically insignificant in columns (1)–
(2). When I look at the results for the girls, those stunting effects are greater in their
magnitude. Column (3) indicates that the girls aged 6–8 born in 1923 are 0.23 cm and 0.53
cm shorter in the non-JMA-SIS7 and JMA-SIS7 areas, respectively. Similarly, column (4)
indicates that the girls aged 9–11 born in 1923 are 0.23 cm and 0.84 cm shorter in each
33Therefore, my method controls for the correlation and heteroskedasticity within clusters. Since aschool is nested in a county, this means that both the dependency across observations of the samebirth cohort within a school and the dependency across schools within same county are allowed in myestimation. In addition, my method can deal with potential heteroskedasticity across clusters. I preferto cluster at the county level than the municipal level because regressors grouped at the county levelare used (Moulton 1986, 1990) and the correlation within the same county over time such as the schoolcohort-specific correlation may be problematic (Bertrand, Duflo, and Mullainathan 2004).
34I herein do not prefer to use the regression specification including the interaction terms betweengender and the key control variables because such a specification postulates a stronger assumption thatthe marginal effects of the other control variables uninteracted with gender are the same between boysand girls. As for the analytical weight for the regressions, I am, unfortunately, unable to use the dataon the number of inspected children for each age and school (Deaton 1997). However, since the numberof schools should have been set to reflect the size of municipalities, the number of inspected children ineach age cell should be largely similar across schools (Ministry of Education 1973). Thus, my analyticalresults should be robust to the weighting.
35See Online Appendix C.1 for the observed growth patterns of students. Although one must be carefulto simply apply the modern growth reference to historical child growth, the height-for-age z-sores of −2in the modern WHO growth reference show similar growth curves to those children (de Onis et al. 2007),supporting that this developmental-stage age bin makes theoretical sense. Despite this, we confirm thatour main results are robust to the use of alternative definitions of the developmental-stage age bin (OnlineAppendix C.2.2).
11
area.
Panel B of Table 3 shows the estimates of the effects on weight, following the same
column layout. Although the estimates in column (1) suggest that the primary school
boys born in 1923 are approximately 0.06 kg lighter than those in the surrounding cohorts,
those listed in column (2) are no longer statistically significant. This result is consistent
with the weak stunting effects on the boys reported in Panel A. Similar to the results on
the girls’ height, the affected birth cohorts among the primary school girls are slightly
lighter than those in the surrounding cohorts. Column (3) indicates that the girls aged
6–8 born in 1923 are 0.1 kg and 0.37 kg lighter in the non-JMA-SIS7 and JMA-SIS7 areas,
respectively. Similarly, column (4) shows that the girls aged 9–11 born in 1923 are 0.13
kg and 0.43 kg lighter in each area.
Overall, these results are consistent with the literature on early-life shocks on human
growth (Mazumder et al. 2010; Rosales-Rueda and Triyana 2019). The Great Kanto
Earthquake of 1923 had negative lasting effects on girls’ growth in the damaged area with
a maximum JMA-SIS of 7. I find that, however, while the boys slightly stunted in their
early primary school ages, the stunting effects were wiped out in the late primary school
ages. I return to this gender difference in the stunting effects in greater detail. Finally,
I do not find any meaningful impacts on the 1924 birth cohorts in a statistical sense. As
discussed, while the 1924 birth cohort includes children who experienced the earthquake
in utero during the first trimester, the proportion of those affected children in this birth
cohort is small compared with that in the 1923 birth cohort. This feature of assignment
can attenuate the estimated effects on the 1924 birth cohort.
3.4 Mechanisms
Next, I extend my discussion to the pathways (i.e., maternal mental and nutritional stress)
through which fetal earthquake exposure may affect children’s health. First, I investigate
what type of disaster relief spending mitigated the adverse effects of the earthquake on
the development of children. Second, I use the market disruption due to damage to the
railway transportation network to identify the potential stress behind these adverse health
effects.
12
3.4.1 Relief Effects
Soon after the earthquake, Chiba prefecture decided to provide disaster relief for affected
counties, namely Awa, Kimitsu, Ichihara, Higashikatsushika, Chosei, and Isumi, and ap-
proximately 12% of people received relief in these counties.36 To investigate the potential
compensating effects of the disaster relief, I use statistics on the county-level relief ex-
penses by expense item in the SRCP (Chiba Prefecture 1925a). Of the relief expenditure,
66% was used to provide food, 31% was for temporary housing, and 0.3% was for med-
ical treatment.37 Panel B of Table 1 shows the summary statistics of the relief expense
variables.
To assess the relief effects, I introduce the interaction terms between the measures of
relief expenses and indicators of affected cohorts into equation 1 as follows:
ysta = κ+ δ0I (YOB=1923)ta + δ1I (YOB=1923)ta × SIS7 gs + δ2I (YOB=1923)ta × Relief gcs
+ δ3I (YOB=1924)ta + δ4I (YOB=1924)ta × SIS7 gs + δ5I (YOB=1924)ta × Relief gcs
+ x′gcsta
θ + ηsa + νt + εsta
(2)
where Relief is the county-level per capita relief expenses (total, food, shelter, or medical
treatment). While medical treatment is obviously necessary for injured people, immediate
hunger relief intervention is the principal emergency response to sustain people in a severe
situation after a crisis. Temporary shelter is also an important emergency facility for
personal safety, climate protection, security, and resistance to infectious diseases after
large-scale disasters.38 Therefore, one can expect the estimates of the parameters of
interest (δ2; δ5) to be positive when the disaster relief had the capacity to compensate for
the negative effects of the earthquake. The regression equation is estimated separately
for each gender and age bin to investigate how relief spending ameliorated the negative
health impacts of the earthquake in detail.
36The relief effort particularly focused on Awa and Kimitsu, which accounted for 85% and 11% oftotal relief expenditure, respectively. Although total relief expenditure was 490,837 yen for the impactedyear, this dropped to 2,790 yen in the following year. This means that the relief was considered to bea temporal and one-shot investment for affected counties in 1923. This feature of public spending maysupport evidence that Japan was able to quickly recover from these shocks (Hunter 2014).
37Although the remaining 5% was used for the other miscellaneous goods, I do not use this categoryfor the regression analysis because the purpose of the relief is unclear and thus it is difficult to interpretthe results.
38See Pingali, Alinovi, and Sutton (2005) and the International Federation of Red Cross and RedCrescent Societies (2013). In Yorho village, for instance, 120 out of 600 households collapsed and thusthe victims had to spend 10 days living outside (Division of Social Affairs, Chiba Prefecture 1933b,p. 748).
13
Table 4 presents the regressions for testing the effects of the disaster relief.39 Panel
A (Panel B) lists the estimates of the relief effects on height (weight). Columns (1)–(4)
present the results for total, food, shelter, and medical treatment expenses, respectively.
Overall, I find that the disaster relief had compensating effects on child stunting to a
certain degree. Column (1) of Panel A shows that while the relief ameliorates the stunting
effects on the primary school boys aged 6–8 born in 1923, it does not have such effects
on the girls’ stunting. Furthermore, column (1) of Panel B indicates that the relief also
ameliorates the weight loss of the primary school boys aged 6–8, while such effects are
not observed for the girls. These results are consistent with the finding from Table 3:
while the boys’ stunting may have been improved throughout the primary school ages,
girls’ stunting was persistent. Comparing the estimates between the boys aged 6–8 and
those aged 9–11 indicates that the relief effects are statistically significantly positive only
at early primary school ages. This difference can be explained by the fact that the growth
rate of the boys becomes higher at late primary school ages than early ages (de Onis et
al. 2007).
To investigate what type of disaster relief can ameliorate the adverse effects of the
earthquake on children, I decompose the relief expenses into three subcategories in columns
(2)–(4). While relief expenses for food, shelter, and medical treatment had positive ef-
fects on both height and weight, the estimated effects are the largest in medical treatment,
followed in order by shelter and food (41.5, 0.38, and 0.17 in the first row in Panel A).
This result seems to be reasonable given the fact that medical treatment was intensively
provided for injured people in devastated areas, whereas relief interventions for food and
shelter were provided more broadly. I next discuss the magnitudes of these relief effects
estimated on the height of the primary school boys. The estimate in column (1a) of Panel
A suggests that a one standard deviation increase in disaster relief (Table 1) might have
increased the affected primary school boys’ height by roughly 0.1 cm at ages 6–8 on aver-
age. Given that the estimated stunting effect on the early primary school boys is 0.2 cm
as discussed, this magnitude implies that disaster relief might have ameliorated roughly
half of the stunting effects.40
39The estimated coefficients on the cohort dummies and area interaction terms (δ0; δ1; δ3; δ4) in equation2 are not reported in the same table because these estimates are largely unchanged from those in Table 3.
40 Similar calculations applied to food (column 2a), shelter (column 3a), and medical treatment (column4a) indicate that the potential compensating effect is roughly 0.1 cm for each expense. These magnitudesare considered to be reasonable. Average relief expenditure on food for treated people in Chiba prefecturewas 3.8 yen, whereas average expenditure for food in peasant households in September 1926 was 7.7 yenper family member. Although this is the most conservative calculation, it suggests that the relief mighthave provided nearly half of the monthly food needed. In this calculation, the number of family members
14
The foregoing results suggest that disaster relief could mitigate child stunting during
primary school ages by remediating adverse shocks in early life. A few points on the
heterogeneity of relief effects do, however, need to be discussed.
First, while my result suggests that the disaster relief might have played a role in nour-
ishing the primary school boys to a certain extent, I find that it had little compensatory
effect on the primary school girls. I discuss this gender bias in relief effects from the timing
dimension: prenatal or postnatal remediation. One plausible explanation of prenatal re-
mediation is the growing body of the literature supporting evidence that male fetuses are
more vulnerable to ambient stress in utero than female fetuses and thus positive selection
into birth is larger for boys than girls (Kraemer 2000). Given the selection effects on the
boys, my results imply that the disaster relief might have further mitigated the stunting
effects by reducing the “scarring” effects on the boys (Valente 2015).41 Another potential
explanation of postnatal remediation is the strong son preference in the family system (ie
seido) in prewar Japan. The prewar Civil Code regarded the first-born son as the next
head of the family (koshu) who had the right to inherit properties and dictatorial power
to allocate household resources (Ramseyer 1996; Hayashi and Prescott 2008).42 Conse-
quently, a gender bias against girls in intrahousehold resource allocations, especially in
educational attainment, was widely observed in prewar Japan (Hijikata 1994).43 From
this perspective, one must be careful about the fact that the 1923 birth cohorts include
children born after the earthquake hit but before (or during) the distribution of the relief.
Since a large part of the disaster relief may have been distributed in and after October
1923, children born in September might have benefitted from the relief in their postnatal
period (Division of Social Affairs, Chiba Prefecture 1933a, p. 420). Thus, the likelihood
of adverse health effects for girls might have been high given the institutions and customs
includes both adults and infants. These data are taken from the agricultural household surveys conductedbetween September 1926 and August 1927. Although the details of the sampling method are unavailable,the number of surveyed households was 670 and those households were sampled from nine prefecturesacross the Japanese archipelago: Yamagata, Saitama, Niigata, Nagano, Aichi, Hyogo, Hiroshima, Ehime,and Fukuoka. See Statistical Bureau of the Cabinet (1929 p. 420).
41The scarring effect indicates a shift of fetal health distribution to the left. Although the improvingeffects on the scarring effects may also be applicable for female fetuses, male fetuses should be moresensitive to nourishing than girls because they are more susceptible than female fetuses. In other words,if I consider the cumulative distribution function of fetal health endowments, male fetuses are more likelyto be distributed on the lower side of the distribution function than female fetuses (see Valente 2015).
42In fact, Article 970 of the Civil Code prioritized the of wedlock boy than the legitimate girl (withinthe same degree of relationship) in inheritance if the boy is recognized by his father (Cabinet OfficialGazette Bureau 1898, pp. 148–149).
43The most extreme case of biased resource allocation is that daughters were often sold by their parents(miuri) in depression periods given the higher return of boys in the labor market (Nakamura 1994,pp. 115–116).
15
in family systems favored boys. Second, neither the immediate hunger relief and shelter
protection nor medical treatment had ameliorating effects on the 1924 birth cohorts. A
potential explanation of this result is, as discussed, the attenuation effects due to the
smaller proportion of the affected children in the 1924 birth cohort than in the 1923 birth
cohort.
My finding on disaster relief for the Great Kanto Earthquake is consistent with the
recent economic history study by Vellore (2018) that reveals that childhood remediation
through New Deal-related spending can mitigate the adverse effects of Dust Bowl exposure
on socioeconomic outcomes and disability in adulthood. Since I use county-level relief
expenses as the relief variable, the estimated effects may underestimate the actual effects
of the relief. Despite this, the present study provides suggestive evidence that remediation
effects through disaster relief can mitigate stunting effects by the end of primary school
ages. While the magnitude of relief effects suggests that medical treatment has greater
remediation effects than food and shelter, the expenses for the latter items might have
been effective in terms of coverage. However, relief effects might be biased by gender
depending on the timing of remediation.
3.4.2 Mental and Nutritional Stress
The disaster relief results suggest that the ameliorating effects of relief might stem from
reducing the risk of maternal nutritional deprivation after the crisis. In this subsection, I
test whether the effects of mental and nutritional stress are more remarkable in the area
that experienced little physical disruption and received no disaster relief.
Mental stress is caused not only by the experience of strong vibrations and a stressful
life in the aftermath of the earthquake but also the physical disruption, whereas nutri-
tional stress is concentrated in physically damaged and market disrupted areas (Harada
et al. 2015). Considering this feature, I begin my analysis by trimming the sample as fol-
lows. First, I exclude all municipalities in the JMA-SIS6 and JMA-SIS7 areas and leave
the JMA-SIS5 area alone because it is difficult to disentangle mental and nutritional stress
in those areas that experienced physical disruption. Second, I drop municipalities that
received any disaster relief to eliminate its compensating effects. The sample reduces to
222 municipalities that include exposed children who experienced strong vibrations in
utero in the limited area with a JMA-SIS of 5 but received no disaster relief.44
44The JMA-SIS ranges from 1 to 7, meaning that a JMA-SIS of 5 is still a very strong vibration.At a scale of 5, people are frightened and feel the need to hold onto something stable. See the Japan
16
For the identification, I sort the locations where suffered market disruption due to dam-
age to railway transportation networks. Regarding railway transport in Chiba prefecture,
there were considerable losses of freight arrivals on the Noda line in Higashikatsushika
county and Kururi line in Kimitsu county.45 In fact, the annual tonnages of freight ar-
rivals on both lines dropped by more than half of those in previous years (Chiba Prefecture
1924, pp. 125–126). Since the main freight items were food, wood, and fuels, this suggests
that the regional markets near both lines were more likely to have been disrupted than
other areas and thus the alimentary deficiency might have occurred after the earthquake
hit (Chiba Prefecture 1925c, pp. 138–145). While railway transport suffered physical
damage, however, marine transport was not impacted by the earthquake. Indeed, Choshi
port, a famous export port for the fish industry located at the western end of the pre-
fecture, restarted to export goods to Tokyo soon after the earthquake (Chiba Prefecture
1924, pp. 150–151).46
Exploiting these facts, I test the pathways by estimating the following regression:
ysta = ϕ+ ζ0I (YOB=1923)ta + ζ1I (YOB=1923)ta × Railgs + ζ2I (YOB=1924)ta
+ζ3I (YOB=1924)ta × Railgs + x′gcsta
φ + ιsa + τt + εsta(3)
where Rail is an indicator variable for municipalities within 10 km of any municipality
including either the Noda line or the Kururi line.47 The estimates (ζ0; ζ2) are expected
to be statistically significantly negative when maternal mental stress to vibrations were
associated with adverse health effects. In addition, the estimates of the coefficients on the
area interaction terms (ζ1; ζ3) are statistically significantly negative when maternal nu-
tritional stress due to disrupted transportation networks had adverse effects on children’s
health. The combination of the estimates provides evidence of which pathway might have
played an important role in disturbing children’s growth in the limited JMA-SIS5 area.
Meteorological Agency (website).45Both lines served several populated provincial towns and villages such as Noda town, Funabashi town,
Chiyoda village, Kisarazu town, Makuta village, Obitsu village, and Kururi town (Chiba Prefecture 1924,pp. 52–53). As described in the “Great Kanto Earthquake” section, the other lines recovered relativelysoon after the earthquake in Chiba prefecture (Ministry of Railways 1927).
46Moreover, marine transport was not a major transportation method in Chiba at that time. In fact,the annual tonnages handled in the marine transport was much smaller than freight by railway transportin Chiba (Chiba Prefecture 1924, pp. 150–151).
47I set the radius to cover neighboring municipalities: 10 km is roughly the median of the 5th percentileof the distances from the origins (i.e., municipalities including railways) to the other municipalities. Mostmunicipalities within the radius are far from other railway networks, making it difficult for them to accessother regional markets. Given the mean distance to the nearest municipality is approximately 3 km, Icheck the sensitivity of my results by changing the diameter to 3 km range and confirm that my resultsremain similar (Online Appendix C.2.2).
17
The regression equation is estimated separately for each gender and age bin.
Panel A of Table 5 presents the results for height.48 Columns (1)–(2) list the estimates
for the boys, whereas columns (3)–(4) list the estimates for the girls. While the estimated
coefficients on the 1923 birth cohort dummies in columns (1), (3), and (4) are statistically
significantly negative, those on the interaction terms are not statistically significant. In
the same columns in Panel B, I find similar effects for the girls’ weight. While these results
may provide evidence of the role of mental stress, my estimates offer less clear arguments
on the stunting effects of nutritional stress.
One might worry about the proxy measure of market disruption used herein. A mar-
ket survey reported that the retail prices of daily food such as rice, wheat, miso, and
soy source increased after the earthquake in the regional markets of Chiba, which might
have temporarily reduced the nutritional intake of mothers.49 However, despite such mar-
ket responses, rural households might have benefitted from conservable vegetables for
self-consumption to compensate for the temporary market disruption.50 If this kind of
unobservable self-protection behavior by households played a role, my estimates would
understate the adverse effects of the disruption. While I acknowledge these potential is-
sues, the result obtained herein is consistent with the growing body of the medical and
economic literature that shows the maternal mental stress caused by the earthquake is as-
sociated with the adverse health outcomes at birth (Yonkers et al. 2014; Kim, Carruthers,
and Harris 2017).
3.5 Robustness
Thus far, I have documented the adverse effects of fetal earthquake exposure on children’s
health. Before discussing the findings of the present study, I conduct a set of exercises
to test the sensitivity of the main results in terms of the following concerns: sorting
issues owing to internal migrations; selection issues because of absenteeism, mortality,
48The estimates of ζ2 and ζ3 are statistically insignificant in all the regressions. To show the resultsconcisely, I do not report these estimates in Table 5.
49Systematic statistics on retail prices around the time of the earthquake are unavailable (Hunter andOgasawara 2019). Although my data on retail prices are also insufficient for use in the regression analyses,the retail prices in some regional markets on September 20 were reported (Division of Social Affairs, ChibaPrefecture 1933a, pp. 293–296). For instance, in the Matsudo market in Higashikatsushika, the maximumprice of rice involuntarily increased from 39 to 45 sen per sho (1.8 liter) after the earthquake hit. Anotherexample shows that in the Kururi market in Kimitsu county, the maximum price of rice increased from36 to 41 sen per sho and the maximum price of miso rose from 80 to 85 sen per kan (3.75 kg).
50Moreover, in the initial stage after the earthquake, there might have been a greater supply of foodin rural areas because of the damaged transport links in Tokyo and Kanagawa prefectures, as describedin the “Great Kanto Earthquake” section.
18
and fertility; the potential impacts of other historical events; and postnatal exposure to
the earthquake. I provide evidence that these concerns do not confound my findings.
Sorting Issues
The nature of the school-level data makes it difficult to capture the sorting of children
due to internal migration across municipalities. If the children exposed the earthquake
in utero in the devastated area moved to other areas after birth, the estimation would
suffer from attenuation bias. In this sense, my estimates of the adverse health effects are
a lower bound of the average treatment effects of fetal earthquake exposure. Therefore, to
understand the potential influence of the internal migration of schoolchildren, I compute
the proportion of children who live in their place of birth to total children at the same
age using the 1930 Population Census (Statistics Bureau of the Cabinet 1931). According
to the census, approximately 94% and 90% of children aged 0–9 and 10–14 were born
in their current municipalities, respectively.51 I regard the figures on children aged 0–9
as plausible for primary schoolchildren because the census figure for children aged 10–14
should be biased downward because of the existence of graduate children aged over 12
who could have jobs in other municipalities.52 This implies that most schoolchildren did
not leave their original places until finishing primary school, making the potential sorting
not a major issue in my analyses. This is indeed consistent with the historical fact that
the internal migration of schoolchildren in the interwar period was limited (Nakagawa
2001, p. 42).53
Selection Issues
First, the potential selection issue might have been arisen from children absent from school
on the date of the physical examination. If the children exposed to the earthquake in utero
have lingering health issues and thus were more likely to be absent from schools for health
reasons compared with other children, my estimates may suffer from downward bias, which
would understate the adverse health effects of fetal earthquake exposure. In addition, if
51The age ranges reported in the census are systematically divided into 0–9 and 10–14 years. Sincethose figures for boys and girls are similar, I present the average figures herein.
52These students were also unlikely to change their primary schools within each municipality (Hijikata1994, pp. 159–165).
53Since the data do not allow me to fully address the potential sorting effects because of the lack ofan appropriate measure of migration, I further test the impacts of potential omitted variable bias byemploying the method proposed by Oster (2019). I find that the suggested bounds of my estimates arenot far from my baseline estimates in Table 3 (Online Appendix C.2.4).
19
the absentees came from poor households and thus were unhealthier than other children,
my estimates would also understate the effects because of systematic positive selection.
Therefore, owing to a similar mechanism to the sorting effects, my estimates present a
lower bound of the true treatment effects of the earthquake. According to the official
reports of Chiba prefecture, however, the absenteeism rates for primary schools were
approximately 0.5% at that time, which is considered to be negligible (Chiba Prefecture
1932, p. 21). As shown in Panel A of Table 6, I therefore confirm that my baseline results
for the primary school children are unchanged if I control for both the municipal-year-level
primary school enrollment rate and the county-year-level primary school attendance rate
as proxies for the proportion of children with health issues.54
Second, the mortality selection is another important selection issue to be discussed.
If unhealthy fetuses and infants were less likely to survive into the sample, my esti-
mates would understate the impacts of the earthquake because survivors should have
been healthier than culled individuals. As discussed, I control for a part of the poten-
tial selection before birth by including the fetal death rate in the regressions. However,
since many unhealthy babies also died within the first 12 months of birth, I further check
whether infant deaths involuntarily increased after the earthquake hit using the time
series plots of infant mortality rates. I find no systematic increasing trend in infant mor-
tality rates after the earthquake in Chiba prefecture (Online Appendix C.2.5). Another
important phenomenon in terms of sample selection may be fertility selection after the
disaster (Dehejia and Lleras-Muney 2004). The fertility rate may involuntarily decrease
during the aftermath of the disaster because some of the population could not afford to
have children in such abnormal circumstances. In other words, children born during the
aftermath of the disaster might be more likely to belong in households that can provide
sufficient compensation for any adverse health effects in the circumstances. Although my
estimates would understate the adverse health effects if such fertility selection behavior
had been common among parents, I do not find any systematic reduction in the crude
birth rate during or after the year in which the earthquake hit (Online Appendix C.2.5).
54Table B.1 in Online Appendix B presents the summary statistics of both rates: across the sam-ple period, the school enrollment rate and school attendance rate were stable around 99% and 95%,respectively.
20
Other Prenatal Shocks and Postnatal Exposure
In Japan, Spanish influenza cases occurred between August 1918 and July 1920 and its
intensity (measured by the number of deaths) spiked in November 1918 and January
1920.55 This feature of pandemics implies that a large proportion of 1919–1920 birth co-
horts in my sample might have been affected by pandemic flu in utero. Another important
event during the sample period that might have influenced children’s health is the First
World War. In Panel B of Table 6, I include two indicator variables for these suspicious
cohorts.56 As shown, the estimated coefficients on the influenza pandemic and wartime
birth cohorts dummies are statistically insignificant in all cases. This result is consistent
with the fact that the influenza epidemics in Chiba prefecture had not been severe during
the epidemics in 1918–1920 (Statistics Bureau of the Cabinet 1921; 1922; 1923). As for
the impacts of the First World War, it is widely accepted in Japanese history that the
war did not change the daily lives of children in Japan and thus did not influence their
growth patterns (Kudo et al., 1976). These findings suggest that there is no significant
difference in these cohorts and reference birth cohorts in this setting.57
The regressions discussed thus far have aimed to capture fetal exposure to the earth-
quake. Evidence of the long-term effects of earthquake exposure suggests that postnatal
exposure, especially by the second year of life, to the great earthquake could also have
adverse effects on human capital accumulation (Caruso and Miller 2015). In Table 7, I
include an indicator of the 1922 birth cohort, which includes children who experienced the
earthquake roughly between 6 months old and 2 years old, in my baseline specification to
test the potential effects of postnatal exposure.58 While I find robust negative effects on
the 1923 birth cohort, I find little evidence of adverse health effects on the 1922 birth co-
55Representative historical studies investigating Spanish influenza in Japan include Rice and Palmer(1993) and Hayami (2006). Although the rice riots of 1918 might also have affected regional food prices,the riots in Chiba prefecture were small and thus had a negligible impact on the regional food prices(Shoji 1957, p. 36).
56 I include a 1919–1920 birth cohort dummy to control for the potential pandemic cohort effects.Regarding the wartime cohort effects, I include an indicator variable for the children in utero in wartimebetween 1915 and 1919 to control for the potential long-term impacts of wartime shocks on healthoutcomes (Lee 2014). When I include these indicators, the coefficients on the 1923 and 1924 birth cohortvariables are all measured relative to the small number of reference cohorts, which loses the advantageof averaging across many reference cohorts. In other words, if the coefficients are estimated relative tofew reference cohorts, one loses the ability to see whether outliers are generated because the earthquakecohort looks unique or because the small set of cohorts serves as a baseline. Given this result, I donot include the indicator variables for the pandemic and wartime cohorts in my baseline specifications(equations 1–3).
57As shown in Online Appendix C.2.3, I find that the estimates in Tables 3–5 are largely unchanged ifI include suspicious cohort indicator variables.
58Children born in January 1922 were exposed to the earthquake at 1 years and 9 months old, whereaschildren born in March 1923 were exposed at 7 months old.
21
hort, implying that postnatal exposure to the earthquake might not matter. This finding
is consistent with the evidence of the recent study by Rosales-Rueda and Triyana (2019),
who show that in-utero exposure to the Indonesian forest fires of 1997 had stunting effects
on children aged 10 and 17 years, while postnatal exposure at 2 years did not have such
persistent stunting effects. As discussed, children born between April and August 1923
included in the 1923 birth cohort experienced postnatal earthquake exposure in infancy.
These data may thus complicate the interpretation of the estimates of the 1923 birth
cohort variables, which could include potential negative effects via the postnatal health
shock. However, my results on the negligible influence of postnatal exposure support
the evidence that the estimated effects on the 1923 birth cohort variables in my baseline
results should capture the adverse effects of prenatal exposure to the earthquake rather
than postnatal exposure.
In Table 7, I also include an indicator of the 1925 birth cohort to run a placebo test.
Since the 1925 birth cohort was not exposed to the earthquake of 1923, the estimated
coefficient on this indicator variable should be statistically insignificant. If the estimates
were statistically significantly negative, the observed stunting effects could be systematic
decreasing trends in the height or weight of children. As shown in Table 7, the estimates
are all statistically insignificant, indicating that the estimated stunting effects do not
capture pre-trends in height and weight.59
4 Discussion
As discussed thus far, the Great Kanto Earthquake had negative consequences for chil-
dren’s health. I find that while both primary school boys and girls were stunted by fetal
earthquake exposure, the stunting effects were much clearer in girls than boys. My finding
on gender bias against girls is consistent with the study by Caruso and Miller (2015) that
finds stronger negative long-term effects of fetal exposure to the Ancash earthquake of
1970 on girls’ years of education. It is also consistent with medical evidence on the culling
mechanism before birth: since male fetuses are more vulnerable in utero than female
fetuses, there may be stronger positive selection into birth for boys than girls (Kraemer
59 If the estimated stunting effects do not capture such trends and can be regarded as the consequencesof fetal earthquake exposure, the mean heights and weights of children born in the surrounding yearsshould be similar after excluding the 1923–24 birth cohorts from the sample. In Online Appendix C.2.6,I confirm that the mean heights and weights of children born between 1918 and 1925 (excluding the1923–1924 birth cohorts) are indeed similar in the statistical sense using the specification including eachsurrounding birth cohort dummy.
22
2000). The ameliorating effects of the disaster relief on primary school boys can also help
explain these gender differences.
I find that the primary school girls aged 9–11 exposed to the earthquake in utero in
the JMA-SIS7 area were roughly 1 cm shorter than those in surrounding cohorts. While
this decline in height accounts for approximately a 0.2 standard deviation change (1/4.6)
in the average height of the primary school girls aged 9–11 (Panel A of Table 1), one must
be careful not to attach too much importance to the small stunting effects. Mazumder
et al.(2010) found that the U.S. birth cohort exposed to the 1919 pandemic influenza
experienced a 0.1 cm decline in final height on average, which is associated with a 5%
higher risk of cardiovascular disease in old age. Although they did not ascertain the causal
links between such stunting and the later-life risk of contracting diseases, a number of
medical and epidemiological studies including randomized intervention trials have found
similar associations between shorter stature in early life and suboptimal function later in
life (Dewey and Begum 2011). Furthermore, a growing body of the literature provides
further evidence on the links between child stunting and lower earnings later in life.
The comprehensive review by McGovern et al. (2017) finds that 0.5 cm child stunting is
associated with a decrease in wages of 2–3% on average. A recent height premium-based
calculation also suggests that almost 2 cm lower height translates into 1.6% lower earnings
for the affected population (Rosales-Rueda and Triyana 2019). While it is difficult to
apply these figures to the present case of prewar Japan since the context, treatment,
and economic environments are different, one must be careful about the possibility that
child stunting has lasting effects on later-life earnings. Another important fact is that the
medical evidence states that girls’ stunting has a risk of lasting adverse health impacts on
their offspring. A growing body of evidence has shown that maternal stunting restricts
the growth of the uterus and placenta and thus increases the risk of intrauterine growth
restriction, which causes insufficient neurological and intellectual development and the
shorter stature of their infants (Black et al. 2008). However, while my results have an
implication for the dynamic adverse health effects of fetal earthquake exposure, it is not
possible to assess these effects without access to data on both mothers and their children.
The long-run impact of maternal stunting on the health and socioeconomic outcomes of
their offspring is thus another avenue for future research.
This study provides suggestive evidence that the disaster relief might have played a
role in nourishing exposed primary school boys. As discussed, the gender bias against
23
girls in relief effects can be explained both by the reduction in scarring effects before
birth and by the potential institutional effects after birth via the prewar Civil Code (e.g.,
Hayashi and Prescott 2008; Valente 2015). The estimates suggest that relief expenses on
food and shelter might have ameliorated roughly half the stunting effects at early primary
school ages. It is also suggested that medical treatment had much greater ameliorating
effects on stunting, even though a limited proportion of people in the devastated area
benefitted from such treatment. These results thus add evidence to the recent literature
on the optimal timing of the remediation of early disadvantage (Heckman 2012; Vellore
2018). Specifically, they show that remediation effects may be able to mitigate stunting
effects by the end of primary school ages. In addition, they highlight the heterogeneity
in remediation effects with respect to types of disaster relief. The results are also in
line with those of Parman (2015), who shows that parents with a child exposed to the
1918 influenza pandemic in utero reallocated their resources to the child’s older siblings,
which led to higher educational attainment for these siblings. My results show that relief
effects can be biased by gender depending not only on the timing of remediation, but
also on parents’ resource allocation based on the strong son preference regulated by the
institution.
Given the unavoidable mental stress of the enormous exogenous shock, the role of
disaster relief indicates that nutritional stress is a plausible pathway of the adverse health
effects in the devastated area. In addition, among municipalities that experienced little
physical damage and disaster relief, I find suggestive evidence that maternal mental stress
due to vibrations might be more likely to be associated with stunting than nutritional
stress due to market disruption. That the earthquake can lead to maternal mental stress
is consistent with previous studies such as Kim, Carruthers, and Harris (2017). My results
further shed light on the role of the different pathways behind the adverse health effects
of fetal exposure to the earthquake. That is, while mental stress is an important pathway
underlying the adverse health effects, nutritional stress can be another pathway in the
physically devastated area, thus enhancing the adverse health effects on exposed children.
5 Conclusion
This study used a catastrophic earthquake from the 1920s to analyze the long-term effects
of a one-off disaster on children’s health. I found that fetal exposure to the Great Kanto
24
Earthquake had stunting effects on primary schoolchildren and that the magnitude of
such effects increased with the degree of earthquake stress. The disaster relief was found
to have compensating effects on stunting among boys by late primary school ages, even
though the magnitude of the effects varied with the types of relief expenses. Finally, while
mental stress might be associated with adverse health effects, nutritional stress might also
have enhanced the effects in the physically devastated area.
My evidence from industrializing Japan is, however, not without its limitations. First,
given the scarcity of individual-level data with the date and place of births in Japan, I
predominantly used school year-age-level datasets. Since this assignment can attenuate
the estimates, those obtained in this study should be considered to be the lower bounds
of the average treatment effects.60 Second, I used data that cover one prefecture in the
Kanto region, Chiba prefecture, implying that I rely on a limited number of observations
in the disrupted area to tease out the effects. Since there was no systematic rule in
editing the school physical examination records across prefectures, obtaining systematic
statistics on the school-level physical examination is difficult. Therefore, some reports are
often incomplete and mostly unavailable for the statistical analyses. Considering these
limitations, applying the exogenous variations due to earthquakes to a comprehensive
child-level dataset with a precise date of birth and place may thus be an avenue for future
work. While I acknowledge these limitations, this study does use a localized measure of
the severity of the 1923 earthquake. Municipal-level geospatial variations in the physical
devastation could improve the assignment by detecting the most impacted area in the
prefecture.
This study contributes to our understanding of the long-term effects of the great
earthquake on children’s health given that the lingering health impacts of earthquakes
on children have been neglected in the literature. It also offers suggestive evidence of
the importance of the remediation of early disadvantage via disaster relief and of the
potentially significant impacts of maternal mental stress on children in utero.
References
[1] Almond, Douglas, and Janet Currie. Killing me softly: The fetal origins hypothesis. Journalof Economic Perspective 25, no.3 (2011): 153–172.
60I set the school-aged children in Chiba prefecture as the benchmark against which these treatmenteffects are calculated given that this study uses data on primary and secondary schools in Chiba prefecture.
25
[2] Almond, Douglas, Lena Edlund, Marten Palme. Chernobyl’s subclinical legacy: Prenatal ex-posure to radioactive fallout and school outcomes in Sweden. Quarterly Journal of Economics124, no.4 (2009): 1729–1772.
[3] Almond, Douglas, and Bhashkar, Mazumder. Fetal origins and parental responses. AnnualReview of Economics 5 (2013): 37–56.
[4] Au Yeung, S. Lin, Li S. Lin, Albert M. Li, et.al. Birth weight and risk of ischemic heartdisease: A Mendelian randomization study. Scientific Reports 6 (2016): 38420.
[5] Balazsi, Laszlo, Laszlo Matyas, and Tom Wansbeek. The estimation of multidimensionalfixed effects panel data models. Econometric Reviews 37, no.3 (2018): 212–227.
[6] Banerjee, Abhijit, Esther Duflo, Gilles Postel-Vinay, et.al. Long-run health impacts of in-come shocks: Wine and phylloxera in nineteenth-century France. Review of Economics andStatistics 92, no.4 (2010): 714–728.
[7] Barker, David J.P. Fetal and infant origins of adult disease. London: BMJ Books, 1992.
[8] Barker, David J.P. Mothers, babies and health in later life (2nd edition). Edinburgh:Churchill Livingstone, 1998.
[9] Barker, David J.P., Johan G. Eriksson, Tom Forsen, and Clive Osmond. Infant growth andincome 50 years later. Archives of Disease in Childhood 90 (2005): 272–273.
[10] Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. How much should we trustdifferences-in-differences estimates? Quarterly Journal of Economics 119, no.1 (2004): 249–275.
[11] Black, Robert E., Lindsay H. Allen, Zulfiqar A. Bhutta, et al. Maternal and child undernu-trition: Global and regional exposures and health consequences. Lancet 371 (2008): 243–260.
[12] Bozzoli, Carlos, Angus Deaton, and Climent Quintana-Domeque. Adult height and child-hood disease. Demography 46, no.4 (2009): 647–669.
[13] Bozzoli, Carlos, and Climent Quintana-Domeque. The weight of the crisis: Evidence fromnewborns in Argentina. Review of Economics and Statistics 96, no.3 (2014): 550–562.
[14] Brown, Ryan, and Duncan Thomas. On the long term effects of the 1918 U.S. influenzapandemic. Working Paper, University of Colorado Denver, 2018.
[15] Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. Bootstrap-based improve-ments for inference with clustered errors. Review of Economics and Statistics 90, no.3 (2008):414–427.
[16] Caruso, German, and Sebastian Miller. Long run effects and intergenerational transmissionof natural disasters: A case study on the 1970 Ancash Earthquake. Journal of DevelopmentEconomics 117 (2015): 134–150.
[17] Case, Anna, and Christina Paxson. Height, health, and cognitive function at older ages.American Economic Review 98, no.2 (2008): 463–467.
26
[18] Currie, Janet, and Tom Vogl. Early-life health and adult circumstance in developing coun-tries. Annual Review of Economics 5 (2013): 1–36.
[19] Datta Gupta, Nabanita, Mette Deding, and Mette Lausten. The effect of low birth weighton height, weight and behavioral outcomes in the medium-run. Economics and Human Biology11, no.1 (2013): 42–55.
[20] Davis, Peter. Estimating multi-way error components models with unbalanced data struc-tures. Journal of Econometrics 106 (2002): 67–95.
[21] Deaton, Angus. The analysis of household surveys: A microeconometric approach to devel-opment policy. Washington, DC: World Bank, 1997.
[22] Dehejia, Rajeev, and Adriana Lleras-Muney. Booms, busts, and babies’ health. QuarterlyJournal of Economics 119, no.3 (2004): 1091–1130.
[23] de Onis, Mercedes, Adelheid W. Onyango, Elaine Borghi, et.al. Development of a WHOgrowth reference for school-aged children and adolescents. Bulletin of the World Health Or-ganization 85, no.9 (2007): 660–667.
[24] de Onis, Mercedes, and Francesco Branca. Childhood stunting: A global perspective. Ma-ternal & Child Nutrition 12, no.S1 (2016): 12–26.
[25] Dewey, Kathryn G., and Khadija Begum. Long-term consequences of stunting in early life.Maternal & Child Nutrition 7, no.S3 (2011): 5–18.
[26] Drixler, Fabian F. Hidden in plain sight: Stillbirth and infanticides in imperial Japan.Journal of Economic History 76, no.3 (2016): 651–696.
[27] Fogel, Robert W. Economic growth, population theory, and physiology: The bearing oflong-term processes on the making of economic policy. American Economic Review 84, no.3(1994): 369–395.
[28] Frankenberg Elizabeth, Jed Friedman, Nicholas Ingwersen, et al. Linear child growth after anatural disaster: a longitudinal study of the effects of the 2004 Indian Ocean tsunami. Lancet389 (2017): s21.
[29] Glynn, M. Laura, Pathik D. Wadhwa, Christine Dunkel-Schetter, et al. When stress hap-pens matters: Effects of earthquake timing on stress responsively in pregnancy. AmericanJournal of Obstetrics and Gynecology 184 (2001): 637–642.
[30] Harada, Nahoko, Jun Shigemura, Masaaki Tanichi, et al. Mental health and psychologicalimpacts from the 2011 Great East Japan Earthquake Disaster: A systematic literature review.Disaster and Military Medicine 1, vol.17 (2015): 1–12.
[31] Hayami, Akira. Spanish influenza struck down to Japan. [in Japanese] Tokyo: Fujiwarashoten, 2006.
[32] Heckman, James J. The developmental origins of health. Health Economics 21, no.1 (2012):24–29.
[33] The HerniaSurge Group. International guidelines for groin hernia management. Hernia 22(2018): 1–165.
27
[34] Hayashi, Fumio, and Edward C. Prescott. The depressing effect of agricultural institutionson the prewar Japanese economy. Journal of Political Economy 116, no.4 (2008): 573–632.
[35] Hibino, Yuri, Jiro Takaki, Yasuhiro Kambayashi, et al. Health impact of disaster-relatedstress on pregnant women living in the affected area of the Noto Peninsula earthquake inJapan. Psychiatry and Clinical Neuroscience 63, no.1 (2009): 107–115.
[36] Hijikata, Sonoko. Kindainihon no gakko to chiikishakai (school and community in prewarJapan). [in Japanese] Tokyo: University of Tokyo Press, 1994.
[37] Hunter, Janet. “Extreme confusion and disorder”? The Japanese economy in the GreatKanto Earthquake of 1923. Journal of Asian Studies 73, no.3 (2014): 753–773.
[38] Hunter, Janet, and Kota Ogasawara. Price shocks in regional markets: Japan’s Great KantoEarthquake of 1923. Economic History Review 72, no.2 (2019): 1335–1362.
[39] Imaizumi Asuka, Kaori Ito, and Tetsuji Okazaki. Impact of natural disasters on industrialagglomeration: The case of the Great Kanto Earthquake in 1923. Explorations in EconomicHistory 60 (2016): 52–68.
[40] International Federation of Red Cross and Red Crescent Societies. Post-disaster shelter:Ten-designs. Geneva: International Federation of Red Cross and Red Crescent Societies,2013.
[41] Karbownik, Krzysztof, and Anthony Wray. Long-run consequences of exposure to naturaldisasters. Journal of Labor Economics 37, no.3 (2019): 949–1007.
[42] Kim, Bongkyun, Celeste Carruthers, and Matthew Harris. Maternal stress and birth out-comes: Evidence from the 1994 Northridge earthquake. Journal of Economic Behavior &Organization 140 (2017): 354–373.
[43] Kraemer, Sebastian. The fragile male. British Medical Journal 321 (2000):1609–1612.
[44] Kudo, Yuko, Masao Shiomoto, Shintaro Takeda, et al. Growth acceleration in Japan asindicated by the maximum growth age in height. Nihoneiseigakuzasshi 31, no.2 (1976): 378–385. [in Japanese]
[45] Law, Catherine M., Michael de Swiet, Clive Osmond, et al. Initiation of hypertension inutero and its amplification throughout life. British Medical Journal 306 (1993): 24–27.
[46] Lee, Chulhee. In utero exposure to the Korean War and its long-term effects on socioeco-nomic and health outcomes. Journal of Health Economics 33 (2014): 76–93.
[47] Mazumder, Bhash, Douglas Almond, Kyung Park, et al. Lingering prenatal effects of the1918 influenza pandemic on cardiovascular disease. Journal of Development Origins of Healthand Disease 1, no.1 (2010): 26–34.
[48] McGovern, Mark E., Aditi Krishna, Victor M. Aguayo, et al. A review of the evidencelinking child stunting to economic outcomes. International Journal of Epidemiology 46, no.4(2017): 1171–1191.
[49] Metzger, Boyd E., Rita A. Vileisis, Veronica Ravnikar, et al. “Accelerated starvation” andthe skipped breakfast in late normal pregnancy. Lancet 319, issue. 8272 (1982): 91–102.
28
[50] Morgan, Ian G., Kyoko, Ohno-Matsui, and Seang-Mei, Saw. Myopia. Lancet 379, no.9827(2012): 1739–1748.
[51] Moroi, Takafumi, and Masayuki Takemura. 1995nen Hyogokennanbujishin niyorukishochoshindo to jyutakuzenkairitsu no kankei (Relationship between seismic intensity inJMA scale and collapse rate of wooden houses for the 1995 Hygoken-Nanbu earthquake). [inJapanese] Jishin 52 (1999): 11–24.
[52] Moulton, Brent R. Random group effects and the precision of regression estimates. Journalof Econometrics 32, no.3 (1986): 385–397.
[53] Moulton, Brent R. An illustration of a pitfall in estimating the effects of aggregate variableson micro units. Review of Economics and Statistics 72, no.2 (1990): 334–338.
[54] Nakagawa, Satoshi. Kokuseichosa karamita 1920–30nendai no jinkoido (Internal migrationof Japan during 1920s and 1930s). [in Japanese] Kokuminkeizaizasshi 184, no.4 (2001): 37–51.
[55] Nakamura, Takafusa. Showakyoko to keizaiseisaku (Showa depression and economic policy).[in Japanese] Tokyo: Kodansha, 1994.
[56] Nyagu, Angelina I, Loganovsky N. Konstantin, Loganovskaja K. Tatiana, et al. Intelligenceand brain damage in children acutely irradiated in utero as a result of the Chernobyl accident.KURRI-KR 79 (2002): 202–229.
[57] Oster, Emily. Unobservable selection and coefficient stability: Theory and evidence. Journalof Business & Economic Statistics 37, no.2 (2019): 187–204.
[58] Otake, Masanori, and William J. Schull. Radiation-related brain damage and growth re-tardation among the prenatally exposed atomic bomb survivors. International Journal ofRadiation Biology 74, no.2 (1998): 159–171.
[59] Okazaki, Tetsuji, Toshihiro Okubo, and Eric Strobl. Creative destruction of industries:Yokohama city in the Great Kanto Earthquake, 1923. Journal of Economic History 79, no.1(2019): 1–31.
[60] Parman, John. Childhood health and sibling outcomes: Nurture reinforcing during the 1918influenza pandemic. Explorations in Economic History 58 (2015): 22–43.
[61] Pereira, Alvaro S. The opportunity of a disaster: The economic impact of the 1755 LisbonEarthquake. Journal of Economic History 69, no.2 (2009): 466–499.
[62] Prinz, Daniel, Chernew Michael, Cutler David, and Frakt Austin. Health and economicactivity over the lifecycle: Literature review. NBER Working Paper No. 24865, Cambridge,MA, July 2018.
[63] Pingali, Prabhu L, Luca Alinovi, and Jacky Sutton. Food security in complex emergencies:Enhancing food system resilience. Disasters 29, no.s1 (2005): S5–S24.
[64] Ramseyer, Mark J. Odd markets in Japanese history. New York, Cambridge UniversityPress, 1996
[65] Rice, Geoffrey W, and Edwina Palmer. Pandemic influenza in Japan, 1918–19: Mortalitypatterns and official responses. Journal of Japanese Studies 19, no.2 (1993): 389–420.
29
[66] Rosales-Rueda, Maria, and Margaret Triyana. The persistent effects of early-life exposureto air pollution: Evidence from the Indonesian forest fires. Journal of Human Resources 54,no.4 (2019): 1037–1080.
[67] Schneider, Eric B., and Kota Ogasawara. Disease and child growth in industrialising Japan:Critical windows and the growth pattern, 1917–39. Explorations in Economic History 69(2018): 64–80.
[68] Schneider, Eric B. Sample selection bias and the historical growth pattern of children. SocialScience History 2019: http://eprints.lse.ac.uk/id/eprint/100826.
[69] Shoji, Kichinosuke. Study on the rice riots. Tokyo, Miraisha: 1957.
[70] Steckel, Richard H., and Nicolas Ziebarth. Trader selectivity and measured catch-up growthof American slave. Journal of Economic History 76, no.1 (2016): 109–137.
[71] Stein, Alan, Rebecca M. Pearson, Sherryl H. Goodman, et al. Effects of perinatal mentaldisorders on the fetus and child. Lancet 384, issue.9956 (2014): 1800–1819.
[72] Takemura, Masayuki. Kantodaishinsai – Daitokyoken no yure wo shiru (The Great KantoEarthquake, knowing ground shaking in greater Tokyo area). [in Japanese] Tokyo: KajimaInstitute Publishing, 2003.
[73] Takemura, Masayuki, and Takafumi Moroi. Chishitsuchosajyo data ni motoduku 1923nenkantojishin no shosaishindobunpu (Detailed seismic intensity distribution due to the 1923Kanto Earthquake based on the data of the report by the Geological Survey of Japan Part 1.Chiba Prefecture). Nihon jishinkogakkai ronbunshu 1, no.1 (2002): 1–26. [in Japanese]
[74] Tang, John P. Railroad expansion and industrialization: Evidence from Meiji Japan. Jour-nal of Economic History 74, no.3 (2014): 863–886.
[75] Tang, John P. The engine and the reaper: Industrialization and mortality in late nineteenthcentury Japan. Journal of Health Economics 56 (2017): 145–162.
[76] Torche, Florencia. The effect of maternal stress on birth outcomes: Exploiting a naturalexperiment. Demography 48, no.4 (2011): 1473–1491.
[77] Tomkins, Andrew, Susan Murray, Patricia Rondo, et al. Impact of maternal infection onfoetal growth and nutrition. SCN News 11 (1994): 18–20.
[78] Valente, Christine. Civil conflict, gender-specific fetal loss, and selection: A new test of theTrivers-Willard hypothesis. Journal of Health Economics 39 (2015): 31–50.
[79] Vellore, Arthi. “The dust was long in settling”: Human capital and the lasting impact ofthe American dust bowl. Journal of Economic History 78, no.1 (2018): 196–230.
[80] Victora, Cesar G, Linda Adair, Caroline Fall, et al. Maternal and child undernutrition:Consequences for adult health and human capital. Lancet 371 (2008): 340–357.
[81] Yonkers, Kimberly A, Megan V. Smith, Ariadna Forray, et al. Pregnant women with post-
traumatic stress disorder and risk of preterm birth. JAMA Psychiatry 71, no.8 (2014): 897–
904.
30
Documents, Statistical Reports, and Database
[1] Awa County Office. Awashinsaishi (History of Awa earthquake). [in Japanese] Chiba: AwaCounty Office, 1926.
[2] Bank of Japan. Nihonginko hyakunenshi (One hundred year history of the Bank of Japan).Tokyo: Bank of Japan, 1986. [in Japanese]
[3] Cabinet Official Gazette Bureau. Houreizensho (Law concerning the application of laws ingeneral). [in Japanese] Tokyo: Cabinet Official Gazette Bureau, 1898. Available online atthe National Diet Library (https://dl.ndl.go.jp/info:ndljp/pid/788007/16) (last accessed onJanuary 5, 2020).
[4] Chiba Prefecture. Chibaken tokeisho (The statistical report of Chiba prefecture, 1923 edition,volume.1). [in Japanese] Chiba: Chiba Prefecture, 1924.
[5] Chiba Prefecture. Chibaken tokeisho (The statistical report of Chiba prefecture, 1923 edition,volume.2). [in Japanese] Chiba: Chiba Prefecture, 1925a.
[6] Chiba Prefecture. Chibaken tokeisho (The statistical report of Chiba prefecture, 1923 edition,volume.3). [in Japanese] Chiba: Chiba Prefecture, 1925b.
[7] Chiba Prefecture. Chibaken tokeisho (The statistical report of Chiba prefecture, 1923 edition,volume.5). [in Japanese] Chiba: Chiba Prefecture, 1925c.
[8] Chiba Prefecture. Seitojido shintaikensa tokei (Statistics of physical examination for stu-dents, 1925–1935 editions). [in Japanese] Chiba: Chiba Prefecture, 1925–1946.
[9] Chiba Prefecture. Chibaken tokeisho (The statistical report of Chiba prefecture, 1930 edition,volume.2). [in Japanese] Chiba: Chiba Prefecture, 1932.
[10] Chiba Prefecture. Chibaken tokeisho (The statistical report of Chiba prefecture, 1935 edi-tion, volume.2). [in Japanese] Chiba: Chiba Prefecture, 1936.
[11] Division of Social Affairs, Chiba Prefecture. Taishodaishinsai no kaiko to sonofukko (Rem-iniscences and reconstructions of the Taisho earthquake, volume 1). [in Japanese] Chiba:Division of Social Affairs, Chiba Prefecture, 1933a.
[12] Division of Social Affairs, Chiba Prefecture. Taishodaishinsai no kaiko to sonofukko (Rem-iniscences and reconstructions of the Taisho earthquake, volume 2). [in Japanese] Chiba:Division of Social Affairs, Chiba Prefecture, 1933b.
[13] Japan Meteorological Agency (database). Available online at the Japan Meteorologi-cal Agency (http://www.data.jma.go.jp/svd/eqdb/data/shindo/index.php) (last accessed onJanuary 9, 2018).
[14] Japan Meteorological Agency (website). Available online at the Japan MeteorologicalAgency (http://www.jma.go.jp/jma/en/Activities/inttable.html) (last accessed on July 28,2017).
[15] Ministry of Education. Gakuseihyakunenshi (The hundred years’ history of modernJapanese education). [in Japanese] Tokyo: Gyosei, 1973.
31
[16] Ministry of Land, Infrastructure, Transport and Tourism of Japan (database). Availableonline at the Ministry of Land, Infrastructure, Transport and Tourism of Japan (http://nlftp.mlit.go.jp/ksj/jpgis/datalist/KsjTmplt-N03.html) (last accessed on May 22, 2017).
[17] Ministry of Railways. Kokuyutetsudoshinsaishi (Japan National Railways’ history of theearthquake). [in Japanese] Tokyo: Ministry of Railways, 1927.
[18] Physical Education Bureau, Ministry of Education. Seitojido shintaikensa tokei (Statisticsof school physical examination, 1929–1938 editions). [in Japanese] Tokyo: Physical EducationBureau, Ministry of Education, 1931–1942.
[19] Social Welfare Bureau of the Cabinet. Taishoshinsaishi (History of the Taisho earthquake).[in Japanese] Tokyo: Social Welfare Bureau of the Cabinet, 1926.
[20] Statistics Bureau of the Cabinet. Taishonananen Nihonteikokushiintokei (Statistics ofcauses of death of the empire of Japan 1918 edition). [in Japanese] Tokyo: Statistics Bu-reau of the Cabinet, 1921.
[21] Statistics Bureau of the Cabinet. Taishohachinen Nihonteikokushiintokei (Statistics ofcauses of death of the empire of Japan 1919 edition). [in Japanese] Tokyo: Statistics Bu-reau of the Cabinet, 1922.
[22] Statistics Bureau of the Cabinet. Taishokyunen Nihonteikokushiintokei (Statistics of causesof death of the empire of Japan 1920 edition). [in Japanese] Tokyo: Statistics Bureau of theCabinet, 1923.
[23] Statistics Bureau of the Cabinet. Taishokyunen Kokuseichosahoukoku (Population censusof Japan, vol.10 Chiba prefecture part, 1920 edition). [in Japanese] Tokyo: Statistics Bureauof the Cabinet, 1924a.
[24] Statistics Bureau of the Cabinet. Jinkodotaitokei (The vital statistics of the empire ofJapan, 1922 edition). [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1924b.
[25] Statistics Bureau of the Cabinet. Jinkodotaitokei (The vital statistics of the empire ofJapan, 1923 edition). [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1925.
[26] Statistics Bureau of the Cabinet. Taishojyuyonen Kokuseichosahokoku (Population censusof Japan, vol.4 Chiba prefecture part, 1925 edition). [in Japanese] Tokyo: Statistics Bureauof the Cabinet, 1926.
[27] Statistics Bureau of the Cabinet. Showagonen Kokuseichosahokoku (Population census ofJapan, vol.4 Chiba prefecture part, 1930 edition). [in Japanese] Tokyo: Statistics Bureau ofthe Cabinet, 1931.
[28] Statistical Bureau of the Cabinet. Kakeichosahokoku (Household survey report, volume.4).[in Japanese] Tokyo: Statistical Bureau of the Cabinet, 1929.
[29] Tokyo City Office. Kantochihoshinsai no keizaikai ni oyohoseru eikyo (Damage to Japan bythe Great Kanto Earthquake). [in Japanese] Tokyo: Isseisha, 1925.
[30] Tokyo City Office. Nyujishibochosa (Survey on the infant mortality rate). [in Japanese]
Tokyo: Tokyo City Office, 1926.
32
Tab
le1:
Sum
mar
ySta
tist
ics
Un
itA
ges
Boy
sG
irls
Pan
elA
:H
eigh
tan
dW
eigh
tM
ean
Std
.D
ev.
Ob
s.M
ean
Std
.D
ev.
Obs.
All
coh
orts
Hei
ght
(cm
)S
chool
-yea
r-ag
e6–
811
3.46
4.37
14,1
3911
2.26
4.33
14,1
45S
chool
-yea
r-ag
e9–
1112
7.47
4.13
14,1
3312
6.89
4.65
14,1
38W
eigh
t(k
g)S
chool
-yea
r-ag
e6–
820.0
71.
7214,1
3919.4
01.
6814,1
45S
chool
-yea
r-ag
e9–
1126.4
92.
1514,1
3326.2
72.
5814,1
38U
nex
pos
edco
hor
tsH
eigh
t(c
m)
Sch
ool
-yea
r-ag
e6–
811
3.45
4.36
12,8
4311
2.25
4.33
12,8
49S
chool
-yea
r-ag
e9–
1112
7.43
4.13
12,8
6412
6.83
4.64
12,8
64W
eigh
t(k
g)S
chool
-yea
r-ag
e6–
820.0
71.
7212,8
4319.4
01.
6812,8
49S
chool
-yea
r-ag
e9–
1126.4
72.
1512,8
6426.2
42.
5812,8
64E
xp
osed
coh
orts
Hei
ght
(cm
)S
chool
-yea
r-ag
e6–
811
3.54
4.42
1,29
611
2.42
4.33
1,29
6S
chool
-yea
r-ag
e9–
1112
7.92
4.10
1,26
912
7.41
4.68
1,27
4W
eigh
t(k
g)S
chool
-yea
r-ag
e6–
820.0
41.
781,
296
19.3
91.
711,
296
Sch
ool
-yea
r-ag
e9–
1126.7
32.
171,
269
26.5
32.
611,
274
Pan
elB
:E
xp
osu
reva
riab
les
and
reli
efex
pen
ses
Un
itM
ean
Std
.D
ev.
Min
Max
NE
xp
osu
reva
riab
les
JMA
-SIS
7[S
IS7
]M
un
icip
al0.
07−
01
344
Rai
lway
dis
rup
tion
[Rai
l]M
un
icip
al0.
14−
01
222
Rel
ief
exp
ense
sp
erca
pit
a(y
en)
Tot
alC
ounty
0.22
10.
724
02.
619
13F
ood
Cou
nty
0.14
90.
502
01.
816
13S
hel
ter
Cou
nty
0.06
80.
205
00.
740
13M
edic
altr
eatm
ent
Cou
nty
0.00
10.
002
00.
007
13
Note
s:P
an
elA
rep
ort
sth
esu
mm
ary
stati
stic
sfo
rth
esc
hool-
yea
r-age-
level
aver
age
valu
esof
hei
ght
(cm
)an
dw
eight
(kg).
Exp
ose
dco
hort
sin
clu
de
the
1923
an
d1924
bir
thco
hort
s,w
her
eas
un
exp
ose
dco
hort
sin
clu
de
all
bir
thco
hort
soth
erth
an
the
exp
ose
dco
hort
s.P
an
elB
rep
ort
sth
esu
mm
ary
stati
stic
sfo
rth
eex
posu
revari
ab
les
an
dre
lief
exp
ense
sp
erca
pit
a.SIS
7is
an
ind
icato
rvari
ab
lefo
rm
un
icip
aliti
esin
the
JM
A-S
IS7
are
a(T
ab
le2).
Rail
isan
ind
icato
rvari
ab
lefo
rm
un
icip
aliti
essa
tisf
yin
gth
efo
llow
ing
con
dit
ion
s:(a
)lo
cate
din
the
JM
A-S
IS7
are
a,
(b)
did
not
rece
ive
any
dis
ast
erre
lief
,an
d(c
)lo
cate
dw
ith
in10
km
of
any
mu
nic
ipality
incl
ud
ing
eith
erth
eN
od
alin
eor
the
Ku
ruri
lin
e.E
ach
relief
vari
ab
leis
defi
ned
as
the
relief
exp
ense
div
ided
by
the
nu
mb
erof
peo
ple
(yen
).S
ou
rces
:B
iolo
gic
al
ou
tcom
ed
ata
are
from
the
SP
ES
(1925–1935
edit
ion
s).
See
Tab
le2
for
the
data
use
dto
calc
ula
teth
eea
rth
qu
ake
inte
nsi
tym
easu
re.
Data
on
the
railw
ay
dis
rup
tion
an
dre
lief
exp
ense
sp
erca
pit
aare
from
the
SR
CP
(1923
edit
ion
).
33
Tab
le2:
Cla
ssifi
cati
onof
Munic
ipal
itie
sby
JM
A-S
IS(s
hin
do)
Nu
mb
erof
Cou
nti
esm
ain
lyM
ean
of
the
dam
aged
Sei
smic
inte
nsi
tysc
ale
mu
nic
ipali
ties
incl
ud
edh
ou
seh
old
s(P
DR
in%
)JM
A-S
IS7
(sh
ind
on
ana)
23
Aw
a,
Kim
itsu
79.8
JM
A-S
IS6
(sh
ind
oro
ku
)46
Aw
a,
Kim
itsu
,Ic
hih
ara
,C
hose
i22.8
JM
A-S
IS5
(sh
ind
ogo
)275
Ch
iba,
Hig
ash
ikats
ush
ika,
0.3
9Im
ba,
Isu
mi,
Ich
ihara
,K
aij
yo,
Kato
ri,
Sanb
u,
Sou
sa
Note
s:T
he
PD
Ris
the
nu
mb
erof
aff
ecte
dh
ou
seh
old
sp
er100
hou
seh
old
s.T
he
JM
A-S
ISra
nges
from
on
e(m
inim
um
)to
seven
(maxim
um
).It
isca
lcu
late
db
ase
don
the
collap
sera
tepre
dic
ted
usi
ng
the
pre
dic
tion
equ
ati
on
of
Moro
ian
dT
akem
ura
(1999):
Collap
seR
ate
=−
1.6
1+
0.4
6×
PD
R+
0.0
051×
PD
R2.
Th
eJM
A-S
ISis
class
ified
base
don
Takem
ura
an
dM
oro
i(2
002):
ara
tegre
ate
rth
an
30%
isd
efin
edas
JM
A-S
IS7
(sh
ind
on
an
a),
gre
ate
rth
an
1%
an
dle
ssth
an
30%
isd
efin
edas
JM
A-S
IS6
(sh
ind
oro
ku
),an
dgre
ate
rth
an
0.1
%an
dle
ssth
an
1%
isdefi
ned
as
up
per
JM
A-S
IS5
(sh
ind
ogo).
No
mu
nic
ipaliti
esex
per
ien
ced
JM
A-S
IS1–4.
Follow
ing
the
offi
cial
class
ifica
tion
,JM
A-S
IS6
incl
ud
esb
oth
6-l
ow
er(r
oku
jyaku
)an
d6-u
pp
er(r
oku
kyo);
JM
A-S
IS5
incl
ud
esb
oth
5-l
ow
er(g
ojy
aku
)an
d5-u
pp
er(g
okyo).
Sou
rces
:D
ata
use
dto
calc
ula
teth
eea
rth
qu
ake
inte
nsi
tym
easu
res
are
from
the
HT
E(1
926)
an
dS
tati
stic
sB
ure
au
of
the
Cab
inet
(1926).
34
Table 3: Effects of Fetal Earthquake Exposure on Height and Weight
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A: Effects on height1923 birth cohort −0.213* −0.196* −0.234* −0.233***
[0.020] [0.044] [0.010] [0.003]1923 birth cohort × SIS7 0.085 0.109 −0.296** −0.604***
[0.250] [0.798] [0.008] [0.005]1924 birth cohort −0.094 −0.086 0.117 −0.107
[0.212] [0.332] [0.236] [0.241]1924 birth cohort × SIS7 0.051 0.025 0.184 0.357
[0.812] [1.000] [0.164] [0.455]Panel B: Effects on weight1923 birth cohort −0.058* −0.041 −0.099** −0.131**
[0.016] [0.314] [0.008] [0.005]1923 birth cohort × SIS7 0.003 0.102 −0.269*** −0.303**
[0.938] [0.438] [0.002] [0.009]1924 birth cohort −0.004 −0.030 0.009 −0.052
[0.864] [0.490] [0.754] [0.051]1924 birth cohort × SIS7 −0.029 −0.006 −0.109 0.035
[0.996] [0.974] [0.204] [0.957]
***, **, and * represent statistical significance at the 0.5%, 1%, and 5% levels based on the p-values from thewild cluster bootstrap resampling method in brackets, respectively. The data are clustered at the 13-countylevel in the bootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: The numbers of observations in columns (1)–(4) are 14, 139, 14, 133, 14, 145, and 14, 138, respectively.All the regressions include controls for the rice yield in the birth year; fetal death rate in the birth year;school enrollment rate of the parental generation; school-age-specific fixed effects; and year fixed effects.The null hypothesis of no joint significance of the affected cohort dummies and those area interaction termsare rejected based on the F -test at the 5% level in most specifications. The null hypothesis of no jointsignificance of the control variables is rejected based on the F -test at the 0.1% level in all the specifications.
35
Tab
le4:
Eff
ects
ofD
isas
ter
Rel
ief
onH
eigh
tan
dW
eigh
tby
Typ
eof
Rel
ief
Exp
ense
Rel
ief
exp
ense
sp
erca
pit
a
Tot
alF
ood
Sh
elte
rM
edic
altr
eatm
ent
(1a)
Exp
ense
s×
(1b
)E
xpen
ses×
(2a)
Exp
ense
s×
(2b
)E
xpen
ses×
(3a)
Exp
ense
s×
(3b
)E
xpen
ses×
(4a)
Exp
ense
s×
(4b
)E
xpen
ses×
1923
bir
thco
hor
t19
24b
irth
coh
ort
1923
bir
thco
hor
t19
24b
irth
coh
ort
1923
bir
thco
hor
t19
24b
irth
coh
ort
1923
bir
thco
hor
t19
24b
irth
coh
ort
Pan
elA
:E
ffec
tson
hei
ght
Boy
sA
ges
6–8
0.11
3[0
.022
]*0.
053
[0.6
22]
0.16
5[0
.012
]*0.
073
[0.6
20]
0.38
1[0
.038
]*0.
208
[0.5
68]
41.4
74[0
.048
]*20
.551
[0.5
64]
Age
s9–
110.
091
[0.9
46]
-0.0
54[0
.784
]0.
139
[0.7
62]
-0.0
77[0
.818
]0.
268
[1.0
00]
-0.1
92[0
.690
]34
.292
[0.9
14]
-22.
279
[0.6
58]
Gir
lsA
ges
6–8
-0.0
12[0
.982
]-0
.023
[0.8
26]
-0.0
11[0
.982
]-0
.026
[0.8
56]
-0.0
91[0
.968
]-0
.127
[0.6
82]
5.76
0[0
.990
]-1
5.00
0[0
.596
]
Age
s9–
110.
072
[0.5
87]
-0.0
42[0
.863
]0.
109
[0.5
63]
-0.0
53[0
.919
]0.
207
[0.6
35]
-0.1
97[0
.709
]35
.767
[0.4
47]
-20.
793
[0.6
67]
Pan
elB
:E
ffec
tson
wei
ght
Boy
sA
ges
6–8
0.02
9[0
.016
]*0.
032
[0.1
88]
0.04
2[0
.024
]*0.
051
[0.1
96]
0.09
7[0
.018
]*0.
080
[0.1
66]
15.9
93[0
.024
]*9.
356
[0.0
72]
Age
s9–
110.
048
[0.8
26]
0.00
5[1
.000
]0.
070
[0.7
10]
0.00
7[1
.000
]0.
155
[0.8
90]
0.01
1[1
.000
]19
.153
[0.6
36]
-0.6
87[0
.978
]
Gir
lsA
ges
6–8
0.00
2[1
.000
]-0
.022
[0.6
70]
0.00
3[1
.000
]-0
.031
[0.6
96]
0.00
9[1
.000
]-0
.085
[0.5
42]
10.1
09[0
.890
]-9
.491
[0.4
70]
Age
s9–
110.
056
[0.4
17]
-0.0
05[0
.995
]0.
079
[0.4
01]
-0.0
06[0
.997
]0.
202
[0.4
41]
-0.0
30[0
.973
]29
.010
[0.4
31]
-7.6
23[0
.923
]
*re
pre
sents
stati
stic
al
sign
ifica
nce
at
the
5%
level
base
don
thep-v
alu
esfr
om
the
wild
clu
ster
boots
trap
resa
mp
lin
gm
eth
od
inb
rack
ets.
Th
ed
ata
are
clu
ster
edat
the
13-c
ou
nty
level
inth
eb
oots
trap
pro
ced
ure
.T
he
nu
mb
erof
rep
lica
tion
sis
fixed
to1,0
00
for
all
the
spec
ifica
tion
s.N
ote
s:E
stim
ate
dco
effici
ents
onI(YOB=1923)×
Relief
an
dI(YOB=1924)×
Relief
ineq
uati
on
2are
rep
ort
edin
the
tab
le.
Th
enu
mb
erof
ob
serv
ati
on
sfo
rb
oys
(gir
ls)
aged
6–8
an
d9–11
are
14,1
39
(14,1
45)
an
d14,1
33
(14,1
38),
resp
ecti
vel
y.A
llth
ere
gre
ssio
ns
incl
ud
eco
ntr
ols
for
the
rice
yie
ldin
the
bir
thyea
r;fe
tal
dea
thra
tein
the
bir
thyea
r;sc
hool
enro
llm
ent
rate
of
the
pare
nta
lgen
erati
on
;sc
hool-
age-
spec
ific
fixed
effec
ts;
an
dyea
rfi
xed
effec
ts.
Th
enu
llhyp
oth
esis
of
no
join
tsi
gn
ifica
nce
of
the
contr
ol
vari
ab
les
isre
ject
edb
ase
don
theF
-tes
tat
the
0.1
%le
vel
inall
the
spec
ifica
tion
s.
36
Table 5: Effects of Fetal Earthquake Exposure on Height and Weight in the LimitedJMA-SIS5 Area
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A: Effects on height1923 birth cohort −0.203* −0.168 −0.248** −0.225*
[0.020] [0.216] [0.008] [0.019]1923 birth cohort × Railway disruption 0.180 0.185 0.241 0.360
[0.732] [0.590] [0.056] [0.767]1924 birth cohort −0.120 −0.066 −0.130 −0.131
[0.248] [0.590] [0.160] [0.261]1924 birth cohort × Railway disruption 0.132 −0.038 −0.024 0.160
[0.480] [0.982] [0.914] [0.525]
Panel B: Effects on weight1923 birth cohort −0.051* −0.008 −0.103* −0.110*
[0.028] [0.996] [0.010] [0.039]1923 birth cohort × Railway disruption 0.081 −0.117 0.059 −0.060
[0.708] [0.480] [0.390] [0.799]1924 birth cohort 0.010 0.015 0.012 −0.052
[0.832] [0.784] [0.732] [0.083]1924 birth cohort × Railway disruption 0.087 −0.127 0.042 0.053
[0.154] [0.540] [0.608] [0.947]
** and * represent statistical significance at the 1% and 5% levels based on the p-values from the wild clusterbootstrap resampling method in brackets, respectively. The data are clustered at the 13-county level in thebootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: The samples include municipalities receiving no disaster relief in the JMA-SIS5 area. The numbers ofobservations in columns (1)–(4) are 9352, 9348, 9353, and 9347, respectively. All the regressions include controlsfor the rice yield in the birth year; fetal death rate in the birth year; school enrollment rate of the parentalgeneration; school-age-specific fixed effects; and year fixed effects. The null hypothesis of no joint significance ofthe control variables is rejected based on the F -test at the 0.1% level in all the specifications.
37
Table 6: Effects of Fetal Earthquake Exposure on Height and Weight:Robustness of including Additional Control Variables
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A-1: Effects on height1923 birth cohort −0.214* −0.197* −0.233** −0.233***
[0.020] [0.038] [0.008] [0.003]1923 birth cohort × SIS7 −0.084 0.109 −0.293* −0.607**
[0.238] [0.824] [0.014] [0.005]Primary school enrollment rate −0.081 0.055 −0.035 0.004
[0.122] [0.470] [0.524] [0.939]Primary school attendance rate −0.026 −0.006 0.042 −0.027
[0.476] [0.836] [0.172] [0.415]Panel A-2: Effects on weight1923 birth cohort −0.058* −0.041 −0.099** −0.131***
[0.018] [0.316] [0.008] [0.005]1923 birth cohort × SIS7 0.003 0.105 −0.268*** −0.304**
[0.926] [0.476] [0.002] [0.009]Primary school enrollment rate −0.019 −0.019 0.017 −0.003
[0.560] [0.340] [0.498] [0.943]Primary school attendance rate −0.004 0.029 0.012 −0.005
[0.794] [0.080] [0.264] [0.781]Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel B-1: Effects on height1923 birth cohort −0.208* −0.197* −0.227** −0.234***
[0.018] [0.040] [0.010] [0.003]1923 birth cohort × SIS7 0.088 0.109 −0.293** −0.607***
[0.230] [0.804] [0.008] [0.005]Wartime birth cohorts 0.022 −0.015 0.097 0.149
[0.612] [0.770] [0.294] [0.131]Influenza pandemic birth cohorts −0.076 −0.004 −0.074 0.066
[0.388] [0.980] [0.340] [0.487]Panel B-2: Effects on weight1923 birth cohort −0.058* −0.040 −0.100** −0.136**
[0.012] [0.320] [0.008] [0.005]1923 birth cohort × SIS7 0.004 0.102 −0.269*** −0.308**
[0.922] [0.430] [0.002] [0.009]Wartime birth cohorts −0.043 0.047 −0.032 0.007
[0.074] [0.142] [0.376] [0.881]Influenza pandemic birth cohorts −0.009 −0.005 −0.008 0.094
[0.764] [0.880] [0.802] [0.097]
***, **, and * represent statistical significance at the 0.5%, 1%, and 5% levels based on the p-values from the wildcluster bootstrap resampling method in brackets, respectively. The data are clustered at the 13-county level in thebootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: The numbers of observations in columns (1)–(4) are 14, 139, 14, 133, 14145, and 14138, respectively. Theestimated coefficients on the 1924 birth cohort dummy and its area interaction term are included in all the regressionsbut are not reported as those estimates are statistically insignificant in all the specifications. All the regressionsinclude controls for the rice yield in the birth year; fetal death rate in the birth year; school enrollment rate of theparental generation; school-age-specific fixed effects; and year fixed effects. The null hypothesis of no joint significanceof the control variables is rejected based on the F -test at the 1% level in all the specifications.
38
Table 7: Effects of Fetal Earthquake Exposure on Height and Weight: Testing thePotential Impacts on Surrounding Cohorts
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A: Effects on height1922 birth cohort 0.056[0.226] -0.046[0.500] 0.056[0.574] 0.003[1.000]1922 birth cohort × SIS7 0.174[0.376] 0.441[0.478] 0.039[0.668] 0.019[1.000]1923 birth cohort -0.187[0.018]* -0.227[0.056] -0.192[0.028]* -0.261[0.009]**1923 birth cohort × SIS7 0.110[0.180] 0.158[0.468] -0.322[0.012]* -0.628[0.003]***1924 birth cohort -0.076[0.306] -0.124[0.338] 0.160[0.138] -0.148[0.231]1924 birth cohort × SIS7 0.077[0.782] 0.073[1.000] 0.158[0.136] 0.332[0.631]1925 birth cohort -0.012[0.898] -0.057[0.532] 0.061[0.334] -0.081[0.331]1925 birth cohort × SIS7 0.053[0.790] -0.050[0.824] -0.273[0.140] -0.326[0.151]
Panel B: Effects on weight1922 birth cohort 0.039[0.262] -0.005[0.898] -0.001[0.974] 0.002[1.000]1922 birth cohort × SIS7 0.085[0.488] 0.308[0.450] -0.087[0.352] -0.104[0.159]1923 birth cohort -0.040[0.022]* -0.054[0.226] -0.095[0.004]*** -0.158[0.003]***1923 birth cohort × SIS7 0.024[0.402] 0.155[0.480] -0.301[0.002]*** -0.325[0.003]***1924 birth cohort 0.010[0.824] -0.052[0.406] 0.016[0.630] -0.089[0.065]1924 birth cohort × SIS7 -0.008[1.000] 0.048[1.000] -0.141[0.216] 0.014[0.983]1925 birth cohort -0.007[0.806] -0.063[0.138] 0.027[0.294] -0.079[0.201]1925 birth cohort × SIS7 0.099[0.486] 0.213[0.482] -0.200[0.056] -0.113[0.327]
***, **, and * represent statistical significance at the 0.5%, 1%, and 5% levels based on the p-values from thewild cluster bootstrap resampling method in brackets, respectively. The data are clustered at the 13-county levelin the bootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: The numbers of observations for each regression reported in columns (1)–(4) are 14, 139, 14, 133, 14, 145,and 14, 138, respectively. All the regressions include controls for the rice yield in the birth year; fetal death ratein the birth year; school enrollment rate of the parental generation; school-age-specific fixed effects; and year fixedeffects. The null hypothesis of no joint significance of the control variables is rejected based on the F -test at the0.1% level in all the specifications.
39
Ibaraki
Saitama
ChibaTokyo
Kanagawa
Yamanashi
Shizuoka
Affected areaMarginally affected area
Figure 1: Affected area and hypocenterNotes: The red circle shows the hypocenter of the earthquake. The affected area includes Tokyo,Kanagawa, and Chiba prefectures. The marginally affected area includes Saitama, Shizuoka,Yamanashi, and Ibaraki prefectures. Sources: Created by the author from Tokyo City Office(1925, p. 161). The location of the hypocenter was based on the official database of the JapanMeteorological Agency (database).
(42.5,98.8](21.1,42.5](3.6,21.1](1.5,3.6](0.6,1.5](0.3,0.6](0.2,0.3](0.0,0.2][0.0,0.0]
(a) Affected households (%)
JMA-SIS 7JMA-SIS 6JMA-SIS 5
(b) Intensity classification
Figure 2: Spatial distribution of affected households in Chiba prefectureNotes: The PDR, which is defined as the number of affected households (collapsed or semi-collapsed dueto the earthquake) per 100 total households, is illustrated in Figure 2a. The classifications of JMA-SIS5,JMA-SIS6, and JMA-SIS7 in Figure 2b include municipalities exposed to a JMA-SIS of 5 (shindo go), 6(shindo roku), and 7 (shindo nana), respectively. Sources: Calculated by the author from the Divisionof Social Affairs, Chiba Prefecture (1933b). Shapefile is based on the database of the Ministry of Land,Infrastructure, Transport and Tourism of Japan (database).
40
Appendices Online Appendices : For onlinepublication only (Supplemental materials for review)
Appendix A Background appendix
0
100
200
300
400
500
600
700
800
900
Num
ber o
f ext
rinsi
c de
aths
1 2 3 4 5 6 7 8 9 10 11 12Month
MaleFemale
(a) Number of extrinsic deaths
0
5
10
15
20
25
30
Prop
ortio
n of
ext
rinsi
c de
aths
(%)
1 2 3 4 5 6 7 8 9 10 11 12Month
MaleFemale
(b) Extrinsic deaths (%)
23456789
101112131415
Num
ber o
f ext
rinsi
c de
aths
per
10,
000
peop
le
1919 1920 1921 1922 1923 1924 1925 1926 1927Year
MaleFemale
(c) Extrinsic death rate
Figure A.1: Extrinsic deaths in Chiba around the Great Kanto EarthquakeNotes: The proportion of extrinsic deaths shown in Figure A.1b is defined as the number of extrinsic deaths relative tototal deaths. The extrinsic death rate shown in Figure A.1c is defined as the number of extrinsic deaths per 10,000 people.Sources: All figures are calculated by the author from the Statistics Bureau of the Cabinet (1922a–1928a).
Figure A.1a confirms the dramatic increase in extrinsic deaths in September 1923.
The striking fact is that the number of female extrinsic deaths is greater than that of
males. This tendency does not change if I divide the number of deaths by total male
and female monthly deaths in 1923 (Figure A.1b). Figure A.1c shows the extrinsic death
rate, defined as the number of extrinsic deaths per 10,000 people, in Chiba between
1919 and 1927. The extrinsic death rate for males was slightly higher than that for
females, reflecting the number of extrinsic deaths in outside workplaces. However, the
small disparity in the male and female death rates of 1923 suggests that women were more
likely to have been affected by the earthquake, which struck households around lunchtime
when most wives had cooked their family’s meal at home. This may have caused such
a gender disparity in the death rate. Figure A.2 shows the spatial distribution of the
number of deaths, missing people, and injured people due to the earthquake per 100
people (see also Online Appendix B). This figure shows a similar but spatially modest
distribution compared with that of affected households in Figure 2a. Indeed, while 28% of
municipalities (97 out of 349) had victims, 53% suffered household damage. The example
picture in Figure B.2 illustrates the devastation in Awa, confirming the validity of the
measure of earthquake stress (see the “Great Kanto Earthquake” section).
Appendix B Data appendix
The school-level primary school dataset used in this study is constructed from the SPES
published by Chiba prefecture between 1925 and 1936.61 The number of primary schools
61The SPES originally consisted of two reports that contained similar information: the Gakko seitojidoshintaikensa tokei published in 1925 and from 1927 to 1934 (data for 1925–1934) and the Seitojidoshintaikensa tokei published in 1936 (data for 1935). I uniformly refer to these publications as the SPES(1925–1935 editions) for simplicity. Some errata in these documents were corrected. For instance, if thesequence of height (in cm) from ages 6 to 11 was 109, 218, 123, 125, 128, and 130, I corrected the secondobservation to 118 because a height of 218 cm is clearly unrealistic for a seven-year-old and thus can beregarded as a typo.
1
(2.40,10.48](0.96,2.40](0.36,0.96](0.23,0.36](0.08,0.23](0.05,0.08](0.03,0.05](0.00,0.03][0.00,0.00]
Figure A.2: Spatial distribution of victims (per 100 people)in Chiba prefecture
Note: The number of victims is defined as the number of deaths, missing people, and injuredpeople due to the earthquake. Source: Calculated by the author from Division of Social Affairs,Chiba Prefecture (1933b).
from 1925 to 1935 was 429, 429, 429, 433, 434, 433, 433, 432, 428, 427, and 422, respec-
tively. According to the SPES and Population Census in 1935, approximately 95% of
primary school-aged children in Chiba were covered in the SPES datasets.62
Figure B.1 illustrates the raw relationship between schoolchildren’s average height and
weight by their year of birth.63 While there is a general increasing trend in height, reduced
growth is observed in the 1923 birth cohort, especially in the JMA-SIS7 area (Figure B.1a
and B.1b). Regarding the weight of children, reduced growth might be observed in the
1923 birth cohort in the JMA-SIS7 area (Figures B.1c and B.1d).
Figure B.3 shows a map of soil compaction in Chiba prefecture, based on boring data.
In this map, the areas with red (blue) meshes were more (less) likely to be affected by
the earthquake. The interesting fact is that the physical disruption in Figure 2a shows
the opposite distribution of affected households. This fact implies that my key measure
of earthquake stress does not depend on soil compaction, which can correlate with the
potential spatial distribution of industries and agricultural production. As discussed in
the “Great Kanto Earthquake” section, the distribution of physical disruption used is
more likely to be dominated by the distribution of the fault plane (Kamogawa teichi
active fault), which snakes across Awa county.
Finally, I list the sources of the documents used to construct the control variables. The
birth year fetal death rate, defined as the number of fetal deaths per 1,000 births. Data
on fetal deaths as well as live births between 1913–1921 are taken from the Statistical
Report of Chiba Prefecture (SRCP) (1913–1921 editions) published by Chiba prefecture
between 1915 and 1922. The missing values for 1922–1924 and 1926–1929 are linearly
62The number of examinees in the primary schools is taken from Physical Education Bureau, Ministryof Education (1942, p. 7; p. 23). The number of children is taken from Statistics Bureau of the Cabinet(1938, pp. 34–35).
63I used Kaitei shichoson binran (Handbook of Municipalities, revised edition) (Bunmeido 1915, pp. 1–27) to match the schools with the municipalities. The confidence intervals for the non-JMA-SIS7 areas aresystematically smaller than those for the JMA-SIS7 area because of the greater number of observationsin the former areas than in the latter area.
2
115.0115.5116.0116.5117.0117.5118.0118.5119.0119.5120.0120.5121.0121.5122.0122.5123.0123.5124.0124.5125.0125.5126.0
Aver
age
heig
ht o
f boy
s ag
es 6
to 1
1 (c
m)
1920 1921 1922 1923 1924Year of birth
SIS7 areaNon-SIS7 areas95% CI95% CI
(a) Height: PS-boys
115.0115.5116.0116.5117.0117.5118.0118.5119.0119.5120.0120.5121.0121.5122.0122.5123.0123.5124.0124.5125.0125.5126.0
Aver
age
heig
ht o
f girl
s ag
es 6
to 1
1 (c
m)
1920 1921 1922 1923 1924Year of birth
SIS7 areaNon-SIS7 areas95% CI95% CI
(b) Height: PS-girls
20.0020.2520.5020.7521.0021.2521.5021.7522.0022.2522.5022.7523.0023.2523.5023.7524.0024.2524.5024.7525.0025.2525.5025.7526.0026.2526.5026.7527.00
Aver
age
wei
ght o
f boy
s ag
es 6
to 1
1 (k
g)
1920 1921 1922 1923 1924Year of birth
SIS7 areaNon-SIS7 areas95% CI95% CI
(c) Weight: PS-boys
20.0020.2520.5020.7521.0021.2521.5021.7522.0022.2522.5022.7523.0023.2523.5023.7524.0024.2524.5024.7525.0025.2525.5025.7526.00
Aver
age
wei
ght o
f girl
s ag
es 6
to 1
1 (k
g)
1920 1921 1922 1923 1924Year of birth
SIS7 areaNon-SIS7 areas95% CI95% CI
(d) Weight: PS-girls
Figure B.1: Average height (in cm) and weight (in kg) of the primary school (PS)students by area and gender
Notes: Figure B.1a and B.1b show the average height of primary boys and girls, respectively. Figure B.1c and B.1d showthe average weight of primary boys and girls, respectively. The SIS7 area refers to the area extremely affected by theearthquake with the maximum seismic intensity scale (JMA-SIS7), whereas the non-SIS7 area refers to the JMA-SIS5–6areas reported in Table 2. Bootstrap percentile confidence intervals are illustrated. The number of replications is fixed to1,000. Sources: Calculated by the author from the SPES (1925–1935 editions).
interpolated by using the values of 1921, 1925, and 1930. Data on fetal deaths and
live births in 1925 and 1930 are taken from the Shichosonbetsu jinkodotai tokei (Vital
Statistics for Municipalities, 1925 and 1930 editions) published by the Statistics Bureau
of the Cabinet in 1927 and 1933, respectively. Data on rice yield (hectoliter per 0.1 ha)
and the number of doctors are taken from the SRCP (1913–1930 editions) published by
Chiba prefecture between 1915 and 1931. Data on the primary school enrollment rate
of the parental generation are obtained from the SRCP (1897–1912 editions) published
between 1899 and 1914.64 In the sensitivity analysis, I added the municipal-level primary
64 As described by Hijikata (1994, p. 13), the school enrollment rate is defined asSchool Enrollment Rate = 100 × (Students aged 6–13 + Graduates aged 6–13)/Children aged 6–13. TheVital Statistics of 1923 report that the average age at first marriage in Chiba was 25 years, implyingthat parents might have had their first child at 26 on average (Statistics Bureau of the Cabinet 1925a,pp. 18–21). Therefore, one can guess that the average age of parents is roughly 26 years. Given the yearof birth of the sampled children described above and primary school entrance age of 6, one may want touse the enrollment rates from 1894–1909, namely the 20-year lagged rate from the year of birth. Since the
3
Figure B.2: Devastation in Funagata town in Awa countySource: Division of Social Affairs, Chiba Prefecture (1933a).
Figure B.3: Soil compaction in Chiba prefectureNotes: Areas with the red (blue) meshes are more (less) likely to be affected by the earthquake.This map was created based on boring data in 2011 (approximately 50,000 observations). Thespectral intensity values are shown in 250-m meshes. Although the data are investigated in 2011,such wide meshes mean that the map could show nearly the same distribution of soil compaction inthe early 20th century. Source: Webpage of Chiba prefecture http://keihatsu.bousai.pref.chiba.lg.jp/hazadmap/ejk/pdf/yure/yure all.pdf, accessed on December 5, 2019.
school enrollment rate and county-level primary school attendance rate. Data on these
rates are taken from the SRCP (1925–1935 editions) published by the Chiba prefecture
between 1927 and 1937.
Appendix C Empirical Analysis Appendix
C.1 Trends in Height and Weight
Figures C.1a and C.1b present the heights of boys and girls by age and measured year,
respectively. Similarly, Figures C.1c and C.1d present the weights of boys and girls,
respectively. The trends of height and weight show near parallel translation over the
measured year for both genders, suggesting that the trends in child development are
data are severely limited before 1896, however, I use the 1897–1912 editions of the SRCP (17-year laggedrate from the year of birth) to establish the primary school entrance rate of the parental generation.
4
Table B.1: Summary statistics of the control variables
Unit Mean Std. Dev. NBaseline control variables
Fetal death rate in the birth year County-birth year 72.05 15.16 208Rice yield in the birth year County-birth year 30.91 4.36 208Coverage of doctors County-birth year 0.08 0.07 208School enrollment rate of the parental generation County-birth year 87.19 12.31 208
Additional control variablesPrimary school enrollment rate in the measured year Municipal-year 99.62 0.38 3769Primary school attendance rate in the measured year County-year 96.09 0.93 143
Notes: The fetal death rate is the number of still births per 1,000 births. Rice yield is the volume of rice yield per1 hectare. Coverage of doctors is the number of doctors per 100 people. The primary school enrollment rate andattendance rate are the shares of enrolled and attended children relative to total school-aged children. Sources:See Online Appendix B.
similar during my sample periods.
C.2 Robustness Exercises
C.2.1 Including Additional Area Interactions
Table C.1 shows the results from the alternative specification of equation 1 including the
additional interaction terms between the affected cohort dummies and indicator variable
for the JMA-SIS6 area. The estimated coefficients on the interaction terms with respect
to the JMA-SIS6 area are statistically insignificant in all cases. This means that the
effects of the earthquake on the affected cohorts are similar in the JMA-SIS5 and JMA-
SIS6 areas. This makes sense because the physical disruption in the JMA-SIS7 area was
massive relative to the JMA-SIS5 and -6 areas (Table 2).
C.2.2 Alternative Definitions of the Developmental-Stage Age Bin and Rail-
way Disruption
I systematically divided the age bins into 6–8 and 9–11 in the main text because the
observed growth patterns of height are mostly straight in the primary school ages, as
shown in Figure C.1. These bins are also consistent with the fact that the height-for-age
z-scores of −2 in the modern WHO growth reference show similar growth curves to those
children (de Onis et al. 2007). In Table C.2, I check the sensitivity to this definition of
developmental-stage age bins by slightly changing the bins to 6–9 and 8–11. The estimated
coefficients are similar to the results reported in Table 3, supporting the robustness of my
baseline results.
In Table C.3, I check the sensitivity of my results from equation 3 using the same
sample used in Panel A of Table 5 but changing the radius to a 1.5 km range. In Panels
A and B of Table C.3, I set Rail in equation 3 as an indicator variable for municipalities
within 8.5 km and 11.5 km of any municipality including either the Noda line or the
Kururi line, respectively. The results are largely unchanged from those reported in Panel
A of Table 5, supporting the robustness of my baseline results.
5
107108109110111112113114115116117118119120121122123124125126127128129130131132133
Hei
ght (
in c
entim
eter
s)
6 7 8 9 10 11Age
1925 1926 1927 19281929 1930 1931 19321933 1934 1935
(a) Boys’ height (6–11)
107108109110111112113114115116117118119120121122123124125126127128129130131132133
Hei
ght (
in c
entim
eter
s)
6 7 8 9 10 11Age
1925 1926 1927 19281929 1930 1931 19321933 1934 1935
(b) Girls’ height (6–11)
151617181920212223242526272829303132
Wei
ght (
in k
ilogr
ams)
6 7 8 9 10 11Age
1925 1926 1927 19281929 1930 1931 19321933 1934 1935
(c) Boys’ weight (6–11)
151617181920212223242526272829303132
Wei
ght (
in k
ilogr
ams)
6 7 8 9 10 11Age
1925 1926 1927 19281929 1930 1931 19321933 1934 1935
(d) Girls’ weight (6–11)
Figure C.1: Trends in the Height and Weight of ChildrenNote: Each figure shows the average height or weight of children in each measured year. Sources: Calculatedby author from the SPES (1925–1935 editions).
6
Table C.1: Effects of fetal earthquake exposure on height and weight:Including additional area interaction terms
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A: Height1923 birth cohort −0.247** −0.219* −0.241** −0.253***
[0.010] [0.028] [0.010] [0.003]1923 birth cohort × SIS6 0.240 0.158 0.043 0.139
[0.158] [0.474] [0.746] [0.371]1923 birth cohort × SIS7 0.119 0.132 −0.290* −0.584***
[0.062] [0.752] [0.034] [0.005]1924 birth cohort −0.113 −0.092 0.132 −0.106
[0.132] [0.364] [0.222] [0.281]1924 birth cohort × SIS6 0.139 0.042 −0.109 −0.007
[0.210] [0.822] [0.402] [0.965]1924 birth cohort × SIS7 0.071 0.032 0.168 0.356
[0.746] [1.000] [0.142] [0.423]Panel B: Weight1923 birth cohort −0.068** −0.055 −0.111** −0.161***
[0.006] [0.136] [0.008] [0.003]1923 birth cohort × SIS6 0.070 0.094 0.082 0.207
[0.142] [0.220] [0.358] [0.235]1923 birth cohort × SIS7 0.013 0.116 −0.257*** −0.273*
[0.646] [0.404] [0.004] [0.015]1924 birth cohort −0.011 −0.044 0.007 −0.051
[0.658] [0.420] [0.866] [0.131]1924 birth cohort × SIS6 0.057 0.100 0.016 −0.005
[0.378] [0.398] [0.834] [0.967]1924 birth cohort × SIS7 −0.021 0.009 −0.106 0.035
[1.000] [1.000] [0.224] [0.885]
***, **, and * represent statistical significance at the 0.5%, 1%, and 5% levels based on the p-values from thewild cluster bootstrap resampling method in brackets, respectively. The data are clustered at the 13-countylevel in the bootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: The numbers of observations in columns (1)–(4) are 14, 139, 14, 133, 14, 145, and 14, 138, respectively.All the regressions include controls for the rice yield in the birth year; fetal death rate in the birth year;school enrollment rate of the parental generation; school-age-specific fixed effects; and year fixed effects. Thenull hypothesis of no joint significance of the control variables is rejected based on the F -test at the 0.1%level in all the specifications.
7
Table C.2: Effects of fetal earthquake exposure on height and weight:Alternative definitions of the developmental-stage age bins
Boys Girls
(1) (2) (3) (4)Ages 6–9 Ages 8–11 Ages 6–9 Ages 8–11
Panel A: Effects on height1923 birth cohort −0.180* −0.173* −0.191*** −0.214***
[0.028] [0.025] [0.004] [0.002]1923 birth cohort × SIS7 0.075 0.134 −0.268* −0.596*
[0.218] [0.735] [0.022] [0.032]1924 birth cohort −0.030 −0.077 0.092 −0.080
[0.512] [0.347] [0.244] [0.234]1924 birth cohort × SIS7 −0.017 −0.031 0.266 0.349
[0.936] [0.100] [0.124] [0.310]Panel B: Effects on weight1923 birth cohort −0.043* −0.042 −0.086** −0.113***
[0.038] [0.207] [0.006] [0.002]1923 birth cohort × SIS7 0.027 0.093 −0.267* −0.264*
[0.200] [0.331] [0.016] [0.048]1924 birth cohort −0.005 −0.026 0.010 −0.046
[0.778] [0.519] [0.784] [0.088]1924 birth cohort × SIS7 −0.043 −0.028 −0.095 0.010
[0.730] [0.991] [0.126] [0.976]
***, **, and * represent statistical significance at the 0.5%, 1%, and 5% levels based on the p-values from thewild cluster bootstrap resampling method in brackets, respectively. The data are clustered at the 13-countylevel in the bootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: The numbers of observations in columns (1)–(4) are 18, 851, 18, 845, 18, 857, and 18, 848, respectively.All the regressions include controls for the rice yield in the birth year; fetal death rate in the birth year;school enrollment rate of the parental generation; school-age-specific fixed effects; and year fixed effects.The null hypothesis of no joint significance of the control variables is rejected based on the F -test at the0.1% level in all the specifications.
8
Table C.3: Effects of fetal earthquake exposure on height and weight in the limitedJMA-SIS5 area: Alternative definitions of railway disruption
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A-1: Effects on height1923 birth cohort −0.182* −0.184 −0.242*** −0.242**
[0.030] [0.162] [0.004] [0.007]1923 birth cohort × Railway disruption (8.5km) −0.078 0.482 0.215 0.705
[0.966] [0.546] [0.156] [0.529]Panel A-2: Effects on weight1923 birth cohort −0.048 −0.020 −0.104** −0.125*
[0.070] [0.922] [0.008] [0.013]1923 birth cohort × Railway disruption (8.5km) 0.050 0.017 0.084 0.141
[0.626] [0.904] [0.196] [0.737]
Panel B-1: Effects on height1923 birth cohort −0.203* −0.161 −0.247** −0.220*
[0.022] [0.218] [0.008] [0.021]1923 birth cohort × Railway disruption (11.5km) 0.161 0.103 0.206 0.274
[0.758] [0.680] [0.090] [0.773]Panel B-2: Effects on weight1923 birth cohort −0.047* −0.004 −0.104** −0.110*
[0.036] [1.000] [0.008] [0.039]1923 birth cohort × Railway disruption (11.5km) 0.031 −0.149 0.061 −0.054
[0.956] [0.256] [0.366] [0.829]
***, **, and * represent statistical significance at the 0.5%, 1%, and 5% levels based on the p-values from the wildcluster bootstrap resampling method in brackets, respectively. The data are clustered at the 13-county level in thebootstrap procedure. The number of replications is fixed to 1,000 for all specifications.Notes: Samples include municipalities receiving no disaster relief in the SIS5 area. The numbers of observations incolumns (1)–(4) are 9352, 9348, 9353, and 9347, respectively. Estimated coefficients on the 1924 birth cohort dummyand its area interaction term are included in all regressions but are not reported as those estimates are statisticallyinsignificant in all specifications. All the regressions include controls for the rice yield in the birth year; fetal deathrate in the birth year; school enrollment rate of the parental generation; school-age-specific fixed effects; and year fixedeffects. The null hypothesis of no joint significance of the control variables is rejected based on the F -test at the 0.1%level in all the specifications.
C.2.3 Including Additional Control Variables
Table 6 presents the results from specifications including additional control variables in
equation 1. In the same way, I show the results from specifications including the same
additional control variables in equations 2 and 3 in Tables C.4 and C.5, respectively. In
Table C.4, I report the results for the boys aged 6–8 given that all the estimates for the
other subsamples are statistically insignificant (see Table 4). As shown in Table C.4, the
results are largely unchanged from those reported in Tables 4.
C.2.4 Potential Influence of Omitted Variables
To investigate the potential influence of omitted variable bias, I present the results when
using the method proposed by Oster (2019). The maximum R-squared from a hypo-
thetical regression of either height or weight on both the observed and the unobserved
9
variables is assumed to be 1.3 × R-squared as recommended. The value for the relative
degree of selection on the observed and unobserved variables is also assumed to be one,
implying that selection into the unobservables is equal to selection into the observables.
Table C.6 shows the main results of Table 3 but adds the suggested bounds in parentheses.
Despite the conservative setting, the estimated bounds suggest weak attenuation bias in
my estimates; nonetheless, they are still close to the estimates in Table 3.
C.2.5 Time Series Plots of the Birth and Mortality Measures
404550556065707580859095
100105110
Num
ber o
f dea
th b
irths
per
100
0 bi
rths
1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929
(a) Fetal death rate
120125130135140145150155160165170175180185190195200205210215220
Num
ber o
f liv
e bi
rths
per 1
000
birth
s
1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929
(b) Infant mortality rate
252627282930313233343536373839404142434445
Num
ber o
f liv
e bi
rths
per 1
000
peop
le
1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929
(c) Crude birth rate
Figure C.2: Fetal death, infant mortality, and fertility rates in Chiba prefectureNotes: The fetal death rate is the number of deaths births per 1000 births. The infant mortality rate is the number ofinfant deaths (i.e., deaths within 12 months of birth) per 1000 live births. The crude birth rate is the number of live birthsper 1000 people. Sources: Statistics Bureau of the Cabinet (1920b–1930b).
Figure C.2 shows the time series plots of the fetal death, infant mortality, and crude
death rates between 1917 and 1929. The data on these rates are obtained from the
1917–1929 editions of the Vital Statistics of Japan (Statistics Bureau of the Cabinet
1920b–1930b). Figure C.2a and Figure C.2b show the decreasing trend of the risk of
fetal deaths and infant mortality during this period, as also shown by Schneider and
Ogasawara (2018). Although the infant mortality rate seems to increase slightly in 1923,
the increment was not greater as that observed in 1918, suggesting it is not systematic
mortality selection due to the earthquake. Figure C.2c shows no clear trend in the fertility
rate and thus does not support the evidence on fertility selection during and after the
earthquake of 1923.
C.2.6 Testing the Cohort Effects of Surrounding Cohorts
If the estimated stunting effects reported in the main text do not simply capture certain
trends, the mean heights and weights of children born in the surrounding years should
be similar when I exclude the 1923–24 birth cohorts from the sample. Each estimate of
the cohort effect in Table C.7 is obtained from a regression for the sample excluding the
1923–24 birth cohorts, which means that I run 80 regressions to obtain these estimates.
As shown, the estimated coefficients are not statistically significant in most cases.
10
Tab
leC
.4:
Eff
ects
ofdis
aste
rre
lief
onth
ehei
ght
and
wei
ght
ofb
oys
aged
6–8:
Rob
ust
nes
sof
incl
udin
gad
dit
ional
contr
olva
riab
les
Hei
ght
Wei
ght
(1)
Tot
al(2
)F
ood×
(3)
Shel
ter
(4)
Med
ical
(5)
Tot
al(6
)F
ood×
(7)
Shel
ter
(8)
Med
ical
Pan
elA
:Sch
ool
enro
llm
ent
and
atte
ndan
cera
tes
Exp
ense
s×
1923
bir
thco
hor
t0.
114∗
0.16
6∗0.
383∗
41.4
13∗
0.02
9∗0.
043∗
0.09
8∗16
.016
∗[0
.024
][0
.010
][0
.036
][0
.044
][0
.016
][0
.020
][0
.018
][0
.022
]P
rim
ary
schoo
len
rollm
ent
rate
-0.0
81-0
.081
-0.0
81-0
.081
-0.0
19-0
.019
-0.0
19-0
.019
[0.1
22]
[0.1
22]
[0.1
22]
[0.1
22]
[0.5
50]
[0.5
50]
[0.5
52]
[0.5
40]
Pri
mar
ysc
hoo
lat
tendan
cera
te-0
.026
-0.0
26-0
.026
-0.0
25-0
.004
-0.0
04-0
.004
-0.0
03[0
.472
][0
.470
][0
.478
][0
.492
][0
.792
][0
.792
][0
.794
][0
.814
]
Pan
elB
:W
arti
me
and
flu
pan
dem
iceff
ects
Exp
ense
s×
1923
bir
thco
hor
t0.
116∗
0.16
8∗0.
388∗
42.3
71∗
0.02
9∗0.
043∗
0.09
8∗16
.095
∗[0
.014
][0
.010
][0
.032
][0
.038
][0
.020
][0
.026
][0
.024
][0
.020
]W
arti
me
bir
thco
hor
ts0.
027
0.02
70.
027
0.02
7-0
.040
-0.0
40-0
.041
-0.0
41[0
.536
][0
.538
][0
.536
][0
.538
][0
.078
][0
.080
][0
.076
][0
.080
]P
andem
icin
fluen
zabir
thco
hor
ts-0
.086
-0.0
86-0
.086
-0.0
86-0
.014
-0.0
14-0
.012
-0.0
13[0
.314
][0
.314
][0
.314
][0
.318
][0
.628
][0
.620
][0
.652
][0
.642
]
*re
pre
sents
stati
stic
al
sign
ifica
nce
at
the
5%
level
base
don
thep-v
alu
esfr
om
the
wild
clu
ster
boots
trap
resa
mp
lin
gm
eth
od
inb
rack
ets.
Th
ed
ata
are
clu
ster
edat
the
13-c
ou
nty
level
inth
eb
oots
trap
pro
ced
ure
.T
he
nu
mb
erof
rep
lica
tion
sis
fixed
to1,0
00
for
all
the
spec
ifica
tion
s.N
ote
s:T
he
nu
mb
erof
ob
serv
ati
on
sin
colu
mn
s(1
)–(4
)an
d(5
)–(8
)are
14,1
45
an
d14,1
38,
resp
ecti
vel
y.T
he
esti
mate
dco
effici
ent
onI(YOB=1923)×Relief
ineq
uati
on
2is
rep
ort
edin
the
tab
le.
Th
ees
tim
ate
dco
effici
ent
onI(YOB=1924)×Relief
ineq
uati
on
2is
incl
ud
edin
all
the
regre
ssio
ns
bu
tn
ot
rep
ort
edas
its
esti
mate
isst
ati
stic
ally
insi
gn
ifica
nt
inall
the
spec
ifica
tion
s.A
llth
ere
gre
ssio
ns
incl
ud
eco
ntr
ols
for
the
rice
yie
ldin
the
bir
thyea
r;fe
tal
dea
thra
tein
the
bir
thyea
r;sc
hool
enro
llm
ent
rate
of
the
pare
nta
lgen
erati
on
;sc
hool-
age-
spec
ific
fixed
effec
ts;
an
dyea
rfi
xed
effec
ts.
Th
enu
llhyp
oth
esis
of
no
join
tsi
gn
ifica
nce
of
the
contr
ol
vari
ab
les
isre
ject
edb
ase
don
theF
-tes
tat
the
0.1
%le
vel
inall
the
spec
ifica
tion
s.
11
Table C.5: Effects of fetal earthquake exposure on height and weight in the limitedJMA-SIS5 area: Robustness of including additional variables in the railway disruption
regressions
Boys Girls
(1) (2) (3) (4)Panel A Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11Panel A-1: Effects on height1923 birth cohort −0.205* −0.167 −0.246** −0.225*
[0.024] [0.216] [0.008] [0.019]1923 birth cohort × Railway disruption 0.184 0.181 0.235 0.361
[0.724] [0.598] [0.086] [0.769]Primary school enrollment rate −0.031 0.059 −0.053 −0.049
[0.522] [0.654] [0.466] [0.663]Primary school attendance rate −0.039 0.012 0.068 −0.063
[0.530] [0.908] [0.054] [0.103]Panel A-2: Effects on weight1923 birth cohort −0.051* −0.008 −0.102** −0.110*
[0.028] [0.994] [0.008] [0.039]1923 birth cohort × Railway disruption 0.081 −0.113 0.057 −0.057
[0.722] [0.496] [0.416] [0.815]Primary school enrollment rate −0.017 −0.035 −0.000 −0.069*
[0.560] [0.326] [0.972] [0.043]Primary school attendance rate 0.002 0.051 0.021 −0.005
[0.946] [0.230] [0.300] [0.859]Boys Girls
(1) (2) (3) (4)Panel B Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11Panel B-1: Effects on height1923 birth cohort −0.197* −0.162 −0.241** −0.233*
[0.018] [0.222] [0.010] [0.015]1923 birth cohort × Railway disruption 0.181 0.187 0.244 0.358
[0.728] [0.588] [0.056] [0.767]Wartime birth cohorts 0.048 0.064 0.208 0.096
[0.552] [0.474] [0.162] [0.253]Pandemic influenza birth cohorts −0.091 −0.070 −0.077 0.167
[0.480] [0.510] [0.528] [0.131]Panel B-1: Effects on weight1923 birth cohort −0.051* −0.006 −0.101* −0.117*
[0.022] [0.990] [0.010] [0.035]1923 birth cohort × Railway disruption 0.081 −0.116 0.060 −0.062
[0.704] [0.480] [0.390] [0.793]Wartime birth cohorts −0.039 0.066 0.001 0.037
[0.310] [0.094] [0.940] [0.635]Pandemic influenza birth cohorts −0.022 −0.015 −0.034 0.149
[0.634] [0.650] [0.520] [0.105]
** and * represent statistical significance at the 1% and 5% levels based on the p-values from the wild clusterbootstrap resampling method in brackets, respectively. The data are clustered at the 13-county level in thebootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: Samples include municipalities receiving no disaster relief in the SIS5 area. The numbers of observations incolumns (1)–(4) are 9352, 9348, 9353, and 9347, respectively. The estimated coefficients on the 1924 birth cohortdummy and its area interaction terms are not reported because those estimates are statistically insignificant inmost specifications. All the regressions include controls for the rice yield in the birth year; fetal death rate inthe birth year; school enrollment rate of the parental generation; school-age-specific fixed effects; and year fixedeffects. The null hypothesis of no joint significance of the control variables is rejected based on the F -test at the0.1% level in all the specifications.
12
Table C.6: Effects of fetal earthquake exposure on height and weight:Sensitivity to the potential omitted variables
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A: Effects on height1923 birth cohort −0.213* −0.196* −0.234* −0.233***
(−0.223) (−0.420) (−0.167) (−0.487)[0.020] [0.044] [0.010] [0.003]
1923 birth cohort × SIS7 0.085 0.109 −0.296** −0.604***(0.155) (0.036) (−0.182) (−0.689)[0.250] [0.798] [0.008] [0.005]
Panel B: Effects on weight1923 birth cohort −0.058* −0.041 −0.099** −0.131**
(−0.016) (0.163) (−0.054) (−0.259)[0.016] [0.314] [0.008] [0.005]
1923 birth cohort × SIS7 0.003 0.102 −0.269*** −0.303**(0.044) (0.043) (−0.220) (−0.347)[0.938] [0.438] [0.002] [0.009]
This table shows the main estimates reported in Table 3 but also presents Oster’s (2019) suggested boundsin parentheses. ***, **, and * represent statistical significance at the 0.5%, 1%, and 5% levels based onthe p-values from the wild cluster bootstrap resampling method in brackets, respectively. The data areclustered at the 13-county level in the bootstrap procedure. The number of replications is fixed to 1,000for all the specifications.Notes: The numbers of observations in columns (1)–(4) are 14, 139, 14, 133,14, 145, and 14, 138, respectively.All the regressions include controls for the rice yield in the birth year; fetal death rate in the birth year;school enrollment rate of the parental generation; school-age-specific fixed effects; and year fixed effects.The null hypothesis of no joint significance of the control variables is rejected based on the F -test at the0.1% level in all the specifications.
13
Table C.7: Effects of fetal earthquake exposure on height and weight: Testing thepotential impacts on surrounding cohorts excluding exposed cohorts
Boys Girls
(1) (2) (3) (4)Ages 6–8 Ages 9–11 Ages 6–8 Ages 9–11
Panel A: Effects on height1918 birth cohort −0.014 0.061 0.006 0.045
[0.802] [0.464] [0.949] [0.658]1919 birth cohort 0.017 −0.029 0.052 0.117
[0.676] [0.648] [0.371] [0.172]1920 birth cohort −0.108* 0.043] −0.118 −0.040
[0.048] [0.496] [0.129] [0.792]1921 birth cohort 0.024 0.024 0.021 −0.038
[0.680] [0.862] [0.737] [0.690]1922 birth cohort 0.076 −0.066 0.029 −0.090
[0.282] [0.526] [0.787] [0.392]1925 birth cohort 0.043 −0.073 0.101 −0.080
[0.542] [0.478] [0.147] [0.520]Panel B: Effects on weight1918 birth cohort −0.046 0.004 −0.055 0.015
[0.160] [0.932] [0.239] [0.796]1919 birth cohort −0.005 0.012 −0.006 0.022
[0.728] [0.492] [0.911] [0.650]1920 birth cohort −0.022 −0.005 −0.011 0.076
[0.080] [0.918] [0.669] [0.362]1921 birth cohort 0.018 0.003 0.034 −0.013
[0.450] [0.928] [0.239] [0.758]1922 birth cohort 0.035 −0.038 −0.012 −0.112
[0.388] [0.534] [0.647] [0.204]1925 birth cohort −0.009 0.039 0.028 0.073
[0.790] [0.428] [0.533] [0.160]
* represents statistical significance at the 5% level based on the p-values from the wild clusterbootstrap resampling method in brackets. The data are clustered at the 13-county level in thebootstrap procedure. The number of replications is fixed to 1,000 for all the specifications.Notes: Each estimate of the cohort effect is obtained from a regression for the sample excludingthe 1923–24 birth cohorts. The numbers of observations for each regression reported in columns(1)–(4) are 11, 548, 11, 580, 11, 555, and 11, 586, respectively. All the regressions include controlsfor the rice yield in the birth year; fetal death rate in the birth year; school enrollment rate of theparental generation; school-age-specific fixed effects; and year fixed effects. The birth cohortsbefore 1918 and after 1926 are not included in these analyses because they include individualsaged under four years, which is insufficient to calculate the mean birth cohort effects. The nullhypothesis of no joint significance of the control variables is rejected based on the F -test at the0.1% level in all the specifications.
14
References, Documents, Statistical Reports, and Database
[1] Bunmeido. Kaitei shichoson binran (Handbook of municipalities, revised edition). [inJapanese] Osaka: Bunmeido, 1915.
[2] Chiba prefecture. Chibaken tokeisho (The statistical report of Chiba prefecture, 1897–1935edition). [in Japanese] Chiba: Chiba prefecture, 1899–1937.
[3] Chiba Prefecture. Seitojido shintaikensa tokei (Statistics of physical examination for stu-dents, 1925–1935 editions). [in Japanese] Chiba: Chiba Prefecture, 1925–1936.
[4] Chiba prefecture (webpage). Available online at the Chiba prefecture (http://keihatsu.bousai.pref.chiba.lg.jp/hazadmap/ejk/pdf/yure/yure all.pdf) (last accessed on December 5,2019).
[5] Division of Social Affairs, Chiba Prefecture. Taishodaishinsai no kaiko to sonofukko (Rem-iniscences and reconstructions of the Taisho earthquake, volume 1). [in Japanese] Chiba:Division of Social Affairs, Chiba Prefecture, 1933a.
[6] Division of Social Affairs, Chiba Prefecture. Taishodaishinsai no kaiko to sonofukko (Rem-iniscences and reconstructions of the Taisho earthquake, volume 2). [in Japanese] Chiba:Division of Social Affairs, Chiba Prefecture, 1933b.
[7] Hijikata, Sonoko. Kindainihon no gakko to chiikishakai (school and community in prewarJapan). [in Japanese] Tokyo: University of Tokyo Press, 1994.
[8] Ministry of Educations. Nihonteikoku monbusho nenpo (The annual report of the Japaneseimperial ministry of education, 32-53 editions). [in Japanese] Tokyo: Ministry of Education,1906–1929.
[9] Ministry of Land, Infrastructure, Transport and Tourism of Japan (database). Availableonline at the Ministry of Land, Infrastructure, Transport and Tourism of Japan (http://nlftp.mlit.go.jp/ksj/jpgis/datalist/KsjTmplt-N03.html) (last accessed on May 22, 2017).
[10] Oster, Emily. Unobservable selection and coefficient stability: Theory and evidence. Journalof Business & Economic Statistics 37, no.2 (2019): 187–204.
[11] Physical Education Bureau, Ministry of Education. Seitojido shintaikensa tokei (Statis-tics of school physical examination, 1938 editions). [in Japanese] Tokyo: Physical EducationBureau, Ministry of Education, 1942.
[12] Schneider, Eric B, and Kota, Ogasawara. Disease and child growth in industrialising Japan:Critical windows and the growth pattern, 1917–39. Explorations in Economic History 69(2018), 64–80.
[13] Statistics Bureau of the Cabinet. Jinkodotai tokei (The vital statistics of the empire ofJapan, 1916–1933 editions). [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1919–1934.
[14] Statistics Bureau of the Cabinet. Nihonteikoku tokeinenkan (Statistical yearbook of theJapanese Empire, vol. 36–53). [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1918–1934.
[15] Statistics Bureau of the Cabinet. Nihonteikoku shiintokei (Statistics of causes of death of theempire of Japan, 1911–1927 editions). [in Japanese] Tokyo: Statistics Bureau of the Cabinet,1915a–1928a.
15
[16] Statistics Bureau of the Cabinet. Nihonteikoku jinkodotai tokei (The vital statistics of theempire of Japan, 1916–1933 editions). [in Japanese] Tokyo: Statistics Bureau of the Cabinet,1919b–1934b.
[17] Statistics Bureau of the Cabinet. Shichoson jinkodotai tokei (The vital statistics for munic-ipalities, 1925 edition). [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1927c.
[18] Statistics Bureau of the Cabinet. Shichoson jinkodotai tokei (The vital statistics for munic-ipalities, 1930 edition). [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1932c.
[19] Statistics Bureau of the Cabinet. Nihonteikoku tokeinenkan (Statistical yearbook of theJapanese Empire, vol. 36–53). [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1918d–1934d.
[20] Statistics Bureau of the Cabinet. Showajyunen kokuseichosa hokoku (Population census ofJapan, vol.2 Chiba prefecture part, 1935 edition). [in Japanese] Tokyo: Statistics Bureau ofthe Cabinet, 1938.
16