Education and Civil Conﬂict in Nepal

Christine Valente

Between 1996 and 2006, Nepal experienced violent civil conﬂict as a consequence of a
Maoist insurgency, which many argue also brought about an increase in female empow-
erment. This paper exploits variations in exposure to conﬂict by birth cohort, survey
date, and district to estimate the impact of the insurgency on education outcomes.
Overall conﬂict intensity, measured by conﬂict casualties, is associated with an increase
in female educational attainment, whereas abductions by Maoists, which often targeted
school children, have the reverse effect. Male schooling tended to increase more rapidly
in areas where the ﬁghting was more intense, but the estimates are smaller in magnitude
and more sensitive to speciﬁcation than estimates for females. Similar results are ob-
tained across different speciﬁcations, and robustness checks indicate that these ﬁndings
are not due to selective migration. JEL codes: I20, J12, O12

Between 1996 and 2006, Nepal experienced violent civil conﬂict as a conse-
quence of a Maoist insurgency. This paper investigates the impact of being
exposed to this insurgency at a young age on education outcomes.
This study makes both an empirical and a methodological contribution to the
growing literature on the impact of civil conﬂict on human capital formation.
First, this paper extends our understanding of the impact of civil conﬂict on edu-
cation to include a conﬂict of moderate intensity. With just over 13,000 casual-
ties and less than 1 percent of the population forcibly displaced, the level of
violence considered in this study was much lower than in conﬂict episodes
considered in previous research.1 Second, two alternative identiﬁcation strategies

Christine Valente is a Lecturer (Assistant Professor) at the University of Bristol, United Kingdom; her
email address is christine.valente@bristol.ac.uk. The author thanks INSEC for sharing their conﬂict data
and Martha Ainsworth, Quy-Toan Do, Helge Holterman, Steve McIntosh, Gudrun Østby, Kati Schindler,
Olga Shemyakina, Helen Simpson, Sarah Smith, Frank Windmeijer, Hassan Zaman, the World Bank’s
Nepal Country Director’s ofﬁce, and participants at two Gender and Conﬂict Research Workshops at the
World Bank (Washington) and Peace Research Institute Oslo for their useful comments. The author also
thanks three anonymous referees for their valuable comments and suggestions. This work was supported
by the World Bank-Norway Trust Fund. The views expressed in this paper are those of the author alone
and do not necessarily reﬂect those of the funding agencies. A supplemental appendix to this article is
available at http://wber.oxfordjournals.org/.
1. The estimated number of individuals forcibly displaced in Nepal is approximately 200,000
(USAID, 2007). Studies reviewed in this paper have considered the impact of conﬂict on education in
Tajikistan (between 50,000 and 100,000 deaths and 10 percent of the country’s population internally
displaced in the two worst years of conﬂict, Shemyakina, 2011b), Guatemala (200,000 deaths during the
worst conﬂict period, Chamarbagwala and Mora ´ n, 2011), Rwanda (nearly 10 percent of the population
killed, Akresh and de Walque, 2008), and Peru ( just under 70,000 deaths, Leo ´ n 2012).

THE WORLD BANK ECONOMIC REVIEW, VOL. 28, NO. 2, pp. 354– 383 doi:10.1093/wber/lht014
Advance Access Publication June 3, 2013
# The Author 2013. Published by Oxford University Press on behalf of the International Bank
for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions,
please e-mail: journals.permissions@oup.com

354
Valente 355

are employed to increase conﬁdence in the reliability of the estimates: the ﬁrst
relies on variation in conﬂict exposure across birth cohorts and geographic areas
in a single survey, as is standard in the literature (e.g., Akresh and de Walque
2008; Shemyakina 2011a; Chamarbagwala and Mora ´ n 2011), and the second
relies on variation in exposure to conﬂict among school-aged individuals
between household surveys and geographic areas.
Educational attainment is generally expected to be adversely affected by expo-
sure to armed civil conﬂict. Direct youth enrollment in the military, limited mo-
bility, and the destruction of schools may all negatively affect the ability of
children to attend school. Increased poverty may drive parents to remove children
from school (to avoid direct costs) and put them to work (to avoid opportunity
costs). Political instability and reduced life expectancy may decrease expected
returns to education and, in turn, reduce investments in human capital.
Moreover, the schooling of girls is often more sensitive to worsening economic
conditions than that of boys. A conﬂict environment may also hinder the func-
tioning of education programs by weakening government institutions and impos-
ing logistical and staff security challenges on local and international NGOs.
However, the general expectation that schooling is disrupted in conﬂict areas
may not be well founded in the particular case of Nepal. National trends do not
indicate an increase in poverty coinciding with the conﬂict but rather a steady
decline in poverty (World Bank 2005). Despite difﬁculties with public service
provision, basic health and education services have been maintained (Armon
et al. 2004). In addition, the insurgency may have had a positive effect on school-
ing outcomes despite the ﬁghting through both intended and unintended conse-
quences of the Maoist presence. The insurgents have been reported to police
teacher absenteeism (Hart 2001; see also Collins 2006 and Devkota and van
Teijlingen 2010, for a similar argument regarding health workers) and have ex-
plicitly opposed caste- and ethnicity-biased traditions; these actions may have
directly beneﬁted both male and female education. In addition, the insurgents
have publicly opposed gender inequality, including gender inequality in access to
schooling. For instance, it has been reported that “the Maoists have taken a
strong stand on this issue – insisting that girls of school age attend the local facil-
ities, even to the point of holding parents accountable and liable to punishment
for the non-attendance of their daughters” (Hart 2001, p. 32). Although the
egalitarian rhetoric has not been followed completely in practice, a number of
women were directly involved in combat, and there is anecdotal evidence of im-
proved conditions for women in areas controlled by the Maoists, such as decreas-
es in polygamy, domestic violence, and alcoholism (Lama-Tamang et al. 2003;
Manchanda 2004; Geiser 2005; Aguirre and Pietropaoli 2008; Arin ˜ o 2008). The
diffusion of the egalitarian Maoist ideology may also have increased the aspira-
tions of young girls for their own education and the aspirations of parents for
their daughters’ education. One unintended aspect of the insurgency may have
contributed to improving schooling outcomes, especially for girls: female labor
356 THE WORLD BANK ECONOMIC REVIEW

force participation increased (Menon and Rodgers 2011), and there is evidence
that when women have more control over household expenditure investments in
children increase, especially for girls (e.g., Thomas 1990; Duﬂo 2003). Thus, in
the case of Nepal, contrary to most episodes of violent conﬂict, the direction of
the effect of the Maoist insurgency on schooling outcomes seems unclear a
priori.
Nonetheless, one particular aspect of the Nepalese conﬂict is likely to have
been unambiguously detrimental to education: the common insurgent practice of
abducting civilians. Parents may have been deterred from sending their children
to school out of fear that they would be abducted by the insurgents (Human
Rights Watch 2004). Quoting ﬁgures from the Informal Sector Service Center
(INSEC), UNESCO (2010) reports that between 2002 and 2006, the Maoists ab-
ducted 10,621 teachers and 21,998 students ( p.8). According to additional data
provided by INSEC, the total number of abductions by Maoist forces during the
conﬂict amounted to more than 85,000. Although most abductees were seeming-
ly returned unharmed after a few days of intensive indoctrination (Becker 2009;
Macours 2011), a number of youths joined the Maoist ﬁghters (in 2003, an esti-
mated 30 percent of Maoist forces were aged 14 –18 years). The indoctrination
sessions held during abductions are likely to have played a part in their recruit-
ment.
In this paper, I exploit differences in the intensity of violence experienced by
individuals born at different times, surveyed at different times, and in different
districts to shed light on the ways in which experiencing the insurgency at a
young age affected educational outcomes.
Individual data from the 2001 and 2006 Demographic and Health Surveys
(DHS) of Nepal are merged with detailed conﬂict data collected by INSEC,
namely, the number of conﬂict fatalities, school destructions, and abductions by
Maoists at the district level.
I ﬁnd that overall conﬂict intensity, as measured by conﬂict casualties, was as-
sociated with an increase in female educational attainment, whereas abductions
by Maoists, which often targeted school children, had the reverse effect. Male
schooling also tended to increase more rapidly in areas where the ﬁghting was
more intense, but the estimates are smaller in magnitude and more sensitive to
speciﬁcation. Similar results are obtained across different identiﬁcation strate-
gies, and robustness checks indicate that these ﬁndings are not due to selective
migration.
In the next section, I review the existing evidence on the impact of armed con-
ﬂict on education outside Nepal with an emphasis on male-female differences. I
then present the Nepalese conﬂict in section II, the data in section III, the empiri-
cal strategy in section IV, and the estimation results in section V. Section VI
concludes.
Valente 357

I . L I T E R AT U R E RE V I E W

A number of cross-country analyses suggest that political instability has large
negative effects on growth but that recovery to equilibrium levels tends to be
rapid (see Blattman and Miguel 2010, for a review). At the microeconomic level,
the results of an emerging body of literature on the impact of war-related destruc-
tion or civil conﬂict on educational attainment show that violent conﬂict often
leads to worse educational outcomes, but estimates vary substantially by conﬂict,
gender, and educational level. Overall, girls in postprimary education appear to
experience the worst effects.
In war-torn Germany and Austria, school-aged individuals exposed to war re-
ceived fewer years of education (Ichino and Winter-Ebmer 2004; Akbulut-Yuksel
2009). In Guatemala, where the worst period of the Guatemalan civil war saw
nearly 200,000 deaths, Chamarbagwala and Mora ´n (2011) ﬁnd that individuals
who were of schooling age in departments that were more affected by the war com-
pleted fewer years of schooling and that this effect was much more marked for
girls. In Bosnia and Herzegovina, Swee (2009) estimates that cohorts of children
exposed to greater conﬂict intensity at the municipal level were less likely to com-
plete secondary schooling, but primary schooling attainment was unaffected.
Shemyakina (2011a) ﬁnds that girls (but not boys) who were of schooling age
during the Tajik civil war were less likely to complete mandatory schooling in
areas severely affected by conﬂict events. Rodriguez and Sanchez (2009) ﬁnd that
in Colombia, children aged 12 years and older who were exposed to more violence
at the municipal level were more likely to drop out of school and enter the labor
market. Leo ´ n (2012) ﬁnds that individuals who were born and raised in Peruvian
districts that were more affected by conﬂict-related violence completed fewer years
of education. Three recent papers, one by Akresh and de Walque (2008) and two
by Annan, Blattman, and colleagues (Annan et al. 2009; Blattman and Annan
2010), illustrate the marked heterogeneity in ﬁndings on the impact of civil conﬂict
across demographic groups and across conﬂicts. Akresh and de Walque (2008) es-
timate that cohorts of children exposed to the extremely violent Rwandan geno-
cide, which killed 10 percent of the country’s population, completed 18.3 percent
fewer years of education. However, contrary to results from Guatemala, for
example (Chamarbagwala and Mora ´n 2011), they ﬁnd that due to the nature of
the conﬂict, nonpoor, male individuals were more negatively affected. Studying
the effect of forced recruitment into the Ugandan Lord’s Liberation Army, Annan
et al. (2009) and Blattman and Annan (2010) ﬁnd dramatically different effects for
men and for women in the opposite direction of those obtained by, for example,
Shemyakina (2011a) and Chamarbagwala and Mora ´n (2011). The abducted men
in their sample, who were abducted, on average, for just over 15 months, experi-
enced much worse educational attainment and labor market outcomes as well as
poorer psychological health (Blattman and Annan 2010). However, these authors
ﬁnd no such effects for female abductees, which they attribute to the lack of oppor-
tunities for women in general (Annan et al. 2009).
358 THE WORLD BANK ECONOMIC REVIEW

II. CONFLICT IN NEPAL

Nepal was an absolute monarchy until 1990. Despite multiparty democratic
elections in 1991, a Maoist insurgency broke out in February 1996 in the Rolpa
district and ended in 2006. The insurgency was initially concentrated in a few
Communist strongholds in Western Nepal, but by the end of the war,
conﬂict-related casualties were recorded in 73 of the 75 Nepalese districts. The
Maoist presence varied from sporadic attacks to the organization of local govern-
ments and law courts. Over the course of the conﬂict, Maoists attacked govern-
ment targets, such as army barracks, police posts, and local government
buildings (Do and Iyer 2010). They were also reported to terrorize, loot, abduct,
and physically assault civilians (Bohara et al. 2006). However, government
security forces also killed civilians and were accused of using children for spying,
torturing, displacing, and summarily convicting civilians (Bohara et al. 2006).
The principal objective of the insurgents was the creation of a constituent as-
sembly to draft a new constitution. Other important stated aims were land redis-
tribution and equality for all castes, language groups, and women.
A crucial moment in the conﬂict was the Maoists’ abandonment of a short-
lived cease ﬁre in November 2001. From that point, the government’s response
intensiﬁed dramatically, involving the Royal Nepal Army and leading to an esca-
lation of violence (see ﬁgure S1 in the supplemental appendix, available at http
://wber.oxfordjournals.org/). Building on opposition to King Gyanendra’s au-
thoritative reaction to the prolonged conﬂict, the Maoists joined forces with
some of the country’s major political parties, leading to the signing of a peace
agreement in November 2006 and the creation of an interim government led by a
power-sharing coalition that included the Maoists.
The intensity of conﬂict varied substantially across the districts of Nepal, as il-
lustrated in ﬁgure 1, which depicts the distribution of districts between the three
terciles of conﬂict deaths per 1,000 inhabitants. One speciﬁc characteristic of the
Nepalese conﬂict that is likely to be particularly relevant for an analysis of educa-
tional outcomes is the insurgents’ practice of abducting civilians, particularly
school children, en masse for short periods of intensive indoctrination. As illus-
trated by ﬁgure S2, there is a positive correlation between the number of abduc-
tions by Maoists and the intensity of ﬁghting as measured by conﬂict-related
casualties, but the relationship is not systematic. Hence, it is possible to consider
the effect of abductions over and above that of overall conﬂict intensity.2
Districts with the highest proportion of abductees among the population are
found in the middle tercile, which may be due to a lesser need for indoctrination
in Maoist strongholds.3 Districts in the top quartile of the distribution of

2. The correlation coefﬁcient between total conﬂict deaths and abductions per 1,000 inhabitants is 0.14.
3. Hutt (2005) also links abductions to the weakness of support for the Maoists: “The Maoists know
that much of their support is hollow and based on fear. Maoist cadres have taken to mounting temporary
abductions of large numbers of school teachers and students, who are taken to remote locations and
subjected to political indoctrination sessions” (Hutt, 2005, p.86).
Valente 359

F I G U R E 1. Conﬂict Intensity across Districts of Nepal

Notes: Author’s calculations are based on casualties recorded in INSEC (2009) and district
population ﬁgures from the 1991 population census (Central Bureau of Statistics, 2009). District
terciles are deﬁned by the distribution of total district casualties per 1,000 inhabitants.

abductions per 1,000 inhabitants that are not also in the top quartile of the distri-
bution of casualties per 1,000 inhabitants tend to be found at the far western or
eastern borders, close to districts characterized by intense ﬁghting (i.e., numerous
casualties).
Several arguments have been advanced to explain the district variation in the
intensity of the insurgency, including geography (Murshed and Gates 2005;
Bohara et al. 2006; Do and Iyer 2010), poverty (Murshed and Gates 2004; Do
and Iyer 2010), a lack of political participation (Bohara et al. 2006), and inter-
group inequality (Murshed and Gates 2005; Macours 2011). Determinants of
district conﬂict intensity are therefore likely to be correlated with the explained
variables of interest, which could give rise to omitted variable bias. As long as
the omitted variables in question are constant over time, the inclusion of district
ﬁxed effects will sufﬁce to remove any bias. If there are time-varying omitted var-
iables correlated with both conﬂict intensity and the explained variables of inter-
est, the inclusion of district ﬁxed effects will not remove all potential biases, and
additional steps must be taken to shed light on the causal impact of the insurgen-
cy. In section IV, I test for the presence of such time-varying omitted variables
and discuss how I address potential threats to identiﬁcation.
Despite the civil conﬂict, Nepal has experienced steady growth in real gross
GDP (5 percent per year between 1995/96 and 2003/04), an additional increase
360 THE WORLD BANK ECONOMIC REVIEW

in disposable income due to substantial ﬂows of remittances from abroad (repre-
senting 12.4 percent of the GDP), a steady decrease in poverty over the period
(from 42 percent in 1995/96 to 31 percent in 2003/04), and an improvement in
human development indicators, such as primary school enrollment (up from 57
percent to 73 percent) and child mortality, which decreased by 5 percent per year
(World Bank 2005; Macours 2011).
However, the positive outlook for Nepal as a whole may mask unequal pro-
gress due to heterogeneous conﬂict intensity across districts. Indeed, national
trends may hide a slower decrease in poverty, or even an increase in poverty, in
more conﬂict-affected areas. In this paper, I exploit variation in the intensity of
exposure to violent conﬂict by birth cohort, survey year, and district to investi-
gate differential changes in primary educational attainment and completed years
of education across districts that experienced varying degrees of violence.

I I I . D ATA

DHS have been conducted in a number of developing countries as part of the
Measure DHS project, a reputable USAID-funded project. The second and third
DHS in Nepal were conducted in 2001 and 2006, respectively, and are nationally
representative repeated cross-sections. The timing of these surveys is particularly
useful because the surveys either preceded or followed the bulk of the ﬁghting.4
For each DHS, a household survey collected the usual individual demographic
and education data as well as household-level socioeconomic information. More
detailed information was then collected from all women and a subset of men of
reproductive age (if ever married, in the case of the 2001 survey). The data used
in this analysis come from the household survey as well as migration information
from the detailed interviews with women in the 2006 DHS.5
Summary statistics by district conﬂict intensity and speciﬁcation subsample
can be found in table S1.
The data used to measure conﬂict intensity are taken from electronic ﬁles pro-
vided by INSEC, an independent, well-regarded human rights NGO based in
Kathmandu with reporters in each of the 75 Nepalese districts who monitor
human rights violations. The INSEC data ﬁles contain the number of
conﬂict-related deaths per month per district of Nepal between February 1996
and December 2006 as well as the total number of school destructions and ab-
ductions by Maoists at the district level, which are used to construct most mea-
sures of exposure to conﬂict used in this paper. Data from INSEC have been
extensively used in the media, international agencies, and government reports
and in a number of academic studies, including those by Bohara et al. (2006) and
Do and Iyer (2010). However, conﬂict deaths and school destructions are easier

4. In 2001, six out of 257 sampling units had to be dropped from the sample for security reasons
(Ministry of Health et al., 2002, p.6).
5. In the 2001 survey, children listed on the household roster cannot be matched to their mothers.
Valente 361

to monitor than abductions, and there are some surprising ﬁgures in the abduction
data provided by INSEC, such as only 284 abductions by Maoists in Rolpa during
the entire conﬂict. A degree of measurement error is likely to affect any conﬂict
event data. If uncorrelated with the actual number of conﬂict events, this classical
measurement error would lead to attenuation bias. However, the measurement
error would have to be both inversely related to the true number of conﬂict events
and very severe for it to lead to a reversal of the sign of the estimated effect of con-
ﬂict. Such a result appears implausible given the degree of consensus on INSEC
conﬂict data. Furthermore, ﬁndings using the number of casualties are consistent
with the simple difference-in-difference calculations in table 1, in which the con-
ﬂict event counts are collapsed into binary indicators. These are more blunt indica-
tors of conﬂict intensity, but they are also less prone to measurement error.
In addition, this study uses two indicators of Maoist control over a given dis-
trict by 2003 based on classiﬁcations reported by Hattlebak (2007). I consider,
in turn, two alternative deﬁnitions of Maoist control. I ﬁrst categorize as under
Maoist control any district that is categorized as such by both Maoists and the
government (Deﬁnition 1). I then apply what Hattlebak (2007) considers a more
reliable classiﬁcation, the government classiﬁcation (Deﬁnition 2).

I V. E M P I R I C A L S T R A T E G Y

I exploit the fact that surveyed individuals have been exposed to varying degrees
of conﬂict intensity according to their district of residence, year of birth, and
whether they were surveyed in 2001 or in 2006.
The baseline estimation strategy is similar to that in much of the literature esti-
mating the impact of civil conﬂict on individual outcomes reviewed in section
I. The strategy exploits differences in exposure to conﬂict by birth year cohort and
district of residence for individuals surveyed at the end of the conﬂict in 2006.
To check the robustness of the baseline results, I use a second identiﬁcation
strategy in which the source of identiﬁcation is the change in the intensity of con-
ﬂict within the district between 2001 and 2006, just before and just after the es-
calation of the conﬂict. By 2006, individuals born in 1991 and 1996, for
example, would have been exposed to the same total amount of conﬂict before
or during their schooling careers (albeit at different times in their lives). Hence, a
comparison of the schooling outcomes of individuals born in 1991 and 1996 in
the 2006 DHS would not be particularly informative. However, individuals born
in 1991 who were observed at age 10 years in the 2001 survey experienced much
less conﬂict by the time their education data were collected in 2001 than individ-
uals born in 1996 and observed at age 10 years in the 2006 survey. Therefore,
comparing their education outcomes at age 10 years is informative.6 This second

6. The variation in exposure between surveys also varies substantially between districts. For instance,
in Mahottari, there were 0.14 additional deaths per 1,000 inhabitants between the two DHS, whereas in
Jumla, there were close to three additional deaths per 1,000 inhabitants during the same period.
362 THE WORLD BANK ECONOMIC REVIEW

approach allows me to use variation in conﬂict intensity over time that would be
discarded in the traditional approach. In addition, the second approach provides
the opportunity for useful checks of the robustness of my ﬁndings to potential
migration and mortality biases.
I consider two outcome variables: a binary indicator for primary schooling
completion and the number of years of education completed. Less than 45
percent (20 percent) of the male (female) adult population surveyed in the 2006
Nepal DHS had completed primary education, so primary schooling completion
is a relevant cutoff in the present context. Given the recent occurrence of the con-
ﬂict, a focus on primary education also has an advantage in that many individu-
als whose primary schooling careers coincided with the conﬂict period are old
enough to have completed their primary education; hence, their long-term
primary schooling outcomes are observed. Finally, given the high prevalence of
voluntary migration in Nepal, it is important to test the robustness of my ﬁndings
to migration bias. I do this by comparing the schooling attainment of children
under 15 years of age surveyed in 2001 and 2006 in a given district. This age
group is appropriate as long as I focus on primary education. As of 2004, 97
percent of Nepali migrants were men aged 15–44 years who typically left their
wives and children behind (Lokshin and Glinskaya 2009). By focusing on chil-
dren under 15 at the time of the survey, the individual is thus both unlikely to
have migrated himself and unlikely to have accompanied a migrant parent. The
DHS did not collect detailed migration data, but it does provide data on the date
of arrival at the current location for women of reproductive age. Before the age
of 15 years, the overwhelming majority of children are still living with their
mothers; hence, I can further test the robustness of my ﬁndings for migration
bias by restricting the sample to children whose mothers had not migrated since
the beginning of the conﬂict.

Speciﬁcation 1: Exploiting Differences in Exposure to Conﬂict between Birth
Year Cohorts within Districts
Similar to previous studies on the impact of conﬂict on educational attainment, I
ﬁrst use data from the postconﬂict DHS (2006) and exploit variations in expo-
sure to conﬂict by birth year cohort and district. In its simplest form, the estimat-
ing equation can be written as follows:

EDUCt t t
ij ¼ dj þ at þ b TOTCONF j þ 1ij ð1Þ

where EDUC t ij is, in turn, a dummy variable equal to one if individual i in district
j born in year t has completed primary education and zero otherwise or the
number of years of education completed by this individual; dj represents district
ﬁxed effects; at represents birth year dummies; and TOTCONF t j is the interac-
tion between a dummy equal to one when the individual belongs to the treated
cohort and the number of conﬂict casualties (per 1,000 inhabitants) in district j
Valente 363

during 1996–2006. In the baseline regressions, I deﬁne the treated cohort as
those aged 5 to 9 years at the beginning of the conﬂict in 1996, whereas the
control cohort includes individuals aged 16 to 19 years at the beginning of the
conﬂict. This choice is discussed in the preliminary analysis at the end of this
section.
Under the assumption that there is no correlation between the number of dis-
trict casualties and unobserved factors varying with district and birth cohort, b is
the causal effect of a one-unit increase in TOTCONF t j on the primary comple-
tion rate or on the number of years of education completed by exposed cohorts.
A one-unit increase in TOTCONF t j roughly corresponds to one standard devia-
tion in the district-level distribution of casualties (0.98). Another way of apprais-
ing the magnitude of b is to consider a one-unit increase in TOTCONF t j as a
move from the district with the least conﬂict to the 53rd district (out of 75) in
order of conﬂict intensity or from the 53rd district to the 69th district in order of
conﬂict intensity—a very large increase in conﬂict intensity.
There are ﬁve “developmental regions” in Nepal, which are relatively homo-
geneous in terms of their level of development (see ﬁgure 1). In equation (1), I im-
plicitly restrict birth cohort effects (at) to be identical across development
regions. To reduce the potential for unobserved cohort-district varying factors to
bias the estimate of the effect of conﬂict exposure, in the main set of results, I
report estimates of equation (1R) in which the birth year intercepts are allowed
R 7
to vary by development region (at ). Here, the effect of conﬂict is identiﬁed by
using the difference in exposure to conﬂict by district and birth year cohort, net
of birth year trends common to all districts within a given development region:

EDUCt R t t
ij ¼ dj þ at þ b TOTCONF j þ 1ij : ð1RÞ

In addition, I estimate variants of equation (1) in which I add regressors captur-
ing speciﬁc aspects of the conﬂict that are likely to have affected schooling out-
comes, namely, the number of school destructions and abductions by Maoist
forces ( per 1,000 inhabitants) during the conﬂict. I also consider whether Maoist
control over the district had an effect on primary attainment by estimating vari-
ants of equation (1) in which I replace TOTCONF t j with an indicator variable
that switches on when the district was under Maoist control at the height of the
conﬂict.
It would be desirable to control for the socioeconomic status of the household
in which an individual was raised. However, household characteristics are not in-
cluded as regressors in this speciﬁcation because these are only observed at the
time of the survey in 2006, 10 years after the beginning of exposure to conﬂict,
and may therefore be caused by the conﬂict. In addition, at the time of the
survey, individuals in the control group were 25 –28 years old, and members of

7. Note that region dummies are subsumed under the district ﬁxed effects, but the interactions
between regions and birth year dummies are not.
364 THE WORLD BANK ECONOMIC REVIEW

the treated group were 14–18 years old. Therefore, it is difﬁcult to imagine
household characteristics that would not depend on the individual’s education
level at the time of the survey. In the second identiﬁcation strategy described
below, I observe school-aged children in their household; hence, I can control for
household characteristics.

Speciﬁcation 2: Comparison of Outcomes in 2001 and 2006 for a Given Age at
Interview
In equation (1), I only use data from the postconﬂict DHS (2006) and exploit var-
iations in exposure to conﬂict according to birth year cohort and district.
However, there is a comparable survey for 2001, just before the conﬂict escalat-
ed, which allows me to estimate the impact of the conﬂict using an alternative
identiﬁcation strategy based on variation in conﬂict exposure by survey date and
district. The idea is to exploit the fact that a child aged, for example, 10 years in
2001 will have experienced much less conﬂict during his lifetime than another
child aged 10 years in 2006 in the same district, and the difference in conﬂict ex-
posure between these two children will also vary across districts. Finding results
similar to those obtained using the traditional identiﬁcation strategy in equation
(1) would bolster conﬁdence in the reliability of my estimates. More speciﬁcally,
I estimate the following:
0
EDUCs s s s
ij ¼ uj þ ls þ g CONFEXPij þ Xij w þ mij

s ¼ 2001; 2006
ð2Þ
X
s
CONFEXPs
ij ¼ CONF jy
y¼birthyi

where EDUC s ij is the primary education completion dummy or the number of
completed years of education of individual i in district j observed in survey year
s, uj represents district ﬁxed effects, ls is a survey dummy equal to one for DHS
2006 and zero for DHS 2001, CONFjy is the number of conﬂict deaths per 1,000
inhabitants that occurred in district j in year y, and CONFEXPs ij is the number of
conﬂict deaths per 1,000 inhabitants in district j that occurred between the indi-
vidual’s birth year and survey year s in which the individual is interviewed,
which I calculate from yearly district death counts. When EDUC s ij is the primary
education indicator, the sample comprises children aged 10–18 years, who may
have completed primary education at the time of the survey. When EDUC s ij is the
number of years of education completed, the sample comprises children aged 5 –
14 years, who are of primary school age at the time of the survey. Under the as-
sumption that there is no correlation between the cumulative number of district
casualties between 1996 and year s and unobserved district-survey-varying
factors, g is the causal effect of a one-unit increase in CONFEXPs ij on the rate of
primary schooling completion (the number of years of education completed) by
Valente 365

the 10- to 18-year-old (5- to 14-year old) group. The magnitude of g can be
directly compared to that of b because both TOTCONFt j and CONFEXPij are
s

expressed in district casualties per 1,000 inhabitants.
Xsij is a set of controls that contains age at interview dummies and their inter-
action with the survey dummy in all speciﬁcations, thus allowing the educational
attainment of each age group to vary independently between the two surveys.8
These covariates are included to control for potential differences in the 2001–
2006 change in the district-level age composition of the 10- to 18-year-old or 5-
to 14-year-old group that, if correlated with conﬂict intensity, could bias the
results. In some variants, Xs ij also includes household characteristics at the time
of the survey (rural location and education of the household head). These charac-
teristics are only measured at the time of the survey and could potentially be
caused by past conﬂict. However, ﬁnding similar results when these observable
household characteristics are included would suggest that potential changes in
the composition of households due to mortality or migration do not drive my
ﬁndings.
To further reduce the potential for unobserved time-varying factors to bias the
estimate of the effect of conﬂict exposure, in the main set of results, I report esti-
mates of equation (2R) in which the coefﬁcients of the survey dummy, the age at
survey intercepts, and their interactions are allowed to vary by development
region. Here, the effect of conﬂict exposure is identiﬁed using the within-district
change in conﬂict exposure at age x between 2001 and 2006, net of 2001–2006
changes in educational attainment at age x common to all districts in a given de-
velopment region:

0
EDUCs R s sR s
ij ¼ uj þ ls þ g CONFEXPij þ Xij w þ mij : ð2RÞ

This second identiﬁcation strategy has three important advantages compared to
the traditional approach based on variation in exposure by birth cohort and dis-
trict for individuals observed at a single point in time. First, it has the advantage
of comparing cohorts that are born only ﬁve years apart but have experienced
very different degrees of conﬂict (i.e., a 10-year-old in 2001 in district j was
exposed to much less conﬂict than a 10-year-old in 2006 in district j but was
born only ﬁve years earlier), which reduces concerns regarding potential con-
founders, including differential migration patterns. Second, when the sample is
restricted to children aged 14 years and under, the concern regarding selection
bias due to voluntary migration decreases because most migrants are men aged
15 –44 years, who typically leave their wives and children behind (Lokshin and
Glinskaya 2009). Third, I can further test the robustness of my ﬁndings to

8. When estimating equation (2) without any of the controls included, g is positive and statistically
signiﬁcant for both females and males in the completed years of education regression and positive and
statistically signiﬁcant for females only in the primary education completion regression (full results are
available upon request).
366 THE WORLD BANK ECONOMIC REVIEW

migration bias by excluding from the sample children surveyed in 2006 whose
mothers moved to their current location after 1996.9
All speciﬁcations are estimated using linear district ﬁxed-effects panel data
models. All models allow for error terms to be correlated in an arbitrary fashion
within a district to avoid overrejection of the null hypothesis of zero treatment
effect due to serial correlation, following Bertrand et al. (2004).

Preliminary Analysis
An inspection of the data shows that although the legally mandated age for be-
ginning schooling is six years old and there are ﬁve years of primary schooling, a
sizeable proportion of children are enrolled in primary school before age six
(70.1 percent at ﬁve years) and until age 14 (16.1 percent) in the 2006 DHS, with
numbers subsequently decreasing sharply (7.9 percent at 15 years and 2.85
percent at 16 years).10 Therefore, an analysis of the long-term effect of conﬂict
on primary schooling completion should consider children aged at least 14 years
at the time of the survey, and control cohorts should have been at least 15 years
at the beginning of the conﬂict (and preferably slightly older). When estimating
variants of equation (1), I therefore deﬁne the treated cohort as comprising indi-
viduals aged ﬁve to nine years at the beginning of the conﬂict in 1996, such that
all treated cohorts are exposed to the conﬂict during most of their potential
primary schooling careers and the youngest exposed cohort is observed at age 14
in the 2006 DHS. In the main regressions, the control cohorts include individuals
aged 16 to 19 years at the beginning of the conﬂict—that is, individuals who
were born too early to have their primary education affected by the conﬂict but
are as close, and therefore as comparable, to the treated cohorts as possible. I
also provide a robustness check in which the control group comprises individuals
aged 18–25 years at the start of the conﬂict. In the baseline speciﬁcation, I
exclude cohorts aged 10 to 15 years in 1996 because the treatment status of these
cohorts is less clear. Many of these individuals could have been enrolled in
primary schooling during the conﬂict, but they were not exposed to conﬂict
during most of their primary schooling careers (see ﬁgures S3a and S3b).
Panel A of table 1 illustrates the basic identiﬁcation strategy used in the base-
line speciﬁcation. This panel shows the difference in the increase in primary
schooling between cohorts exposed (row (1)) and not exposed (row (2)) to con-
ﬂict during their primary schooling careers in districts experiencing above-
median conﬂict casualties (columns (1) and (4)) compared to districts experienc-
ing below-median conﬂict casualties (columns (2) and (5)). Women born too
early to be affected by the conﬂict during their primary schooling years have a

9. The education data used in this paper come from the DHS household datasets. Information on date
of arrival at the present location was only collected in individual interviews with women aged 15 –49
years. The same exclusion could not be implemented for the 2001 DHS because individuals listed on the
household roster cannot be matched to their mothers.
10. In the 2001 DHS, 44.4 percent of ﬁve-year-olds, 24.6 percent of 14-year-olds, and 13.6 percent of
15-year-olds were enrolled in primary schooling.
T A B L E 1 . Preliminary Difference-in-Difference Calculations, Completion of Primary Schooling
Female Male

Primary Education Rate by

Number of Casualties in District Number of Casualties in District

(1) (2) (3) (4) (5) (6)
High Low Difference High Low Difference

Panel A: Binary DiD Experiment
(1) Age 5 to 9 in 1996 0.67 0.64 0.03 0.80 0.79 0.01
(2) Age 16 to 19 in 1996 0.27 0.43 2 0.16 0.63 0.70 2 0.07
Difference 0.40 0.21 0.19 (0.045)*** 0.17 0.09 0.08 (0.038)*

Panel B1: Placebo DiD Experiment 1
(1) Age 16 to 19 in 1996 0.27 0.43 2 0.16 0.63 0.70 2 0.07
(2) Age 26 to 29 in 1996 0.13 0.20 2 0.07 0.50 0.53 2 0.03
Difference 0.14 0.23 2 0.09 (0.045)** 0.13 0.17 2 0.04 (0.047)
Panel B2: Placebo DiD Experiment 2
(1) Age 16 to 19 in 1996 0.27 0.43 2 0.16 0.63 0.70 2 0.07
(2) Age 20 to 23 in 1996 0.15 0.32 2 0.17 0.58 0.62 2 0.04
Difference 0.12 0.11 0.01 (0.040) 0.05- 0.08 2 0.03 (0.047)

Valente
(Continued )

367
368
TABLE 1. Continued

THE WORLD BANK ECONOMIC REVIEW
Female Male

Primary Education Rate by

Number of Abductions in District Number of Abductions in District

(1) (2) (3) (4) (5) (6)
High Low Difference High Low Difference

Panel C: Binary DiD Experiment
(1) Age 5 to 9 in 1996 0.70 0.62 0.08 0.82 0.77 0.05
(2) Age 16 to 19 in 1996 0.33 0.37 2 0.04 0.61 0.70 2 0.09
Difference 0.37 0.25 0.12 (0.053)** 0.21 0.07 0.14 (.038)***
Panel D1: Placebo DiD Experiment 1
(1) Age 16 to 19 in 1996 0.33 0.37 2 0.04 0.61 0.70 2 0.09
(2) Age 26 to 29 in 1996 0.15 0.19 2 0.04 0.53 0.52 0.01
Difference 0.18 0.18 0.00 (0.051) 0.08 0.18 2 0.10 (0.049)**
Panel D2: Placebo DiD Experiment 2
(1) Age 16 to 19 in 1996 0.33 0.37 2 0.04 0.61 0.70 2 0.09
(2) Age 20 to 23 in 1996 0.22 0.27 2 0.05 0.62 0.59 0.03
Difference 0.11 0.10 0.01 (0.041) 2 0.01- 0.11 2 0.12 (0.045)***

Notes: District casualties are expressed per 1,000 inhabitants. “High” and “Low” refer to above-median or below-median district totals per 1,000 inhabi-
tants. Standard errors clustered at the district level are in parentheses. All ﬁrst differences (i.e., row (1) – row (2) for a given conﬂict category) are statistically
signiﬁcant, except for values marked with -. DiD indicates difference-in-difference. *p , 0.10, **p , 0.05, ***p , 0.01.
Source: INSEC 2009, Central Bureau of Statistics 2009. Education data are based on Nepal DHS 2006.
Valente 369

much lower rate of primary schooling in high-conﬂict districts compared to
women in low-conﬂict districts. In contrast, primary schooling completion is
slightly higher in high-conﬂict districts compared to low-conﬂict districts among
the cohort of women who were entering primary school around the beginning of
the conﬂict, resulting in an additional increase in female primary schooling of 19
percentage points between the two cohorts in high-conﬂict areas compared to
low-conﬂict areas. A qualitatively similar but less dramatic effect is observed
among men.
To shed light on the direction of the potential biases due to differential precon-
ﬂict trends, I conduct several control or “placebo” experiments in which conﬂict
exposure is artiﬁcially assigned to cohorts who were too old to be affected by the
conﬂict. In panel B1, I compare the change in primary schooling attainment
between cohorts aged 16 –19 years and cohorts aged 26–29 years at the begin-
ning of the conﬂict in above-median versus below-median conﬂict intensity dis-
tricts. The difference-in-difference is negative, and for females, it is statistically
signiﬁcant. This ﬁnding indicates that rates of primary schooling completion
were improving more slowly in areas where more conﬂict occurred in 1996–
2006 when comparing cohorts that were potentially enrolled in primary school
(i.e., aged 5 –14 years) in the 1972–1984 period and cohorts that were potential-
ly enrolled in the 1982–1994 period. If this trend had continued during the con-
ﬂict period, the estimates presented in this paper would be a lower bound of the
true effect of conﬂict; that is, the positive coefﬁcient of the conﬂict variable
would be an overly conservative estimate, especially for females.
The ideal placebo experiment would be based on the actual cohorts involved
in the experiment of interest in the absence of conﬂict. Such a test is clearly not
feasible. However, it is possible to conduct an additional placebo experiment
based on cohorts born immediately before the period relevant to the experiment
of interest to check for differences in trends as close as possible to the period of
interest. The results of this additional placebo experiment comparing cohorts
aged 16 –19 years at the beginning of the conﬂict with those aged 20–23 years
are shown in panel B2. During this period immediately preceding the conﬂict, I
cannot reject the hypothesis that the evolution of primary schooling was parallel
in districts with above- and below-median conﬂict casualties for both males and
females.
In panel C, the experiment is conducted by replacing the below- and above-
median casualty categories with below- and above-median Maoist abduction cat-
egories. The results in panel C show that primary schooling has progressed more
rapidly in districts with above-median Maoist abductions. However, two-thirds
of districts classiﬁed as high (low) conﬂict based on the median number of casu-
alties are also classiﬁed as high (low) conﬂict based on the median number of ab-
ductions. Therefore, these simple two-by-two calculations may capture the effect
of overall conﬂict intensity, the effect of Maoist abductions, or both. In the re-
gressions that follow, I disentangle the effect of overall conﬂict intensity and
Maoist abductions by including both conﬂict variables.
370 THE WORLD BANK ECONOMIC REVIEW

Panels D1 and D2 show results for tests that replicate the placebo tests for
panels B1 and B2, where the deﬁnition of high- and low-conﬂict districts based
on above- and below-median conﬂict casualties is replaced with the deﬁnition
based on above- and below-median district abductions. Females experienced
similar preconﬂict primary education trends in high- and low-abduction districts.
Male cohorts found in high-abduction districts experienced slower progress in
preconﬂict primary education relative to low-abduction districts. If the same
trends continued for cohorts considered in the experiment of interest, then the
effect of exposure to Maoist abductions would tend to be biased downward for
males (i.e., to be more negative) but not for females. On the contrary, in the re-
gression analysis, I ﬁnd that after controlling for district casualties, the education
of females suffered from Maoist abductions, but that the education of males did
not. Therefore, the difference in male trends observed in panels D1 and D2 does
not drive my conclusions.

V. R E S U L T S

The preliminary analysis in section IV suggested that primary schooling comple-
tion rates tended to increase more rapidly during armed conﬂict in areas that
experienced a high intensity of conﬂict, especially for girls. In tables 2 to 4, I
present estimates of the impact of exposure to conﬂict on educational outcomes
to determine whether this striking conclusion of the preliminary analysis is
conﬁrmed when using more detailed information on the intensity of conﬂict, con-
trolling for unobserved heterogeneity between individual districts and between
regions over time and using different identiﬁcation strategies.
Table 2 reports ﬁndings on the impact of conﬂict exposure on primary school-
ing completion. The ﬁrst two columns present estimates of the long-term effect
of conﬂict intensity on primary schooling completion using the baseline speciﬁ-
cation (equation (1R)). The last four columns indicate the robustness of these
ﬁndings through comparison of the change in primary completion rates for the
10- to 18-year-old group for districts with varying degrees of conﬂict intensiﬁca-
tion between the 2001 and 2006 DHS (equation (2R)). The last two columns
include controls for rural location and the educational attainment of the house-
hold head.
The results in the ﬁrst column indicate that areas with more ﬁghting witnessed
a larger increase in female primary education attainment. Casting this result in
terms of the distribution of conﬂict violence, an increase in violence of one stan-
dard deviation of the district-level distribution of casualties during the conﬂict
(0.98 casualties per 1,000 inhabitants) increases female primary schooling at-
tainment by 5.6 percentage points. This is roughly the effect of a move from the
5th to the 75th percentile of the district-level conﬂict distribution of total casual-
ties. The sign and order of magnitude of this effect is conﬁrmed when comparing
cohorts born only ﬁve years apart but exposed to very different levels of conﬂict
using equation (2R). Across all speciﬁcations in table 2, conﬂict exposure does
T A B L E 2 . Impact of Conﬂict Intensity Measured by Casualties on Primary Schooling Completion
(1) (2) (3) (4) (5) (6)
Primary Primary Primary
Education - Education - Primary Education - Education - Male Primary Education Primary Education
Explained Variable and Sample Female Male Female 10–18 10 – 18 - Female 10– 18 - Male 10– 18

Speciﬁcation Eq. (1R) Eq. (1R) Eq. (2R) Eq. (2R) Eq. (2R) Eq. (2R)

¼ 1 if 5 – 9 in 1996 Â District 0.0555** 0.0241
casualties during 1996– 2006 (0.0272) (0.0251)
(TOTCONF t j)
District casualties before survey 0.0764* 0.0036 0.0811** 0.0094
(CONFEXP s ij ) (0.0431) (0.0368) (0.0368) (0.0338)
¼ 1 if rural 2 0.1094*** 2 0.0622***
(0.0205) (0.0154)
¼ 1 if head has primary education 0.0543*** 0.0684***
(0.0133) (0.0115)
¼ 1 if head has secondary education 0.2127*** 0.2046***
(0.0120) (0.0139)
¼ 1 if head has higher education 0.3249*** 0.2597***
(0.0251) (0.0253)
Panel variable District District District District District District
Included dummies:

Valente
Year of birth Yes Yes No No No No
Region Â Year of birth Yes Yes No No No No
DHS 2006 No No Yes Yes Yes Yes

371
(Continued )
372
THE WORLD BANK ECONOMIC REVIEW
TABLE 2. Continued
(1) (2) (3) (4) (5) (6)
Primary Primary Primary
Education - Education - Primary Education - Education - Male Primary Education Primary Education
Explained Variable and Sample Female Male Female 10–18 10 – 18 - Female 10– 18 - Male 10– 18

Speciﬁcation Eq. (1R) Eq. (1R) Eq. (2R) Eq. (2R) Eq. (2R) Eq. (2R)

Age at interview No No Yes Yes Yes Yes
Region Â DHS 2006 No No Yes Yes Yes Yes
DHS 2006 Â Age at interview No No Yes Yes Yes Yes
Region Â Age at interview No No Yes Yes Yes Yes
DHS 2006 Â Region Â Age No No Yes Yes Yes Yes
at interview
Observations 3,823 3,055 9,595 9,267 9,584 9,255
No. of clusters 75 75 69a 69a 69a 69a
R-squared 0.1106 0.0368 0.2077 0.3021 0.2602 0.3372
p value male vs. femaleb 0.345 0.074 0.044

Notes: All speciﬁcations are estimated using the panel ﬁxed-effects estimator and include a constant. District casualties are expressed per 1,000 inhabi-
tants. Columns (1) and (2): Sample only includes individuals surveyed in the Nepal DHS 2006 and aged 5 – 9 years (treatment group) or 16 – 19 years (control
group) at the beginning of the conﬂict in 1996. Columns (3) to (6): Sample only includes individuals surveyed in Nepal DHS 2001 or 2006 aged 10 to 18
years at the time of the survey. Standard errors clustered at the district level are in parentheses. *p , 0.10, **p , 0.05, ***p , 0.01.
a
DHS data collection was somewhat affected by the conﬂict in 2001. Hence, contrary to DHS 2006, four districts were not covered: Dolpa, Jajarkot,
Rolpa, and Rukhum. The small districts of Manang and Mustang were not surveyed, but these districts did not experience any casualties during the conﬂict.
b
p value of an F test of equality between the reported treatment effects for males and females.
Source: INSEC 2009, Central Bureau of Statistics 2009, Nepal DHS 2001, and Nepal DHS 2006.
T A B L E 3 . Impact of Alternative Conﬂict Variables on Primary Schooling Completion
(1) (2) (3) (4) (5) (6) (7) (8)
School School Maoist Maoist Maoist Control Maoist Control Maoist Control Maoist Control
Destructions Destructions Abductions Abductions Female – Male – Female – Male –
Female Male Female Male Deﬁnition 1 Deﬁnition 1 Deﬁnition 2 Deﬁnition 2

Primary Primary Primary Primary Primary Primary Primary Primary
Explained Variable Education Education Education Education Education Education Education Education

Speciﬁcation Eq. (1R) Eq. (1 R) Eq. (1 R) Eq. (1R) Eq. (1R) Eq. (1R) Eq. (1R) Eq. (1R)

¼ 1 if 5 – 9 in 0.0539* 0.0301 0.0638** 0.0231
1996 Â District (0.0302) (0.0291) (0.0275) (0.0259)
casualties during
1996– 2006
(TOTCONF t j)
¼ 1 if 5 – 9 in 0.4176 2 1.7848
1996 Â District (3.1298) (2.3395)
schools destroyed
2002– 2006
¼ 1 if 5 – 9 in 2 0.0022*** 0.0002
1996 Â Maoist (0.0005) (0.0010)
Abductions during
1996– 2006
¼ 1 if 5 – 9 in 0.0916 0.2009***
1996 Â District (0.0590) (0.0529)
controlled by

Valente
Maoists
(Deﬁnition 1)

(Continued )

373
TABLE 3. Continued

374
(1) (2) (3) (4) (5) (6) (7) (8)
School School Maoist Maoist Maoist Control Maoist Control Maoist Control Maoist Control

THE WORLD BANK ECONOMIC REVIEW
Destructions Destructions Abductions Abductions Female – Male – Female – Male –
Female Male Female Male Deﬁnition 1 Deﬁnition 1 Deﬁnition 2 Deﬁnition 2

Primary Primary Primary Primary Primary Primary Primary Primary
Explained Variable Education Education Education Education Education Education Education Education

Speciﬁcation Eq. (1R) Eq. (1 R) Eq. (1 R) Eq. (1R) Eq. (1R) Eq. (1R) Eq. (1R) Eq. (1R)

¼ 1 if 5 – 9 in 0.1456*** 0.0923*
1996 Â District (0.0457) (0.0506)
controlled by
Maoists
(Deﬁnition 2)
Panel variable District District District District District District District District
Included dummies:
Year of birth Yes Yes Yes Yes Yes Yes Yes Yes
Region Â Year of Yes Yes Yes Yes Yes Yes Yes Yes
birth
Observations 3,823 3,055 3,823 3,055 3,823 3,055 3,823 3,055
No. of clusters 75 75 75 75 75 75 75 75
R-squared 0.1106 0.0370 0.1118 0.0368 0.1105 0.0415 0.1130 0.0382
p value male vs. 0.466 0.0134 0.208 0.456
femalea

Notes: All speciﬁcations are estimated using the panel ﬁxed-effects estimator and include a constant. School destructions and abductions by Maoists are
expressed per 1,000 inhabitants. Sample only includes individuals surveyed in the Nepal DHS 2006 and aged 5 – 9 years (treatment group) or 16 – 19 years
(control group) at the beginning of the conﬂict in 1996. Deﬁnition of a district controlled by Maoists based on matches between People’s Army and govern-
ment classiﬁcations as of 2003 (Deﬁnition 1) or government classiﬁcation (Deﬁnition 2), according to Hattlebak (2007). Standard errors clustered at the dis-
trict level are in parentheses. * p , 0.10, ** p , 0.05, *** p , 0.01.
a
p value of an F test of equality between the reported treatment effects for males and females.
Source: INSEC 2009, Central Bureau of Statistics 2009, Nepal DHS 2006,
Valente 375

not appear to signiﬁcantly affect male primary schooling completion, although
the gender difference in the conﬂict effect is only statistically signiﬁcant in
columns (3) to (6).
These results are robust to including controls for household characteristics,
suggesting that the results are not driven by a change in household composition
due to, for example, selective mortality or migration (columns (5) and (6)).
Table S2 presents speciﬁcations similar to those in table 2 but replaces the indica-
tor for primary schooling completion with years of education completed. Similar
results are obtained, indicating that an increase in violence of one standard devia-
tion increases female educational attainment by 0.6 years.
Table S3 presents three different speciﬁcations to further check the robustness
of the baseline results in the ﬁrst two columns of table 2 to the following changes
in speciﬁcation: restricting birth year ﬁxed effects to be identical across the ﬁve
development regions of Nepal, changing the control cohort, and replacing the
number of casualties with its natural logarithm. The results in table S3 conﬁrm
that primary education progressed more rapidly during the conﬂict in districts ex-
periencing more casualties and that this effect was more robust across speciﬁca-
tions for females.
Next, I investigate whether speciﬁc aspects of the conﬂict had different effects
on primary schooling completion (table 3). First, I use INSEC data on the total
number of school destructions per district to test whether these destructions had
a negative effect on primary schooling completion despite the overall positive
impact of the insurgency (columns (1) and (2)). For both genders, I ﬁnd a statisti-
cally insigniﬁcant effect, which is likely because a district-level analysis lacks the
power to identify the effect of school destructions. School destructions were a
rare and isolated aspect of the conﬂict11 that could be expected to have had a
large effect on schooling at a disaggregated level but not at the district level.
Second, I use INSEC data on the total number of abductions by Maoists per dis-
trict to test whether a larger number of abductions, often targeting school chil-
dren, had an adverse effect on schooling. The results in columns (3) and (4)
indicate that abductions had a negative effect on female primary schooling. An
increase in the number of abductions ( per 1,000 inhabitants) by one standard
deviation of the district-level distribution (16.82) decreases female primary
schooling attainment by 3.7 percentage points. In other words, the effect of a
move from the 5th to the 75th percentile of the district-level distribution of total
abductions yields a 1.6 percentage point decline in female primary completion.
Third, I test whether primary schooling completion improved more in districts
controlled by Maoists where the insurgents were better able to affect schooling
provision according to their ideology (columns (5) to (8)). There is no clear-cut
deﬁnition of insurgent control, with discrepancies between the classiﬁcations
used by the People’s Army and the government (Hatlebakk 2007). Therefore, I
use two alternative classiﬁcations. The choice of deﬁnition affects the magnitude

11. According to the data provided by INSEC, 76 schools were destroyed.
376
THE WORLD BANK ECONOMIC REVIEW
T A B L E 4 . Impact of Conﬂict Intensity on Completed Years of Education, 5- to 14-year-olds
(1) (2) (3) (4)
Years of Education Years of Education Years of Education Female 5 – 14 – Years of Education Male 5 – 14 –
Explained Variable and Sample Female 5– 14 Male 5 – 14 Mother here since 1996 Mother here since 1996

Speciﬁcation Eq. (2R) Eq. (2R) Eq. (2R) Eq. (2R)

District casualties before survey 0.2795** 0.1224 0.2633** 0.0755
(CONFEXP s ij ) (0.1147) (0.1207) (0.1172) (0.1391)
¼ 1 if rural 2 0.3916*** 2 0.1559** 2 0.4703*** 2 0.1770**
(0.0788) (0.0664) (0.0914) (0.0703)
¼ 1 if head has primary 0.2210*** 0.2301*** 0.2051*** 0.2381***
education (0.0428) (0.0361) (0.0453) (0.0374)
¼ 1 if head has secondary 0.8263*** 0.7064*** 0.8619*** 0.7242***
education (0.0431) (0.0386) (0.0509) (0.0388)
¼ 1 if head has higher education 1.1291*** 1.0828*** 1.3829*** 1.2016***
(0.1035) (0.0846) (0.1137) (0.1124)
Panel variable District District District District
Included dummies:
DHS 2006 Yes Yes Yes Yes
Age at interview Yes Yes Yes Yes
Region Â DHS 2006 Yes Yes Yes Yes
DHS 2006 Â Age at interview Yes Yes Yes Yes
Region Â Age at interview Yes Yes Yes Yes
DHS 2006 Â Region Â Age at Yes Yes Yes Yes
interview
Observations 11,793 12,116 9,772 9,959
No. of clusters 69a 69a 69a 69a
R-Squared 0.5062 0.6077 0.4909 0.5996
p value male vs. femaleb 0.079 0.065

Notes: All speciﬁcations are estimated using the panel ﬁxed-effects estimator and include a constant. District casualties are expressed per 1,000 inhabi-
tants. Sample only includes individuals surveyed in the Nepal DHS 2001 and 2006 and aged 5– 14 years at the time of the survey. In columns (3) and (4), the
2006 sample is restricted to individuals whose mothers were interviewed individually and whose mothers reported having lived in their current place of resi-
dence as of 1996. The same exclusion could not be implemented for the 2001 DHS because individuals listed on the household roster cannot be matched to
their mothers. Standard errors clustered at the district level are in parentheses. * p , 0.10, ** p , 0.05, *** p , 0.01.
a
DHS data collection was somewhat affected by the conﬂict in 2001. Hence, contrary to DHS 2006, four districts were not covered: Dolpa, Jajarkot,
Rolpa, and Rukhum. The small districts of Manang and Mustang were not surveyed, but these districts did not experience any casualties during the conﬂict.
b
p value of an F test of equality between the reported treatment effects for males and females.
Source: INSEC 2009, Central Bureau of Statistics 2009, Nepal DHS 2001, and Nepal DHS 2006.

Valente
377
378 THE WORLD BANK ECONOMIC REVIEW

and signiﬁcance of estimates, but the overall message is that primary schooling
has tended to become more prevalent over time for both genders in areas con-
trolled by the Maoists.
Tables S4 and S5 replicate the analysis in table 3 with birth year ﬁxed effects
restricted to be identical across development regions of Nepal (tables S4 and S5)
and the control cohort replaced with individuals aged 18–24 years at the begin-
ning of the conﬂict (table S5). The same conclusions apply as those drawn from
the set of preferred results in table 3.
In table 4, I turn to the estimated effect of an increase in conﬂict intensity
between 2001 and 2006 on completed years of education of children of primary
schooling age (5–14 years) at the time they were surveyed (as per equation (2R)).
Column (1) of table 4 indicates that an increase in casualties since birth by one
standard deviation increases the completed years of education by just over one-
quarter of a year for girls aged 5 to 14 years in 2006 compared to girls from the
same district who were the same age when surveyed in 2001, before the conﬂict
escalated. For boys, the coefﬁcient of interest is less than half the magnitude of
that for girls and is signiﬁcantly different from the estimated conﬂict effect for
girls (at the 10 percent signiﬁcance level). The estimates are very similar when re-
stricting the 2006 sample to children whose mothers had not moved since 1996
(columns (3) and (4)), which conﬁrms that changes in composition due to migra-
tion patterns are not driving these ﬁndings. In table S6, I repeat the analysis in
table 4 but restrict the age intercepts and survey year dummy to be identical
across Nepal’s development regions. The results for the female sample are almost
identical, but estimates for the male sample are now nearly as large as those for
the female sample and are statistically signiﬁcant. Echoing the ﬁndings for
primary schooling completion, overall, these results conﬁrm that primary educa-
tion progressed more rapidly during the conﬂict in districts that experienced
more casualties, and this effect is more robust across speciﬁcations for females.
In conclusion, the results presented in this section provide no support for the
hypothesis that the Nepalese civil conﬂict had a negative effect on schooling
overall. There is a robust positive effect of the intensity of the insurgency on
female educational attainment, but there is less of an effect for male educational
attainment. There is also evidence of a decrease in female primary schooling
completion where insurgents were more prone to abductions, holding overall
conﬂict intensity constant.

VI. CONCLUDING REMARKS

Despite experiencing a substantial civil conﬂict between 1996 and 2006, Nepal
has surprisingly enjoyed one of the best periods in its history in terms of econom-
ic growth and poverty reduction. At present, however, little is known about
whether this period of development at the aggregate level hides disparities at a
more disaggregated level due to the wide variation in conﬂict intensity across the
country.
Valente 379

In this paper, I exploit variation in exposure to conﬂict by birth cohort, survey
date, and district to estimate the impact of conﬂict intensity on schooling out-
comes.
I ﬁnd no support for the hypothesis that civil-conﬂict-related violence, as mea-
sured by the number of conﬂict casualties, had a negative effect on the quantity
of schooling attained by children of either gender. On the contrary, there is
robust evidence that female primary schooling attainment increased in districts
that experienced more conﬂict deaths relative to districts with fewer conﬂict
deaths. This result holds irrespective of whether one compares (within a given
district) (i) the completion of primary education for cohorts exposed and not
exposed to the conﬂict and observed at the end of the conﬂict or (ii) years of edu-
cation completed by a given age for cohorts observed before (2001) and after
(2006) a sharp escalation of the conﬂict and that were therefore exposed to very
different degrees of conﬂict. It is also robust to a number of changes in speciﬁca-
tions. In particular, robustness checks indicate that changes in household compo-
sition due to conﬂict-induced migration patterns do not drive this ﬁnding.
However, one aspect of the Nepalese civil conﬂict that is particularly relevant
to schooling outcomes had adverse consequences on female primary schooling:
the widespread insurgent practice of abducting civilians, many of whom were
school children.
The ﬁndings reported in this paper echo the positive changes observed for
Nepal as a whole during the conﬂict period in terms of economic growth, educa-
tion, and child health. The present analysis shows that the progress in education
observed at the country level does not hide a slower increase in districts where
more ﬁghting occurred, but the insurgent practice of abducting civilians adverse-
ly affected female educational outcomes. The estimates presented in this paper
are in line with the existing qualitative literature on the Nepalese civil conﬂict,
which consistently reports mixed conclusions with respect to the impact of the
conﬂict on education and female empowerment (e.g., Hart 2001; Lama-Tamang
2003; Manchanda 2004; Pettigrew and Shneiderman 2004; Geiser 2005; Aguirre
and Pietropaoli 2008; Arin ˜ o 2008; Falch 2010).
Education, particularly female educational attainment, appears to have bene-
ﬁted from the societal changes induced directly or indirectly by the insurgency
more than it was adversely affected by the loss of income and other disruptions
caused by the conﬂict. Data limitations prevent a more detailed analysis of the
channels through which the conﬂict affected education beyond the distinction
between the effect of conﬂict as a whole and that of abductions. However, poten-
tial mechanisms suggested by the existing anthropological and peace studies liter-
ature include Maoist efforts to remove barriers to schooling for all children from
the lower castes and to reduce teacher absenteeism (e.g., Hart 2001;
Lama-Tamang 2003), which could have beneﬁted both male and female educa-
tion; the Maoist inﬂuence in encouraging or coercing parents to send girls to
school (Hart 2001); and the Maoists’ effect on female empowerment. Although
the exact ﬁgure is contested, a substantial share of the guerillas in the Maoist
380 THE WORLD BANK ECONOMIC REVIEW

ranks was female. Many more females were involved in the Maoist movement
without direct participation in combat, such as by disseminating propaganda
(Lama-Tamang et al. 2003; Pettigrew and Shneiderman 2004), and even larger
numbers may have been inﬂuenced by the Maoist discourse on gender equality.
In addition, there is anecdotal evidence of an improvement in the condition of
women in areas controlled by the Maoists, such as decreases in polygamy, do-
mestic violence, and alcoholism, as well as greater support for women to divorce
their husbands (Lama-Tamang et al. 2003; Manchanda 2004; Geiser 2005;
Arin˜ o 2008). Although the insurgents’ rhetoric was often in contrast with their
actual practice (Pettigrew and Shneiderman 2004), the presence of females in
their ranks and the propaganda promoting female autonomy may have increased
female bargaining power within the household as well as female aspirations.
According to Hart (2001), “girls and women are strongly encouraged to gain an
education and to participate in society generally and in activities connected to
the ‘People’s War’ in particular. This directly challenges their traditional role and
apparently stimulates girls to consider leading lives beyond marriage and the
home (Hart 2001, p.35)”. Furthermore, an unintended consequence of the con-
ﬂict has been that women have adopted roles typically reserved for men.
Women’s involvement in the labor market increased as a consequence (Menon
and Rodgers 2011). The rise in female labor market participation may have in-
creased returns to female schooling and motivated girls to obtain more education
and parents to invest more in their daughters’ education. Increased female earn-
ings are also likely to improve the ability of mothers to inﬂuence the way house-
hold resources are spent. Moreover, there is evidence that when women have
more control over household expenditures, such as because their own earnings
make up a larger share of the household’s income, investments in children in-
crease; this is especially the case for girls (e.g., Thomas 1990; Duﬂo 2003), al-
though this may not be the case in all contexts (Quisumbing and Maluccio 2003;
Gitter and Barham 2008). Finally, the nature of the occupations of women
outside the home also changed. In many areas, women were reported to take on
leadership roles in local institutions, including schools (Pettigrew and
Shneiderman 2004). This improvement in female representation in local institu-
tions may have contributed to increased education, especially for girls.
Data limitations prevent rigorous tests of the role played by these different po-
tential channels in explaining the ﬁnding that education, particularly female edu-
cation, increased more in areas where the ﬁghting was more intense.12 Future
research aiming to disentangle the role of each of the channels through which the
insurgency may have improved educational outcomes would be valuable.
An issue beyond the scope of this paper is the important question of the effect
of civil conﬂict on the quality of education, which is potentially large (for a
review, see Shemyakina and Valente 2011). Data limitations have thus far

12. See the appendix for some insights based on self-reported measures of female empowerment
available in the DHS.
Valente 381

precluded quantitative research on the impact of conﬂict on the quality of school-
ing, but there is growing evidence that cognitive skills, rather than completed
years of education, matter for individual earnings and economic growth (e.g.,
Hanushek and Woessmann 2008). Therefore, even where the number of years of
education completed is not adversely affected by civil conﬂict, such conﬂict may
have deleterious effects on human capital if the quality of learning deteriorates.
From an international perspective, this paper contributes to unpacking the
complexity that lies behind the generic term civil conﬂict. The idiosyncrasies of
each conﬂict highlight the need for additional research on the impacts of different
conﬂicts to shed light on the range of potential effects rather than a focus on
extreme, but thankfully rare, instances.
From a policy perspective, the present ﬁndings call for measures that aim to
protect school children and teachers from being directly targeted by combatants.
As shown in this paper, even where primary education systems appear very resil-
ient to surrounding violence, direct targeting of schools, however mild (e.g., brief
abductions of pupils and teachers for indoctrination purposes), has adverse
effects on schooling, especially for girls.

REFERENCES

Aguirre, D., and I. Pietropaoli. 2008. “Gender Equality, Development and Transitional Justice: The Case
of Nepal.” The International Journal of Transitional Justice 2: 356–77.
Akbulut-Yuksel, M. 2009. “Children of War: The Long-run Effects of Large-scale Physical Destruction
and Warfare on Children.” IZA Working Paper No.4407. Institute for the Study of Labor, Bonn,
Germany.
Akresh, R., and D. de Walque. 2008. “Armed Conﬂict and Schooling: Evidence from the 1994 Rwandan
Genocide.” World Bank Policy Research Paper No.4606. World Bank, Washington, DC.
Annan, J., C. Blattman, D. Mazurana, and K. Carlson. 2009. “Women and Girls at War: ‘Wives’,
Mothers, and Fighters in the Lord’s Resistance Army.” HICN Working paper 63. Households in
Conﬂict Network, Brighton, UK.
˜ o, M. V. 2008. “Nepal: A Gender View of the Armed Conﬂict and Peace Process.” Quaderns de
Arin
Construccio´ de Pau No.4. Escola de Cultura de Pau, Barcelona, Spain.
Armon, J., C. Berry, D. Duncan, R. Calder, S. Clapham, and M. Harvey. 2004. “Service Delivery in
Difﬁcult Environments: The Case of Nepal.” Department for International Development, Asia Policy
Division.
Becker, J. 2009. “Child Recruitment in Burma, Sri Lanka, and Nepal.” In S. Gates, and S. Reich eds.,
Child Soldiers in the Age of Fractured States, 108– 20. Pittsburg, PA: University of Pittsburg Press.
Bertrand, M., E. Duﬂo, and S. Mullainathan. 2004. “How Much Should We Trust
Differences-in-Differences Estimates?” Quarterly Journal of Economics 1191: 249– 75.
Blattman, C., and J. Annan. 2010. “The Consequences of Child Soldiering.” Review of Economics and
Statistics 92 (4): 882–98.
Blattman, C., and E. Miguel. 2010. “Civil War.” Journal of Economic Literature 48 (1): 3–57.
Bohara, A., N. Mitchell, and M. Nepal. 2006. “Opportunity, Democracy, and the Exchange of Political
Violence: A Subnational Analysis of Conﬂict in Nepal.” Journal of Conﬂict Resolution 50 (1):
108–28.
382 THE WORLD BANK ECONOMIC REVIEW

Central Bureau of Statistics [Nepal]. 2009. Nepal Statistical Year book 2009. Central Bureau of Statistics,
Kathmandu. http://cbs.gov.np/?page_id=1233, last accessed May 22, 2013.
´ n. 2011. “The Human Capital Consequences of Civil War: Evidence
Chamarbagwala, R., and H. E. Mora
from Guatemala.” Journal of Development Economics 94: 41– 61.
Collins, S. 2006. “Assessing the Health Implications of Nepal’s Ceaseﬁre.” Lancet 368: 907– 8.
Devkota, B., and E. R. van Teijlingen. 2010. “Understand Effects of Armed Conﬂict on Health Outcomes:
The case of Nepal.” Conﬂict and Health 4: 20.
Do, Q-T., and L. Iyer. 2010. “Geography, Poverty and Conﬂict in Nepal.” Journal of Peace Research 47
(6): 735–48.
Duﬂo, E. 2003. “Grandmothers and Granddaughters: Old-Age Pensions and Intrahousehold Allocation in
South Africa.” World Bank Economic Review 17 (1): 1–25.
Falch, A. 2010. “Women’s Political Participation and Inﬂuence in Post-Conﬂict Burundi and Nepal.”
Peace Research Institute Oslo Working Paper. Oslo, Norway.
Geiser, A. 2005. “Social Exclusion and Conﬂict Transformation in Nepal: Women, Dalit and Ethnic
Groups.” Working Paper 5/2005. Swisspeace, Bern, Switzerland.
Gitter, S. R., and B. L. Barham 2008. “Women’s Power, Conditional Cash Transfers, and Schooling in
Nicaragua.” World Bank Economic Review 22 (2): 271– 90.
Hanushek, E. A., and L. Woessmann 2008. “The Role of Cognitive Skills in Economic Development.”
Journal of Economic Literature 46: 607–68.
Hart, J. 2001. “Conﬂict in Nepal and its Impact on Children.” A Discussion Document Prepared for
UNICEF Regional Ofﬁce South Asia, Refugee Studies Centre, Oxford.
Hatlebakk, M. 2007. “LSMS Data Quality in Maoist Inﬂuenced Areas of Nepal.” No 6, CMI Working
Papers. CMI (Chr. Michelsen Institute), Bergen, Norway.
Human Rights Watch. 2004. “Between a Rock and a Hard Place: Civilians Struggle to Survive in Nepal’s
Civil War.” http://www.hrw.org/en/reports/2004/10/06/between-rock-and-hard-place?print, last
accessed May 20, 2013.
Hutt, M. 2004. “Nepal and Bhutan in 2004: Two Kings, Two Futures.” Asian Survey 45 (1): 83– 87.
Ichino, A., and R. Winter-Ebmer 2004. “The Long-Run Educational Cost of World War II.” Journal of
Labor Economics 22 (1): 57– 86.
INSEC. 2009. Electronic data ﬁles. Informal Sector Service, Kathmandu.
Lama-Tamang, M. S., S. M. Gurung, D. Swarnakar, and S. R. Magar 2003. “Social Change in Conﬂict
Areas: Assessment Report.” Prepared for UK Department for International Development (DFID)
Nepal, Kathmandu.
´ n, G. 2012. “Civil Conﬂict and Human Capital Accumulation: The Long-Term Effects of Political
Leo
´ .” Journal of Human Resources 47 (4): 992–1022.
Violence in Peru
Lokshin, M. M., and E. Glinskaya 2009. “The Effect of Male Migration on Employment Patterns of
Women in Nepal.” World Bank Economic Review 23 (3): 481– 507.
Macours, K. 2011. “Increasing Inequality and Civil Conﬂict in Nepal.” Oxford Economic Papers 63 (1):
1– 26.
Manchanda, R. 2004. “Maoist Insurgency in Nepal. Radicalizing Gendered Narratives.” Cultural
Dynamics 162 (3): 237– 58.
Menon, N., and Y. Rodgers 2011. “War and Women’s Work: Evidence from the Conﬂict in Nepal.”
World Bank Policy Research Working Paper No. 5745. World Bank, Washington, DC.
Ministry of Health [Nepal], New ERA, and ORC Macro. 2002. Nepal Demographic and Health Survey
2001. Calverton, Maryland: Family Health Division, Ministry of Health; New ERA; and ORC Macro.
Murshed, S. M., and S. Gates 2005. “Spatial-Horizontal Inequality and the Maoist Insurgency in Nepal.”
Review of Development Economics 9(1): 121– 34.
Pettigrew, J., and S. Shneiderman 2004. “Women and the Maobaadi: Ideology and Agency in Nepal’s
Maoist Movement.” Himal South Asia 17 (1): 19 –29.
Valente 383

Quisumbing, A. R., and J. A. Maluccio 2003. “Resources at Marriage and Intrahousehold Allocation:
Evidence from Bangladesh, Ethiopia, Indonesia, and South Africa.” Oxford Bulletin of Economics
and Statistics 65: 283–327.
´guez, C., and F. Sa
Rodrı ´ nchez 2009. “Armed Conﬂict Exposure, Human Capital Investments and Child
Labor: Evidence from Colombia.” HiCN Working Paper No. 68. Households in Conﬂict Network,
Brighton, UK.
Shemyakina, O. 2011a. “The Effect of Armed Conﬂict on Accumulation of Schooling: Results from
Tajikistan.” Journal of Development Economics 95 (2): 186– 200.
——— 2011b. “Labor Market, Education and Armed Conﬂict in Tajikistan.” World Bank Policy
Research Working Paper No. 5738. World Bank, Washington, DC.
Shemyakina, O., and C. Valente 2011. “Education and Violent Conﬂict in South Asia: Challenges and
Solutions.” Background paper prepared for the World Bank Regional Study on “Improving the
Quality of Learning in South Asia.” Mimeograph, University of Bristol, UK.
Swee, E. L. 2009. “On War and Schooling Attainment: The Case of Bosnia and Herzegovina.” HiCN
Working Paper No. 57. Households in Conﬂict Network, Brighton, UK.
Thomas, D. 1990. “Intra-Household Resource Allocation: An Inferential Approach.” Journal of Human
Resources 25 (4): 635–64.
UNESCO. 2010. Education under Attack. Paris, France: United Nations Educations Scientiﬁc and
Cultural Organization.
USAID. 2007. Nepal – Humanitarian Assistance Fact Sheet No. 1, June. http://pdf.usaid.gov/pdf_docs/
PNADJ147.pdf, last accessed May 20, 2013.
Wallensteen, P., and M. Sollenberg 2001. “Armed Conﬂict 1989–2000.” Journal of Peace Research 38
(5): 629– 44.
World Bank. 2005. “Nepal - Resilience amidst Conﬂict: An Assessment of Poverty in Nepal, 1995–96 and
2003–04.” Report No. 34834-NP. World Bank, Washington, DC.