The Impact of Child Labor on Children’s Educational Performance:
Evidence from Rural Vietnam
Abstract: In 1998, nearly one-third of Vietnamese children engaged in non-housework labor
supply, 95% of these working children residing in rural areas. This paper investigates the impact
of child labour on children’s educational outcomes in rural Vietnam using the 1998-Vietnam
Living Standard Survey. The paper finds that child labor lowers children’s academic performance
and the negative impact is bigger for girls.
JEL classifications: J82; I21; N35; C30
Keywords: Child labor; children’s educational performance; Vietnam; instrumental variable;
joint Tobit-ordered probit model
Published in Journal of Asian Economics 36 (2015): 1-13
1
1. Introduction
Child labor has long been a critical issue. Over the last decades, child labor in developing
countries has been of particular concern. This is not only because child labor is a moral issue but
also because of its significant impact on children’s development, a decisive factor for a country’s
future growth and development (Glewwe and Jacoby, 1994; Edmonds, 2007).
Vietnam is an interesting country to study child labor, for several reasons. First, Vietnam is
a developing country in Asia, a region harboring the largest number of child workers.1 In 1998,
nearly one-third of Vietnamese children aged 6 to 15 years (more than 5 million) were engaged in
non-housework labor supply. Second, Vietnam underwent more than 30 years of war (1945-1975)
and had a long period as a centrally planned economy with stagnant economic performance. By
the end of the period – in 1985 – Vietnam was one of the poorest countries in the world. The
country’s renovation, which started in 1986, has made significant progress in poverty reduction.2
However, in 1998 around 45% of households in rural Vietnam still lived under the poverty line
(World Bank, 1999) and poverty often compels parents to send their children to work for the
family to survive. Third, until the 1990s about 80% of the country’s population still lived in rural
areas and two-thirds of laborers worked in agriculture, mostly self-employed on household farms
(Dollar and Glewwe, 1998). These characteristic provide various opportunities and pressures for
children to engage in child labor within the household.
1
According to the International Labour Organization (2002), there were 211 million children aged 5 to 14 at
work in economic activity in the world in 2000. This accounted for nearly 20 percent of all children in this age group.
The Asia-Pacific region had the largest number of child workers (127.3 million), followed by Sub-Saharan Africa (48
million) then Latin America and the Caribbean (17.4 million). In relative terms, Sub-Saharan Africa had the highest
proportion of working children (29%) followed by Asia and the Pacific (19%) then Latin America and the Caribbean
(16%).
2
See Dutta (1995) and Plummer (1995) for a comprehensive picture of the Vietnamese economy in the initial
years of its transition process.
2
Vietnam has made significant progress in promoting education over the last decades.
Primary education became almost universal in 2000, which is a remarkable achievement compared
with other developing countries with the same level of per capita income. Between 1991 and 1999,
the number of lower secondary enrolments more than doubled, from 2.7 million to 5.8 million and
the number of upper secondary enrolments tripled, from 0.7 million to 2 million. The enrolment
rates in lower secondary education and upper secondary education were 74% and 49% in 2000,
respectively (Vietnam’s Ministry of Education and Training, 2001; Asian Development Bank,
2003). Demand for education also increased substantially in the 1990s as the Vietnamese become
richer (Glewwe and Jacoby, 2004) which reflects the fact that Vietnam is a society in which
education is highly valued and respected. Despite its low income level, Vietnam had a relatively
well educated population (Glewwe and Jacoby, 1998). Nowadays, education is placed as a top
priority in the development strategy of the Vietnamese government in the course of the country’s
industrialization and modernization.3 The share of education expenditure increased from 2.2% of
GDP and 10% of the government budget in 1992 to 3.5% of GDP and 17% of the government
budget in 1998 (World Bank, 2000).
Given the importance of early childhood for children’s success in their later life (Becker and
Tomes, 1986), understanding the impact of child labor on children’s outcomes is particularly
important. To date, most studies on child labor in Vietnam have focused on investigating the
evolution of child labor during the country’s transition period, or the impact of child labor on
children’s health. Influential papers are Edmonds and Turk (2002), Edmonds (2005), Edmonds
and Pavcnik (2005), O'Donnell et al. (2005) and Beegle et al. (2009). However, little is known
about the consequences of child labor in Vietnam on children’s educational outcomes. To our
knowledge, Beegle et al. (2009) is the only study which investigates the problem. Using panel data
3
See Chapter 3 – Article 35 of the National Constitution issued in 1992, the Education and Training
Development Strategy to year 2010 issued in 2001 and the Education Law issued in 2005.
3
from the Vietnam Living Standard Surveys carried out in 1993 and 1998, the authors found that
five years subsequent to the child labor experience, there were significantly negative impacts on
children’s school enrolment and grade attainment. Even for a developed country such as Portugal,
child labor is found to increase the student’s probability of repeating a grade (Goulart and Bedi,
2008).
Using unusually rich information from the 1998-Vietnam Living Standard Survey (VLSS),
this paper investigates the impact of child labor on children’s educational performance. In contrast
to Beegle et al. (2009) we use a direct evaluation of students’ study results in class as an indicator
for children’s educational performance, which is classified into one of four levels ranking from
poor, average, good to excellent.4 In Vietnam, if a student gets a poor academic score, he or she
will have to repeat the grade. Therefore this measure using the ranking of a student’s academic
performance is broader than grade repetition, which is often used to study the impact of child labor
on children’s educational outcome.
Because there are unobserved factors that simultaneously affect child labor and a student’s
academic achievement, leading to the endogeneity of child labor in the educational performance
equation, we use the average wage of unskilled female workers within a commune as an
instrument for child labor. We estimate a system of simultaneous tobit and ordered probit
equations standing for the determinants of child labor and children’s educational performance,
respectively. This estimation is different from previous studies on the impact of child labor on
children’s educational performance which either estimate the two equations separately or use a biprobit or multinomial probit model to estimate the simultaneous determinants of child labor and
4
Children’s educational scores are ranked according to their mark. In each subject, a 10 point scale is applied
with 10 being the highest and 0 being the lowest. A mark under 5 is classified as poor; a mark from 5 to under 6.5 is
classified as average; a mark from 6.5 to under 8 is good and a mark from 8 to 10 is excellent.
4
children’s educational success, treating child labor as a binary outcome.5 Our estimation allows us
to investigate child labor by total hours of child work including hours doing housework and nonhousework labor supply instead of using a dummy variable. Thus we measure the true workload
that a child has to bear.
The estimation results suggest that child labor has a negative impact on children’s academic
performance. The negative impact is bigger for girls. Household economic conditions are one of
the most important determinants of child labor and children’s educational performance. Both
mother’s and father’s education have a strong impact on children’s academic performance. There
is a bias against girls in the child work burden. Among school characteristics, teachers’
qualifications and school infrastructure, such as the presence of electricity, a library and a good
blackboard are significant in enhancing students’ study results. The academic environment within
a commune also has a significant impact on children’s educational performance.
The remainder of the paper is structured as follows. Section 2 describes the data and sample.
Section 3 presents an overview of child labor in Vietnam and children’s academic scores by
activities. Section 4 provides the econometric framework and variable descriptions. Section 5
explains the estimation results. The conclusions are presented in the final section.
2. Data and sample
We use data from the Vietnam Living Standard Survey (VLSS) undertaken in 1997/1998 by
the Vietnamese General Statistics Office, with technical assistance from the World Bank. The
1998-VLSS is the most comprehensive wave of the VLSS. The survey provides information on
both students’ academic performance in their previous grade and their labor activity. Specifically,
the academic outcome takes one of four values: excellent, good, average or poor. Children’s labor
5
The child labour variable equals one if a child reports any positive hours of work and zero if a child does not
work.
5
activity provides detailed information about total hours of housework and non-housework labor
supply over the past 12 months. This characteristic makes our study possible.6
Besides the 1998 VLSS, there have been six other waves of the VLSS carried out in 1993,
2002, 2004, 2006, 2008 and 2010. Unfortunately, none of these waves provides information as
suitable for our study purpose as the 1998 wave. For example, the wave in 1993 provides
relatively detailed information on education and child work participation but does not have
information on students’ academic achievement. The wave in 2006 provides information on the
academic scores of students but does not have sufficient information on child labour in selfemployed agricultural activities. Other waves do not have information on either academic records
or self-employed agricultural activities.
The 1998 VLSS includes five sets of questionnaires relating to households, communities,
schools, prices and clinics. The household questionnaire included 1,561 questions about individual
and household demographics, education, health, labor activities, income, expenditure, credit and
savings. The sample was 6,002 households in 194 communes of 61 provinces of Vietnam and was
chosen to be nationally representative with sample weights attached. The community, school, price
and clinic questionnaires included 737 questions and collected information about community,
school and clinic characteristics, and price levels of basic commodities within the commune.
These questionnaires were administered in 156 communes in all rural and semi-urban areas.
We restrict our sample to rural Vietnam for several reasons. First, while 35 percent of
children aged 6-15 years in rural areas engage in non-housework labor supply, the rate is only 6
percent in urban areas. About 95 percent of working children in Vietnam reside in rural areas.
Second, while the information on community and school characteristics in the 1998-VLSS is
6
It is argued that children’s abilities accumulate over time. Current academic score reflects the study and work
patterns of several previous years. Since we only have information about children’s working hours over the past 12
months, we assume that current work patterns are indicative of the past as well.
6
nationally representative for rural areas, this information was only collected for semi urban areas
(small towns). Third, the sample of VLSS only includes the registered residence, so children of
unregistered migrants and street children are omitted from the sample (O’Donnell and Rosati,
2005). While this omitted sample problem is significant in urban areas, it is not significant in rural
areas because most children in rural Vietnam work with their family in household agricultural
production.
We further restrict our sample to school children aged 6 to 15 years. Six is the age when
children start school in Vietnam and child labor for children under 15 years is of special concern to
the International Labor Organization. In addition, the Vietnamese Labor Code prohibits the
employment of children under the age of 15 years. There are 5,476 children in this group. Children
aged 6 with no information on academic achievement were dropped from the sample, resulting in
the loss of 160 observations.
3. Child labor in Vietnam and children’s academic score by activities
3.1 Child labor in Vietnam
Table 1 breaks down the incidence of child work in Vietnam in 1998 by urban-rural. Of all
rural children aged 6-15 years, 63 percent only studied, 22 percent combined both work and study,
11 percent worked only, and 4 percent reported neither work nor study.
[Table 1 about here]
Among working children in rural areas, the average daily hours of non-housework labor
supply was 2.76 hours. Together with more than one hour spent doing housework, those children
have to work up to four hours a day. Figure 1 further illustrates rural children’s labor force
participation and average hours of non-housework labor supply by their ages. The Figure show
that both children’s labor force participation and average hours of non-housework labor supply
increase as they get older.
7
[Figure 1 about here]
Given that children in Vietnam spend 4–5 hours per day on weekdays (Monday to Friday)
studying in class, working may substantially constrain the time left for doing homework and selfstudying. Long working hours may tire children and reduce their ability to concentrate when
studying. Thus working is expected to have a negative impact on children’s academic
performance.
3.2 Children’s academic score by activities
Table 2 breaks down rural children’s academic attainment by their working status. It can be
seen that, while about 11 percent of children who only study obtain excellent results, the rate is
only 5 percent among children who have to combine both work and study. While about 49 percent
of children who only study get average results, the rate is 56 percent among children who work
and study. Overall, the percentages of children who receive good and poor results are about the
same in the two groups.
[Table 2 about here]
Simple descriptive statistics can only inform us about correlations between factors. We need
a method that facilitates inferences about the true impact of child labor on children’s academic
performance. Such an empirical strategy is described in the next section.
4. Empirical strategy
4.1. Variable descriptions
When estimating the impact of child labor on students’ academic achievement, the main
statistical issue is endogeneity. That is because of both unobserved heterogeneity and simultaneity.
For example, a child with high innate ability can be a successful student despite work. Another
example is that a parent’s preference for education also affects their decision on whether to send
8
their children to work or to school. Parents with high expectations for their children’s education
will do more to help their children succeed at school, such as spending more time or money on
their children’s private tutoring, helping them do their homework, buying more books or reading
more stories for them to develop their literacy skills (Dang, 2007). In addition, household
economics condition may affect both child labor and children’s academic performance. Because of
poverty, parents have to send their children to work for the family to survive (Basu and Van, 1998;
Edmonds, 2007). However children from poor households also tend to have less nutritious food,
less favourable investments for their education and thus achieve lower study results. These factors,
if not controlled for, could lead to biased and inconsistent estimates.
To partly overcome the problem, we exploit the unusually rich information of the 1998
VLSS data set to include a set of variables observed at individual, household, school, commune
and regional levels as explanatory variables. At the individual level, we include variables
reflecting student’s age, age squared, sex and ethnicity. In Vietnam, children have to take the
national exam in their last grade at each school level.7 During the years of their last grades,
children tend to put more effort into studying and parents tend to create more favourable study
conditions for their child, thus reducing the children’s workload. To capture these effects, we
include a dummy variable indicating whether a child is in his or her last grade. We also include the
standardized height-for-age z-scores to partly evaluate the impact of children’s nutritional status
on their academic performance.
The data design allows us to match a child with their parents and thus include mother’s and
father’s years of schooling as additional explanatory variables. These variables are important in
many ways. First, children from a family with better educated parents may inherit a better genetic
7
The current formal education in Vietnam consists of 12 years of basic education, including 5 years of primary
education, 4 years of secondary education and 3 years of high school. Children in our sample – aged 6 to 15 – take the
national exam for primary education at grade 5 and the national exam for secondary education at grade 9.
9
ability. Second, better educated parents usually have higher preferences for their child’s education.
Third, better educated parents can assist their children better in their school work and thus help
them succeed at school. Accordingly, this variable will capture a part of the unobserved
heterogeneity of a child’s ability and the parent’s preference for their child’s education, all of
which are highly important in determining child labor and children’s educational success. Because
temporary paternal absence has direct influences on children’s outcome (Booth and Tamura,
2009), a dummy variable indicating whether both parents are present in the household is included.
At the household level, we include variables that reflect the household economic conditions
and the demographic structure of children within the household. We would like to use household
per capita income to measure household economic status. However, due to measurement errors in
measuring rural income (Deaton, 1997), we use household per capita expenditure instead. Other
variables reflecting the demographic structure of children within the household are the number of
siblings by age groups and the average birth space between siblings.
To control for the heterogeneity of schools and identify which school characteristics are
significant in determining child labor as well as student’s academic achievement, several school
characteristics are used. These are variables indicating whether the school has electricity, a library,
enough chairs and tables, and the proportion of rooms with a good blackboard. Variables reflecting
teacher characteristics are the proportion of experienced teachers having more than five years of
teaching and the proportion of qualified teachers according to the Vietnamese Ministry of
Education and Training (MOET) standard.8
8
According to the MOET teacher’s standard, qualified teachers in primary and lower secondary education are
expected to have graduated from a college-level training institution with 3-years training period and upper-secondary
school teachers are expected to have graduated from a university-level training institution with 4-years training period.
10
At commune level, variables indicating the average distance to the school and the
percentage of educated people within the commune are included. If children live far from school,
going to school takes time and effort; children may choose to stay at home and work. As a result,
the further the distance from school, the higher the probability of missing classes. Following Card
(1993), we expect a negative correlation between the distance to school and children’s academic
score. In contrast, we expect a positive association between the percentage of educated people
within the commune and children’s educational performance. Six regional dummies are also added
to control for heterogeneity across seven regions.
To solve the endogeneity problem of child labor when estimating the academic performance
equation, we need instruments which are highly correlated with child labor but do not directly
influence children’s academic success.9 As discussed in Edmonds (2007), for developing
countries, a perfect instrument is the shadow value of child time. Theoretically, the shadow value
of child time to a household depends on the amount of money that an hour of child work brings to
the family and the marginal utility of an additional money unit to the family. This suggests that
children are more likely to participate in child work than in study if their household is living under
the subsistence level or there are various opportunities for a child to work and the return from an
hour of child work is high (Basu and Van, 1998; Edmonds, 2007). With this in mind, some
variables are thought of as instruments such as: the per capita cultivated land area of the
household, the presence of local factories within 10km of the commune, and the average wage of
unskilled female workers within the commune.10 The relevance of these instruments was checked
9
When studying the effect of school year work on high school achievement for students in the US, Tyler
(2003) used the child labour law which varies across states as an instrument for school year work.
10
We first estimated two different model specifications with the average wage of unskilled female workers and
then the average wage of unskilled male workers as instrumental variables. The result showed that while the average
wage of unskilled female workers is highly statistically significant in determining hours of work for children of both
11
through their significance in the first stage estimation of child labor determinants (See AppendixA). Among them, only the average wage of unskilled female workers within a commune is
statistically significant. As a result, this variable is used as an instrument. In addition, we also find
that the correlation between the average wage of unskilled female workers within a commune and
a student’s educational performance is very small, only at 0.0073 and this correlation is not
significant even at up to a 60% level. The average wage of unskilled female workers within a
commune satisfies the conditions for being a good instrumental variable therefore we used it as an
instrument for child labor when estimating the impact of child labor on children’s educational
success.
4.2. Estimation model
In our observed sample, children’s average daily working hours are censored: about 40% of
children report zero hours of work, while students’ academic scores are discrete and orderedranking. Consequently a system of two simultaneous equations consisting of a tobit equation for
child labor and an ordered-probit equation for children’s educational success is estimated.11
(1a)
sexes, the average wage of unskilled male workers is not statistically significant in determining hours of work for
girls. Our empirical result is consistent with Dinopoulos and Zhao (2007)’s theoretical investigation that under certain
restrictions, child labor and labor supply of unskilled female workers are perfectly substitutable.
11
It is argued that a parent’s decision on whether to send a child to work or to school also depends on the
child’s educational ability. For this reason, previous academic performance should be also included in the current
child labour equation. However, it is challenging to find a variable that affects children’s academic achievement
without affecting child labour in order to get an unbiased and consistent estimate of the impact of previous academic
performance on current child labour in Equation 1b. Our major interest here is the impact of child labour on children’s
educational performance. With an instrumental variable for child labour, we can obtain an unbiased and consistent
estimate of the impact of child labour on children’s educational performance in Equation 1a. We leave the question of
how children’s previous academic performance affects their labour force participation for further studies.
12
(1b)
Where:
is a latent variable for student
’s academic score achieved in the past year (
), which is
ordinal and observed in one of four ranking values: 12
if
(a child has poor study result)
if
(a child has average study result)
if
(a child has good study result)
if
(a child has excellent study result)
: are three cut points of the ordered-probit model.
is a latent variable for the actual value (
indicating average daily hours of child work over
the past year, excluding schoolwork at home:
12
if
(a child does work)
if
(a child does not work)
Children’s educational scores were collected from the self-reported information. This self-reported
information may be biased because a child tends not to report the true academic score if the score is bad. For the case
of Vietnam, as discussed in Dang (2007), there are two reasons that this bias due will not happen. First, in contrast to
many countries, in Vietnam students’ test scores are public. Teachers let all students in the class know the scores of
their friends and name those who score the highest to encourage students to emulate these successes. Second, children
are asked about their academic results in the presence of adult family members so it is unlikely that they will misreport
their results.
13
is a vector of characteristics observed at individual, household, school, commune and regional
levels;
is a vector consisting of the constant term, the vector of observed characteristics
the instrument for child labor;
are parameters to be estimated;
and
assumed to follow a bivariate normal distribution with
and
are error terms
normalised to 1,
is the correlation coefficient between the error terms of the two equations, and the variancecovariance matrix is given by:
.13
5. Estimation results and interpretations
Table 3 reports the determinants and impact of child labor on children’s educational
performance in rural Vietnam. The estimated sample includes children of both sexes. Panel A
presents the results of the joint equations estimates, Panel B presents the results of the single
equations estimate.
In the joint equations estimate, presented in Panel A, the estimated correlation coefficient
between the error terms of the two equations – rho (
– is positive and highly statistically
significant. In addition, while the impact of child labor on children’s academic performance is
positive and not significant in the single equation estimate, it is negative and highly significant in
the joint equations estimate. These results suggest that there is bias in the estimated impact of child
13
When
: the system can be estimated by single equations, when
: a single equation estimation will
be biased and the two equations should be estimated jointly. For further details of the system of simultaneous
equations and its applications, see Amemiya (1974), Dang (2007), Cameron and Cobb-Clark (2008), Nguyen, Liu and
Booth (2012). Our system estimation is programmed in the aML (applied multi-level) statistical package developed by
Lillard and Panis (2003). All optimization in aML uses Full Information Maximum Likelihood.
14
labor on children’s educational performance when the correlation between unobserved factors is
not controlled for.
Figure 2 plots the predicted probability of falling into one of four academic rankings in
response to hours of child work, keeping all other observed characteristics at their mean. The
horizontal axis indicates children’s working hours ranging from 0 to 12. With 12 as the maximum
hours of child work 1, 2, 3.2, 4.1 and 6.1 hours correspond to the 50th, 75th, 90th, 95th and 99th
percentiles, respectively. The Figure suggests that more hours of child work relate to a lower
probability that a child accomplishes good or excellent study results and a higher probability that a
child falls into a poor or average academic category.
[Table 3 and Figure 2 about here]
It can also be seen from Table 3 that the gender dummy is highly significant with negative
estimated coefficients in the child labor equations of all model specifications suggesting that other
things being equal, girls have to work more. There is also a significant difference in academic
achievement between boys and girls. The change in predicted probability of falling into each
academic category in response to a change in children’s gender from girl to boy suggests that,
ceteris paribus, girls have a higher probability of achieving a good or excellent academic category
and a lower probability of reporting a poor academic category. Note that, while girls have to work
more they are more successful at school. This could be consistent with other studies showing that
girls tend to develop earlier than boys and thus perform better than boys at early ages (Epstein et
al., 1998). However, another possible explanation for our result is sample selection. Because our
sample only includes those who are attending school with their academic score available, those
who drop out of school are not observed. For children of primary school ages (from 6 to 10 years),
this sample selection is not a problem because primary education is compulsory in Vietnam and
the enrolment rates for children at these ages are high (over 96 percent) for both sexes. However,
15
as children get older, the gender bias against girls in rural households’ investment decisions leads
to a higher school drop-out rate for girls than for boys. The enrolment rates for girls and boys in
rural Vietnam at lower secondary education level in 1998 are 74 percent and 84 percent,
respectively. Therefore, those girls who remain at school might be a group of higher educational
ability.
We estimated the impact of child labor on children’s educational performance separately for
children of each sex. The estimation results are shown in Table 4. Consistent with the pooled
sample estimate, the estimations separately for children of each sex show that the estimated
coefficients of hours of child work in the academic achievement equation are negative. However,
while the impact of child labor on children’s academic performance is highly statistically
significant for girls in all model specifications, the impact is not significant for boys in
specifications of model 1 and 3.
Figure 3 further plots the predicted probability of achieving one of four academic rankings
in response to the child work burden, keeping all other observed characteristics at their mean, for
boys and girls separately. The absolute slope of the graph – illustrating the change in the
probability of getting a poor, good or excellent academic result in response to hours of child work
– is always higher for girls than for boys. The results suggest that child labor has a bigger negative
impact on girls’ academic performance.
[Figure 3 about here]
Turning to the impact of other explanatory variables, household economic conditions,
measured as household per capita expenditure, play an important role in determining child labor.
Other things being equal, while children from poor households have to work significantly longer
hours, these children achieve better academic results. Ceteris paribus, children from the minority
16
ethnic group have to work more and these children have significantly lower academic results than
children from the majority ethnic group.
Children have to work more as they get older, however the rate of increase declines with
children’s age. We find no evidence of a reduction in working hours during the children’s final
grade of each educational level. However, the dummy variable for the last grades is highly
significant in determining children’s academic achievement. Specifically, the predicted probability
of falling in each academic ranking in response to a change in a child’s last grade shows that there
is a significant improvement in a student’s academic performance in the final grade. This may
arise from the fact that once children are sent to work because of household poverty, the economic
value of child labor contributes to the survival of the family. So despite recognizing the
importance of children’s academic performance in their final grades, there is nothing that
households can do to reduce their child’s work burden if the family is to survive economically.
However students themselves recognise the importance of academic performance in their final
grades and thus make their own effort to succeed in these years.
Consistent with the literature, we find a highly significant and positive impact of father’s
and mother’s education on children’s academic achievement. Additionally, the presence of both
parents in the house has a positive impact on children’s educational performance. Perhaps when
both parents are in the house, they have more time to observe, support and supervise their children
in studying and thus help them have better academic achievements.
Turning to the demographic structure of children within the household, we see that while
children in households with more young siblings (0-5 years) have to work more, children in
households with more older siblings (12 & over years) have to work less. Children in households
with larger birth spaces between siblings have to work less.
17
It is remarkable that children’s height is highly significant in determining their academic
performance.14 Taller children have a higher probability of accomplishing either a good or an
excellent academic result and a lower probability of falling into a poor academic category.
Because children’s height reflects their nutritional status (O’Donnell et al., 2009; Gørgens et al.,
2012), our result suggests that children’s nutritional status plays an important role in improving
children’s academic performance in rural Vietnam.
Does better school quality reduce child labor and enhance students’ learning? Several school
characteristics, such as the presence of electricity, a library and the proportion of rooms with a
good blackboard are found to have significantly positive impacts on the students’ academic
achievement. However, the adequacy of a school’s chairs and tables is not significant in
determining the students’ academic performance. In terms of teacher characteristics, a higher
proportion of experienced teachers significantly reduces child labor. As expected, the probability
of falling into an average and a poor academic category increases with the children’s distance from
school.
Children who live in mountainous areas have to work more than comparable children who
live in the inland delta. However, geographical characteristics are not significant in determining
children’s educational outcome. Notably, the proportion of educated people within the commune
has a strong impact on a student’s academic success. The probability of achieving good or
excellent academic results increases remarkably in communes with a higher proportion of
educated people. Perhaps education in those communities with a higher proportion of educated
people is valued more highly. Parents there have a higher preference for their child’s schooling;
children are more ambitious to succeed at school. It may also be that children in communities with
14
Children’s height-for-age z-score is calculated using version 3.1 of the ANTHRO program released by the
World Health Organization in June 2010.
18
a higher proportion of educated people can easily find people who have the ability to help them
with difficult study problems.
6. Conclusions
This paper evaluates the impact of child labor on children’s academic performance in rural
Vietnam using the 1998 Vietnam Living Standard Survey. We use the average wage of unskilled
female workers within a commune as an instrument for child labor. We estimate a system of
simultaneous tobit and ordered probit equations standing for the determinants of child labor and
children’s educational performance, respectively. We find that child labor has a negative impact
on the students’ academic performance. The negative impact is bigger for girls.
Among other observed characteristics, household economic condition is one of the most
important determinants of child labor. While children from poor households have to work harder,
these children have a higher probability of being successful at school. There is a bias against girls
with regard to the child work burden and households’ educational investment decision. Children
from the minority ethnic group have to work more and these children achieve lower academic
results than children from the majority group. Therefore, rural child support programs should
particularly focus on educational programs to alleviate the gender bias in households’ educational
investment decisions and on socio-economic programs supporting children in the minority ethnic
group.
Because children’s nutritional status plays an important role in improving their academic
performance, child support programs which provide nutritious food for children attending school
in poor rural communities will help to improve children’s academic results. Our results also have
important policy implications for community and school investments. In particular, providing
electricity, a library and a good blackboard are important in enhancing student’s study results.
19
Acknowledgements
For helpful comments on earlier drafts of this paper, we would like to thank Professor Alison
Booth, Professor Robert Breunig, Associate Professor Tue Gorgens, Dr. Ha Trong Nguyen, Dr.
Mathias Sinning and Dr. Yuji Tamura. Any errors are our own.
20
References
Amemiya, Takeshi. 1974. Multivariate Regression and Simultaneous Equation Models when the
Dependent Variables are Truncated Normal. Econometrica 42 (6): 999-1012.
Asian Development Bank. 2003. Education Sector in Vietnam - ADB's Road Map Toward
Secondary Education for All. Manila.
Basu, Kaushik, and Pham Hoang Van. 1998. The Economics of Child Labor. The American
Economic Review 88 (3): 412-427.
Becker, Gary S., and Nigel Tomes. 1986. Human Capital and the Rise and Fall of Families.
Journal of Labor Economics 4 (3): S1-S39.
Beegle, Kathleen, Rajeev Dehejia, and Roberta Gatti. 2009. Why should we care about child
labor? The education, labor market, and health consequences of child labor. The Journal
of Human Resource 44 (4): 871-889
Booth, Alison, and Yuji Tamura. 2009. "Impact of Paternal Temporary Absence on Children Left
Behind." Working paper. No 7440. Centre for Economic Policy Research - The
Australian National University.
Cameron, Lisa A., and Deborah Cobb-Clark. 2008. Do Coresidency and Financial Transfers from
the Children Reduce the Need for Elderly Parents to Works in Developing Countries?
Journal of Population Economics 21 (4): 1007-1033.
Card, David. 1993. "Using Geographic Variation in College Proximity to Estimate the Return to
Schooling." Working paper. No. 4483. The National Bureau of Economic Research.
Dang, Hai-Anh. 2007. The determinants and impact of private tutoring classes in Vietnam.
Economics of Education Review 26 (6): 683-698.
Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconometric Approach to
Development Policy. Baltimore, The Johns Hopkins University Press.
Dinopoulos, Elias, and Laixun Zhao. 2007. Child Labor and Globalization. Journal of Labor
Economics 25 (3): 553-579.
Dollar, David, and Paul Glewwe. 1998. "Poverty and Inequality in the Early Reform Period." In
Household Welfare and Vietnam's Transition, eds. David Dollar, Paul Glewwe and Jennie
Litvack. Washington, DC: The World Bank.
Dutta, M. 1995. Vietnam: Marketization and internationalization of its economy. Journal of
Asian Economics 6 (3): 311-326.
Edmonds, Eric V. 2005. Does Child Labor Decline with Improving Economic Status? The
Journal of Human Resources 40 (1): 77-99.
———. 2007. "Chapter 57 Child Labor." In Handbook of Development Economics, eds. T. Paul
Schultz and John A. Strauss: Elsevier.
Edmonds, Eric V., and Nina Pavcnik. 2005. The effect of trade liberalization on child labor.
Journal of International Economics 65 (2): 401-419.
Edmonds, Eric V., and Carrie Turk. 2002. "Child Labor in Transition in Vietnam." Policy
Research Working Paper. No. 2774. The World Bank.
21
Epstein, Debbie, Jannette Elwood, Valerie Hoy, and Janet Maw, eds. 1998. Falling Boys? Issues
in Gender and Achievement. Florence: Taylor & Francis.
Glewwe, Paul, and Hanan Jacoby. 1994. Student Achievement and Schooling Choice in LowIncome Countries: Evidence from Ghana. The Journal of Human Resources 29 (3): 843864.
Glewwe, Paul, and Hanan G. Jacoby. 1998. "School Enrollment and Completion in Vietnam: An
Investigation of Recent Trends." In Household Welfare and Vietnam's Transition, eds.
David Dollar, Paul Glewwe and Jennie Litvack. Washington, DC: The World Bank.
———. 2004. Economic Growth and the Demand for Education: Is There a Wealth Effect?
Journal of Development Economics 74 (1): 33-51.
Gørgens, Tue, Xin Meng, and Rhema Vaithianathan. 2012. Stunting and selection effects of
famine: A case study of the Great Chinese Famine. Journal of Development Economics
97 (1): 99-111.
Goulart, Pedro, and Arjun S. Bedi. 2008. Child Labour and Educational Success in Portugal.
Economics of Education Review 27 (5): 575-587.
International Labour Organization. 2002. Every Child Counts - New Global Estimates on Child
Labour. Geneva.
Lillard, L.A, and C.W.A Panis. 2003. "aML Multilevel Multiprocess Statistical Software,
Version 2.0. Econ Ware." Los Angeles, CA.
Nguyen, Ha Trong, Amy Y. C. Liu, and Alison L. Booth. 2012. Monetary Transfers from
Children and the Labour Supply of Elderly Parents: Evidence from Vietnam. The Journal
of Development Studies 48 (8): 1177-1191.
O'Donnell, Owen, Ángel López Nicolás, and Eddy Van Doorslaer. 2009. Growing richer and
taller: Explaining change in the distribution of child nutritional status during Vietnam's
economic boom. Journal of Development Economics 88 (1): 45-58.
O'Donnell, Owen, Furio C. Rosati, and Eddy van Doorslaer. 2005. Health Effects of Child Work:
Evidence from Rural Vietnam. Journal of Population Economics 18 (3): 437-467.
Plummer, Michael G. 1995. The emerging new tiger: Economic reform and development in
Vietnam. Journal of Asian Economics 6 (3): 307-309.
Tyler, John H. 2003. Using State Child Labor Laws to Identify the Effect of School-Year Work
on High School Achievement. Journal of Labor Economics 21 (2): 381-408.
Vietnam's Ministry of Education and Training. 2001. Vietnamese Education and Training
Development Strategy to 2010 for the Cause of Industrialization and Modernization.
Hanoi.
World Bank. 1999. Vietnam Development Report 2000 - Attacking Poverty. Hanoi: Joint Report
of the Government-Donor and NGO Working Group.
———. 2000. Vietnam - Managing Public Resource Better - Public Expenditure Review 2000.
Hanoi: Poverty Reduction and Economic Management Unit - East Asia and Pacific
Region.
World Health Organization. 2010. "ANTHRO Program for Assessing Growth and Development
of the World's Children - Version 3.1." Geneva.
22
Table 1: Children’s Activities across Different Characteristics in 1998
By Rural-Urban
Activity
All sample
Rural
Urban
Work only
9%
11%
3%
Study only
67%
63%
87%
Work and study
19%
22%
5%
No activity
4%
4%
5%
Total
100%
100%
100%
Source: VLSS 1998, own calculations.
Notes:
(1) Samples are weighted.
(2) A child is considered as working if he or she has a positive number of hours of non-housework labor
supply
Table 2: Children’s Academic Score by Activities in Rural Areas in 1998
All sample
By Gender
Girls
Study
only
Work
and
study
Study
only
Work
and
study
Study
only
Work
and
study
Poor
4%
3%
3%
3%
4%
4%
Average
49%
56%
43%
52%
55%
59%
Good
36%
36%
41%
39%
32%
33%
Excellent
11%
5%
13%
7%
9%
4%
Total
100%
100%
100%
100%
100%
100%
Academic
ranking
Source: VLSS 1998, own calculations.
Note:
Boys
See Table 1
23
Table 3: Determinants and Impact of Child Labor on Children’s Educational Outcome in Rural Vietnam in 1998
VARIABLES
Panel A
Model 1
Academic score
Hours worked
Age
Age squared
Last grade
Male
Minority
Father's years of schooling
Mother's years of schooling
Parents living in the house
Number siblings age 0-5
Number siblings age 6-11
Number siblings age 12-15
Number siblings age 16-18
Number siblings age 19 & over
Birth space
Height for age z scores
Hours worked
-0.122 **
(0.053)
Mean wage of unskilled workers
Individual & household
characteristics
Ln of household per capita expenditure
Model 2
Academic score
0.319 ***
(0.046)
-0.002
(0.061)
0.000
(0.003)
0.141 ***
(0.045)
-0.238 ***
(0.036)
-0.271 ***
(0.058)
0.047 ***
(0.006)
0.034 ***
(0.006)
0.060
(0.060)
Model 3
Hours worked
-0.110 **
(0.053)
0.137 ***
(0.045)
-0.212 ***
(0.073)
1.149 ***
(0.100)
-0.035 ***
(0.005)
0.090
(0.074)
-0.335 ***
(0.056)
0.298 ***
(0.091)
-0.002
(0.010)
0.023 **
(0.011)
-0.077
(0.095)
Academic score
Hours worked
-0.101 *
(0.061)
0.156 ***
(0.044)
0.236 ***
(0.049)
-0.006
(0.058)
0.000
(0.003)
0.122 ***
(0.046)
-0.248 ***
(0.037)
-0.295 ***
(0.057)
0.048 ***
(0.006)
0.030 ***
(0.007)
0.085
(0.069)
-0.036
(0.028)
-0.088 ***
(0.023)
-0.091 ***
(0.025)
-0.030
(0.035)
-0.016
(0.031)
-0.018
(0.021)
0.084 ***
(0.019)
Panel B
Model 4
-0.180 **
(0.074)
1.164 ***
(0.102)
-0.033 ***
(0.005)
0.092
(0.074)
-0.343 ***
(0.056)
0.218 **
(0.089)
-0.006
(0.010)
0.007
(0.011)
-0.064
(0.102)
0.093 *
(0.048)
0.025
(0.036)
-0.174 ***
(0.041)
-0.293 ***
(0.056)
-0.083
(0.057)
-0.064 *
(0.036)
0.016
(0.030)
School & community characteristics
24
Academic score
0.169 ***
(0.046)
0.288 ***
(0.049)
-0.034
(0.063)
0.001
(0.003)
0.172 ***
(0.047)
-0.239 ***
(0.039)
-0.217 ***
(0.064)
0.049 ***
(0.007)
0.029 ***
(0.007)
0.109 *
(0.062)
Hours worked
0.017
(0.015)
-0.202 ***
(0.077)
1.100 ***
(0.105)
-0.033 ***
(0.005)
0.091
(0.077)
-0.338 ***
(0.058)
0.263 ***
(0.098)
-0.003
(0.011)
0.025 **
(0.011)
-0.074
(0.096)
0.172***
(0.050)
0.308***
(0.047)
-0.073
(0.059)
0.001
(0.003)
0.168***
(0.050)
-0.218***
(0.037)
-0.242***
(0.064)
0.050***
(0.007)
0.027***
(0.007)
0.114*
(0.064)
-0.210***
(0.076)
1.076***
(0.099)
-0.031***
(0.005)
0.092
(0.076)
-0.344***
(0.059)
0.255**
(0.100)
-0.004
(0.011)
0.028**
(0.011)
-0.140
(0.109)
% rooms good blackboard
Have enough chairs & tables
Have electricity
Have library
Share of qualified teachers
Share of experienced teachers
Distance to school
Coastal
Hills & mountain
Have road that car can travel on
% educated people within commune
Regional dummies
YES
Constant
YES
YES
-5.261 ***
(0.747)
Cut1 (Alpha1)
Cut2 (Alpha2)
Cut3 (Alpha3)
0.784 *
(0.436)
2.891 ***
(0.435)
4.286 ***
(0.436)
Sigma
-5.389 ***
(0.790)
-0.254
(0.478)
1.872 ***
(0.479)
3.275 ***
(0.482)
1.697 ***
(0.036)
Rho
Log Likelihood
Observations
Observations left-censored at 0
0.189 **
(0.074)
-10400.151
4,367
YES
4,367
1,665
0.159 **
(0.066)
-0.046
(0.063)
0.154 ***
(0.050)
0.127 ***
(0.044)
0.138
(0.085)
-0.035
(0.104)
-0.016 *
(0.008)
0.064
(0.078)
0.014
(0.055)
0.028
(0.057)
1.751 ***
(0.412)
YES
1.996 ***
(0.584)
4.139 ***
(0.584)
5.542 ***
(0.588)
4,252
1,599
0.159 **
(0.066)
-0.046
(0.063)
0.154 ***
(0.050)
0.127 ***
(0.044)
0.138
(0.085)
-0.035
(0.104)
-0.016 *
(0.008)
0.064
(0.078)
0.014
(0.055)
0.028
(0.057)
1.751 ***
(0.412)
YES
-5.815 ***
(0.976)
1.665 ***
(0.035)
0.172 **
(0.072)
-10074.752
4,252
0.177 *
(0.103)
-0.166 *
(0.100)
0.371 ***
(0.083)
-0.016
(0.067)
0.060
(0.128)
-0.427 ***
(0.153)
0.001
(0.015)
0.013
(0.133)
0.266 ***
(0.087)
0.010
(0.094)
0.922
(0.662)
YES
-5.584***
(0.953)
1.917***
(0.579)
4.082***
(0.581)
5.500***
(0.583)
1.655 ***
(0.037)
0.165 **
(0.082)
-9210.041
3,919
0.177 *
(0.103)
-0.166 *
(0.100)
0.371 ***
(0.083)
-0.016
(0.067)
0.060
(0.128)
-0.427 ***
(0.153)
0.001
(0.015)
0.013
(0.133)
0.266 ***
(0.087)
0.010
(0.094)
0.922
(0.662)
YES
3,919
1,492
1.647***
(0.025)
-3638.190
3,919
-5575.005
3,919
1,492
Source: VLSS 1998, own estimations.
Notes:
(1) Huber-White robust standard errors accounting for clustered sample at commune levels are in parentheses, P values: significance * = 10%; ** = 5%; *** = 1%.
(2) The two equations in models 1 to 3 are jointly estimated. The two equations in model 4 are estimated separately.
(3) Inland Delta is the base group for geographical characteristics. Six regional dummies are included to control for heterogeneity across seven regions.
(3) Model 1 controls for the most basic individual characteristics and regional dummies; Model 2 adds to Model 1 the number of siblings by age groups, the average birth
space between siblings and the standard height-for-age z scores; Model 3 adds to Model 1 school and community characteristics.
25
Table 4: Impact of Child Labor on Children’s Educational Performance in Rural Vietnam in 1998 for Girls and Boys separately
VARIABLES
GIRLS
BOYS
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
Academic score
Academic score
Academic score
Academic score
Academic score
Academic score
equation
equation
equation
equation
equation
equation
-0.166 **
-0.146 *
-0.185 **
-0.086
-0.146 *
-0.037
(0.080)
(0.081)
(0.083)
(0.069)
(0.081)
(0.083)
Other explanatory variables
YES
YES
YES
YES
YES
YES
Observations
2,019
1,967
1,823
2,348
2,285
2,096
Hours worked
Source: VLSS 1998, own estimations.
Notes:
(1) Huber-White robust standard errors accounting for clustered sample at commune levels are in parentheses
P values: significance * = 10%; ** = 5%; *** =1%.
(2) Results are extracted from the joint equations estimates. Model 1 controls for the most basic individual characteristics and regional dummies; Model 2 adds to Model
1 the number of siblings by age groups, the average birth space between siblings and the standard height-for-age z scores; Model 3 adds to Model 1 school and
community characteristics. For the list of variables in each model specification, see Table 3.
26
Figure 1: Rural Children’s Labor Force Participation and Non-housework Labor Supply by Age in 1998
Source: VLSS 1998, own calculations.
Notes:
(1) Samples are weighted by sample weights
(2) A child is considered as working if he or she has a positive number of hours of non-housework labor supply
27
.4
0
.2
Predicted Probability
.6
.8
Figure 2: Predicted Probabilities of Falling in Each Academic Category in Response to Hours of Child Work
0
1
2
3.2
4.1
6.1
12
Daily Hours of Child Work
Excellent
Good
Average
Poor
Source: VLSS 1998, own estimations.
Note:
(1) Sample includes rural children.
(2) These predicted probabilities are calculated from the result of the joint equations estimate reported in Table 3 - model 3; other variables are evaluated at their
mean.
28
Figure 3: Predicted Probabilities of Falling in Each Academic Category in Response to Hours of Child Work for Girls and Boys
.6
0
.2
.4
Predicted Probability
.4
0
.2
Predicted Probability
.6
.8
Boys
.8
Girls
0
1
2
3.3 4.3
6.6
Daily Hours of Child Work
Excellent
Good
Average
12
Poor
0
1
2
3.2 4
6
Daily Hours of Child Work
Excellent
Good
Average
12
Poor
Source: VLSS 1998, own estimations.
Note:
(1) Sample includes rural children.
(2) These predicted probabilities of children falling in each academic category are calculated from the results of the joint equations estimates reported in Table 4
- model 3; other variables are evaluated at their mean
29
Appendix A: First Stage Estimation of Hours Worked per Week on Primary Control Variables Plus Various Instruments
Model 1
Model 2
Model 3
0.000
0.000
0.000
(0.000)
(0.000)
(0.000)
0.015
0.013
0.035
(0.061)
(0.061)
(0.067)
0.141***
0.161***
0.173***
(0.047)
(0.048)
(0.049)
YES
YES
YES
Pseudo R-squared
0.136
0.143
0.147
No. of observations
4,367
4,252
3,919
F statistic
F (3, 4349) = 3.078
F(3, 4227) = 3.986
F(3,3890) = 4.254
Two-sided P value
[Prob> F(3, 4349)] = 0.026
[Prob> F(3, 4227)] = 0.007
[Prob> F(3,3890)] = 0.005
F statistic
F (2, 4349) = 0.105
F(2, 4227) = 0.190
F(2,3890) = 0.136
Two-sided P value
[Prob> F (2, 4349)] = 0.901
[Prob> F(2, 4227)] = 0.827
[Prob> F(2,3890)] = 0.873
Explanatory Variables
Instrumental variables:
Per capita cultivated land area
Presence of factory located within 10km of the commune
Mean wage of unskilled female workers within the commune
Other explanatory variables
Test of the hypothesis H : all the instrumental variables are jointly zero
Test of the hypothesis H : the first two instrumental variables are jointly
zero
Source: VLSS 1998, own estimations.
Notes:
(1) Standard errors are in parentheses, P values: significance * = 10%; ** = 5%; *** =1%.
(2) Model 1 controls for the most basic individual characteristics and regional dummies; Model 2 adds to Model 1 the number of siblings by age groups, the average birth
space between siblings and the standard height-for-age z scores; Model 3 adds to Model 1 school and community characteristics.
30
Appendix B: Summary Statistics of Variables
VARIABLES
All sample
By Gender
Girls
Boys
Dependent Variables:
Hours worked
Academic score
1.254
1.429
1.332
1.504
1.186
1.363
0.197
0.200
0.194
7.637
10.297
112.969
0.182
0.187
7.626
6.194
0.926
0.524
1.503
1.031
0.356
0.277
2.068
-1.896
7.636
10.151
109.812
0.182
0.182
7.658
6.314
0.921
0.572
1.563
1.013
0.360
0.273
2.076
-1.835
7.638
10.424
115.718
0.183
0.191
7.598
6.089
0.929
0.483
1.450
1.047
0.352
0.281
2.062
-1.949
0.770
0.888
0.748
0.422
0.777
0.733
3.169
0.079
0.479
0.823
0.744
0.779
0.893
0.738
0.421
0.784
0.736
3.183
0.088
0.476
0.828
0.744
0.762
0.884
0.757
0.423
0.771
0.730
3.158
0.071
0.482
0.820
0.745
0.196
0.165
0.098
0.105
0.132
0.164
0.196
3,826
0.194
0.172
0.092
0.106
0.132
0.161
0.194
1,781
0.197
0.159
0.103
0.104
0.132
0.166
0.197
2,045
Instrumental variable
Average wage of unskilled workers within
the commune
Explanatory Variables:
Individual & household characteristics
Ln of household expenditure per capita
Age
Age squared
Last grade
Minority
Father's years of schooling
Mother's years of schooling
Parents living in the house
No siblings age 0-5
No siblings age 6-11
No siblings age 12-15
No siblings age 16-18
No siblings age 19 & over
Birth space
Height for age z score
School & community characteristics
Share of rooms with good blackboard
Have enough chairs & tables
Have electricity
Have library
Share of qualified teachers
Share of experience teachers
Distance to school
Coastal
Hills & mountain
Have road that car can travel on
Share of educated people within commune
Regional dummies
Northern Upland
Red River Delta
North Central Coast
South Central Coast
Central Highlands
South East
Mekong River Delta
No of observations
Source: VLSS 1998, own calculations.
Notes:
(1) Samples include rural children aged 6 to 15 years and are weighted.
(2) Hours of child work is the average daily working hours over the past 12 months, including housework and nonhousework labor supply.
31