Caste, Gender and School Enrollment: Evidence from the...
Transcript of Caste, Gender and School Enrollment: Evidence from the...
Caste, Gender and School Enrollment: Evidence from the Nepalese LivingStandard Survey
Margarita Pivovarova∗
This draft: June, 2011
Abstract
This paper explores the role of community effects - ethnic composition of the community andaggregate level of village development - on the probability of school enrollment among the low castegirls in rural Nepal. Using data from the Nepalese Living Standards Surveys 1996 and 2004 andemploying difference-in-difference framework, I find that girls from the underprivileged groups (lowcastes) who live in the villages with the high proportion of upper castes households are more likely tobe enrolled in school, thus implying the presence of positive externalities. This finding is at odds withthe almost uniform agreement about the negative impact ethnic heterogeneity has on the provisionof public goods and performance of the members of underprivileged groups within heterogeneouscommunities. A number of potential channels through which this positive effect might operate areanalyzed: improvements in the school and village infrastructure, increased labour market opportuni-ties through expansion of tourism industry, and peers effects. The empirical findings suggest that theprevalence of the high caste households in the community, along with the overall village developmentand infrastructure, plays an important role in improving school participation and progression amongthe most vulnerable group of children - girls from the low caste households.
JEL CLASSIFICATION: J15; J16; O15KEYWORDS: caste, gender, education, development, heterogeneous community, social norms,
Nepal.
∗PhD Candidate, Department of Economics, University of Toronto. e-mail: [email protected]. I am indebted tomy supervisor, Dwayne Benjamin, for his patience and support, and for enlightening ideas he generously shared with me. Ialso would like to thank Gustavo Bobonis for his helpful suggestions and discussion. Sincere thanks go to seminar participantsat the University of Toronto Department of Economics. I thank the Central Bureau of Statistics Nepal for providing me withthe data.
1
1 Introduction
Nepal, a small landlocked country in South Asia, for the last two decades has experienced shift to a new
social, political and economic society. Previous research has found that transition may affect different
population groups disproportionately. For instance, low caste girls in India were able to exploit new
labor market opportunities better than the boys (Munshi and Rosenzweig 2006), while female wages
in transition in Russia have fallen relative to male wages tremendously (Brainerd 2000). The focus of
this paper is on the dynamic changes in educational outcomes for the most underprivileged group of
Nepalese population - girls from the low castes households, and on the determinants of these outcomes
at the community level1.
Although Nepalese educational statistics has improved a lot for the past three decades, the participation
of the low caste girls remains on extremely low levels compared to the girls from other castes and boys
from the low castes: 53 percent versus 74 percent for other girls and 71 percent for the low caste boys
2. However, in communities where proportion of advantaged castes is above the average for Nepal,
disadvantaged girls are more likely to attend school. This finding is surprising and contradictory for at
least two reasons. First, socio-economic literature has established a broad link between discrimination,
social identity and behavior. Discrimination affects structure of opportunities open to different social
groups. Experimental findings show that when caste identity is publicly revealed in a mixed caste group,
a significant caste gap emerges (Hoff and Pandey 2005a, 2005b, 2006, Kochar 2004) and members of
the disadvantaged castes tend to underperform even in simple tasks. Second, there is almost uniform
agreement among scholars about the direct effect of ethnic heterogeneity on public goods provision both
at macro (Easterly and Levine 1997, Alesina, Baqir and Easterly 1999) and micro level. Ethnic diversity
is negatively correlated with levels of publicly provided public goods (Alesina and La Ferrara 2005).
Higher levels of local ethnic heterogeneity are associated with lower primary school funding and worse
school facilities in Kenya (Miguel and Gugerty 2005), lower access to public goods in India (Banerjee,
Iyer, Somanathan 2004), social heterogeneity negatively affects the maintenance of community projects
1The effect of the caste status on educational outcomes has been extensively explored in studies based on Indian data. See,for instance, Deshpande (2000), Dreze and Kingdon (2001), Dostie and Jayaraman (2006) among others.
2Nepalese society is divided into groups called castes which are linked to one or more traditional occupations. For moredetailed information on castes in Nepal and their classification, see Appendices C1 and C2.
2
(Khwaja 2006). At the same time, heterogeneity has been found to be associated with greater willingness
to contribute privately to a public good (M.Schundeln 2007). The results of the present research are
also in line with Banerjee and Somanathan (2007) who find that in rural India, the population share of
Brahmans (highest caste in Indian hierarchy) in a constituency is positively correlated with access to
primary, middle and secondary schools, to post offices and to piped water.
This paper seeks to explore the relationship between heterogeneity of the community and individual
outcomes from a different angle: instead of looking at the provision of public goods at the community
level in the presence of caste diversity, I will analyze the demand for one of such publicly provided goods
– education – in highly fragmented along caste lines Nepalese society3.
I will rely on two rounds of nationally representative household survey to analyze the determinants of
school enrollment among Nepalese children. More specifically, I will provide evidence that contrary to
the findings in the literature on heterogeneity, social identities and economic outcomes, low caste girls’
participation in school and their school progression are positively associated with the proportions of upper
caste households in the community.
To preview the results of the empirical analysis and provide intuition for the quantitative exercise in this
paper, figure 1 in Appendix B depicts the relationship between the enrollment rates of the low caste girls
and caste compositions of the community where they live. I disaggregate the data by the proportion
of advantaged castes below the median (low intensity) and above the median (high intensity). In 2004,
attendance rates of the low caste girls in the communities with larger proportion of the upper castes
households were consistently higher than those in communities where proportion of high castes was
below the median. That pattern does not hold for the sample of high caste girls in 2004, low caste boys
in 2004, and for the low caste girls in 1996. The raw data suggest that there are changes happening
across time and across communities shifting the attitude towards education among low caste households.
I propose several explanations for the observed trends in school enrollment among the low caste girls
and try to discriminate between those explanations. Among the possible channels are the better school
quality and overall village infrastructure in the communities where the majority belongs to the high
3The effect of the community caste structure in the Indian context has been explored by Dostie and Jayamaran (2006) whofind that caste composition matters but only if one does not control for unobserved heterogeneity within communities
3
castes; response of the households to either decrease in the child wages, or increase in the return to
education; pro-educational and anti-discriminating policies of the Maoist insurgents. I understand that
one or more of the proposed channels are complementary to each other and thus cannot be distinguished
in the empirical analysis. In such cases, I use auxiliary data to provide additional evidence in favour or
against the competing hypothesis. The possible explanations are: (1) response of the household to the
increased returns to education (Munshi and Rozenzweig); (2) response to decrease in current wages for
child labour which lower opportunity cost of going to school; (3) village development, such as better
infrastructure, school availability; (4) peer pressure; (5) pro-educational and non-discriminating policy
of the Maoist insurgents; (6) government policy targeting underprivileged members of Nepalese society,
development programs.
The findings of the the present research indicate that not only an individual’s identity matters – belong-
ing to the disadvantaged caste reduces the probability of school enrollment – but also community caste
composition might significantly affect chances of school participation. The probability of being enrolled,
attend school, or complete primary school conditional on enrollment for low caste girls is higher if they
live in a community where the proportion of upper-caste households is higher. Part of the positive effect
is indeed channeled through better village infrastructure. However, when unobserved heterogeneity be-
tween communities is taken into account, the positive effect vanishes and thus can be attributed to the
social norms prevalent in some of the communities. For instance, perception of education as a high-
valued good might increase the willingness of all households in the community to send their children to
school. [Add]
This paper essentially consists of two parts: the next section describes the data used in the present re-
search, and section 3 presents largely descriptive analysis of the recent developments in the Nepalese
education based on the survey data, as well as detailed statistics for the sample of rural Nepalese children
whose participation in schooling is the subject of the present research. Section 4 outlines empirical strat-
egy based on a series of the OLS regressions. Results of the empirical analysis are presented in section
5. Section 6 concludes with the final remarks.
4
2 Data
The empirical analysis in this paper is based on the data from two rounds of the Nepal Living Stan-
dard Survey (NLSS). The NLSS is a nationally representative survey of households and communities
conducted between June 1995 and June 1996 (NLSS I) and April 2003 and April 2004 (NLSS II) by
the Nepal Central Bureau of Statistics with the assistance of the World Bank. Both rounds use similar
modules to collect data on the demographic composition of the interviewed households, the labour status
of the household members, their health and educational achievements, and various sources of household
income. The NLSS I sample includes information on 3,373 households in 274 primary sampling units
(PSUs), while the NLSS II sample is based on 326 PSUs enumerating 3,912 households.
The sample for the empirical analysis is limited to the rural areas of Nepal for two reasons: 86% of the
Nepalese population live in rural areas, while the remaining 14% live mostly in Kathmandu (capital of
Nepal) and other emerging urban areas. Second, school characteristics and indicators of village devel-
opment have been recorded in a community questionnaire only for rural areas. I further limit the sample
to children 6-15 years - which corresponds to primary, lower secondary and secondary school as only a
small fraction of children continue to higher education (grades 11 and 12). I have to drop observations
for which one or more variables used in the empirical analysis are missing. My final sample consists of
3117 rural children in 1996 NLSS, of whom 1591 are boys and 1526 are girls, and 3776 rural children in
2004 NLSS of whom 1944 are boys and 1832 are girls.
3 Descriptive analysis
For the recent decades, the government of Nepal has implemented various policy measures targeted at im-
provements in educational sector: free primary education for all, government-provided school facilities,
teachers salaries and educational materials, scholarships for poor and female students. The literacy rates
in Nepal increased significantly from 24% in 1981 to 33% in 1990 and to 51% in 20044. However, Nepal
still lags behind its South Asian neighbours in literacy and enrollment rates in post-primary education
levels (secondary, high secondary and tertiary). Table 2 in Appendix A provides descriptive analysis of
4Data from the Central Bureau of Statistics, Nepal. Literacy rate is defined as proportion of population 6 years age andolder who can read and write.
5
the trends in primary (Grades 1 to 5), lower secondary (Grades 6 to 8) and secondary (Grades 9 and 10)
education in Nepal along several dimensions — gender, social status, ethnicity, income, and also captures
regional and geographical differences in those trends. The admission to school begins when child is 6
years old, but some of the children enter pre-school at 5 years. The education system consist of 10 years
of primary and secondary school with an option to continue to high secondary after passing the nation-
wide examination and obtaining the School Leaving Certificate. Overall, net enrollment rates5increased
by 17 percent for primary school, 13 percent for lower secondary, and 9 percent for secondary educa-
tion. Still, about quarter of all primary school aged children were out of school and only 17 percent of
14-15 years old were enrolled in secondary school. Similar to other developing countries, children in
Nepalese urban areas were more likely to attend school, both at the primary and secondary levels, and
this difference was more pronounced at higher levels of education. While the gender gap in enrollment
rates into primary schools has been reduced significantly – from 21 percent in 1994 to 9 percent in 2004
for primary school children, the majority of out-of-school children are girls.
As evident from the table, the differences in the enrollment rates are strongly related to the consump-
tion level for both primary and secondary school. The difference is significant - children from wealthy
households are by 35 percent more likely to attend primary school, and almost eighteen times more likely
then poor children to attend secondary school. Similarly, there are disparities between ethnic and caste
groups6. Highest rates of enrollment at all levels of education are among Hindu groups with lowest rates
for Muslims minorities. More pronounced are differences in the enrollment rates among castes: dominant
groups such as Brahman and Newar have their children enrolled in primary school almost uniformly. The
lowest rates are among Dalits - historically peasants and laborers, also called ”untouchables” low castes.
Table does not include enrollment rates for the rest of the 9 castes group - Hills Dalits, Muslims, Janajatis
from Hills and Tarai and other non-caste groups. Enrollment rates for those groups are roughly 70% for
primary, 20% for lower secondary and 11% for secondary levels, thus being comparable to average rates
in Nepal. Gross enrollment rates also increased among boys and girls and for both low caste children
5Net enrollment rate is the fraction of children in the age group who attend school, gross enrollment rate is the fraction ofchildren out of all children at any level of education.
6Nepal is the only Hindu state in the world with 84 percent of the population identifying themselves as Hindu, 10 percentof Janajatis or Indigenous population (communities that have their own ethnic language other than Nepali and observe theirown distinct customs and tradition than the Hindu), and 4 percent of Muslims.
6
and children from the middle and high castes. While both boys and girls from the low castes were more
likely to attend school in 2004 than in 1996, the proportional increase in the enrollment rates among the
girls was much higher.
Table 3 summarizes individual, household and village level variables describing the sample for present
research, as well as indicators of school quality. The main dependent variable is an indicator of school
enrollment at the time of the survey as reported by an individual or his or her parent. In 1996, 73 percent
of the rural boys in my sample attended school and only 50 percent of the rural girls reported to be
enrolled in school. In 2004, 83 percent of boys and 69 percent of girls attended school. The enrollment
rates among girls in my sample experienced larger proportional increase relative to boys. Among the
individual controls are indicators of caste and gender, and age of the child. The mean age of children
in my sample was 10.2 years, proportion of low caste children was 15 percent in 1996 and 19 percent
in 2004, children from the high caste households was about 43 percent in 1996 and roughly 40 percent
in 2004 7. The set of the household controls includes completed level of education for both parents - to
some extent, this measure capture preference of parents for education. In 1996 sample, the mean years of
completed education for mother and fathers of the sampled children were 0.42 and 2.5 respectively, thus
reflecting the striking gender inequality in access to education among the adult Nepalese population. The
situation was only slightly better in 2004: mothers had on average 0.8 years of education, while fathers
had 3.4 years. Expenditure per capita and land ownership measure child’s family welfare. The average
level of annual expenditure for rural households in 1996 was not different from the average for Nepal:
the mean expenditure per capita in Nepal was 14, 809 rupees, while for the rural household with at at
least one child of 6-15 years in my sample this figure was 16, 557 rupees a year. In 2004, the mean per
capita expenditure deflated by the 1996 prices was 15, 977 rupees, while the corresponding number for
Nepal was 22, 555 rupees. The divergence of rural incomes from the Nepalese average in 2004 reflects
growth of manufacturing and service sectors in the urban areas and higher return to labor in those sectors
in which the majority of the urban population have been employed in 2004. For households living in the
rural areas not only the availability of school matters, but also proximity of the school. The distance to
7The coding of the caste and ethnicity differs among two waves of the survey in that there was less number of ethnic/castegroups identified in 1996, and for some of the households in 1996 sample ethnicity is missing.
7
the nearest primary school captures the direct costs of attending school. In 1996, an average household
in my sample had a school within 23 minutes of walking time, in 2004, that number was 18.5 minutes.
Village level variables in Table [3] capture variation in infrastructure and caste composition across com-
munities. I use village development indicators - availability of electricity, public standpipe, and all
weather road - as a measure of the demand for skilled labor in a given community, and fraction of landless
households and proportion of households below the national poverty line as a measure of overall level of
wealth in a village. Descriptive statistics reveal that compared to 1996, overall infrastructure in Nepalese
villages improved: the proportion of villages with accessible all-weather roads increased from 22% to
69%, fraction of households connected to electricity source increased from 22% to 53%. Public stand-
pipes became available in 58% of the communities compared to 50% in 1996. The fraction of households
who live below the national poverty line decreased by 9 percent and was 30 percent in 2004. This reduc-
tion in poverty rate is consistent with the national trend in general8. For rural areas, landholding is an
important indicator of wealth. Between two waves of the survey, the proportion of landless households
did not change among the sampled villages: on average, 16 percent of households did not own any land
in 1996 and 2004 surveys. The members of such households are usually employed by land owners in the
same or neighbor village, or are engaged in traditional crafts, or other non-agricultural activities.
The main focus of the analysis in this paper is on two variables which describe the ethnic composition of
the community: the caste fractionalization index (CFI) and the proportion of advantaged castes (PHC)9.
Next set of the village level variables measures school quality and quantity. Those measures include
indicators of whether school is a permanent structure, availability of the school supplies (desks and
chairs), number of days school has been closed over the past year and presence of a female teacher in
school. Quantity indicators include the number of primary, secondary or incomplete schools normalized
by the number of households in a village. I also use district level variables – the number of schools,
8NLSS I has found Nepal poverty rate to be 42% , while NLSS II recorded reduction by 11% to 31%. In rural areas,poverty declined from 43% to 35%
9Caste fractionalization index is computed asCFI = 1−∑I
i=1
(ni
N
)2, where I is a number of different ethnic group/castes
in a village, ni is a number of household that belongs to ethnic group/caste i, and N is a total number of households in avillage. The proportion of high caste households is the ratio of the number of households from the advantaged castes to thetotal number of households in the community/village. Both indicators take values between 0 and 1. The closer CFI to 1, themore heterogeneous the community is.
8
pupils and teachers in a district – to construct pupil-teacher ratio at an aggregate level 10.
4 Theoretical Considerations and Empirical Specification
4.1 Theoretical Consideration
To be completed
4.2 Empirical Specification
This section presents a series of ordinary least squares regressions aimed to analyze the relationship
between individual and community characteristics and schooling decision in households where at least
one child currently attends school or has attended school in the past, allowing for this relationship to
differ between boys and girls. The dependent variable is either a school attendance indicator that equals
one if the child attends school during the survey year, or a continuous measure of relative education
level represented by the grade-for-age z-score which is a normalized measure of school progression
for each age-gender group. From the theoretical point of view, schooling choice is determined by the
intersection of the demand and supply curves. In the following specification, demand side variables
are represented by the individual and household characteristics: age of the child, gender, distance to
school and per capita expenditure, mother and father completed years of education; variables representing
village infrastructure. From supply side, I use variables describing school quality: number of primary
and secondary school per household, pupil-teacher ratio, suitability and quality of the school building,
availability of desks and chairs in school. The estimation considers the relationship between community
level indicators and children’s schooling status as follows:
Yihc = α+ βXi + γXh + δXc + εihc (1)
where Yihc is an outcome of interest for child i in household h from community c, Xi is a set of individual-
level controls, Xh is a set of household level characteristics, and Xc is a set of community characteristics.
The estimation allows for low caste × gender effects by the interaction of low caste dummy and the
female indicator.
Yihc = α+ βXi + β1Female + β2Low caste + β3Female × Low caste + γXh + δXc + εihc (2)
10The administrative unit in Nepal is district. There are 75 districts in Nepal.
9
The inclusion of the caste × gender interaction causes the coefficient of the low caste dummy to be inter-
preted as an average change in the outcome associated with the caste status of a child. The interpretation
of the coefficient is as follows: β2 is an estimate of the effect associated with the low caste status for
boys, β3 is the extra increment in the outcome associated with being female in addition to the effect for
boys.The corresponding effect of low caste status for girls is given by the sum of coefficients β2 + β3,
after controlling for age and gender differences. If I believe that the children from disadvantaged caste
are less likely to attend school at any given age and have less years of completed education on average
compared to the children from upper and middle caste households, then β2 < 0 and β2 + β3 < 0 . This
approach also allows me to test whether the effect of the low caste status is the same for boys and girls,
or whether β3 = 0.
The relationship between propensity of the low caste girls to attend school and composition of the com-
munity where they live can be depicted in a standard difference-in-difference table.
School Attendance
PHC<median PHC>median Difference11
LC girls 2004 0.44 0.70 -0.26***
[0.03] [0.04] [0.05]
LC girls 1996 0.37 0.45 -0.08
[0.05] [0.03] [0.06]
Difference 0.07 0.25 -0.18
Z-score
PHC<median PHC>median Difference
LC girls 2004 -0.43 0.05 -0.48***
[0.05] [0.07] [0.09]
LC girls 1996 -0.67 -0.49 -0.19*
[0.05] [0.08] [0.10]
Difference 0.24 0.54 -0.30
This table demonstrates that in 1996, the propensity to attend school for the low caste girl was the same
whether she lived in the village where the majority of the households was from the high castes or not. In
2004, the girls from the disadvantaged castes were more likely to attend school if they lived in a village
10
where the fraction of high castes was above average. Not only the propensity to attend school was higher,
but these girls progressed better as measured by their z-score.
This observation suggest the use of the difference-in-difference strategy where I will exploit potentially
exogenous measure of caste composition.
To analyze the association between own caste, gender and caste composition of the village, the last
specification allows for caste composition×caste×gender effects by the interaction of female and low
caste dummies with continuous variable PHC which measures the proportion of high castes in a village12:
Yihc = α+ βXi + β1Female + β2Low caste + β3Female × Low caste
+β4PHC + β5Female× PHC + β6PHC × Low caste
+β7Female × Low caste × PHC + γXh + δXc + εihc
(3)
As before, the interpretation of the coefficients after controlling for community and gender×low caste
effects is as follows: β6 is and estimate of the effect associated with changes in proportion of high
caste households in a village for low caste boys on schooling outcomes, β7 is an extra increment in the
outcome associated with being female for that of boys, and the compound effect of increasing proportion
of advantaged castes in community for low caste girls is given by β6 + β7. If I believe that the low caste
girls are more likely to attend school and get more years of education when they live in the communities
where proportion of privileged castes is higher, then I expect the sum of coefficients β6 and β7 to be
positive. This approach will also allow me to test whether the girls from the low castes are better off
living in those communities than boys (β7 = 0). Another specification uses village fixed effects, thus
controlling for unobservables at the community level:
Yihc = α+ βXi + γXh + δXc + FEc + εihc (4)
where FEc is a community c fixed effect and all the interactions terms are as in (3). The inclusion of
fixed effects allows me to account for the unobserved village heterogeneity.
The last step is to analyze the relationship between schooling indicators and individual, household and
community characteristics separately for boys and for girls to see whether the aggregation hinders gender
12This specification also includes interaction of CFI - caste fractionalization index with female and low caste dummies.
11
specific effects.
4.3 Identifying assumption
In this section I will discuss why the variation in caste composition of a village maybe considered ex-
ogenous. Appendix [] provide very brief discussion of caste system in Nepal and classification of castes
into three main groups. The caste barriers in Nepal are very rigid and members of low caste by no means
may assume middle or high status. The intercaste marriages in Hindu system was not allowed in order to
keep the characteristics of a caste pure. The change in a caste status of individual might happen through
marriage: if a member of high caste marries low caste, then he or she accepts the lower status. Another
avenue through which the caste composition of the community might change is migration. Since the
movement of individuals from one caste into another is impossible, the caste composition of a village
may change because of migration. To prove my first point, I use data from the survey to analyze the
total number of married individuals and their ethnicity/caste. In 2004, there were 8811 couples for whom
I have data on caste/ethnicity. Out of those 8811 there are only 26 intercaste marriages, or merely 0.3
percent. There was only one case when husband and wife were from the low and upper castes, the rest
are unions between middle and high castes. In 1996, ethnicity/caste was recorded only for the head of
the household, thus explicitly assuming that all the members of the household belong to the same caste.
Also, in the community questionnaire in both survey years, ethnicity of the village residents is reported
at the level of household, again assuming that only members of one caste/ethnicity may cohabitate13.
The second argument requires more detailed analysis on the patterns of migration within Nepal across
districts and villages.
I am using data on migration on both aggregate level - district migration, and household level - probability
of individuals to migrate depending on their social status and controlling for observables. The statistics
at the district level is: number of individuals who migrated during 1996-2006 on average 1.2 individuals
per thousand of population in a district. Three districts have no statistics for migration, but there is
no reason to suspect that these three district have had abnormal migration during the reported period.
13I plan on using Demographic and Health Survey to provide more detailed evidence on the inter caste marriages. Thissurvey is available for 1987, 1996, 2001 and 2006 years.
12
Another way to look at the problem is to show that intercaste marriages are rare. After the marriage,
woman moves to the village/household of her husband and given that husband and wife are from the
same caste, new marriages do not affect caste composition of a given village. Table 1 and Table 2 in
Appendix provide regression results for migration. Numbers for the intercaste marriage in NLSS I and II
are respectively:Another way to show identification assumption holds is to use marriage statistics from
both waves of the survey [can reference here UBC paper - for instance, such and such provide indirect
evidence that there is no caste based migration and almost 99 percent of all marriages are within castes]
Figure 2 in Appendix plots the distribution of ever migrated individuals by gender and caste. Overall, 37
percent in 2004 sample reported ever migrated, and 70 percent of those who ever migrated are women.
This is not surprising given that after the marriage women move to the residence of their husband. This
assumption is confirmed in the data: out of all female who ever migrated, only 7 percent have never
been married and majority of those female are either children or young women below 32 years old (95th
percentile). Those statistics together with the absence of inter caste marriages imply that migration is
unlikely to affect the caste composition of the village or community. One potential threat is a non-
random displacement during the war. However, 72 percent of those who migrated stated family reason
as the primary reason for their decision to migrate, and not disasters or extreme circumstances. For
comparison, similar findings from 1996 confirms migration and marriage patterns. Interestingly, that in
1996, caste was recorded only for the head of the household, which assumes that all members of the
household then belong to the same caste. Also, question about migration is only asked from the head of
the household. Out of 3,372 households in the survey, about 12 percent of the heads of the households
reported that they have ever migrated. Only 15 percent of them are women. The distribution across caste
looks similar: among high, middle and low caste from 11 (middle) to 14 (high) percent of the heads ever
migrated14.
14Population monograph on Nepal based on the 2001 Census provides the following statistics: the volume of within districtmigration in 2001 was 13.2 percent of total population. However, the data collected in 1996 at a smaller smaller spatial level– Village Development Committees – indicated that 32.7 of the native-born population has ever migrated from their place ofbirth. The main reason for migration reported by 27 percent of the migrants was marriage.Source: Bal Kumar KC, PopulationMonograph of Nepal, Central Bureau of Statistics Nepal, 2002
13
5 Discussion of Results
The idea of the empirical estimation is to look at how the magnitude and significance of the interaction
coefficient which measures the marginal change in schooling outcomes for the low caste girls changes
if they live in communities with higher proportion of advantaged castes. By introducing different sets
of controls into regression equation, I am able to track the potential channels for the “spillover” effect
described in the introduction and plotted in Figure [1]. First two columns of Table [7] report the results
for the baseline regression (Eq. 1) where I only include individual and household controls. Since Nepal
is an extremely poor country, it is not surprising that the coefficient on the expenditure variable is not
only statistically significant, but also large in magnitude, implying an increase in the probability of en-
rollment by 14 percent with every income increase by 10 percent. Father education is also an important
determinant of both boys and girls enrollment, which confirms finding from the previous studies. The
low level of mother’s education is reflected in the insignificant impact of that variable on schooling de-
cision. Distance to school, though has been found to be an important determinant of school attendance
in a number of studies, is economically and statistically insignificant in the present analysis, most likely
due to the satisfactory supply of schools and given that all children in my sample live sufficiently close
to the primary school. Coefficients of interest - female and low caste dummies - have expected negative
sign and are large in magnitude conforming patterns from the data: low caste girls are 20% less likely
to attend school comparing to the boys of the same age and caste. However, the impact of belonging
to the low caste on the grade-for-age z-score does not differ by gender among children from the low
castes: on average, grade-for-age z-score is by 0.16 standard deviation lower for both and girls (the mean
grade-for-age z-score for low caste boys and girls are -0.30 and -0.34 respectively).
Next, I include proportion of high castes (PHC) and caste fractionalization index (CFI) as additional in-
dependent variables to test whether children in villages with higher share of advantaged castes perform
better in terms of school participation and grade-for-age z-score (Columns 3 and 4 of Table [7]). The
coefficient of PHC is positive and large in magnitude and statistically significant. However, level of
heterogeneity in the community has even larger positive effect on the probability of enrollment for all
rural children independent of gender. Since the supply of schools and school quality seem to be indepen-
14
dent of the level of caste diversity in the data, positive effect of caste fractionalization might be channeled
through greater demand for schooling in heterogeneous communities. The main purpose of this empirical
exercise is, however, to see how participation of the low caste girls depends on the ethnic composition of
the community. This is done by estimating Eq. 4 with interaction terms. Results are reported in Columns
(5) and (6) in Table [7]. Presence of upper castes in the community has strong positive impact on low
caste girls’ enrollment, while low caste boys are not subject to this influence: one standard deviation
increase in proportion of high castes is associated with 5.3% increase in probability of enrollment for low
caste girls relative to boys from the low castes (as measured by the magnitude and significance of the
coefficient for the interaction of female and PHC variables).
The two last columns of Table [7] present estimation results for the full sample of children with additional
inclusion of the village development indicators. If participation in school is associated with the better
developed village infrastructure, then inclusion of the community-level controls should have reduced the
magnitude and significance of the caste composition effect. However, there is only slight decrease in the
magnitude of the coefficient which measures the marginal change in low caste girls school enrollment:
the implied effect is equal to 7% increase in enrollment rates for one standard deviation increase in
proportion of the high caste households.
Table [1] reports summary statistics on village development indicators to rule out the most obvious reason
why the propensity to attend school for low caste girls is higher in the communities with high fraction
of advantaged caste - households from high caste might care about the overall quality of life and quality
of schooling. Thus, the development and infrastructure of the common property might be better in those
villages. As can be seen in the table, almost all indicators are numerically comparable across two types
of the communities15. The data do not seem to indicate differential trends in the village development -
that, for instance, villages with the high fractions of Brahmins and Chetry households have experienced
faster development in the years between two survey waves.
15For the purpose of the descriptive analysis I define communities where proportion of high caste households is below themedian as low intensity, and where fraction of high caste households is above the median as high intensity. The mean andmedian for the proportion of high caste in sampled Nepalese villages are 0.32 and 0.23 respectively. The number of villageswhere proportion of high caste is above the mean is 98 out of 229 (or 43%), the number of villages where more than half ofthe households belong to the high caste is 65 out of 229 (28%). The village development indicators numerically are equivalentwhether I compare villages above and below the mean or median.
15
Table 1: Village Development Indicators, by intensity of PHCNLSSI NLSSII
PHC<median PHC>median PHC<median PHC>medianNumber of primary schools 0.025 0.032 0.025 0.031(normalized by the number of hhs)Number of secondary schools 0.011 0.018 0.010 0.013(normalized by the number of hhs)Accessible road 0.27 0.17 0.71 0.64HHs are connected to electricity 0.25 0.22 0.53 0.54Fraction of landless HHs 0.20 0.11 0.19 0.12Public standpipe 0.33 0.66 0.44 0.73Number of observations 103 100 114 115
Village development indicators have significant impact on the school participation of rural children and
are interesting to look at on their own16 (this is also true for Indian villages, see Dostie and Jayaraman
(2006) who find positive effect of village road on boys participation). The availability of all weather
road and public standpipe are associated with increase in probability of enrollment for both boys and
girls17. This is an anticipated result since children are responsible for collecting wood and fetching water
in Nepal. Access to clean water saves time and all-weather roads reduce time walking to the forest and
collecting woods, thus reducing opportunity cost of schooling. Presence of the electricity source in a
village also has a positive impact on both girls and boys’ enrollment. This might reflect the fact that
when the village is electrified, children spend less time in the kitchen where they usually maintain labor
intensive and inefficient stoves (those stoves use wood and are very hard to maintain; moreover, they
are very harmful for the health of those present at home throughout the day). Village electrification also
makes the use of the efficient technologies at home more feasible. As expected, higher proportion of
landless households in a village has a negative impact on schooling outcomes. This continuous measure
serves both a proxy for aggregate village welfare and income inequality. In rural Nepal, land is the main
household asset providing living, and 80% of the population depend on agriculture for their income and
employment. The landownership serves as a measure of wealth. Thus, on the aggregate level, villages
where only a few household do not own land, are on average wealthier, with better living conditions and
16Results for the village level indicators estimation are not reported here, but are available upon request.17This is also true in the context of Indian villages, see Dostie and Jayaraman (2006) who find positive effect of village road
on boys participation
16
better school facilities.
To further investigate the differential impact of community caste composition on low caste girls en-
rollment and grade progression, I estimate Equations (2)-(4) separately for boys and girls. Coefficient
estimates for the low caste girls 2004 sample are presented in Table [8]. One notable feature of the
regression analysis in Table [8] is that despite the number of included left-hand side variables, the pos-
itive effect on low caste girls’ enrollment from living in a community with the prevalence of high caste
households remains statistically significant and economically important implying roughly 11 percent bet-
ter chances of attending school and 13.5 points increase in the grade-for-age z-score for every standard
deviation increase in proportion of high caste in a village. This finding suggest that better infrastructure
is not the only channel through which participation in school and grade progression are related to the
community caste composition. The second likely channel as hypothesized in the introduction maybe
peer pressure and social attitudes prevalent in the diverse communities. To test this hypothesis, I need
more detailed data on school attendance across villages with different caste composition. For instance,
I need to know whether children from different castes attend same school, or there are several school
available in the same village and those school ate attended by children from different social groups. The
alternative channel mentioned in the introduction is the labour market externalities. The mechanism sug-
gested and empirically tested by Munshi and Rosenzweig (2006) in Bombay, India, works as follows: in
the presence of the new labour market opportunities, boys from the low caste continued to be channeled
into their traditional occupations, while girls from the same caste were not binded by the caste occupa-
tional constraints. Girls from the low caste were found to be more responsive to the new labour market
opportunities opened by the modernization of Indian society.
One of the new opportunities opened in Nepal for the last two decades was the expansion of the tourism
industry. However, for the period between two waves of the survey from 1996 to 2004 the flow of tourists
did not increase, but rather slightly decreased (385,297 in 2004 and 393,613 in 1996) due to the ongo-
ing civil conflict and associated security concerns. The main purposes of visit reported by those who
traveled were trekking (26%), holiday/pleasure (8%), and business (4%). Although the absolute earnings
from foreign tourism in Nepal have almost doubled from 1995 to 2004, the share of tourism industry in
Nepalese GDP remained the same - 3.8%. Number of travel and trekking agencies increased roughly by
17
50%. However, female population is remains underrepresented in these sectors: only 3% of those em-
ployed are females. Similar pattern is observed in the service sector (hotels and lodges which traditionally
employ more female labour): only 13% percent of the employees are women18. These evidences suggest
that the increased school enrollment among low caste girls can not be solely explained by the expansion
and attractiveness of the tourism industry in Nepal. Female representation in the industry remains at low
levels and the gender composition of the labour force in the tourism industry has been stable for the past
14 years. Data from the NLSS I and II imply that in 1996 only in 3% of the surveyed villages households
owned a lodge (in 7 villages out of 205 rural villages in the sample). In 2004, this number doubled to 8%,
representing 19 villages out of 226. Only in 1 village out of 7 in 1996 the lodges were primarily used
by foreigners rather than locals (14%). In 2004, households reported that 74% of the lodges were used
solely by locals, 10% both by locals and foreigners, and 14% mostly by foreigners. Thus, since 1996
the number of households owning lodges doubled. This finding is in accordance with the aggregate data
for Nepal on the number of hotels and lodges. I have also checked whether the presence of lodges in a
village servicing foreigners is related to the caste composition of the community. It turns out that there is
no direct relationship between the two: in 2004, half of the villages where households reported owning
a lodge belonged to the communities with high intensity of PHC (above the median), and half - to the
low intensity PHC villages. Back in 1996, only quarter of low intensity PHC villages reported to have
lodges. The growth in the number of lodges servicing foreign tourists was mostly through the increase
within the communities with lower proportions of high castes. These evidences from the data need to be
treated with caution. Even though the surveys were designed to be representative for Nepal, the sampling
procedure might not fully reflect the aggregate picture of lodges’ construction and maintenance. One
implication of the survey data to the current analysis is that there is no direct evidence of the impact that
expansion of the tourism industry in Nepal might have had on the low caste girls’ enrollment.
There is a valid concern that the communities where proportion of privileged households is higher are
less likely to be affected by the 1996- 2006 People’s war in Nepal. People’s war has affected Nepal
unevenly: not all of the 75 administrative districts experienced violence and damage to the infrastructure.
During the armed conflict, children often do not go to school for security reasons (long and unsafe
18Source: Nepal Tourism Statistics 2009 (Annual Statistical Report).
18
walk to school, danger of attack), or schools may be closed for the period of the conflict, or the school
building might be destroyed. If districts that were subject to the conflict at a lesser degree, also have
higher fractions of advantaged caste households, then the effect that I find for the low caste girls who
live in those communities can not be attributed to the caste composition of the village. To address this
concern, I provide correlations between the proportion of high caste households in a community and
(1) the number of causalities per 1000 population from 1996 to 2004, (2) number of deaths caused by
maoists, (3) number of deaths caused by the state. The corresponding correlations are: 0.077, 0.077 and
0.067 19. These correlations are positive and small, thus implying that the association between low caste
girls attendance and proportion of high caste households in a community cannot be explained only by the
absence of the armed conflict20.
To test the impact of the supply side variables on the demand for schooling, I use available indicators of
school quality and school supply21. Only one indicator has persistent positive impact on girls’ enrollment
- presence of at least one female teacher in school. Potential positive influence of female teachers on girls
enrollment has been recognized by the Nepalese government in 1997 in the Education for All program
which requires at least one female teacher in multi-teachers schools. Although the number of female
teaches has been steadily increasing until 2003 and it has been reported that on average there was one
female teacher in each primary school, in reality, there were more than 10,000 schools that did not have
a single female teacher22.
The results of the empirical analysis are robust to changes in the specification of regression equation to
probit or logit models, as well as the villages fixed effects model. In all the specifications the marginal
effect of community caste composition (PHC) on low caste girls attendance is positive and significant.
Another potential reason for the higher propensity of school enrollment in the high caste villages is the
less need in child labour. Child labour and schooling have been found to be substitutes in the household
production function. Clearly, if a child is working, he or she has less time left to study. Going to school
involves not only time spent in school, but also time at home doing homework and other extra curriculum
19The data on causalities across Nepalese districts were kindly provided by L. Iyer, Harvard Business School.20Pivovarova and Swee (2011) find no effect of conflict on schooling using the same data, Valente (2011) finds small
positive effect of war on girls’ education independent of caste.21Results are available upon request22Source: Ministry of Education and Sports (2004) “School Level Educational Statistics of Nepal“.
19
activities. To check the patterns of school enrollment and other activities of children, I can simply contrast
proportions of children who go to school and who is engaged in child labour across different communities.
There is a well-known challenge with estimating impact of child labour on schooling and in the absence
of the right instrument this problem cannot be [ ] in the empirical analysis. The best I can do here is to
compare the hours of work for children in high caste communities and the rest. I find that the difference in
hours worked is statistically significant with almost one hour magnitude difference. On average, children
in the high caste communities work 2.2 hours while in the rest of the communities average hours are 3.2.
Table [9] reports results for two other samples: 6-15 years old girls in 1996 and low caste boys in
2004. The same pattern as in Figure [1] emerges as a result of empirical analysis: there is no significant
association between caste composition of the community and school attendance and grade-for-age z-
score for the low caste boys in 2004 and low caste girls in 1996. It is interesting to note, that in 1996 the
gender gap in enrollment rates was much more pronounced than the caste gap: probability of enrollment
for girls was on average 30% lower that for the boys controlling for individual, family and community
characteristics. Low caste status did not have any impact on girls’ or boys enrollment as indicated by the
insignificance of the coefficient for the low caste dummy and low caste*female interaction.
Finally, I run falsification tests using high caste girl dummy in a place of low caste girl dummy. Results
do not hold, moreover, interaction PHC*high caste*female is always negative and never significant23. In
other words, girls from the high caste households on average have the same probability to be enrolled in
school independent of the caste composition of the community compared to other girls and boys from
the high castes.
6 Conclusion
In this paper, I attempted to discriminate among several competing hypotheses in order to explain a con-
tradicting observation from the recent educational data in Nepal: significant increase in the enrollment
rates among young girls from the low castes if they live in communities where majority of the households
belongs to the privileged castes. I analyzed the determinants of school participation in rural Nepal for all
children controlling for family background characteristics and village development indicators with the
23Results are not reported here and available upon request
20
emphasis on the girls from the disadvantaged households. While I find that improved village infrastruc-
ture has positive effect on the propensity to attend school for both boys and girls in rural Nepal, these
findings cannot explain why in communities with larger proportion of high caste households low caste
girls are more likely to attend school and have relatively higher grade-for-age score compared to the boys
from the low castes and girls from other caste groups. My findings suggest that the positive effect of
the community caste composition operates not only through the better developed village infrastructure
(regular supply of electricity and piped water, presence of all-weather road and quality school facilities
in the community). I also analyzed the trends in the tourism industry in Nepal and compared them to
the village-level developments to assess the impact the expansion in the tourism sector might have had
on the employment opportunities of the low caste girls. Both the survey and aggregate Nepal data point
out that even though the tourist industry has experienced growth for the recent two decades, the relative
proportion of females employed in the hotels and services for the tourists did not increase. The third
channel – peer effects and pressure – can not be analyzed with the available data.
The results of the empirical analysis show that the households’ decision to send children to school is
determined both by the adequate supply and quality of the schools, overall level of village development,
and ethnic and caste composition of the community.
My findings suggest that the schooling decisions among rural Nepalese households are subject to ag-
gregate level influences. Thus, in the villages with developed infrastructure, children are more likely to
attend school, and this result is stronger for girls. Older girls whose participation is especially low in
rural Nepal, are responsive to such improvements on the village level as regular supply of piped water
and electricity. Apart from the economic development indicators, important role belongs to the social
attitudes in the community as measured by the village ethnic and caste composition.
In shaping development policy targeted at attracting and keeping girls at school in a country where most
of the population lives in rural communities, it is important to account for social attitudes and tradi-
tions. Another implication is that providing one girl with the scholarship will not change much while
the overall level of development is low. Girls still be forced out of school to fill the needs of families
and communities. In order to promote education in Nepalese rural communities, it is necessary to act at
both individual and village level, closely work with village development committees, and target the most
21
vulnerable groups - poor and low caste children.
References
[1] Alesina A., A. Devleeschauwer, W. Easterly, S. Kurlat, and R. Wacziarg (2003): Fractionalization,
Journal of Economic Growth Vol. 8 (2), 155-94.
[2] Alesina A., R.Baqir and W. Easterly (1999): Public Goods and Ethnic Divisions,”The Quarterly
Journal of Economics, Vol.114(4), 1243-1284.
[3] Alesina, A. and E. La Ferrara (2005): “Ethnic Diversity and Economic Performance,”Journal of
Economic Literature, Vol. 43, 721-61.
[4] Banerjee, A. and R. Somanathan (2001): “Caste, Community and Collective Action: The Political
Economy of Public Good Provision in India,” Manuscript, MIT.
[5] Banerjee, A. and R. Somanathan (2007) “The political economy of public goods: Some evidence
from India,” Journal of Development Economics, Vol. 82(2), 287-314.
[6] Becker, Gary S., (1962): Investment in Human Capital: A Theoretical Analysis, Journal of Political
Economy, Vol. 70, 9-49.
[7] Bista, D.B. (1972): People of Nepal , Ratna Pustak Bhander.
[8] Bista, D.B. (1990): Fatalism and Development: Nepal’s Struggle for Modernization, Sangam
Books, London.
[9] Dostie, B. and R. Jayaraman (2006): ”Determinants of School Attendance in Indian Villages,”
Economic Development and Cultural Change, Vol.54(2), 405-421.
[10] Dreze, J. and Geeta G. Kingdon (2001): ”School Participation in Rural India,” Review of Develop-
ment Economics, Vol.5(1), 1-24.
22
[11] Easterly, W. and R. Levine (1997): Africas Growth Tragedy: Policies and Ethnic Divisions, Quar-
terly Journal of Economics, Vol.112(4), 1203-1250.
[12] Hoff, K. and P. Pandey (2006): “Discrimination, Social Identity, and Durable Inequalities,” Ameri-
can Economic Review Vol. 96 (2), 206-211.
[13] King E. M. and M. A. Hill, eds, Women’s education in developing countries: barriers, benefits, and
policies, Johns Hopkins University Press, 1993. Published for the World Bank.
[14] Khwaja, A. (2006): “Can Good Projects Succeed in Bad Communities?”, Manuscript, Harvard
University.
[15] Lewis, M. A. and M. E. Lockheed (2006): Inexcusable Absence: Why 60 Million Girls Still Arent
in School and What to Do About It. Washington, DC: Center for Global Development.
[16] Lewis, M. A. and M. E. Lockheed (2008): ”Social Exclusion and the Gender Gap in Education,”
The World Bank Policy Research Working Paper WPS 4562.
[17] Miguel, E. and M. K. Gugerty (2005): “Ethnic Diversity, Social Sanctions, and Public Goods in
Kenya,” Journal of Public Economics, 89 (11-12), 2325-2368.
[18] Moffitt, (2001): ”Policy Interventions, Low Level Equilibria, and Social Interactions” in
S.N.Durlauf and H.P. Young, eds. Social Dynamics, Cambridge, Massachusetts: The MIT Press.
[19] Munshi, K. and M. Rosenzweig (2006):”Traditional Institutions meet the Modern World: Caste,
Gender and Schooling Choice in a Globalizing Economy.” American Economic Review Vol 96, n.4
[20] Orazem, P. and E.King (2008): Schooling in Developing Countries: The Roles of Supply, Demand
and Government Policy, Chapter 55 in T. P. Schultz and John Strauss, eds. Handbook of Develop-
ment Economics, Volume 4. Amsterdam: North Holland.
[21] Pivovarova, M. and E.L.Swee (2011):“Quantifying the Microeconomic Effects of War: How Much
Can Panel Data Help?” University of Toronto mimeo.
23
[22] Schultz, T. P. (1993b): Returns to Womens Education, In Womens Education in Developing Coun-
tries: Barriers, Benefits, and Policies, eds. Elizabeth M. King and M. Anne Hill. Baltimore, MD:
Johns Hopkins University Press.
[23] Schundeln, M.(2007): “Ethnic Heterogeneity and the Private Provision of Public Goods,” Harvard
University mimeo.
[24] Valente, C.(2011):“What Did the Maoists Ever Do for Us? Education and Marriage of Women
Exposed to Civil Conflict in Nepal,” The University of Sheffield, Department of Economics Working
Paper 2011009.
24
A Statistical Tables
Table 2: Net Enrollment Rates for Primary and Lower Secondary Schools, 2004(1996), %Indicator Primary Lower Secondary SecondaryAll Nepal 74.27 (57) 31.67 (19) 17.27 (9)Urban 83.14 (71) 47.89 (37) 33.48 (23)Rural 71.79 (56) 26.20 (18) 12.09 (8)GenderBoys 78.95 (67) 33.71 (23) 19.05 (13)Girls 69.50 (46) 29.34 (14) 15.48 (6)Development RegionEast 75.95 (59) 31.13 (26) 18.92 (17)Central 66.02 (51) 35.27 (18) 16.76 (10)West 84.72 (70) 34.11 (19) 17.22 (8)Mid West 80.23 (52) 21.69 (18) 14.62 (2)Far West 74.07 (47) 26.32 (9) 19.28 (0)Ecological ZoneMountains 76.95 (47) 29.77 (17) 10.34 (2)Hills 80.45 (65) 35.39 (21) 16.56 (11)Tarai 68.10 (51) 28.05 (18) 19.52 (9)Quintile1 (poorest) 52.42 (37) 7.86 (6) 2.07 (2)2 71.83 (49) 17.36(9) 4.85 (4)3 82.14 (59) 27.83(18) 13.99(6)4 82.85 (74) 39.94(30) 17.37(10)5 (most well-off) 87.85 (78) 58.97(41) 37.14(29)Ethnic groupHindu 76.76 (60) 35.73 (24) 21.22 (14)Janajatis 75.07 (55) 29.29 (21) 14.04 (12)Muslims 51.01 (47) 5.33 (4) 1.00 (4)CasteBrahman 85.04 (70) 45.39 (29) 27.96 (16)Newar 89.80 (76) 57.14 (37) 24.39 (36)Tarai Middle Castes 62.28 (53) 24.43 (18) 14.86 (6)Dalits (Tarai) 36.00 (42) 7.69 (7) 4.76 (0)Conflict IntensityHigh Intensity 75.63 (56) 27.26 (19) 12.10 (6.2)Low Intensity 72.75 (61) 36.76 (27) 22.97 (21)
Source: Author’s calculation based on NLSS I and NLSS II (2004)
25
Table 3: Descriptive Statistics, Children 6-15 years, Rural NepalNLSSI NLSSII
Boys Girls Boys GirlsPanel A: Individual and Household ControlsSchool Attendance 0.73 (0.44) 0.50 (0.49) 0.83 (0.38) 0.69 (0.45)Years of Schooling 2.64 (2.54) 1.74 (2.27) 3.28 (2.55) 2.69 (2.52)Grade-for-age Z-score -0.16 (1.01) -0.23 (0.96) 0.16 (0.96) 0.22 (0.98)Age 10.2 (2.84) 10.1 (2.83) 10.3 (2.9) 10.23 (2.9)Mother’s years in school 0.41 (1.6) 0.43 (1.6) 0.83 (2.2) 0.82 (2.1)Father’s years in school 2.56 (3.6) 2.49 (3.5) 3.37 (3.9) 3.5 (3.9)Per capita expenditure 17.0 (10.4) 16.0 (9.2) 16.0 (8.9) 15.9 (9.0)Below poverty line 0.41 (0.49) 0.43 (0.49) 0.31 (0.46) 0.34 (0.47)Land owner 0.88 (0.32) 0.90 (0.30) 0.87 (0.34) 0.87 (0.34)Distance to primary school 22.9 (25.0) 24.2 (26.0) 18.4 (17.7) 18.9(17.3)Low Caste 0.15 (0.34) 0.15 (0.34) 0.19 (0.39) 0.19 (0.39)High Caste 0.43 (0.49) 0.43 (0.49) 0.39 (0.48) 0.41 (0.49)Number of observations 1591 1526 1944 1832
Panel B: Village Level ControlsAccessible road 0.22 (0.41) 0.69 (0.47)HHs are connected to electricity 0.22 (0.42) 0.53 (0.5)Fraction of landless HHs 0.16 (0.26) 0.16 (0.26)Public standpipe 0.5 (0.5) 0.58 (0.23)Fraction of HHs below poverty line 0.39 (.25) 0.30 (0.23)
Panel C: School Level ControlsNumber of primary schools 2.75 4.0Number of secondary schools 1.4 1.9Number of incomplete schools 1.8 2.0
One randomly selected primary schoolNumber of days school closed 7.77 17.4School is a permanent structure 0.24 0.47School has electricity 0.11 0.11Desks provided for all students 0.28 0.37Chairs provided for all students 0.19 0.42Female teacher in school 0.35 0.60Number of years school has been in operation 21 27.52Number of observations 203 229
Administrative data, by districtPupil-teacher ratio, primary school, by district 37.25 35.50Number of teachers per primary school, by district 3.8 3.8Number of pupils per primary school, by district 141.1 131.4Number of observations 72 72
26
27
Tabl
e4:
Eff
ecto
fCom
mun
ityC
aste
Com
posi
tion
onSc
hool
Enr
ollm
enta
ndG
rade
-for
-Age
Z-s
core
,All
Chi
ldre
n20
04D
epen
dent
Var
iabl
e(1
)(2
)(3
)(4
)(5
)(6
)(7
)(8
)at
tend
ance
zsco
reat
tend
ance
zsco
reat
tend
ance
zsco
reat
tend
ance
zsco
re
Low
cast
e-0
.04
-0.1
6**
-0.0
5*-0
.19*
**-0
.06
-0.2
4*-0
.07
-0.2
8**
(0.0
3)(0
.07)
(0.0
3)(0
.06)
(0.0
7)(0
.13)
(0.0
7)(0
.13)
Hig
hca
ste
0.08
***
0.23
***
(0.0
2)(0
.05)
Gen
der
-0.1
1***
0.06
*-0
.11*
**0.
06*
-0.1
5***
0.06
-0.1
5***
0.06
(0.0
2)(0
.03)
(0.0
2)(0
.03)
(0.0
6)(0
.10)
(0.0
6)(0
.10)
Fath
ered
ucat
ion
0.01
***
0.04
***
(0.0
0)(0
.01)
Mot
here
duca
tion
-0.0
00.
01(0
.00)
(0.0
1)L
and
owne
rshi
p0.
15**
*0.
29**
*(0
.03)
(0.0
7)D
ista
nce
tosc
hool
-0.0
0***
-0.0
1***
(0.0
0)(0
.00)
Exp
endi
ture
0.14
***
0.38
***
(0.0
2)(0
.04)
Fem
ale*
Low
Cas
te-0
.09*
*-0
.09
-0.0
8**
-0.0
8-0
.04
-0.0
2-0
.04
-0.0
1(0
.04)
(0.0
7)(0
.04)
(0.0
7)(0
.10)
(0.1
6)(0
.10)
(0.1
6)PH
C0.
08**
0.26
***
-0.0
10.
15-0
.02
0.13
(0.0
4)(0
.08)
(0.0
4)(0
.09)
(0.0
4)(0
.09)
CFI
0.12
**0.
34**
*0.
13**
0.38
***
0.10
0.28
**(0
.05)
(0.1
0)(0
.06)
(0.1
2)(0
.06)
(0.1
3)L
owca
ste*
PHC
0.12
0.30
0.15
0.37
*(0
.10)
(0.2
2)(0
.10)
(0.2
1)L
owca
ste*
Fem
ale*
PHC
0.21
*0.
110.
20*
0.08
(0.1
2)(0
.24)
(0.1
2)(0
.24)
Fem
ale*
PHC
0.10
**0.
090.
11**
0.09
(0.0
5)(0
.10)
(0.0
5)(0
.10)
Obs
erva
tions
3776
3776
3776
3776
3776
3776
3760
3760
R2
0.22
20.
238
0.22
90.
250
0.23
60.
252
0.24
30.
263
Rob
usts
tand
ard
erro
rsar
ein
pare
nthe
ses,
and
are
clus
tere
dby
the
LSM
Ssa
mpl
ing
unit:
***s
igni
fican
tat1
%,*
*sig
nific
anta
t5%
,*si
gnifi
cant
at10
%.
Sam
ple
cons
ists
ofal
lrur
alch
ildre
n6-
15ye
ars
old
for
who
min
form
atio
non
left
-han
dsi
deva
riab
les
isno
tmis
sing
.D
epen
dent
vari
able
sar
esc
hool
atte
ndan
ce(o
dd-n
umbe
red
colu
mns
)an
dag
e-fo
r-gr
ade
z-sc
ore
(eve
n-nu
mbe
red
colu
mns
).A
llre
gres
sion
sin
clud
efu
llse
tof
indi
vidu
alan
dfa
mily
cont
rols
,an
dag
edu
mm
ies.
Col
umns
(1)
and
(2)
repo
rtes
timat
esof
equa
tion
[2]
with
incl
usio
nof
indi
vidu
alan
dfa
mily
leve
lco
ntro
lson
ly.
Col
umns
(3)
and
(4)
pres
ent
regr
essi
onco
effic
ient
whe
nPH
Can
dC
FIar
ein
clud
ed,c
olum
ns(5
)and
(6)-
estim
ates
ofeq
uatio
n[4
]with
inte
ract
ion
term
s,an
dco
lum
ns(7
)and
(8)-
with
villa
ge-l
evel
cont
rols
.
28
Tabl
e5:
Eff
ecto
fCom
mun
ityC
aste
Com
posi
tion
onSc
hool
Enr
ollm
enta
ndG
rade
-for
-Age
Z-s
core
,Rur
alG
irls
2004
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
atte
ndan
cezs
core
atte
ndan
cezs
core
atte
ndan
cezs
core
atte
ndan
cezs
core
Low
cast
e-0
.10*
*-0
.22*
**-0
.11*
**-0
.24*
**-0
.09
-0.2
5*-0
.10
-0.2
9*(0
.04)
(0.0
7)(0
.04)
(0.0
7)(0
.08)
(0.1
5)(0
.09)
(0.1
6)Fa
ther
educ
atio
n0.
02**
*0.
05**
*(0
.00)
(0.0
1)M
othe
redu
catio
n0.
000.
03**
(0.0
0)(0
.01)
Lan
dow
ners
hip
0.17
***
0.30
***
(0.0
4)(0
.08)
Dis
tanc
eto
scho
ol-0
.00*
*-0
.01*
**(0
.00)
(0.0
0)E
xpen
ditu
re0.
18**
*0.
42**
*(0
.03)
(0.0
5)PH
C0.
13**
0.30
***
0.07
0.23
*0.
060.
23**
(0.0
5)(0
.11)
(0.0
6)(0
.12)
(0.0
5)(0
.11)
CFI
0.11
*0.
27**
0.13
*0.
28*
0.10
0.19
(0.0
7)(0
.13)
(0.0
8)(0
.16)
(0.0
8)(0
.16)
Low
cast
e*PH
C0.
34**
*0.
39*
0.36
***
0.43
**(0
.11)
(0.2
1)(0
.11)
(0.2
1)
Obs
erva
tions
1832
1832
1832
1832
1832
1832
1825
1825
R2
0.23
50.
282
0.24
40.
293
0.25
10.
295
0.25
90.
307
Rob
usts
tand
ard
erro
rsar
ein
pare
nthe
ses,
and
are
clus
tere
dby
the
LSM
Ssa
mpl
ing
unit:
***s
igni
fican
tat1
%,*
*sig
nific
anta
t5%
,*si
gnifi
cant
at10
%.S
ampl
eco
nsis
tsof
allr
ural
girl
s6-1
5ye
arso
ldfo
rwho
min
form
atio
non
left
-han
dsi
deva
riab
lesi
snot
mis
sing
.Dep
ende
ntva
riab
lesa
resc
hool
atte
ndan
ce(o
dd-n
umbe
red
colu
mns
)and
age-
for-
grad
ez-
scor
e(e
ven-
num
bere
dco
lum
ns).
All
regr
essi
ons
incl
ude
full
seto
find
ivid
ual
and
fam
ilyco
ntro
lsan
dag
edu
mm
ies.
Col
umns
(1)
and
(2)
repo
rtes
timat
esof
equa
tion
[2]
with
incl
usio
nof
indi
vidu
alan
dfa
mily
leve
lco
ntro
lson
ly.C
olum
ns(3
)and
(4)p
rese
ntre
gres
sion
coef
ficie
ntw
hen
PHC
and
CFI
are
incl
uded
,col
umns
(5)a
nd(6
)-es
timat
esof
equa
tion
[4]w
ithin
tera
ctio
nte
rms,
and
colu
mns
(7)a
nd(8
)-w
ithvi
llage
-lev
elco
ntro
ls.
29
Tabl
e6:
Eff
ecto
fCom
mun
ityC
aste
Com
posi
tion
onSc
hool
Enr
ollm
enta
ndG
rade
-for
-Age
Z-s
core
,Rur
alG
irls
1996
and
Rur
alB
oys
2004
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
atte
ndan
cezs
core
atte
ndan
cezs
core
atte
ndan
cezs
core
atte
ndan
cezs
core
atte
ndan
cezs
core
atte
nd3
zsco
re
Low
cast
e-0
.01
0.07
-0.0
40.
14-0
.02
-0.2
1-0
.04
-0.0
7-0
.07
-0.2
4*-0
.07
-0.2
9**
(0.0
7)(0
.21)
(0.0
8)(0
.17)
(0.1
2)(0
.19)
(0.1
2)(0
.19)
(0.0
8)(0
.13)
(0.0
8)(0
.13)
Gen
der(
Fem
ale=
1)-0
.32*
**-0
.28*
**-0
.32*
**-0
.30*
**(0
.04)
(0.0
9)(0
.04)
(0.0
9)L
owca
ste*
Fem
ale
0.00
-0.1
00.
04-0
.02
(0.0
7)(0
.10)
(0.0
8)(0
.13)
PHC
AST
E0.
10**
*0.
150.
08**
0.11
0.13
**0.
22**
0.09
0.16
*0.
010.
150.
000.
14(0
.03)
(0.0
9)(0
.04)
(0.1
0)(0
.06)
(0.1
0)(0
.05)
(0.1
0)(0
.04)
(0.1
0)(0
.04)
(0.1
0)C
FI0.
12**
0.14
0.11
**0.
100.
24**
*0.
41**
*0.
22**
*0.
37**
*0.
14**
0.41
***
0.12
*0.
30**
(0.0
5)(0
.13)
(0.0
5)(0
.13)
(0.0
7)(0
.13)
(0.0
7)(0
.13)
(0.0
6)(0
.12)
(0.0
6)(0
.14)
Low
Cas
te*P
HC
0.08
-0.1
40.
07-0
.25
0.03
0.15
-0.0
9-0
.12
0.12
0.31
0.14
0.38
*(0
.14)
(0.2
3)(0
.15)
(0.2
4)(0
.17)
(0.2
6)(0
.20)
(0.3
0)(0
.10)
(0.2
2)(0
.10)
(0.2
2)L
owca
ste*
Fem
ale*
PHC
-0.0
60.
33-0
.12
0.22
(0.1
8)(0
.29)
(0.2
0)(0
.32)
Fem
ale*
PHC
0.03
0.06
0.02
0.05
(0.0
5)(0
.10)
(0.0
5)(0
.10)
Obs
erva
tions
3113
3113
2805
2805
1522
1522
1363
1363
1944
1944
1935
1935
R2
0.26
40.
232
0.27
20.
231
0.26
60.
276
0.28
50.
280
0.18
60.
215
0.19
20.
227
Rob
usts
tand
ard
erro
rsar
ein
pare
nthe
ses,
and
are
clus
tere
dby
the
LSM
Ssa
mpl
ing
unit:
***s
igni
fican
tat1
%,*
*sig
nific
anta
t5%
,*si
gnifi
cant
at10
%.
Sam
ple
cons
ists
ofal
lrur
algi
rls
6-15
year
sol
dfo
rwho
min
form
atio
non
left
-han
dsi
deva
riab
les
isno
tmis
sing
.D
epen
dent
vari
able
sar
esc
hool
atte
ndan
ce(o
dd-n
umbe
red
colu
mns
)an
dag
e-fo
r-gr
ade
z-sc
ore
(eve
n-nu
mbe
red
colu
mns
).A
llre
gres
sion
sin
clud
efu
llse
tof
indi
vidu
alan
dfa
mily
cont
rols
and
age
dum
mie
s.C
olum
ns(1
)an
d(2
)re
port
estim
ates
ofeq
uatio
n[2
]w
ithin
clus
ion
ofin
divi
dual
and
fam
ilyle
vel
cont
rols
only
.C
olum
ns(3
)an
d(4
)pr
esen
tre
gres
sion
coef
ficie
ntw
hen
PHC
and
CFI
are
incl
uded
,col
umns
(5)a
nd(6
)-es
timat
esof
equa
tion
[4]w
ithin
tera
ctio
nte
rms,
and
colu
mns
(7)a
nd(8
)-w
ithvi
llage
-lev
elco
ntro
ls.
30
B Figures
Figure 1: Enrollment rates among Nepalese girls 6-15 years
31
Figure 2: Migration patterns in 2004
1162
773
1911
1388
3381
2447265
806
8481699
916
2139
0
1000
2000
3000
4000
5000
LC male LC female HC male HC female MC male MC female
Distribution of ever-migrated individuals by caste and gender
Migrated
Never migrated
32
C Additional Information
C.1 Social Exclusion in Nepal
A small landlocked country in South Asia, bordered with China and India, Nepal is one of the 49 least
developed countries in the world, ranked as low as 145 out of 179 countries by Human Development
Index and 99 out of 135 developing countries by Human Poverty Index 24.For the last decade, GDP
growth barely topped the rate of population growth - growth rate of real GDP for 1999-2008 was on
average 3.9% accompanied by 2.2% population growth. Nepal’s pace of growth and poverty reduction
has lagged behind that of other South Asian countries although share of population living below poverty
line decreased from 42% in 1995/1996 to 31% in 2003/2004 25. Nepal has a very diverse society in
several dimensions. Although the majority of the population belongs to the Hindu religion, there are
deep caste divisions in the Nepalese society, and discrimination and human right abuses against the lower
castes are not uncommon. Traditionally, political and economic power was consolidated by interlinking
it with the Hindu caste system.26 The priestly Brahmans were at the top of the ritual order, with the
Kshatriya (kings and warriors) just beneath them and in command of the political order; next came
the Vaishya (merchants) and the Sudra (peasants and labourers). Beneath everyone were occupational
groups, considered ”impure”, and ”untouchable” or acchut. Officially abolished in 1963, caste-based
discrimination, while diluted, remains even today. About twenty Dalit caste groups exist in Nepal at
present. As per Nepalese Census 2001, low occupational castes (Dalits) represents about 14% of the
total Nepal population. Dalits are discriminated on the basis of caste and ”untouchability”. Dalit women
suffer both from discrimination from high caste and within the same caste group.
Dalits have been relegated to do caste-based work as black/goldsmith, tailors, shoemakers and street
cleaners. All these occupations are considered to be of low social status. Persistent poverty and lack of
other means (most of the Dalits do not own any land) force Dalits to continue their traditional occupations.
Dalit women and children often work in the households of their landlords of help in the traditional jobs
of Dalits. Those Dalits who work in bonded labor or forced labor are not even earning from their work,
24United Nations, Human Development Indicator 2008.25Nepal Living Standard Survey I (1996) and II (2004).26This section draws heavily from the World Bank report on Caste, Gender and Ethnic Exclusion in Nepal, 2006.
33
but getting in-kind payment (usually it is food grains). Even when Dalits work for wages they face
differential treatment compared to other caste and ethnic group, with Dalit women earning less than Dalit
men. Therefore, when Dalits change their traditional occupation to wage labor, it does not necessarily
mean that they improve their economic status. The lack of financial resources and modern technology
skills prevents Dalits from participating in slowly modernized Nepalese economy. There exists wide-
spread practice of social and cultural discrimination against Dalits. They are prohibited from entry to
the houses, temples and other public spaces. Untouchability is practiced in schools. There are numerous
evidence from the field surveys and case studies that Dalit children refuse to enrol in school, particularly
if the school lies outside their immediate neighborhood. It should be noted that in case of polarized
communities not only physical distance but also social distance matters for the cost of schooling. In
the school context, discrimination can stem from both peers and teachers: Dalit children are offered
segregated seating in class and are subject to physical and mental bulling.27 Financial support in the form
of Dalit scholarship seems not to bring desired results: Dalits children prefer not to apply for the stipend
which demonstrates their low status in a society.
Studies also show that incidence of caste-base discrimination is prevalent in western region of the country,
implying positive correlation with the degree of development in the region.
Nepals new Constitution (1990) established a more inclusive state. It describes Nepal as ”multi-ethnic,
multi-lingual and democratic” and declares that all citizens are ”equal irrespective of religion, race, gen-
der, caste, tribe or ideology”. However, it also retained some ambiguities by declaring Nepal a Hindu
Kingdom and Nepali as the only official language, denying women the right to pass their citizenship to
their children and explicitly protecting traditional practices, which has been used to bar Dalits (low caste,
occupational caste and ”untouchables” now call themselves Dalits) from temples and to permit continued
caste discrimination.
The archaic caste system has left its mark on education, and continues to influence it today. Traditionally
confined to the elite, education has only recently been recognized as a fundamental right, and in many
areas this attitude has yet to become the norm when considering girl children. The state assumed re-
sponsibility for the education system in the 1970s; previously locally run schools were turned over to a
27UNESCO report ”Winning People’s Will for Girl Child Education” 2005.
34
centralized educational administration. Public education expanded rapidly thereafter. To help poor and
socially excluded children access the kind of education that will open opportunities for them, Nepalese
government committed to provide equal access to educational resources for all excluded groups girls,
linguistic minorities, Dalits and Janajatis. In an effort to reform the system, in 2001 the parliament passed
the Seventh Amendment of the Education Act, allowing management of local public schools to be handed
over to School Management Committees (SMCs). The rules require at least one woman member but do
not mandate Dalit or Janajati representation. Participation of both Dalits and women in the SMCs re-
mains low. Schools with female teachers tend to attract more female students. For that reason the policy
of having at least one female teacher per school in multi-teacher schools was established over a decade
ago, and the Nepal Education for All programme requires at least two female teachers in such schools.
However, neither policy has yet been fully implemented. Not surprisingly that the excluded groups are
also under represented in higher education with Dalits being less than one percent of those with Bache-
lor degree and above and this is largely due to exclusion at the lower levels. Although the government
of Nepal has taken steps to remedy these inequities by subsidizing education and reserving positions in
institutions on tertiary level and in the public sector for the low castes, a large caste-gap in education and
income continues to persist in both rural and urban Nepal today.
C.2 Caste Groups in Nepal
Table below presents caste division in Nepal. NLSS II provides a list of 103 ethnic group living in Nepal.
I use classification of caste in Hindu and Newar communities described in Bista (1972) to construct
measures of caste fractionalization.
35
High CastesChhetri Brahman NewarThakuri Sanyasi KayashtaBaniya Rajput MarwadiJaine Nurang Bengali
Middle CastesKalwar Teli YadavSudi Sonar LoharKoiri Kurmi KanuHaluwai Hajam/Thakur BadheRajbhar Kewat Mallah NumharKahar Lodha Bing/BandaBhediyar Mali Kamar DhuniaMagar Tamang RaiGurung Limbu SherpaBhote Walung BuansiHyolmo Gharti/Bhujel KumalSunuwar Baramu PahariAdivasi Janajati YakkhaShantal Jirel DaraiDura Majhi DunuwarThami Lepcha ChepangBote Raji HayuRaute Kasunda TharuDhanuk Rajbans TajpiriyaGangai Dhimal MecheKisan Munda Santhal/Satar/DhangadKoche Pallarkatta/Kusbadiya
Low castesKami Damai SarkiGaine Badi ChamarMusahar Tatme BantarDhusadadh/Paswan Khatway DomChidimar Dhobi HalkhorMuslim Churoute Bhujel/GhartiNuniya Baantar
C.3 Education system in Nepal
The formal education system in Nepal was established in 1971 and is divided into five levels: 1) pre-
primary or early childhood education for 3 to 5 years of age; 2) primary education (Grade 1-5) for
children from 6 to 10 years; 3) lower secondary education (Grade 6-8) for eleven to thirteen year-old
36
children; 4) secondary education (Grade 9-10) for fourteen and fifteen year-old; 5) higher secondary
education (Grade 11-12) for sixteen and seventeen year-old. Secondary education generally refers to
Grade 6-12. Before the Interim Constitution of 2007, only primary education was provided free charge
for all children enrolled in community schools and textbooks are also provided at no cost, and secondary
education was free only for female students and children from low caste 28, poor29, and ethnic minorities.
The main budget source for financing education in Nepal is the government. The government provides
teachers salaries, management costs, and program costs. At the decentralized level, district, municipal-
ity and village development committees provides support in terms of physical infrastructure and also
teachers’ salaries. In addition, schools collect fees from students. In the case of community lower
secondary/secondary schools, students pay NPR 500 30 (equivalent of US$ 7.06) or more per year in
community schools. Fees for private schools vary. In addition, each school collects various kinds of fees
for activities such as sports, special training, as well as for repair and maintenance of facilities such as
libraries and laboratories. There are mainly two types of institutions providing primary and secondary
education in Nepal: community schools and institutional, or private, schools. Community schools are
run by the government and receive regular grants from the government. Institutional schools are pri-
vately managed and do not receive regular government funding. In Nepal, 17 percent of the students are
enrolled in private schools at all levels of education. In Nepal, transition from primary to lower secondary
education depends on the results of the final examinations that is conducted at the end of Grade 5. In
order to be admitted to Grade 6, students are required to obtain at least 32% in the examination. As for
the higher secondary education (Grade 11-12), admission is decided based on the results of the admis-
sion test at the end of Grade 10, or School Leaving Certificate (SLC). SLC test is considered nation-wide
examination.28Nepali society is based on the Hindu caste system. Dalits are considered one of the lowest and underprivileged groups
called ”untouchables” and have been deprived of basic rights,including the right for education.29poor are defined as those whose family income falls below the poverty line.30Mean nominal expenditure per capita in NLSS II was 15848 rupees and GDP per capita in current US$ in 2004 was 240
(Penn World Tables 6.1).
37
C.4 Survey design
The first round of Nepalese Living Standards Survey (NLSS I) was conducted by the Central Bureau
of Statistics (CBS) in 1995/96. The goal of the survey was to collect information on the extent, nature
and determinants of poverty covering different aspects of household welfare, including consumption,
income, housing, access to facilities, education, health, employment, access to credit and remittances.
The second round of the survey (NLSS II) was originally planned for 2002/03, but was implemented one
year later in 2003/04 with the intention to track changes in the living standards and access the impact
of different government policies and programs targeted on poverty and other indicators of economic and
social development in a country. NLSS II has two parts - cross-section to estimate trends and levels of
the socio-economic indicators, and a smaller panel survey in order to track changes for the last eight
years. Both NLSS I and II followed methodology of the World Bank Living Standards Survey applied
in more than 50 developing countries. The general characteristics of the World Bank suveys comparing
to the specific purpose survey are smaller sample size, intergrated household questionnaire, inclusion of
panel survey part and community questionnaire. The basis for the NLSS II sample frame was provided
by the 2001 Population Census of Nepal. The design of the NLSS I and II was based on the two-stage
stratification method. First, the sample of 334 primary sampling units (PSU, or villages, or communi-
ties) was selected from six strata using probability proportional to size (PPS) sampling method with the
number of households as a measure of size. On a second stage, 12 households from each of 334 PSU
units were selected by systematic sampling from all households listed with the total sample size of 4,008
households. The final size of the survey consists of 3, 912 households in 326 PSUs for cross-sectional
sample, and 1165 households in 95 PSUs as some of the wards could not be reached due to the military
conflict in the area (in total, 12 PSUs could not be reached even after repeated attempts, 8 of them in
cross-section and 4 in panel sample).
38