Policy Research Working Paper 8927 Unfortunate Moms and Unfortunate Children Impact of the Nepali Civil War on Women’s Stature and Intergenerational Health Lokendra Phadera Poverty and Equity Global Practice June 2019 Policy Research Working Paper 8927 Abstract This paper analyzes the long-term health impacts of Nepal’s is detrimental to her child’s health. Mothers exposed to 1996–2006 civil conflict. It exploits the heterogeneity in conflict during their childhood have more children and conflict intensity across villages and birth cohorts to docu- live in less wealthy households, likely reducing their ability ment long-term health and intergenerational impacts. The to invest during their children’s critical period of physical analysis finds that childhood exposure to conflict and, in development. The finding points to a potential trade-off particular, exposure starting in infancy, negatively impacts between the quantity and quality of children. The paper attained adult height. Each additional month of exposure uses information on monthly conflict incidents at the vil- decreases a women’s adult height by 1.36 millimeters. The lage level, which allows identifying identify the effects of impacts are not limited to first-generation. The analysis also exposure to conflict more accurately than prior studies. finds that a mother’s exposure to conflict in her childhood This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at lphadera@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Unfortunate Moms and Unfortunate Children: Impact of the Nepali Civil War on Women's Stature and Intergenerational ∗ Health † Lokendra Phadera Poverty and Equity Group, World Bank Keywords: Civil Conict; Adult Height; Intergenerational Health; Nepal JEL Classication: I12; J13; O12 ∗ I am grateful for guidance from Benjamin Crost, Richard Akresh, Rebecca Thornton, Hope Michelson, Mary Arends-Kuenning, and Alex Winter-Nelson. Mateus Nogueira Meirelles De Souza and Bhim B. Phadera provided excellent research assistance. I am thankful for comments from Aine McCarthy, Derek Headey, Jennifer Alix-Garcia, the participants of the 16th MWIEDC, and the IFPRI's PHND Brown Bag Seminar. † E-mail: lphadera@worldbank.org 1 Introduction The environmental conditions experienced in utero and in early life have profound inuence on human biology and long-term health (Golden, 1994; Martorell et al., 1994; Forsdahl, 1977; Barker, 1992; Bateson et al., 2004; Gluckman et al., 2007, 2008). Similarly, early life conditions have lasting and signicant impacts on adult economic outcomes (see reviews by Strauss and Thomas, 2008; Cur- rie and Vogl, 2013). These ndings are highly relevant in the context of civil conicts, which cause considerable human suering, death and property destruction. Despite the potential for conict to contribute to lasting health impacts, the empirical evidence on long-term and intergenerational eects of conict on health is limited. 1 Lack of conict data at detailed geographic scale has made it dicult to measure precisely the consequences of conicts. In this paper, I investigate the impacts of early childhood exposure to Nepal's 1996-2006 civil conict on women's nal adult height and on second-generation health using the 2016 Nepal De- mographic Health Survey (NDHS) and village-level variation in conict intensity. By exploiting detailed geographic information on conict incidents (village-level conict intensity), I am able to identify the eects of exposure to conict more accurately than previous research. 2 The literature on the consequences of war, including in Nepal 3 has thus far mostly focused on conict variation at 1 Previous studies have documented long-run eects at the cross-country level, suggesting conicts have large and negative immediate eects on overall economic growth. However, the recovery to equilibrium is rapid (see review by Blattman and Miguel, 2010). Correlated with war exposure, the literature has extensively documented long-run eects of exposures to stress on mental health (Persson and Rossin-Slater, 2018), height and diseases (Bozzoli et al., 2009), birth weight (Camacho, 2008; Quintana-Domeque and Ródenas-Serrano, 2017), and education and socioeconomic status (Almond, 2006). Previous research on conict has mostly focused on human capital accumulation during or shortly after conict (Akresh and De Walque, 2008; Bundervoet et al., 2009; Valente, 2014; Akresh et al., 2014; Pivovarova and Swee, 2015). A few recent studies have focused on the long-term impact of conict on human capital accumulation, with some nding no eect (Miguel and Roland, 2011) and others (Akresh et al., 2012; León, 2012; Justino et al., 2014; Palmer et al., 2016; Akbulut-Yuksel, 2017) nding signicant negative impacts. Domingues and Barre (2013) nds that the exposure to Mozambican civil war during childhood had adverse eect on women's adult height-for-age z-score. 2 A common approach in the literature is to exploit conict variation at a broader regional level, which has the potential to misclassify one's exposure to conict and induce measurement errors. For instance, using detailed GPS data on distance between survey villages and conict sites, Akresh et al. (2014) show that substantial number of households in Eritrea were misclassied as being in non-conict region in the Akresh et al. (2012) paper that used less precise regional conict data and there are signicant dierences in the estimated eects of the Eritrea-Ethiopia conict between the two measures of the conict: eects are 87-188% larger using the GPS based measure than the regional based measure of the conict. 3 Previous studies of Nepal's civil war are mostly focused on understanding the causes of the war: geographical terrain (Murshed and Gates, 2005; Bohara et al., 2006; Do and Iyer, 2010; Menon and van der Meulen Rodgers, 2015), economic exclusion and poverty (Murshed and Gates, 2005; Onesto, 2005; Do and Iyer, 2010), inequality (Murshed and Gates, 2005; Macours, 2011; Nepal et al., 2011), and lack of political representation (Murshed and Gates, 2005; Bohara et al., 2006; Macours, 2011). The little evidence documenting the consequences of the war, thus far, is focused on district level disaggregation and results are mixed - little to zero impact on human capital accumulation (Valente, 2014; Pivovarova and Swee, 2015), increased miscarriages (Valente, 2015) and positive impact 1 a broader regional level (see Valente, 2014, 2015; Pivovarova and Swee, 2015; Akresh et al., 2012). Moreover, this study extends the literature on legacies of war by documenting the long-term eect of exposure to conict on women's nal adult height. Along with Akresh et al. (2018), which examines the impacts of the Biafran war using the variation in exposure to the war by ethnicity, this is among the rst papers to document the intergenerational transmission of the impact of early childhood conict exposure on second generation health. Nepal experienced a decade-long violent civil conict between 1996 and 2006, which resulted in more than 13,000 fatalities, destroyed considerable infrastructure, hindered delivery of basic services, and generated pervasive and strong feelings of fear, insecurity, and stress among its citizens. I use the Informal Sector Service Center's (INSEC) records of conict victims to create a casualty-level data set with exact geographical locations (villages) and dates of incidents. I merge the village- level conict intensity data with the 2016 Nepal Demographic Health Survey (NDHS), which is a nationally representative survey of the female population aged 15 to 49. I limit the analytical sample to women who were either born before or who were in utero at the start of the war in February 1996 to assess the lasting impact of conict on height. In 2016, these women were old enough (20 to 49) to be sampled in the ever-married NDHS 2016 women sample and had documented survey responses about their children that can be used for second-generation analysis. Limiting the sample to women born before the conict's start date also reduces potential confoundedness through selective fertility and migration (further discussed in empirical strategy section 4). I rely on the biomedical literature that height development in human beings is characterized by rapid growth during the rst three years of life, followed by lower level of constant growth and then a secondary growth spurt during adolescence (Figure A.5), and classify women into three treatment cohorts, namely: ages 0 to 3, 4 to 8, and 9 to 15 at the start of the war in February 1996. While women in age cohort 0 to 3 would have been exposed to conict throughout their entire three stages of growth, the age cohorts 4 to 8 and 9 to 15 would have been exposed only in their latter stages of growth. I dene women ages 16 to 21 in 1996 as a comparison group because they would have passed their pubertal ages and attained full adult height by the time of the conict's start in 1996. I also include women ages 22 to 29 as a on women's employment (Menon and van der Meulen Rodgers, 2015). Libois (2016), on the other hand, using conict measurement at more detailed geographical area (distance from the conict sites) nds signicant negative immediate impact on consumption and income. Failure to capture the substantial conict heterogeneity across villages within district, thus, may explain the little or no impact of conict at the district level. 2 second control group to validate parallel-trend dynamics in dierence-in-dierence specication. This research makes two primary contributions to the literature on the legacies of war. First, using the variation in exposure to conict, as measured by months of war, by birth cohort and village of residence, I nd that conict and, in particular, exposure starting in infancy, has a highly signicant and negative impact on women's nal adult height. Findings are robust across model specications and measures of conict. In validating the dierence-in-dierences estimation strategy used, I nd no evidence of presence of non-parallel dynamics nor of selective migration and fertility. These results are important given increasing evidence of the lasting impacts of stunting and slow growth in height early in life on overall physical, biological and cognitive development, school achievement, economic productivity and maternal reproductive outcomes (see review by Dewey and Begum, 2011). Additionally, given the established literature on the existence of a height premium (Persico et al., 2004; Case and Paxson, 2008; Vogl, 2014; Bargain and Zeidan, 2017), these results have important economic importance. A second contribution: I nd that the mothers' exposure to conict is detrimental for their children's health, especially child weight as measured by weight-for-height, weight-for-age and BMI z-scores. As with the rst generation impacts, the second generation results are robust to alternative measures of conict intensity, including the one where I dene a mother's exposure to conict at her district of birth. I nd strong evidence that the women exposed to conict during childhood have more children and live in poorer households as adults. The combination of these two factors may result in meaningful decreases in parental ability to invest in children. Increased fertility, and, hence, burden of feeding more mouths may negatively aect children's health. However, other unobserved factors such as stress and genomic changes may also inuence the intergenerational transmission. This paper links to three important strands of economics literature. First, the established literature of in utero and early life shocks on adult outcomes (see review by Almond and Currie, 2011) and that insucient or lack of parental investment during critical periods of child development can lead to irreversible damage (Cunha and Heckman, 2007). Second, the paper is related to the literature providing strong positive intergenerational human capital transmission (Currie and Moretti, 2003, 2007; Almond et al., 2012; Justino et al., 2014; Bhalotra and Rawlings, 2013). Third, combining the negative second generation health impacts and nding indicating increased fertility among the rst generation point to the Becker and Lewis's (1973) child quality-quantity trade-o 3 hypothesis. 4 The reminder of the paper is organized as follows. Section 2 provides the background on Nepal's civil conict, major events that helped shape the war, and the physical and economic costs of the war. Conict intensity from the INSEC's database and individual data from the 2016 NDHS are discussed in section 3. Section 4 presents empirical strategies for evaluating both the rst and second-generation impacts. Empirical results along with identication validation and potential mechanisms for intergenerational transmission are presented in section 5 and concluding remarks are presented in section 6. 2 Background Nepal is a landlocked country between India and China. Because of its highly mountainous and rugged terrain and lack of adequate infrastructure and economic development, most parts of the country remain remote, and access to basic services remains unattainable to many. With two- thirds of its 30 million inhabitants (estimated as of July 2017) relying on agriculture and a quarter living under the poverty line, Nepal is one of the least developed nations in the world (CIA World Factbook, 2019) and is among the lowest in health, sanitation, primary education and electricity in South Asia. Figure A.1 shows the administrative divisions of Nepal before the implementation of a new constitution in 2015. The country was divided into ve geographically homogeneous development regions, which were further divided to form 75 districts. Districts were further divided into rural (village development committee, VDC) and urban (municipalities) areas, which were the lowest level of administrative units. At the time of the 2011 population census, Nepal consisted of 3,914 VDCs and 58 municipalities (Central Bureau of Statistics, 2012). I calculate conict intensity at the level of these 3,972 local administrative areas. 5 4 Along these lines, Nepal et al. (2018) nd that Nepal's civil conict increased women's desired and actual fertility during the conict by 22 percent. 5 For convenience, throughout the paper I refer to these local administrative units as villages although some are technically urban municipalities. 4 2.1 Conict in Nepal For most of modern history, Nepal was governed by absolute monarchs. In the early 1990s several political parties launched pro-democracy street protests, known as the Jana Andholan (People's movement), leading to the emergence of multi-party democracy and the introduction of a new constitution. Despite participating in the 1991 legislative democratic elections and winning 9 of the 205 parliamentary seats, the Communist Party of Nepal Maoist (CPN-M) launched an armed struggle, the so-called People's War, against the state on February 13, 1996 (or in Nepali calendar 2052 Falgun 1 Bikram Sambat). A week before the conict's start, the CPN Maoist submitted a 40-point memorandum to the government and warned of armed militant struggle if demands were not met. Demands included drafting a new constitution through an election of a constituent assembly; land redistribution; and political equality for all castes, language groups, and women. The government refused to meet the demands of the Maoists. In response, the Maoists attacked an agricultural bank and three police posts in rural western Nepal and formally launched the People's War". Over the following decade, the insurgency developed into an entrenched and brutal country-wide civil war. By the end of the insurgency, conict related killings were recorded in 73 of the 75 Nepalese districts. Figure 1 presents the timeline of the war including major events that shaped the conict and monthly casualty numbers. As part of the Maoists' strategy, in the early years of the insurgency they launched a guerrilla warfare mostly harassing police forces and garnering support in a few rural areas with communist strongholds (Thapa and Sijapati, 2004). Nepal's remote terrain, under- development, extreme rural poverty, deeply rooted caste and ethnic discrimination, sentiments of political and economic exclusion among rural communities, and lack of government presence in rural areas propelled the Maoists' cause further (Onesto, 2005). The initial inability of the government to recognize the underlying problems that fueled the conict and to acknowledge the connection between armed conict and political, economic, and social grievances of the period enabled small communist political elites to mobilize a large base and eventually challenge the government militarily and politically (Kreuttner, 2008). The year 2001 was a crucial moment for the insurgency. In June 2001, the killing of King Birendra along with most of his immediate family members in a royal massacre shocked the nation. 5 The King's brother, Gyanendra, was then crowned King. A conspiracy theory emerged centering on Gyanendra's possible involvement in the massacre and questioning the ndings of the ocial investigation further destabilized the country, increasing distrust in the government and the King (Thapa and Sijapati, 2004). A state of emergency was declared in November 2001 and the Royal Nepal Army (RNA) ocially became involved in the war after the Maoists walked away from a two-month long ceasere and attacked an RNA barrack. Thereafter, the conict intensied and extended geographically. As illustrated in Figure 1, most killings occurred after 2001. However, the insurgency drastically changed its course after King Gyanendra, citing prolonged conict and growing attacks by the Maoists, dismissed the elected government, placed major political gures under arrest, and assumed direct control over the country in February 2005. Joining the widespread disapproval of the King's actions, the Maoists formed a pact with seven major political parties to present a common front against the monarchy. This eventually led to the signing of a peace accord in November 2006 (or 2063 Mangshir Bikram Sambat), the formation of an interim seven party plus the Maoists coalition government, and an ocial end to the war. At the time of the signing of the peace agreement, the death toll of the war had reached more than 13,000 (Table 1). The CPN Maoist's presence across the country over the course of the conict varied greatly. While the Maoists had a weak presence in urban areas  failing to control even a single city or a district headquarters  they dominated rural Nepal. In October 2003, they declared control over 80% of rural areas (Onesto, 2005) and in many places established fully functional local governments and law courts of their own. They also, however, selectively targeted government forces, attacking army barracks and police posts in urban areas and destroying local government buildings (Do and Iyer, 2010). There were widespread human rights violations and abuses throughout the insurgency by both the government forces and the CPN Maoists (OHCHR, 2012). Physical assault, abduction, and torture of civilians, and looting of individual properties by the Maoists were reported extensively throughout the conict (Bohara et al., 2006). The security forces, on the other hand, were the major perpetrators of sexual violence, arbitrary arrests and disappearances of civilians and were accused of murder, torture, mutilation, and other cruel and inhumane treatment of civilians to extract information from anyone they deemed appropriate (OHCHR, 2012). 6 2.2 Consequences of the Civil Conict and Mechanism The conict had widespread impact on economic development and severely hampered delivery of government services. The Maoists' unocial motto of Destruction before construction" was very popular among its cadres and was heavily advertised (Nepal, 2004). Maoists destroyed key infrastructure linking urban areas to their rural strongholds and sabotaged public delivery systems. Maoists often targeted rural bridges that linked rural to urban areas and district headquarters, and in many parts of the country, destroyed health posts, drinking water systems, public communication systems, and schools (Jha, 2008). Between 1996 and 2003, physical infrastructure worth at least $250 million was destroyed (Mahat, 2006) and the cost of the conict was estimated at $66.2 billion (Ra and Singh, 2005). The conict in Nepal is likely to have aected adult health and economic outcomes in multiple ways including direct physiological and mental stress, nutritional shocks and reduced access to health care. 6 First, the conict severely aected the delivery of government services in rural areas; in particular, decreasing health care delivery. Hundreds of community health posts were destroyed; several health care workers were killed; many ed their posts; and cold-chain delivery of vaccines became impossible (Singh, 2004). Second, the conict likely led to direct physiological and mental stress on residents, especially rural residents. As reported in Table 1, by the end of the conict, more than 13,000 people had lost their lives, and more than 1,500 people were either disappeared, injured, or disabled. Among the family members of these casualties (estimated between 400,000 and 500,000), many suered from mental and psychological trauma (Media Foundation, 2011). Third, many lost their sources of income, were displaced, widowed, and many children were orphaned. Fourth, the complete disruption of public delivery systems came as a major shock to nutrition and food security, especially in the northern region that had relied heavily on the government-subsidized rations. 6 See Akresh et al. (2018) for the discussion of mechanisms through which stress and inadequate resources during civil conict are likely aect health. 7 3 Data One of the major impediments to analysis of conict is the lack of conict data at suciently granular geographic scale. Most of the literature on the legacies of war therefore is focused on assessing the eects of conict intensity at relatively high geographical levels. For example, previous studies exploring the determinants of Nepal's civil conict (Murshed and Gates, 2005; Bohara et al., 2006; Do and Iyer, 2010; Macours, 2011) and the consequences of the conict (Valente, 2014; Pivovarova and Swee, 2015; Menon and van der Meulen Rodgers, 2015) exploit variation in intensity of the insurgency at district level or even higher geographical units. Dening conict variables at a broader geographical level measures individuals' exposure to conict less precisely and can create measurement error. Moreover, signicant heterogeneity in geography, socioeconomic status and development among areas within districts can create diculty in addressing the association between the determinants of war at teh district level and explained variables of interest, likely leading to omitted variable bias. In contrast, this analysis uses detailed and geographically granular conict intensity data; i.e. village-level insurgency. As households within a village tend to be highly homogeneous in socioe- conomic status and ethnicity and live in the same geographical terrain, this paper is able to avoid many of the concerns induced by the imprecise measurement of individuals' exposure to conict. Additionally, combining these high-resolution conict data with the 2016 Nepal Demographic Health Survey (NDHS 2016) allows me to explore the impacts of the conict on the children of individuals who were exposed to the war in their own childhoods. 3.1 Conict Data I use the Informal Sector Service Center's (INSEC) records of the conict victims to create con- ict intensity variables. INSEC is an active Nepalese non-governmental human rights organization. Throughout the war, INSEC documented human rights violations and abuses extensively and its archive of the casualties provides detailed information on each victim's demographic, social and economic characteristics. The database is considered the most reliable data source on casualties of the conict. Numerous studies including Do and Iyer (2010); Nepal et al. (2011); Valente (2014, 2015) and Libois (2016) have used the database. I extract demographic, educational achievement, 8 social and economic characteristics, and political aliation of each victim. Most importantly, I extract exact geographical location (village) and the date of the incident. Table 1 reports descriptive statistics of the war fatalities. In total the INSEC data set contains information on 14,982 victims; most (13,210) are fatal casualties. More than 60% of the casualties are perpetrated by the state. The CPN Maoists deny exploiting the grievances among ethnic groups regarding political, social and economic exclusion to advance their agenda. However, the majority of their cadres belonged to ethnic groups such as Magars, Gurungs, and Dalits of the hills and mountains. It is, therefore, not surprising that the majority of casualties among the Maoists are ethnic minorities (60% not reported) and more than half of the total victims are also from the minority groups. Apart from attacking security forces, the Maoists also frequently targeted upper caste civilians, especially Bramins and Chhetris; these upper castes the Maoists labeled as counterrevolutionary elements. The average age of the victims is 28.3 years and almost 90% are male. Many were actively involved in politics- 54% are aliated to either the rebel party or other political parties. The conict intensity varied greatly over time. I summarize the number of monthly casualties and major events that helped shape the insurgency in Figure 1. As illustrated, the period after 2001 when the country was under the state of emergency and the RNA was actively involved was the bloodiest. The conict lasted for total of 131 months from February 1996 to November 2006. Using information on each victim's village and date of incident, I dene months of warv in a village, v, as the baseline conict intensity variable, constructed as follow: 131 conictv 1 = months of warv = 1(casualtyvm ) m=1 (1) with 1(casualtyvm ) = 1 if casualtyvm >0 and subscripts v and m index a village and a month since the beginning of the war i.e. m takes the value of 1 for February 1996 and 131 for November 2006. Variable casualtyvm is number of casualties in a village v in a month m. Villages in Nepal experienced dierent levels of conict as illustrated in Figure 2, which depicts the number of months each village experienced conict out of the total 131 months of the war. The intensity of the conict varied substantially across villages within districts. Dening the conict 9 intensity as in Equation (1), however, may create a possibility of under measuring the conict intensity  for example, a village could be under the siege of the Maoists, reducing access to public services but without any casualties; similarly, one-o destruction of infrastructure could have longer- term ramications. Unfortunately, we do not have records on infrastructure damages during the conict. Nonetheless, total months of exposure to conict and casualty count is used extensively in the literature to measure conict intensity and assuming classical measurement error, because of the tendency to attenuate towards zero, the estimated coecients will provide lower bound of the conict impacts. Additionally, I use several other measures of conict intensity as below: 131 conictv 2 = number of casualtiesv = casualtyvm (2) m=1 131 conictvN 1 = months of warvN = 1(casualtyvN m ) (3) m=1 131 conictvN 2 = number of casualtiesvN = casualtyvN m (4) m=1 131 conictv 50a = months of warv 50 = 1(casualtyv50m ) (5) m=1 131 conictv 50b = number of casualtiesv 50 = casualtyv 50m (6) m=1 Conict intensity based on Equation (2), measure of total casualties in a village over the duration of the war, is illustrated in Figure A.2. Again, the measure exhibits signicant variation across villages and the intensity pattern is highly similar to Figure 2. The next four Equations (3) to (6) are dened at higher geographic level with the consideration for potential spatial spillovers  conict in nearby villages may induce stress, limit one's access to health care or other services. While Equations (3) to (4) are months of war and casualty counts, respectively, in a village including in its contiguous neighboring villages (N), Equations (5) to (6) report months of war and casualty count in a village including the villages around 50-kilometer radius from the center of the village. 10 3.2 Individual Data I also use the 2016 Nepal Demographic Health Survey (NDHS 2016) in the analysis. The survey was implemented by New ERA under the aegis of the Ministry of Health of Nepal and was funded by the United States Agency for International Development (USAID). The data collection took place between June 19, 2016 and January 31, 2017. The NDHS 2016 is a nationally representative survey of the female population ages 15 to 49. The sampling frame for the survey was based on the updated version of the 2011 Nepal Population Census. After the implementation of the 2015 constitution, based on the population several VDCs and Municipalities within districts were merged to form rural development areas and urban areas. The old Village Development Committees (VDC) in the rural and enumeration areas (EAs) in urban places essentially form primary sampling units (PSU) for the 2016 NDHS. In the nal sample, 383 clusters or PSUs were selected with probability proportional to their population size (see Ministry of Health, 2017). Figure A.3 illustrates the coverage of the survey. All 75 districts except Manang and Mustang were sampled; however, these two districts also had zero casualties during the conict and are excluded from the baseline analysis. As illustrated in Figure 3 (months of war) and Figure A.4 (casualty count), conict intensity varied signicantly across the 383 NDHS villages (clusters). Within the selected DHS clusters, 30 randomly selected households were interviewed, and all women aged 15 to 49 who were permanent residents or visitors who stayed in the household the night before were eligible for the interview. A sub-sample of about half of the households were selected for biomarker information. All children aged 0 to 59 months and women 15 to 49 years in these households were administered the anthropometry, hemoglobin, and blood pressure measurements. I limit the analytical sample to those that are born before or were in utero during the start of the conict in February 1996 so that they are old enough to be sampled in the ever-married NDHS 2016 women sample and have children that can be incorporated in the second-generation analysis. The nal analytical sample size is 4,421 women ages 20 to 49 at the time of the survey (0 to 29 at the start of the conict) and their 2,168 under age 5 children. Table 2 summarizes women's health outcomes and their exposure to conict and demographic characteristics. I divide women into two cohorts: ages 0 to 15 (treatment) and 16 to 29 (control) at the time of the start of the conict. While the women in the former cohort would have still been in 11 a period of physical growth during the conict years, the latter cohort would have already gained full adult height (detail discussion is presented in empirical strategy section 4). On average, women in the sample are 151.6 centimeters tall, with the younger treated cohort 0.46 cm taller on average. Similarly, compared to the control cohort, treated cohorts are less likely to have had any incidence of pregnancy loss, report having had fewer live births, were slightly younger at rst birth, attained more years of education, and were less likely to be employed at the time of the survey. However, there is no dierence in economic status (wealth index) between the two groups. By construction, the older cohort faced zero level of conict during the rst 15 years of life (Panel B.1). Balance in lifetime exposure to conict between the two sets of women (Panel B.2) is reassuring for the empirical strategy used in section 4, which implies that the two groups do not come from dierent types of sampled clusters. On average, women are 33.4 years old with treatment and control cohorts being 27.9 and 41.8 years old respectively (Panel C). Women in treatment groups are more likely to live in a household with a female head, are more likely to be of lower caste, and are less likely to be from Eastern Development region. All other controls are balanced between the groups. Summary statistics of the children sample is presented in Table 3. Children are divided into treated and control group by their mother's age at the start of the conict. 47% of the children are girls and on average tend to be the second child (Panel C). There is no dierence in control variables (Panel C) and health outcome variables between the two groups except treated children are slightly taller. Panel B reports mother's exposure to conict. Again, it is reassuring that there is no disparity in their mothers' lifetime conict experience. 4 Empirical Strategy Women surveyed in the NDHS 2016 experienced dierent levels of exposure to conict intensity according to their village of residence and year of birth. The identication strategy exploits this variation, specically, the variation in exposure to the conict during the individual's critical period of physical growth. Height development in humans is characterized by three distinct stages. There is a rapid growth during the rst three years of life, followed by a lower level of constant gain in height until the 12 start of adolescence and then a second growth spurt during adolescence ending in gaining full adult height (Tanner et al., 1966a,b; Beard and Blaser, 2002; Bozzola and Meazza, 2012). Figure A.5 demonstrates height velocity curves for a typical boy and a typical girl. Under adequate nutritional and environmental conditions, the height growth rate is highest during infancy, 26 centimeters per year, and progressively declines until around age three, then stabilizes around 6 centimeters per year until the start of puberty (Beard and Blaser, 2002; Bozzola and Meazza, 2012). Pubertal height spurt among girls starts after age nine, peaks at about 12 years, and stops around the age of 15 (Figure A.5). I borrow these stylized facts from the clinical and bio-medical literature to establish causality. NDHS collects individuals' month and year of birth, which I use to create the age of women at the start of the conict in February 1996. Figure 4 presents cohorts by age at the start of the conict and potential exposure to conict at dierent stages of their physical growing periods. As demonstrated in Figure A.5, girls past their pubertal age i.e. cohorts aged 16 to 21 and 21 to 29 in 1996 (control 1 and 2 respectively) would have gained full adult stature by the time of the start of the conict, and the eect of the conict should be minimal or zero on their nal height. I use cohort 16 to 21 as the main comparison group and use cohort 22 to 29 as a control placebo experiment in a dierence-in-dierence specication. Based on their growing phases at the beginning of the conict, I create three conict-exposed cohorts. Although in the important phase of adolescence height spurt, girls aged 9 to 15 (treatment 3) would have faced conict only in their third phase of height growth. While girls aged 4 to 8 (treatment 2) would have been exposed to conict in the second and third stages of growth, cohort 0 to 3 (treatment 1) would have been exposed to conict through the entire growing period (all three stages). 7 In the baseline specication, I dene conict intensity at the village level, hence, all ve cohorts from any given village would have been exposed to the same total amount of conict during their lifetimes. However, the exposure would have started at dierent times of their lives. 7 These ve cohorts aged 0 to 3, 4 to 8, 9 to 15, 16 to 21, and 22 to 29 at the start of the conict in 1996 would become 20 to 23, 24 to 28, 29 to 35, 36 to 41, and 42 to 49 at the time of the NDHS 2016 survey. 13 4.1 First Generation Impact To explore the impact of early childhood exposure to conict on adult outcome, I employ the following estimation strategy: Yimntcvdr =βc (conictv × λc ) + conictv + λc + αt + ηm (7) T δ v + γr + Xi + ωn + εimntcvdr where Y is an outcome of a woman i born in month m and year t, interviewed in month n, and residing in village v, district d and development region r. While women's adult stature is the main outcome of interest, I also explore conict's impact on women's reproductive health, sexual behavior, educational attainment, employment, and wealth. The independent variable of interest, conictv × λc , is constructed as a vector of age-cohort specic coecients. The baseline conict intensity variable is months of war in a village; however, I also estimate the same equation using all the other conict variables dened in the data section. Equation (7) also includes the main conict variable, conictv , and cohort xed eects, λc , as part of the independent variables. While αt are year of birth xed-eects, ηm are month of birth xed eects added to control for any seasonality - whether women were born in peak or lean season. While δv are village (NDHS cluster) xed eects, ε is a random, idiosyncratic error term.8 Equation (7) also includes ve development-region-specic trends, T, γr to isolate variance in a cohort's outcome in deviation from the long-run trend in her development region of residence. The ve development regions were relatively homogeneous in terms of development, geographical terrain and ethnic composition before the war. The socioeconomic status of the household is likely to play a signicant role in child development and would be desirable to control for in the regression. Albeit observed 10 years after the end of the civil war during the time of the survey, household characteristics may still be inuenced by the conict. Xi , therefore, includes only variables that are time-invariant. Contrary to the Maoists' denial, ethnicity played an important role in the insurgency. Xi , therefore, includes indicators for belonging to a high caste. In addition, month of the survey interview xed eects, ωn , are included in the regressions to control for the variation in seasonality due to the timing of the survey. βc s in Equation (7) are the main coecients of interest. Under a standard dierence-in dierences 8 I also estimate equation 6 using district xed eects instead and the results are robust. All the standard errors are clustered at the village level to allow for the correlation among error terms within village. 14 model assumption and, in particular, under the assumption that there is no correlation between village level conict and unobserved factors varying with village and birth year cohort within the development region, βc coecients indicate the causal impact of early childhood exposure to civil conict on adult stature. While interpreting the results, given the set of xed eects in Equation (7), βc do not identify the eects at a national level. Rather eects are identied due to women's exposure to conict by village of residence and birth year cohort net of birth year trends common to all the villages within the development region. The goal of the paper is to measure the total eect of the conict on one's life and specication 7 does exactly that. Rather than measuring the impact of exposure to conict at a specic period of one's life, it measures the cumulative impact of exposure during one's entire growth period. 9 Studies exploring the determinants of the Nepal's civil conict have advanced several arguments regarding the insurgency heterogeneity across Nepal including geographical terrain (Murshed and Gates, 2005; Bohara et al., 2006; Do and Iyer, 2010; Menon and van der Meulen Rodgers, 2015) economic exclusion and poverty (Murshed and Gates, 2005; Do and Iyer, 2010), inequality (Macours, 2011; Nepal et al., 2011), and lack of political representation (Murshed and Gates, 2005; Bohara et al., 2006; Macours, 2011). These determinants of variation in insurgency intensity are, therefore, likely to be correlated with the outcomes of interest, threatening the validity of the identication. However, all these studies have focused on the determinants of the conict at the district level. Therefore, the application to village level conict are at most minimal because unlike districts, villages in Nepal are highly homogeneous in terms of ethnic composition, socioeconomic status, and geography. A major advantage over previous studies of the conict is that this paper uses a detailed geographical level conict intensity allowing for the inclusion of village-level xed-eects, which eliminates any village-level time-invariant factors. The timing of the beginning of the conict in women's lives within a village therefore forms the comparison divide in the estimation strategy. 9 A typical approach in the literature is to measure conict based on one's exposure at specic age and use xed eects models to identify ages or age-periods during which the exposure was most critical as below: Yimntvdr =β0 + β1 × Exposure during 0 to 3 years + β2 × Exposure during 4 to 8 years+ β3 × Exposure during 9 to 15 years + αt + ηm + δv + γr T + Xi + ωn + εimntvdr Conict variables are dened as woman's exposure to conict during her age of 0 to 3, 4 to 8 and 9 to 15 years and all other variables have the same meanings as in Equation (7). The results are presented in Tables A.1 to A.2. While important in identifying what part of one's life was important, the specication does not identify the overall impact of the conict. 15 Selective fertility and endogenous migration are other major concerns regarding the identication strategy. As discussed in the data section, I limit the analytical sample to those who were already born or in utero during the start of the conict in 1996. Purely on identication prospective, it helps limit the potential confoundedness between the explained variables of interest and selective fertility and migration. For instance, the strategy helps mitigate the scenario in which after grasping the seriousness of the war, couples that are highly concerned about their children's health in high conict areas may choose to delay having children or migrate to low conict areas to start a family. However, it could be true for the periods prior to the start of the conict that in anticipation of the war concerned couples may have delayed having children or may have migrated. Detailed robustness checks are presented in the empirical results section. 4.2 Intergenerational Health Impacts The gap between the start of the Nepal's civil conict and the time of the NDHS 2016 survey is sucient enough that I can explore the impacts the conict had on the children of women who were exposed to the war in their childhood. Anthropometric measures were collected for children under the age of 5 at the time of the survey and hence I limit the second-generation sample to children under 5 in 2016. I employ the same strategy as in Equation (7) and add child specic controls to estimate. Following is the estimation equation: Yjklnimtcvdr =βc (mother's conict exposurev × mother's cohort(λc )) + mother's conict exposurev + λc + αt + ηm + δv (8) T + γr + +Xi + µk + θl + πn + Xj + εjklnimtcvdr where Y is a health outcome of a child j whose anthropometrics were measured in month n, was born in month k and year l to a woman i who was born in month m and year t, and resides in village v, district d and development region r. Child health endowment is dened as function of all mother's controls and exposure to war as dened in section 4.1 and child specic characteristics. θl are child's birth-year xed eects. As with mothers, child month of birth xed eects, µk , are included to account for season of birth. Similarly, child anthropometric measures are sensitive to the timing of the measurement; in particular child weight, hence, Equation (8) also includes month 16 of measurement xed eects, πn . ε is a random, idiosyncratic error term and all the standard errors are clustered at the village level to allow for correlation among error terms within villages. Xj is a vector of time-invariant child controls  dummy variables equal to one if the child is a girl and if the child is a twin and child birth order xed eects. As in Equation (7), βc s are the coecients of interest and have the same meaning as in Equation (7) but identify to the impact of mother's childhood exposure to conict on her child's outcomes. The equation under the standard assumptions of dierence-in-dierences models provides estimates of the causal impact of conict on second-generation health. 5 Results In this section, rst I present the impact of childhood exposure on adult stature and establish validity for the identication strategy. Second, I report impact on health and economic outcomes that are very important to women's well-being, but also provide additional explanation for the impact on second-generation health. Finally, I present the health impacts on the children of women who were exposed to the war in their childhoods. 5.1 Impact on Women's Stature Table 4 presents the impact of early childhood exposure to conict on adult stature using dierence-in-dierences Equation (7). The conict intensity variable used is months of war in the village of residence. While the outcome variable, height, in columns 1 to 3 is measured in centimeters, columns 4 to 6 present height-for-age standard deviation (HAZ). The possibility of non-parallel dynamics" in the dierence-in-dierences estimation could be problematic. Because of dierence in overall trends (health, education, poverty, environmental etc.), changes in adult stature could vary systematically across villages and, in particular, there could be mean reversion. Given the data structure, I can, however, test for the identication assumption. Besides the control group (aged 16 to 21 in 1996), women in age cohort 22 to 29 in 1996 would have gained full adult height by the start of the conict, hence, the changes in adult height between these two cohorts should not dier systematically across villages. Age cohort 22 to 29 is therefore included in all the regressions as a control to validate the identication assumption. 17 Village xed eects are included in all specications. I start estimation with no additional con- trols (column 1) and progressively include extra controls. Column 3 is the full baseline specication. The estimated dierences-in-dierences for cohort 22 to 29 are close to 0 in size and statistically not dierent from 0 across all specications and measures of height. This provides strong evidence that the dierence-in-dierences coecients of interests are not driven by inappropriate identication assumptions. Exposure to civil conict only during the pubertal spurt appears to have no signicant eect on adult height. Across all specications, conict had statistically zero impact on height of the women aged 9 to 15 in 1996 compared to women in the control group. On the surface, this nding is slightly at odds with Akresh et al.'s (2018) analysis of Biafran war, where they nd conict intensity during women's adolescent years to have signicantly negative impact on their adult stature. However, as illustrated in Figure 1, Nepal's conict started as a small-scale rebellion and only after 2001 developed into countrywide brutal civil war. Unlike the other two treatment cohorts in the paper, most women in this age group would have escaped the most brutal phase of the civil conict during their growing period. In addition, my identication strategy diers from Akresh et al.'s (2018) in that women in this age group start facing conict only during their adolescent years. Women aged 4 to 8 in 1996, on the other hand, would have just entered or already be in their adolescent spurt years when the conict started to intensify in 2001.Therefore, it is not surprising to see the cumulative violence, which was at a lower level in the second growth stage and intense in the third stage, has signicant and negative impact on their adult stature. Eects vary between 0.67 to 0.71 millimeters (0.011 to 0.012 sd) across the specications. The coecient in the full model can be interpreted as: an additional month of exposure to civil war during the latter two stages of the growing period decreases nal adult height by 0.71 millimeters or 0.012 sd. Besides being a period of a rapid growth, the rst three years are also the most sensitive period to environmental inuences on height in human beings (Schmidt et al., 1995). Early childhood height at ages 2 (Luo and Karlberg, 2000) and 5 (Satyanarayana et al., 1986), is a strong predictor of nal adult height. Similarly, the pubertal growth spurt plays an important role in determining the nal adult stature (Case and Paxson, 2008). Women in the age group 0 to 3 in February 1996, who would have been exposed to conict during all three phases of growth, therefore would suer the most from the conict compared to the other age groups. On average, girls aged 0 to 3 in 1996 suered 18 a reduction in adult height of 1.36 millimeters or 0.023 standard deviations due to an additional month of exposure to conict during their entire period of physical development (Columns 3 and 6). The eect size is twice as big as the eect on the cohort aged 4 to 8 (0.071mm and 0.012 sd) that experienced same level of violence but only after the age of 3, hence signifying the importance of environmental inuences on height during the rst three years. The result is highly signicant and robust across all specications and for both measures of height ranging from 1.22 to 1.36 mm and 0.021 to 0.023 sd reduction in adult height. Figure 5 presents the impact of exposure to conict starting at dierent ages using the baseline specication. Again, conict exposure in rst 8 years of life reduce adult height, however, ages 0 to 3 are the only ages that are statistically signicant. As discussed in the data section, I dene the conict intensity variable multiple ways and results using the baseline specication are presented in Table 5. Conict intensity used in column 1 is the number of casualties in one's own village of residence. Consistent with the earlier ndings, among women who were aged 0 to 3 and 4 to 8 in 1996, increased casualties in the village of residence signicantly decreased their nal adult stature. In columns 3, 4 and 5, I dene war intensity as months of war including the contiguous neighboring villages and within a 50-kilometer radius from the village center. Again, the results are robust, especially among age cohort 0 to 3. Height has long been recognized as an important factor inuencing individuals' professional and personal success. Results presented in this section, thus, have important economic signicance. Taller workers receive a wage premium. An additional inch of height is associated with 1 to 3 percentage increase in earning among the British and American adults (Persico et al., 2004; Case and Paxson, 2008). There is even greater height premium in lower income settings: an additional centimeter of height is associated with a 2 percent increase in hourly earnings both in Mexico (Vogl, 2014) and Indonesia (Bargain and Zeidan, 2017). Using the results from Indonesia and Mexico, a quick back-of-the-envelope calculation implies that an additional month of conict exposure starting at infancy is associated with a decrease in hourly earnings of 0.27 percent. 5.2 Identication Validation In this section, I present additional evidence in support of the estimation strategy. In addition to the possible presence of non-parallel dynamics discussed in the earlier subsection, selective migration 19 and fertility are other major threats to the identication strategy. The conict intensity variable is dened at the level of an individual's village of residence at the time of the survey. Systematic sorting by economic and physical status between the stayers at high conict villages and movers from the high conict to low conict villages is of concern, which will lead to overestimation of the impact of conict. Unfortunately, I do not observe women's village of birth. However, the 2016 NDHS asked each individual how long she has been residing at the place where she was surveyed and if not her entire life, the name of the district from which she migrated. Twenty-six percent of the women in the sample are living in a dierent district. Although the survey did not collect the information, marriage is the most likely reason for women's migration, as most women in Nepal move to their husband's home permanently from their maternal home, which is also likely to be their place of birth. I dene conict at the level of women's districts of birth (district they moved from) and estimate the same specication as Equation (7) except with district of birth xed eects and region of birth trends and present the results in Table 6. Columns 1 to 3 examine the presences of dierences in migration patterns between the control and treated cohorts. The dierence-in-dierences across the specications are zero, suggesting no selective migration. Columns 4 to 9 re-estimate the main results from Table 4. The dierences-in-dierences estimates, although reduced in magnitude, are statistically signicant and in the same direction as in the main specication for age cohort 0 to 3. Villages within district are highly diverse and dening conict at the district level takes away that variation. Additionally, I limit the sample to women living in the village where they were surveyed their entire life and estimate Equation (7). The results are presented in Table A.3. The eect sizes are comparable to the baseline results in Table 4. These results, overall, suggest there is minimal or no selective migration among the women in the sample. Limiting the sample to women already born at the start of the conict limits the scope for confoundedness through selective fertility. However, prior to the start of the conict couples that were highly concerned about their children's health may have delayed having children in anticipation of the war. If true, we expect to see signicantly dierent level of births between high and low conict areas periods just before the start of the war. However, the conict lasted for a decade and concerned couples may have had to wait for the full decade to have a child  a highly unlikely scenario. To formally examine this issue, I use the information on district of birth and year of 20 birth from every individual observed during the 2001 Nepal Population Census and calculate yearly district birth rates. As reported in Figure 6, there is no dierence in birth rates between the districts experiencing above and below the median conict intensity (months of war). Table A.4 reports regression results for the same test and we see no dierence in periods long before and just before the start of the conict. 5.3 Impact on Fertility, Education, Employment and Wealth Estimates in Table 7 show the impact of conict on reproductive health and fertility using the baseline specication. Early childhood exposure to conict has no signicant impact on probability of miscarriage or of stillbirth among women in age cohorts 9 to 15 and 4 to 8 in 1996. Women expe- riencing conict during their entire growth period, however, are signicantly more like to have had a stillbirth or miscarriage: each additional month of conict exposure increases the risk of stillbirth by 0.03 and of miscarriage by 0.06 percentage respectively. Valente (2015) nds similar results that pregnancies that were exposed to Nepal's conict were more likely to result in a miscarriage. There is no signicant impact on the probability of abortion. Women exposed to conict during their early growing periods have signicantly more live births, 0.024 and 0.013 more births per month of exposure among cohorts aged 0 to 3 and 4 to 8 in 1996 respectively. The impact of conict on the number of infant deaths is zero (column 5). Impacts of early childhood exposure to conict on other fertility outcomes are presented in Table 8. The conict has statistically no impact on the likeliness of contraceptive use (column 1), number of women's sex partners, age at rst sexual intercourse and age at rst birth. There are also no signicant dierences between women in control and treatment groups on their age at rst cohabitation with their domestic partner and number of marriages and unions. Additionally, there is no dierence in the smoking or chewing tobacco habits of women in treatment groups and control group. Table 9 shows impact on women's human capital accumulation, employment and wealth. While there is no signicant impact on years of education, women in age cohort 0 to 3 are signicantly less likely to have completed a school leaving certicate (SLC). The lack of results in years of schooling are consistent with Valente (2014) and Pivovarova and Swee (2015). Both examine the conict's impact on human capital accumulation. Similarly, there is no impact on the probability 21 of being employed. However, women exposed to conict are signicantly more likely be employed in agricultural sector. As a measure of a household's cumulative living standard, the DHS reports a wealth index using ownership of easy to collect assets, materials used in housing construction, and access to type of water and sanitation facilities (see Rutstein and Johnson, 2004). Women exposed to conict early in their childhood are likely to live in households with poorer living conditions (column 7). The wealth factor score is lower by 977 and 675 per additional month of exposure to conict among women aged 0 to 3 and ages 4 to 8 in 1996 respectively. In addition to providing information on women's adult living conditions, outcomes discussed in this subsection provide a window to what types of households the children of individuals exposed to conict in early childhood are living in. Therefore, these eects are likely to provide explanations for the second-generation health impacts presented in the next subsection. 5.4 Impact on Intergenerational Health The intergenerational impacts of conict on health are presented in Table 10. As discussed in section 4.2, the estimation strategy used is the same as for the rst generation, except I add child-specic controls. The sample consists of children under the age 5 whose anthropometrics were measured. I also include children born to women aged 22 to 29 in 1996 to provide support for the identication validation required by dierence-in-dierences assumptions. The falsication test supports that identication strategy, as estimates for all outcomes for children born to the experimental control cohort of mothers are statistically zero. Although statistically imprecise, mothers' exposures to conict negatively aects child's height (column 1: table 10). Children born to women in all the three treatment groups are shorter by 0.005 to 0.011 standard deviations for their age. Child development in terms of weight gain is signicantly hampered by mother's exposure to conict in her childhood. Compared to children born to the control cohort of mothers, an additional month of mother's exposure to conict during her entire growth period decreases her child's weight for height z-scores by 0.030 standard deviations (5.2 percent less than control mean). Exposure to conict starting at older age is even more severe for second-generation weight for height. Women in cohorts age 4 to 8 and 9 to 15 in 1996 have children 0.039 and 0.041 standard deviations lighter for their height (6.7 and 7.1 percent less than the control mean). All the coecients are estimated precisely. Column 3 reports the conict's 22 impact on second-generation weight-for-age. Mother's exposure during her childhood, again, has signicant negative impact on child weight-for-age, especially among children born to women in the cohort aged 9 to 15 in 1996. Additionally, maternal war exposure is strongly associated with signicant lower body mass of children (column 4). As with the weight-for-height, the impact size is increasing with the age at which the mother started experiencing war. Compared to the control, mothers exposed in all three stages, the latter two stages, and the nal adolescence stage (aged 0 to 3, 4 to 8, and 9 to 15 at the start of conict), have children with lower BMI by 0.031, 0.040, and 0.044 standard deviations respectively. Alternatively, an extra month of exposure led to a decrease in children's BMI by 0.030 to 0.044 standard deviations. These results are consistent with Akresh et al. (2018), in that the adolescent exposure to conict has strongest impacts on second-generation health. Table A.5 presents the same results using other conict measures discussed in section 3 and the results are robust across all conict denitions. As a robustness check, Table 11 presents the intergenerational impact using conict intensity based on mother's district of birth. The estimate shows stronger results than when dening mater- nal exposure to conict at the level of village of residence. Compared to the children of women in the control group, weight-for-height z-scores are signicantly less among children of women in the treatment group. The impact size between the treatment groups are highly comparable. Similarly, impacts on weight for age and BMI have similar strong negative impacts. These coecients are smaller than those reported in table 11 but are more precisely estimated. In both the specica- tions, we observe negative but statistically zero impact on child height. However, at this stage of physical growth, children may be too young to develop stunting and negative impacts on weight measurements provide strong indication for future stunting. Channels for intergenerational transmission of maternal exposure to conict to child may vary greatly. In addition to unobserved factors such as the physiological stress and genomic changes, the intergenerational transmission may be working through the maternal health, education or wealth endowment or through early childhood investment (Cunha and Heckman, 2007). Results presented in section 5.3 are likely to explain some of these channels. Nepalese women exposed to conict during their childhood development are more likely to have had pregnancy losses and at the same time have more live births (Table 7). Although I nd no evidence for sexual behavioral changes (Table 8), women are less likely to have completed SLC, and most work on their own farms (Table 9). 23 Additionally, highly signicant in terms of parental ability to invest in children, exposed women have signicantly less wealth. Combined with having more children, this drastically decreases parents' ability to invest in children during their critical period of development. 6 Concluding Remarks This paper exploits variation in conict at a detailed geographical level to establish causality between early childhood exposure to conict and women's nal stature. Additionally, along with Akresh et al. (2018), this paper is among the rst to document the intergenerational transmission of the impact of early childhood conict exposure on second generation health. Nepal experienced a decade-long violent civil conict between 1996 and 2006, which resulted in more than 13,000 fatal casualties, signicant infrastructure damages, and severe hindrance in delivery of basic services and generated extreme fear, sense of insecurity and stress among its citizens. Considered the most reliable, I use INSEC's database on the conict casualties and create an individual level victims' data set with exact geographical location (village) and dates of the incidents. This allows me to exploit variation in conict intensity at the village level for identication. Fueled by international remittances, Nepal enjoyed consistent economic growth and poverty reduction (Uematsu et al., 2016) during the period of conict. The country also made signif- icant improvement in other dimensions of development including health (Headey and Hoddinott, 2015) and non-income based multidimensional poverty (OPHI, 2013). These aggregate development trends, however, may mask disparities at a more disaggregated level due to signicant variation in conict intensity. The little research documenting the consequences of the war thus far is focused on district-level disaggregation, and results are mixed - little to zero impact on human capital accu- mulation (Valente, 2014; Pivovarova and Swee, 2015) and positive impact on women's employment (Menon and van der Meulen Rodgers, 2015). Libois (2016), on the other hand, using conict mea- surement at a more detailed geographical level (distance from the conict sites) nds signicant negative immediate impact on consumption and income. Failure to capture the substantial conict heterogeneity across villages within district may explain the lack of evidence of conict eects at the district level. In contrast, I exploit variation in early childhood exposure to conict by birth cohort and village 24 of residence to estimate the impact of conict intensity, as measured by months of war, on adult height. Using the 2016 NDHS women sample, I nd that conict and, in particular, exposure starting very early in one's growing period, has highly signicant and negative impact on women's nal adult height. Findings are robust across (i) model specications and (ii) measures of conict. In validating the dierence-in-dierences estimation strategy used, I nd no evidence of presence of non-parallel dynamics nor of selective migration and fertility. These results are aligned with the biomedical literature that early childhood conditions are highly signicant in determining nal height  early life stunting increases the risk of being short as an adult (Golden, 1994; Martorell et al., 1994). Conditions in the fetal period and early years after birth are profound in inuencing human biology and long-term health. Responses to lack of adequate nutrition of developing fetus may be coded permanently, which is likely to increase later life health hazards (Barker, 1992). Similarly, early years of life are highly susceptible to environmental inuence on height (Schmidt et al., 1995). Nepalese children growing up during the conict experienced substantial levels of physiological and mental stress, and for many, mostly in rural areas under the control of the Maoists, it was a major nutritional shock and reduced access to healthcare. These are likely mechanisms at work and describe the results presented in the paper. The sucient time gap between the start of the conict and the time of the NDHS 2016 allows me to explore the impacts of the conict on children of the women who were exposed to conict in their childhood. I nd that the mothers' exposure to conict is detrimental for their children's health. Although imprecise, impacts on children's height-for-age z-scores are negative. Results for children's weight-for-height, weight-for-age and BMI z-scores, on the other hand, are precisely estimated and again negative. Results are robust to alternative measure of conict intensity (mother's district of birth). To explore possible intergenerational transmission mechanisms, I investigate the impacts on mother's economic and fertility outcomes. I nd that women exposed to conict during childhood have more children and live in a household with signicantly less wealth. The combined eect of the two likely results in drastic decreases in the parental ability to invest in children. 25 References Akbulut-Yuksel, M. (2017). War during childhood: The long run eects of warfare on health. Journal of health economics, 53:117130. Akresh, R., Bhalotra, S., Leone, M., and Osili, U. O. (2012). War and stature: growing up during the nigerian civil war. American Economic Review, 102(3):27377. Akresh, R., Bhalotra, S., Leone, M., and Osili, U. O. (2018). First and second generation impacts of the biafran war. NBER Working Paper 23721, National Bureau of Economic Research. Akresh, R., Caruso, G. D., and Thirumurthy, H. (2014). Medium-term health impacts of shocks experienced in utero and after birth: Evidence from detailed geographic information on war exposure. NBER Working Paper 20763, National Bureau of Economic Research. Akresh, R. and De Walque, D. (2008). Armed conict and schooling: Evidence from the 1994 Rwandan genocide. The World Bank. Almond, D. (2006). Is the 1918 inuenza pandemic over? long-term eects of in utero inuenza exposure in the post-1940 us population. Journal of political Economy, 114(4):672712. Almond, D. and Currie, J. (2011). Killing me softly: The fetal origins hypothesis. Journal of economic perspectives, 25(3):15372. Almond, D., Currie, J., and Herrmann, M. (2012). From infant to mother: Early disease environment and future maternal health. Labour Economics, 19(4):475483. Bargain, O. and Zeidan, J. (2017). Stature, skills and adult life outcomes: Evidence from indonesia. The Journal of Development Studies, 53(6):873890. Barker, D. J. (1992). The fetal origins of adult hypertension. Journal of hypertension. Supplement: ocial journal of the International Society of Hypertension, 10(7):S3944. Bateson, P., Barker, D., Clutton-Brock, T., Deb, D., D'udine, B., Foley, R. A., Gluckman, P., Godfrey, K., Kirkwood, T., Lahr, M. M., et al. (2004). Developmental plasticity and human 26 health. Nature, 430(6998):419. Beard, A. S. and Blaser, M. J. (2002). The ecology of height: the eect of microbial transmission on human height. Perspectives in Biology and Medicine, 45(4):475498. Becker, G. S. and Lewis, H. G. (1973). On the interaction between the quantity and quality of children. Journal of political Economy, 81(2, Part 2):S279S288. Bhalotra, S. and Rawlings, S. (2013). Gradients of the intergenerational transmission of health in developing countries. Review of Economics and Statistics, 95(02):660672. Blattman, C. and Miguel, E. (2010). Civil war. Journal of Economic literature, 48(1):357. Bohara, A. K., Mitchell, N. J., and Nepal, M. (2006). Opportunity, democracy, and the exchange of political violence: A subnational analysis of conict in nepal. Journal of conict resolution, 50(1):108128. Bozzola, M. and Meazza, C. (2012). Growth Velocity Curves: What They Are and How to Use Them, pages 29993011. Springer New York, New York, NY. Bozzoli, C., Deaton, A., and Quintana-Domeque, C. (2009). Adult height and childhood disease. Demography, 46(4):647669. Bundervoet, T., Verwimp, P., and Akresh, R. (2009). Health and civil war in rural burundi. Journal of human Resources, 44(2):536563. Camacho, A. (2008). Stress and birth weight: evidence from terrorist attacks. American Economic Review, 98(2):51115. Case, A. and Paxson, C. (2008). Stature and status: Height, ability, and labor market outcomes. Journal of political Economy, 116(3):499532. Central Bureau of Statistics (2012). National population and housing census 2011. National report, Kathmandu, Nepal. CIA World Factbook (2019). The world factbook - nepal. https://www.cia.gov/library/ 27 publications/the-world-factbook/geos/np.html#. Accessed: April 05, 2019. Cunha, F. and Heckman, J. (2007). The technology of skill formation. American Economic Review, 97(2):3147. Currie, J. and Moretti, E. (2003). Mother's education and the intergenerational transmission of human capital: Evidence from college openings. The Quarterly journal of economics, 118(4):1495 1532. Currie, J. and Moretti, E. (2007). Biology as destiny? short-and long-run determinants of intergen- erational transmission of birth weight. Journal of Labor economics, 25(2):231264. Currie, J. and Vogl, T. (2013). Early-life health and adult circumstance in developing countries. Annu. Rev. Econ., 5(1):136. Dewey, K. G. and Begum, K. (2011). Long-term consequences of stunting in early life. Maternal & child nutrition, 7:518. Do, Q.-T. and Iyer, L. (2010). Geography, poverty and conict in nepal. Journal of Peace Research, 47(6):735748. Domingues, P. and Barre, T. (2013). The health consequences of the mozambican civil war: An anthropometric approach. Economic Development and Cultural Change, 61(4):755788. Forsdahl, A. (1977). Are poor living conditions in childhood and adolescence an important risk factor for arteriosclerotic heart disease? Journal of Epidemiology & Community Health, 31(2):9195. Gluckman, P. D., Hanson, M. A., and Beedle, A. S. (2007). Early life events and their consequences for later disease: a life history and evolutionary perspective. American Journal of Human Biology, 19(1):119. Gluckman, P. D., Hanson, M. A., Cooper, C., and Thornburg, K. L. (2008). Eect of in utero and early-life conditions on adult health and disease. New England Journal of Medicine, 359(1):6173. Golden, M. (1994). Is complete catch-up possible for stunted malnourished children? European 28 journal of clinical nutrition, 48(1):5871. Headey, D. D. and Hoddinott, J. (2015). Understanding the rapid reduction of undernutrition in nepal, 20012011. PloS one, 10(12):e0145738. Jha, H. B. (2008). Economics of conict and peace with focus on nepal. Dhobighat, Lalitpur. Centre for Economic and Technical Studies (CETS). Paper submitted to "New Dynamics of Development: Challenges and Prospects". Justino, P., Leone, M., and Salardi, P. (2014). Short- and Long-Term Impact of Violence on Education: The Case of Timor Leste. World Bank Economic Review, 28(2):320353. Kreuttner, T. R. (2008). The maoist insurgency in nepal, 1996-2001: Implicating for us counterin- surgency doctrine. https://apps.dtic.mil/docs/citations/ADA505200. León, G. (2012). Civil Conict and Human Capital Accumulation: The Long-term Eects of Political Violence in Perú. Journal of Human Resources, 47(4):9911022. Libois, F. (2016). Households in Times of War : Adaptation Strategies during the Nepal Civil War. Working Papers 1603, University of Namur, Department of Economics. Luo, Z. and Karlberg, J. (2000). Critical growth phases for adult shortness. American journal of epidemiology, 152(2):12531. Macours, K. (2011). Increasing inequality and civil conict in Nepal. Oxford Economic Papers, 63(1):126. Mahat, P. S. (2006). Socio-economic transformation of rural areas for peace and democracy. Martorell, R., Khan, L. K., and Schroeder, D. G. (1994). Reversibility of stunting: epidemiological ndings in children from developing countries. European journal of clinical nutrition, 48:S4557. Media Foundation (2011). Healing the wounds: Stories from nepal's transitional justice pro- cess. Technical report, Media Foundation, Kathmandu, nepal. https://www.mfnepal.org/ publication/healing-wounds-2010/. 29 Menon, N. and van der Meulen Rodgers, Y. (2015). War and Women's Work: Evidence from the Conict in Nepal. Journal of Conict Resolution, 59(1):5173. Miguel, E. and Roland, G. (2011). The long-run impact of bombing vietnam. Journal of development Economics, 96(1):115. Ministry of Health (2017). Nepal Demographic and Health Survey 2016. Technical report, Ministry of Health, Nepal, Kathmandu, Nepal. Murshed, S. M. and Gates, S. (2005). SpatialHorizontal Inequality and the Maoist Insurgency in Nepal. Review of Development Economics, 9(1):121134. Nepal, A., Halla, M., and Stillman, S. (2018). Violent conict and the child quantity-quality tradeo. Discussion Paper Series 11690, IZA Institute of Labor Economics. Nepal, L. J. (2004). destruction before construction": civil war grips western nepal, making hu- manitarian aid dicult, but essential. Nepal, M., Bohara, A. K., and Gawande, K. (2011). More Inequality, More Killings: The Maoist Insurgency in Nepal. American Journal of Political Science, 55(4):886906. OHCHR (2012). Nepal Conict Report. Technical report, United Nations Oce of the High Commissioner for Human Rights, Geneva. Available at https://www.ohchr.org/Documents/ Countries/NP/OHCHR_Nepal_Conflict_Report2012.pdf. Onesto, L. (2005). Dispatches From the People's War in Nepal. Pluto Press. OPHI (2013). Country Brieng: Nepal Multidimensional Poverty Index (MPI) At a Glance. Tech- nical report, Oxford Poverty and Human Development Initiative, University of Oxford, Manseld Road, Oxford OX1 3TB, UK. Avaialble at: http://www.ophi.org.uk/wpcontent/uploads/ Nepal-2013.pdf. Palmer, M., Nguyen, C., Mitra, S., Mont, D., and Groce, N. (2016). The long-term impact of war on health. HiCN Working Papers 216, Households in Conict Network. 30 Persico, N., Postlewaite, A., and Silverman, D. (2004). The Eect of Adolescent Experience on Labor Market Outcomes: The Case of Height. Journal of Political Economy, 112(5):10191053. Persson, P. and Rossin-Slater, M. (2018). Family Ruptures, Stress, and the Mental Health of the Next Generation: Reply. American Economic Review, 108(4-5):12561263. Pivovarova, M. and Swee, E. L. (2015). Quantifying the microeconomic eects of war using panel data: Evidence from nepal. World Development, 66:308  321. Quintana-Domeque, C. and Ródenas-Serrano, P. (2017). The hidden costs of terrorism: The eects on health at birth. Journal of Health Economics, 56:4760. Ra, S. and Singh, B. (2005). Measuring the economic costs of conict. Working Paper Series 2, Asian Development Bank,. Rutstein, S. O. and Johnson, K. (2004). The dhs wealth index. DHS Comparative Reports 6, Calverton, Maryland: ORC Macro. Satyanarayana, K., Prasanna Krishna, T., and S Narasinga Rao, B. (1986). Eect of early childhood undernutrition and child labor on growth and adult nutritional status of rural indian boys around hyderabad. Human nutrition. Clinical nutrition, 40:1319. Schmidt, I., Jørgensen, M. H., and Michaelsen, K. F. (1995). Height of conscripts in europe: is postneonatal mortality a predictor? Annals of human biology, 22(1):5767. Singh, S. (2004). Impact of long-term political conict on population health in nepal. CMAJ, 171(12):14991501. Strauss, J. and Thomas, D. (2008). Health over the life course. volume 4 of Handbook of Development Economics, chapter 54, pages 33753474. Elsevier. Tanner, J., Whitehouse, R., and Takaishi, M. (1966a). Standards from birth to maturity for height, weight, height velocity, and weight velocity: British children, 1965. i. Archives of disease in childhood, 41(219):454471. 31 Tanner, J. M., Whitehouse, R. H., and Takaishi, M. (1966b). Standards from birth to maturity for height, weight, height velocity, and weight velocity: British children, 1965. ii. Archives of Disease in Childhood, 41(220):613635. Thapa, D. and Sijapati, B. (2004). A kingdom under siege : Nepal's Maoist insurgency, 1996 to 2003. The Printhouse ; Zed Books : Distributed in the USA exclusively by Palgrave Macmillan Kathmandu : New York, updated ed. edition. Uematsu, H., Shidiq, A. R., and Tiwari, S. (2016). Trends and drivers of poverty reduction in nepal : a historical perspective. Policy Research Working Paper Series 7830, The World Bank. Valente, C. (2014). Education and Civil Conict in Nepal. World Bank Economic Review, 28(2):354 383. Valente, C. (2015). Civil conict, gender-specic fetal loss, and selection: A new test of the triverswillard hypothesis. Journal of Health Economics, 39:31  50. Vogl, T. S. (2014). Height, skills, and labor market outcomes in mexico. Journal of Development Economics, 107:84  96. 32 7 Figures and Tables Figure 1: Conict timeline Source: Author's calculation based on the INSEC's archive on the conict victims. 33 Figure 2: Conict intensity heterogeneity: Months of war Source: Author's calculation based on the INSEC's archive on the conict victims. 34 Figure 3: Conict intensity heterogeneity: Months of war and NDHS 2016 clusters Source: Author's calculation based on the INSEC's archive on the conict victims. 35 Figure 4: Cohorts by age at the start of the war in 1996 Age 0-3 (Treatment 1) Age 4-8 (Treatment 2) Age 9-15 (Treatment 3) Age 16-21 (Control 1) 36 Age 22-29 (Control 2) 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 Date of birth Note: Conict started on February 13, 1996 or Falgun 1, 2052 Bikram Sambat (11th month of year 2052 according to Nepali calendar). Age in the gure refers to the age at the beginning of the Nepal's civil war. Figure 5: Impact on women's adult height (cm) by age at start of the war .4 Treatment 3 Treatment 2 Treatment 1 .2 0 −.2 Estimate 95% CI −.4 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 Age at start of the conflict Note: Figure presents coecients on interaction between exposure to months of war and age at start of the civil war in main specication. 37 Figure 6: Births per 1000 district population using the 2001 Nepal population census 40 30 Births per thousand 20 10 0 1966 1971 1976 1981 1986 1991 1996 Birth year High conflict Low conflict 95% CI Note: High conict areas are dened as districts that experienced above median level of conict intensity (months of war). Birth rates are calculated using individuals observed in the 2001 Nepal population census and their year of birth. 38 Table 1: Characteristics of the victims of civil war in Nepal Total casualties 14982 Political aliation (%) Killed 13210 Nepali Congress 3.19 Disappeared 998 CPN-UML or ML 1.50 Injuried 774 CPN Maoist (rebel) 48.32 Other parties 0.91 Perpetrator No aliation 46.07 State 9208 Maoist 5302 Occupation (%) Other 472 Agriculture 21.01 Wage laborer 2.27 Age (mean) 28.34 Employed 1.47 Female (%) 11.10 Teacher 1.68 Police 11.92 Social caste (%) Army 6.53 Bramin or Chettrey 44.76 Lawyer 0.05 Janajati, Aadibashi or Dalit 46.82 Doctor 0.04 Madeshi or Muslim 6.25 Politician 43.89 Other 2.17 Social worker 0.16 Rights activists 0.03 Education (%) Sports personality 0.05 Bachelor's degree or more 2.61 Driver 0.23 Intermediate 7.30 Student 5.50 Secondary school 26.26 Journalist 0.03 Lower secondary school 21.99 Businessman 1.57 Primary school 14.32 Ex-security personnel 0.01 Literate 15.05 Other, not clear 3.56 Illiterate 12.47 Source: Author's calculation based on the INSEC's archive on the conict victims (http: //www.insec.org.np/victim/). Note: While classes 8 to 10 are dened as secondary school, 6 and 7 are lower secondary school. Nepali Congress (Democratic) and Nepali Congress are combined as one as the former was formed due to a vertical split of Nepali Congress into two in 2002. However, the parties merged into one in 2007. Similarly, Communist Party of Nepal - Marxist Leninists (CPN-ML) was reunited with the Communist Party of Nepal - Unied Marxist Leninists (CPN-UML) in 2002 but few members refused to go along the merger forming a new party with the same name. Party's sister organizations and student wings are also accounted for while assigning party aliation. 39 Table 2: Summary statistics of women ages 20 to 49 at time of the survey Age at start of the war in 1996 Variables 0 to 29 0 to 15 16 to 29 Dierence (All women) (Treatment) (Control) Panel A: Outcomes Height (cm) 151.56 151.75 151.28 0.46*** [5.50] [5.45] [5.56] (0.17) Weight (kg) 52.04 51.36 53.05 -1.69*** [10.05] [9.52] [10.73] (0.31) Body mass index 22.63 22.29 23.14 -0.85*** [4.06] [3.86] [4.30] (0.12) Pregnancy loss (yes=1) 0.28 0.27 0.30 -0.04*** [0.45] [0.44] [0.46] (0.01) Total births 2.88 2.25 3.81 -1.56*** [1.65] [1.20] [1.77] (0.04) Age at rst birth 19.74 19.54 20.04 -0.51*** [3.19] [2.99] [3.45] (0.10) Years of education 3.83 5.05 2.04 3.01*** [4.25] [4.30] [3.48] (0.12) Employed last 12 months (yes=1) 0.72 0.69 0.76 -0.07*** [0.45] [0.46] [0.43] (0.01) Wealth index factor score 3879.78 3274.44 4776.24 -1501.79 [96436.27] [93601.64] [100507.35] (2957.08) Panel B: Exposure to conict B1: During 0-15 years Months of war 1.72 2.87 0.00 2.87*** [4.63] [5.71] [0.00] (0.14) Number of casualties 3.12 5.22 0.00 5.22*** [10.76] [13.53] [0.00] (0.32) Months inc. neighboring villages 6.22 10.43 0.00 10.43*** [11.08] [12.73] [0.00] (0.30) B2: Whole life Months of war 5.45 5.49 5.39 0.10 [8.27] [8.28] [8.26] (0.25) Number of casualties 10.41 10.55 10.20 0.35 [20.85] [21.01] [20.60] (0.64) Months inc. neighboring villages 18.67 18.71 18.61 0.10 [14.42] [14.40] [14.45] (0.44) Panel C: Controls Current age 33.48 27.85 41.83 -13.99*** [8.07] [4.45] [3.93] (0.13) Female headed HH (yes=1) 0.34 0.37 0.30 0.07*** [0.47] [0.48] [0.46] (0.01) Hight caste (yes=1) 0.39 0.37 0.41 -0.04*** [0.49] [0.48] [0.49] (0.01) Rural (yes =1) 0.86 0.86 0.86 0.00 [0.35] [0.34] [0.35] (0.01) Eastern region 0.19 0.18 0.21 -0.02** [0.39] [0.39] [0.40] (0.01) Central region 0.24 0.24 0.25 -0.00 [0.43] [0.43] [0.43] (0.01) Western region 0.21 0.21 0.21 -0.00 [0.41] [0.41] [0.41] (0.01) Mid-western region 0.21 0.22 0.20 0.02 [0.41] [0.42] [0.40] (0.01) Far-western region 0.14 0.14 0.13 0.01 [0.34] [0.35] [0.34] (0.01) Number of women 4,421 2,639 1,782 4,421 Note: Standard deviations are in brackets and standard errors are in parentheses and signi- cance levels are denoted as follows: * p<0.10, ** p<0.05, *** p<0.01. 40 Table 3: Summary statistics of children under 5 at time of the survey in 2016 Mother's age at start of the war in 1996 Variables 0 to 29 0 to 15 16 to 29 Dierence (All women) (Treatment) (Control) Panel A: Outcomes Height/age sd -1.55 -1.52 -1.85 0.33*** [1.34] [1.33] [1.39] (0.11) Weight/age sd -1.35 -1.33 -1.50 0.17* [1.07] [1.06] [1.16] (0.09) Weight/height sd -0.65 -0.65 -0.62 -0.03 [1.10] [1.09] [1.15] (0.09) Body mass index sd -0.51 -0.51 -0.45 -0.06 [1.11] [1.11] [1.14] (0.09) Panel B: Mother's exposure to conict B1: During 0-15 years Months of war 3.17 3.43 0.00 3.43*** [6.01] [6.17] [0.00] (0.49) Number of casualties 5.83 6.29 0.00 6.29*** [14.28] [14.74] [0.00] (1.16) Months inc. neighboring villages 11.79 12.73 0.00 12.73*** [13.10] [13.16] [0.00] (1.04) B2: Whole life Months of war 4.69 4.73 4.10 0.63 [7.42] [7.41] [7.64] (0.61) Number of casualties 8.85 8.92 8.09 0.82 [18.74] [18.55] [20.92] (1.53) Months inc. neighboring villages 17.15 17.20 16.54 0.66 [13.73] [13.73] [13.83] (1.12) Panel C: Controls Child sex (girl=1) 0.47 0.48 0.42 0.06 [0.50] [0.50] [0.49] (0.04) Child birth order number 2.36 2.16 4.97 -2.81*** [1.57] [1.25] [2.53] (0.11) Child is a twin (yes =1) 0.01 0.01 0.02 -0.01 [0.11] [0.11] [0.14] (0.01) Female headed HH (yes=1) 0.33 0.33 0.30 0.03 [0.47] [0.47] [0.46] (0.04) Hight caste (yes=1) 0.36 0.37 0.32 0.04 [0.48] [0.48] [0.47] (0.04) Rural (yes =1) 0.89 0.89 0.93 -0.03 [0.31] [0.31] [0.26] (0.03) Eastern region 0.18 0.18 0.24 -0.06* [0.39] [0.38] [0.43] (0.03) Central region 0.27 0.27 0.28 -0.01 [0.44] [0.44] [0.45] (0.04) Western region 0.19 0.20 0.16 0.04 [0.40] [0.40] [0.37] (0.03) Mid-western region 0.22 0.22 0.22 0.00 [0.42] [0.42] [0.41] (0.03) Far-western region 0.13 0.14 0.11 0.03 [0.34] [0.34] [0.31] (0.03) Number of children 2,168 2,007 161 2,168 Note: Standard deviations are in brackets and standard errors are in parentheses and signicance levels are denoted as follows: * p<0.10, ** p<0.05, *** p<0.01. 41 Table 4: Impact on rst generation adult stature by age at start of the civil war Height in cm Height for age sd (HAZ) (1) (2) (3) (4) (5) (6) Age 22 to 29 × Months of war 0.011 0.011 0.014 0.002 0.002 0.002 (0.033) (0.034) (0.035) (0.006) (0.006) (0.006) Age 9 to 15 × Months of war -0.026 -0.026 -0.027 -0.004 -0.004 -0.005 (0.026) (0.026) (0.025) (0.004) (0.004) (0.004) Age 4 to 8 × Months of war -0.068** -0.067** -0.071** -0.012** -0.011** -0.012** (0.031) (0.031) (0.032) (0.005) (0.005) (0.005) Age0 to 3 × Months of war -0.122*** -0.127*** -0.136*** -0.021*** -0.021*** -0.023*** (0.033) (0.032) (0.032) (0.006) (0.005) (0.005) Observations 4,421 4,421 4,421 4,418 4,418 4,418 Adjusted R-squared 0.031 0.031 0.043 0.031 0.031 0.043 Birth year and month xed eects Yes Yes Yes Yes Regional trends and other controls Yes Yes Village xed eects Yes Yes Yes Yes Yes Yes Number of clusters 383 383 383 383 383 383 Control mean (Age 16 to 21) 151.4 151.4 151.4 151.4 151.4 151.4 Note: *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Comparison cohort is age 16 to 21 and cohort 22 to 29 is a second comparison group that serves as a placebo test. Other controls include an indicator for high caste and month of interview xed-eects. 42 Table 5: Impact on rst generation adult height (cm) by alternative measure of conict Casualty count Including contiguous villages Including villages within 50 km Conict variable = (1) (2) (3) (4) (5) Own village Months of war Casualty count Months of war Casualty count Age 22 to 29 × Conict 0.005 0.022 0.004 0.011 0.001 (0.014) (0.022) (0.006) (0.016) (0.001) Age 9 to 15 × Conict -0.013 -0.006 -0.002 -0.011 0.000 (0.009) (0.017) (0.006) (0.013) (0.001) Age 4 to 8 × Conict -0.030*** -0.013 -0.006 -0.006 0.000 (0.011) (0.020) (0.006) (0.016) (0.001) Age 0 to 3 × Conict -0.050*** -0.068*** -0.019*** -0.036* -0.001 43 (0.012) (0.023) (0.007) (0.020) (0.001) Observations 4,421 4,421 4,421 4,421 4,421 Adjusted R-squared 0.042 0.042 0.042 0.040 0.040 Birth year and month xed eects Yes Yes Yes Yes Yes Regional trends and other controls Yes Yes Yes Yes Yes Village xed eects Yes Yes Yes Yes Yes Number of clusters 383 383 383 383 383 Control mean (Age 16 to 21) 151.4 151.4 151.4 151.4 151.4 Note: *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Comparison cohort is age 16 to 21 and cohort 22 to 29 is a second comparison group that serves as a placebo test. Other controls include an indicator for high caste and month of interview xed-eects. Table 6: Impact of conict at district of birth Migration Height in cm Height for age sd (HAZ) (1) (2) (3) (4) (5) (6) (7) (8) (9) Age 22 to 29 × Months of war 0.001 0.001 0.001 -0.008 -0.007 -0.006 -0.000 0.000 0.000 (0.001) (0.001) (0.001) (0.019) (0.019) (0.018) (0.003) (0.003) (0.003) Age 9 to 15 × Months of war -0.001 -0.001 -0.001 -0.010 -0.008 -0.007 -0.002 -0.001 -0.001 (0.001) (0.001) (0.001) (0.012) (0.011) (0.011) (0.002) (0.002) (0.002) Age 4 to 8 × Months of war -0.002 -0.001 -0.001 0.001 0.004 0.001 0.000 0.001 -0.000 (0.001) (0.001) (0.001) (0.013) (0.013) (0.012) (0.002) (0.002) (0.002) Age 0 to 3 × Months of war -0.001 -0.001 -0.001 -0.030** -0.030** -0.030** -0.005** -0.005** -0.005** 44 (0.001) (0.001) (0.001) (0.012) (0.013) (0.013) (0.002) (0.002) (0.002) Observations 4,421 4,421 4,421 4,421 4,421 4,421 4,418 4,418 4,418 Adjusted R-squared 0.064 0.066 0.123 0.008 0.008 0.019 0.008 0.007 0.019 Birth year and month xed eects Yes Yes Yes Yes Yes Yes Regional trends and other controls Yes Yes Yes District xed eects Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of clusters 74 74 74 74 74 74 74 74 74 Control mean (Age 16 to 21 0.253 0.253 0.253 151.4 151.4 151.4 -2.056 -2.056 -2.056 Note: Conict is dened at district level i.e. district of birth. *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at the district of birth. All ages in the table refer to age at the start of the war. Comparison cohort is age 16 to 21 and cohort 22 to 29 is a second comparison group that serves as a placebo test. Other controls include an indicator for high caste and month of interview xed-eects. Table 7: Impact on rst generation reproductive health by age at start of the civil war (1) (2) (3) (4) (5) Ever had a Ever had a Ever had an Total live Number of stillbirth miscarriage abortion births infant deaths Age 22 to 29 × Months of war 0.002 0.001 -0.007** -0.004 -0.000 (0.001) (0.002) (0.003) (0.008) (0.001) Age 9 to 15 × Months of war 0.001 0.000 -0.004 0.002 0.000 (0.001) (0.002) (0.002) (0.006) (0.001) Age 4 to 8 × Months of war 0.002 0.001 -0.005* 0.013* 0.000 (0.001) (0.002) (0.003) (0.007) (0.001) Age0 to 3 × Months of war 0.003*** 0.006* -0.002 0.024*** -0.001 (0.001) (0.003) (0.004) (0.007) (0.002) Observations 4,421 4,421 4,421 4,421 4,421 Adjusted R-squared 0.022 0.018 0.061 0.466 0.003 Birth year and month xed eects Yes Yes Yes Yes Yes Regional trends and other controls Yes Yes Yes Yes Yes Village xed eects Yes Yes Yes Yes Yes Number of clusters 383 383 383 383 383 Control mean (Age 16 to 21) 0.0764 0.164 0.142 3.520 0.003 Note: *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Comparison cohort is age 16 to 21 and cohort 22 to 29 is a second comparison group that serves as a placebo test. Other controls include an indicator for high caste and month of interview xed-eects. 45 Table 8: Impact on sexual behavior and marriage by age at start of the civil war (1) (2) (3) (4) (5) (6) (7) Ever Number Age at rst Age at Age at Number of Smoke or use of sex sexual rst rst marriages chew contraceptive partners intercourse birth cohabitation or unions tobacco Age 22 to 29 × Months of war -0.001 -0.003 -0.026 -0.030* 0.001 -0.020 -0.005** (0.002) (0.003) (0.019) (0.017) (0.002) (0.020) (0.002) Age 9 to 15 × Months of war -0.001 0.001 0.018 0.016 0.000 0.032* -0.002 (0.001) (0.001) (0.015) (0.016) (0.001) (0.018) (0.002) Age 4 to 8 × Months of war -0.002 -0.003 0.012 0.009 0.000 0.024 -0.001 (0.002) (0.003) (0.019) (0.020) (0.001) (0.020) (0.002) Age0 to 3 × Months of war 0.001 0.000 -0.028 -0.009 0.000 0.002 0.000 46 (0.003) (0.001) (0.025) (0.023) (0.001) (0.021) (0.002) Observations 4,421 4,421 4,421 4,420 4,420 4,421 4,421 Adjusted R-squared 0.110 -0.013 0.144 0.146 0.039 0.108 0.155 Birth year and month xed eects Yes Yes Yes Yes Yes Yes Yes Regional trends and other controls Yes Yes Yes Yes Yes Yes Yes Village xed eects Yes Yes Yes Yes Yes Yes Yes Number of clusters 383 383 383 383 383 383 383 Control mean (Age 16 to 21) 0.872 1.065 17.56 17.50 1.059 19.89 0.195 Note: *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Comparison cohort is age 16 to 21 and cohort 22 to 29 is a second comparison group that serves as a placebo test. Other controls include an indicator for high caste and month of interview xed-eects. Table 9: Impact on education, employment and wealth by age at start of the civil war Education Employment (1) (2) (3) (4) (5) (6) (7) Years Completed Employed Professional Agri - manual wealth index of SLC in last sales own work factor schooling 12 months clerical etc farm score Age 22 to 29 × Months of war -0.015 -0.003 -0.005 -0.007** 0.006** 0.001 -182.432 (0.025) (0.002) (0.003) (0.003) (0.003) (0.002) (440.164) Age 9 to 15 × Months of war 0.003 -0.001 -0.003 -0.001 0.004** -0.003 -504.890 (0.018) (0.002) (0.002) (0.002) (0.002) (0.002) (323.558) Age 4 to 8 × Months of war 0.007 -0.003 0.001 -0.005* 0.004* 0.002 -675.200* (0.025) (0.003) (0.003) (0.003) (0.002) (0.003) (367.459) Age0 to 3 × Months of war -0.020 -0.006** -0.002 -0.004 0.005* -0.001 -977.280** 47 (0.031) (0.003) (0.003) (0.004) (0.003) (0.004) (393.707) Observations 4,421 4,421 4,421 3,188 3,188 3,188 4,421 Adjusted R-squared 0.440 0.215 0.259 0.300 0.382 0.165 0.712 Birth year and month xed eects Yes Yes Yes Yes Yes Yes Yes Regional trends and other controls Yes Yes Yes Yes Yes Yes Yes Village xed eects Yes Yes Yes Yes Yes Yes Yes Number of clusters 383 383 383 381 381 381 383 Control mean (Age 16 to 21) 2.472 0.0831 0.765 0.200 0.722 0.0781 3484 Note: School leaving certicate (SLC) is a national exam that everyone takes at the end of grade ten. DHS uses asset ownership such as televisions and bicycles, materials used for housing construction, and types of water access and sanitation facilities to calculate wealth index using factor analysis (see Rutstein, and Johnson, 2004 for detail). *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Comparison cohort is age 16 to 21 and cohort 22 to 29 is a second comparison group that serves as a placebo test. Other controls include an indicator for high caste and month of interview xed-eects. Table 10: Second generation health impact (1) (2) (3) (4) Height for Weight for Weight for Body mass age sd height sd age sd index sd (HAZ) (WHZ) (WAZ) (BMIZ) Mother's age 22 to 29 × 0.048 -0.032 0.005 -0.045 Mother's exposure to conict (0.046) (0.032) (0.040) (0.028) Mother's age 9 to 15 × -0.010 -0.041** -0.034* -0.044*** Mother's exposure to conict (0.033) (0.017) (0.020) (0.015) Mother's age 4 to 8 × -0.005 -0.039** -0.025 -0.040** Mother's exposure to conict (0.030) (0.017) (0.022) (0.017) Mother's age 0 to 3 × -0.011 -0.030** -0.025 -0.031** Mother's exposure to conict (0.027) (0.015) (0.020) (0.014) Observations 2,165 2,163 2,169 2,164 Adjusted R-squared 0.131 0.114 0.132 0.110 Mother and other controls Yes Yes Yes Yes Children contrls Yes Yes Yes Yes Village xed eects Yes Yes Yes Yes Number of clusters 373 373 373 373 Control mean (Mother's age 16 to 21) -1.836 -0.580 -1.472 -0.423 Note: Sample is children aged 0 to 59 months at the time of the survey. *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. Mother's age in the table refers to mother's age at the start of the war. Comparison cohort is children born to mothers whose age was 16 to 21 at the start of the war. Children born to mother's cohort 22 to 29 is a second comparison group that serves as a placebo test. Mother's controls are mother's years of birth xed eects, mother's month of birth xed eects, region specic trends. Household controls are indicator for high caste, female headed households, and whether residing in a rural area. Child controls are indicator if child is a girl, a twin, birth order xed eect, and xed eects for child years of birth, month of birth, and month of anthropometric measurements. Reported outcomes are z-scores based on the WHO anthropometric measurement standards. 48 Table 11: Second generation health impact by mother's district of birth (1) (2) (3) (4) Height for Weight for Weight for Body mass age sd height sd age sd index sd Mother's age 22 to 29 × 0.016 -0.007 -0.001 -0.011 Mother's exposure to conict (0.013) (0.008) (0.009) (0.008) Mother's age 9 to 15 × -0.006 -0.014** -0.016** -0.015** Mother's exposure to conict (0.007) (0.006) (0.006) (0.006) Mother's age 4 to 8 × -0.003 -0.013** -0.012** -0.015** Mother's exposure to conict (0.006) (0.006) (0.006) (0.006) Mother's age 0 to 3 × -0.003 -0.013** -0.012** -0.016*** Mother's exposure to conict (0.006) (0.006) (0.005) (0.005) Observations 2,165 2,163 2,169 2,164 Adjusted R-squared 0.181 0.094 0.130 0.095 Mother and household controls Yes Yes Yes Yes Children controls Yes Yes Yes Yes Mother's district of birth xed eects Yes Yes Yes Yes Number of clusters 74 74 74 74 Control mean (Age 16 to 21) -1.836 -0.580 -1.472 -0.423 Note: Conict is dened at district level i.e. mother's district of birth. Sample is children aged 0 to 59 months at the time of the survey. *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. Mother's age in the table refers to mother's age at the start of the war. Comparison cohort is children born to mothers whose age was 16 to 21 at the start of the war. Children born to mother's cohort 22 to 29 is a second comparison group that serves as a placebo test. Mother's controls are mother's years of birth xed eects, mother's month of birth xed eects, region specic trends. Household controls are indicator for high caste, female headed households, and whether residing in a rural area. Child controls are indicator if child is a girl, a twin, birth order xed eect, and xed eects for child years of birth, month of birth, and month of anthropometric measurements. Reported outcomes are z-scores based on the WHO anthropometric measurement standards. 49 A Appendix Figure A.1: Administrative map of Nepal Note: The map represents administrative areas before the 2015 constitution when Nepal was divided into 5 de- velopment regions, 75 districts and about 4000 rural (village development committees) and urban (municipalities) areas. 50 Figure A.2: Conict intensity heterogeneity: Casualty count 51 Figure A.3: Nepal Demographic Health Survey 2016 Coverage 52 Figure A.4: Conict intensity heterogeneity: Casualty count and NDHS 2016 clusters 53 Figure A.5: Typical growth velocity curve Source: Adopted from Tanner, Whitehouse, and Takaishi (1966). Archives of disease in childhood vol. 41,220 (1966): 613-35. 54 Table A.1: Impact on rst generation adult stature using alternative specication Height in cm Height for age sd (1) (2) (3) (4) (5) (6) Months of war during 0 to 3 years -0.026 -0.011 -0.103 -0.008 -0.007 -0.021 (0.334) (0.338) (0.361) (0.056) (0.057) (0.061) Months of war during 4 to 8 years -0.130** -0.171*** -0.173*** -0.022** -0.029*** -0.029** (0.056) (0.064) (0.067) (0.009) (0.011) (0.011) Months of war during 9 to 15 years 0.009 -0.052* -0.060** 0.002 -0.009* -0.010** (0.024) (0.028) (0.028) (0.004) (0.005) (0.005) Observations 4,421 4,421 4,421 4,418 4,418 4,418 Adjusted R-squared 0.027 0.029 0.041 0.027 0.029 0.041 Birth year and month xed eects Yes Yes Yes Yes Regional trends and other controls Yes Yes Village xed eects Yes Yes Yes Yes Yes Yes Number of clusters 383 383 383 383 383 383 Outcome mean 151.6 151.6 151.6 -2.034 -2.034 -2.034 Note: *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Other controls include an indicator for high caste and month of interview xed-eects. Specication used is Yimntvdr = β0 + β1 × Exposure during 0 to 3 years + β2 × Exposure during 4 to 8 years + β3 × Exposure during 9 to 15 years + αt + ηm + δv + γr T + Xi + ωn + εimntvdr , where Y is a height of a woman i born in a month m and year t, and residing in village v. district d and development region r. Conict variables are dened as women's exposure to conict during 0 to 3 years, 4 to 8 years, and 9 to 15 years of age. All other variables have same meaning as the main specication equation 7. Reported in the table are the estimated β1 , β2 , and β3 coecients. 55 Table A.2: Impact on rst generation adult height (cm) using alternative specication and alternative measure of conict Casualty count Including contiguous villages Conict variable = (1) (2) (3) Own village Months of war Casualty count Months of war during 0 to 3 years -0.033 -0.003 -0.007 (0.148) (0.084) (0.040) Months of war during 4 to 8 years -0.069*** -0.101*** -0.027*** (0.019) (0.034) (0.010) Months of war during 9 to 15 years -0.021** -0.019 -0.006 (0.010) (0.018) (0.005) Observations 4,421 4,421 4,421 Adjusted R-squared 0.041 0.041 0.041 Birth year and month xed eects Yes Yes Yes Regional trends and other controls Yes Yes Yes Village xed eects Yes Yes Yes Number of clusters 383 383 383 Outcome mean 151.6 151.6 151.6 Note: *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Other controls include an indicator for high caste and month of interview xed-eects. Specication used is Yimtvdr = β0 + β1 × Exposure during 0 to 3 years + β2 × Exposure during 4 to 8 years + β3 × Exposure during 9 to 15 years + αt + ηm + δv + γr T + Xi + εimtvdr , where Y is a height of a woman i born in a month m and year t, and residing in village v. district d and development region r. Conict variables are dened as women's exposure to conict during 0 to 3 years, 4 to 8 years, and 9 to 15 years of age. All other variables have same meaning as the main specication equation 6. Reported in the table are the estimated β1 , β2 , and β3 coecients. 56 Table A.3: Impact on rst generation adult height (cm) - women living in the same district as birth Height in cm Height for age sd (1) (2) (3) (4) (5) (6) Age 22 to 29 × Months of war -0.026 -0.028 -0.023 -0.004 -0.005 -0.004 (0.047) (0.046) (0.047) (0.008) (0.008) (0.008) Age 9 to 15 × Months of war -0.003 -0.001 -0.004 -0.000 -0.000 -0.001 (0.039) (0.039) (0.039) (0.007) (0.007) (0.007) Age 4 to 8 × Months of war -0.069* -0.066 -0.074* -0.011* -0.011 -0.012 (0.041) (0.043) (0.044) (0.007) (0.007) (0.007) Age0 to 3 × Months of war -0.145*** -0.148*** -0.162*** -0.024*** -0.025*** -0.027*** (0.046) (0.046) (0.046) (0.008) (0.008) (0.008) Observations 3,243 3,243 3,243 3,242 3,242 3,242 Adjusted R-squared 0.024 0.023 0.032 0.025 0.023 0.033 Birth year and month xed eects Yes Yes Yes Yes Regional trends and other controls Yes Yes Village xed eects Yes Yes Yes Yes Yes Yes Number of clusters 380 380 380 380 380 380 Control mean (Age 16 to 21) 151.3 151.3 151.3 -2.089 -2.089 -2.089 Note: Sample is limited to women living in the same district as their birth. *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. All ages in the table refer to age at the start of the war. Comparison cohort is age 16 to 21 and cohort 22 to 29 is a second comparison group that serves as a placebo test. Other controls include an indicator for high caste and month of interview xed-eects. 57 Table A.4: Control experiment of district birth rates (yearly births) Conict: Months of Conict: Casualties per 1000 war in district district 1991 population (1) (2) (3) (4) (5) (6) Births between 1966 to 1974 × Conict -0.009 -0.009 -0.009 -0.016 -0.016 -0.016 (0.006) (0.006) (0.006) (0.136) (0.138) (0.139) Births between 1981 to 1987 × Conict -0.007 -0.007 -0.007 -0.142 -0.142 -0.142 (0.011) (0.011) (0.011) (0.179) (0.182) (0.183) Births between 1988 to 1992 × Conict 0.026 0.026 0.026 0.426 0.426 0.426 (0.022) (0.022) (0.022) (0.397) (0.403) (0.406) Births between 1993 to 1996 × Conict 0.043 0.043 0.043 0.732* 0.732 0.732 (0.026) (0.027) (0.027) (0.434) (0.441) (0.444) Observations 2,325 2,325 2,325 2,325 2,325 2,325 Adjusted R-squared 0.447 0.437 0.809 0.446 0.436 0.808 District xed eects Yes Yes Yes Yes Birth year xed eects Yes Yes Number of clusters 75 75 75 75 75 75 Control mean 15.69 15.69 15.69 15.69 15.69 15.69 (Births between 1975 to 1980) Note: Yearly births are calculated using individuals observed in the 2001 Nepal population census and their year of birth normalized to 1000 districts inhabitants in 2001. *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at district level. 58 Table A.5: Second generation health impact using alternative measure of conict Casualty count Months of war Casualty count Conict variable = own village including contiguous villages including contiguous villages (1) (2) (3) (4) (5) (6) (7) (8) (9) HAZ WHZ BMIZ HAZ WHZ BMIZ HAZ WHZ BMIZ Mother's age 22 to 29 × 0.028 -0.009 -0.015 0.063*** -0.003 -0.015 0.023*** -0.003 -0.007 Mother's exposure to conict (0.024) (0.019) (0.019) (0.022) (0.016) (0.016) (0.007) (0.006) (0.006) Mother's age 9 to 15 × -0.009 -0.010*** -0.011*** -0.001 -0.018* -0.020* -0.002 -0.006** -0.007** Mother's exposure to conict (0.009) (0.004) (0.004) (0.014) (0.010) (0.010) (0.005) (0.003) (0.003) Mother's age 4 to 8 × -0.005 -0.011* -0.010* -0.004 -0.018* -0.020** -0.001 -0.007** -0.007** Mother's exposure to conict (0.008) (0.006) (0.006) (0.013) (0.010) (0.010) (0.004) (0.003) (0.003) Mother's age 0 to 3 × -0.008 -0.005 -0.005 -0.000 -0.013 -0.016 -0.001 -0.005* -0.005* Mother's exposure to conict (0.006) (0.005) (0.005) (0.013) (0.010) (0.010) (0.004) (0.003) (0.003) 59 Observations 2,165 2,163 2,164 2,165 2,163 2,164 2,165 2,163 2,164 Adjusted R-squared 0.229 0.119 0.116 0.231 0.119 0.116 0.232 0.120 0.117 Mother and household controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Child controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Village xed eects Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of clusters 373 373 373 373 373 373 373 373 373 Control mean (Age 16 to 21) -1.836 -0.580 -0.423 -1.836 -0.580 -0.423 -1.836 -0.580 -0.423 Note: Sample is children aged 0 to 59 months at the time of the survey. *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Treatment on the treated estimates are reported based on a dierence-in-dierence specication. Standard errors are clustered at village level. Mother's age in the table refers to mother's age at the start of the war. Comparison cohort is children born to mothers whose age was 16 to 21 at the start of the war. Children born to mother's cohort 22 to 29 is a second comparison group that serves as a placebo test. Mother's controls are mother's years of birth xed eects, mother's month of birth xed eects, region specic trends. Household controls are indicator for high caste, female headed households, and whether residing in a rural area. Child controls are indicator if child is a girl, a twin, birth order xed eect, and xed eects for child years of birth, month of birth, and month of anthropometric measurements. Reported outcomes are z-scores based on the WHO anthropometric measurement standards.