Documentos de Académico
Documentos de Profesional
Documentos de Cultura
DO
D CT
OC OR
TO AL
RA LD ER
DIISSSSE TA
RT AT ON
TIIO ON
NO N
CH
““C LD
HIIL LA
DL BO
AB UR
OU OV
R,, PPO ER
VE TY
RT AN
YA DH
ND UM
HU AN
MA CA
NC TA
APPIIT L””
AL
A
A SSttuuddyy ooff C
Caam
mbbooddiiaa
encouraged me and provides valuable advice during my study and graciously supported me
Econometrics seminars. Your advice, patient and care have greatly contributed to my
I am very grateful to Professor Terukazu SURUGA and Dr. Keiichi OGAWA, who
comments and suggestions have shaped my dissertation further. I am also grateful to all the
Professors and staff members in GSICS, Kobe University for providing me chances to get
advice and support in my professional career as well as my academic work such as editing
Finally, I am very grateful to my mother Ms. Chhor Sathon, my wife Ms. Chusana
K, HAN and my son Phousanak HAN, and all my brothers and sisters, who always provide
me with their love and warmth, courage and strengths to endure all difficulties in life.
Phoumin HAN
i
TABLE OF CONTENTS
Page
Chapter I: INTRODUCTION
Chapter III: UNDERLYING ROOT CAUSES OF CHILD LABOUR: Empirical Evidence and
3.1 Introduction 50
3.2 Existing Literatures and Alternative Hypotheses 54
3.2.1. Poverty and Altruism 54
3.2.2. Wealth Paradox 57
3.2.3. Parents’ Occupation 59
ii
3.2.4. Parents’ Education 60
3.2.5. Gender 61
3.2.6. Number of Children 61
3.2.7. Social Infrastructure 63
3.2.8. Other Factors 63
3.3 Empirical Framework 64
3.3.1. Empirical Model 64
3.3.2. Data Used in this Study 68
3.3.3 Results of Empirical Estimate 71
3.4 Conclusion 84
Chapter IV: CHILD’S WAGE, HUMAN CAPITAL AND HOURS WORKED TRADE-OFF:
4.1 Introduction 97
4.2 Theoretical Framework 102
4.2.1 Rate of Return to Human Capital Investment 102
4.2.2 Hours Worked of Children and Schooling Trade-off 104
4.3 Empirical Framework 107
4.3.1 Rate of Return to Human Capital of Child Labour 107
4.3.2 Human Capital Formation of Children and Hours Worked Trade-off 107
4.3.3 Dataset and Variables 111
4.4 Results of Empirical Estimate 114
4.4.1 Result Estimates on Rate of Return to Human Capital of Child Labour 114
4.4.2 Result Estimates on Hours Worked and Its Schooling Trade-off 116
4.5 Conclusion 123
iii
Chapter Appendix IV: MULTINOMIAL LOGIT MODEL OF CHILD ACTIVITY 126
Chapter V: HEALTH EFFECTS OF CHILD LABOUR AND ITS DETERMINANTS: Evidence and
iv
Annex 5.4 Probit Coefficient Estimates, the Estimated Value of “Reading 172
Newspaper”
Annex 5.5 Estimated Value of Total Household Income 173
Annex 5.6 Result of First Principal Component for Variable of “Social 174
Capital”
Annex 5.7 Matrix of Covariates Correlation Used in the Model 174
Graph 5.1 Body Mass Index-for-Age Percentiles (Girls, 2 to 20 Years) 177
Graph 5.2 Body Mass Index-for-Age Percentiles (Boys, 2 to 20 Years) 178
Graph 5.3 Stature-for-Age Percentiles (Girls, 2 to 20 Years) 179
Graph 5.4 Stature-for-Age Percentiles (Boys, 2 to 20 Years) 180
v
LIST OF TABLES
vi
Age: SAGE Index (Sample Consists of Children Aged 6-14 Years Old)
4.5 Regression Coefficient Estimates of Years of Child’s Schooling (Sample 121
Consists of Children Aged 6-14 Years Old)
5.1 Basic Characteristics of the Villages Surveyed 142
5.2 Frequency, Row and Column Percentage of Children Aged 5-14, by 143
Categories of Children Operating Machinery and Applying Pesticide
5.3 Frequency, Row and Column Percentage of Children Aged 5-14, by Child 143
Labour and Sex
5.4 Frequency, Row and Column Percentage of Children Aged 5-14, by Sex 144
and BMI-Age Percentile Category
5.5 Frequency, Row and Column Percentage of Children Aged 5-14, by 144
Village and BMI-Age Percentile Category
5.6 Descriptive Statistics (Sample Size of Children in Age Group 5-14 Years 155
Old)
5.7 Hours Worked by Children in the Past 7 Days by Age and Sex 162
5.8 Probit Coefficient Estimates of “Self-Health Assessment” 163
5.9 Regression Coefficient Estimates of “Standardized MBI for Age” 165
LIST OF FIGURES
LIST OF MAPS
vii
C
Chhaapptteerr II
IIN
NTTR
ROOD
DUUC
CTTIIO
ONN
developing countries. The International Labour Organization estimated in 1995 that nearly
250 million children work all over the world. Africa and Asia together account for over 90
percent of total child employment. A later estimate by the ILO in 2002 (ILO, 2002),
confirmed that there were some 211 million children aged between 5 to 14 years engaged
in economic activities in the world. The Asian-Pacific region harbours the largest number
of child workers in the 5-14 age groups, accounting for 127.3 million in total. It is followed
by Sub-Saharan Africa and Latin America and the Caribbean with 48 million and 17.4
million, respectively.
In response to the above issue of child labour, the world leaders have made a
2
commitment to meet children’s rights to survival, health, education, protection and
1
Age is generally used to define child labour by international instruments; they accord the rights and
protection of a child for those under age 18 (the UN Convention on the rights of the child, and the ILO Worst
Forms of Child Labour Convention No. 182). The ILO Minimum Age Convention, 1973 (No. 138) states that
ratifying Members shall raise progressively the minimum age for admission to employment or work to a level
consistent with the fullest physical and mental development of young persons. The minimum age shall not be
less than 15 years, although developing countries, whose economies, educational and administrative facilities
are insufficiently developed, may initially permit children of 12-14 years of age to carry out light work of
certain types and under certain conditions.
2
The Convention on the Rights of the Child: In 1989, world leaders decided that children needed a special
convention just for them because people under 18 years old often need special care and protection. The
leaders also wanted to make sure that the world recognized that children have human rights too. The
Convention sets out these rights in 54 articles and two Optional Protocols. It spells out the basic human rights
that children everywhere have: the right to survival; to develop to the fullest; to protection from harmful
influences, abuse and exploitation; and to participate fully in family, cultural and social life. The four core
principles of the Convention are non-discrimination; devotion to the best interests of the child; the right to
life, survival and development; and respect for the views of the child. Every right spelled out in the
Convention is inherent to human dignity and the harmonious development of every child. The Convention
protects children's rights by setting standards in health care; education; and legal, civil and social services.
1
3
Development Goals (MDGs) . Both the declaration and the MDGs were later reaffirmed in
the 2005 World Summit. Based on fundamental human rights, they provide a framework
for the member countries of the United Nations to work coherently towards a series of
achieve the MDGs on education and the elimination of child labour are mutually
reinforced. On the one hand, education is a key element in the prevention of child labour.
There is broad consensus that the single most effective way to monitor the flow of school
age children into work is to extend and improve access to school, so that families have the
opportunity to invest in their children’s education and it is worthwhile for them to do so.
4
On the other hand, child labour is one of the main obstacles to “Education for All” , since
children who are working full time cannot go to school or cannot complete their studies
(UCW, 2007). For the purpose of poverty reduction, the improvement of education and
health condition are regarded as the most important elements to achieve the MDGs.
The common understanding is that poverty is the seed-bed of child labour. Poor
parents send their children to work for reasons of economic expediency; the consequent
denial of education sets in motion a mutually reinforcing cycle liable to be passed down
poverty; schools are often prohibitively expensive, of poor quality or inaccessible. Cultural
pressures can undermine perceptions of the long term value of education, especially for
3
The Millennium Development Goals were adopted five years ago by all the world's Governments as a
blueprint for building a better world in the 21st century. The Millennium Development Goals (MDGs)
consist of eight goals – which range from halving extreme poverty to halting the spread of HIV/AIDS and
providing universal primary education, all by the target date of 2015 – form a blueprint agreed to by all the
world’s countries and the entire world’s leading development institutions. They have galvanized
unprecedented efforts to meet the needs of the world’s poorest.
4
Educational for All is an international commitment first launched in Jomtien, Thailand in 1990 to bring the
benefits of education to “every citizen in every society.” Partners comprised a broad coalition of national
governments, civil society groups, and development agencies. In response to slow progress over the decade,
the commitment was reaffirmed in Dakar, Senegal in April 2000 and then again in September 2000, when
189 countries and their partners adopted two of the EFA goals among the eight Millennium Development
Goals (MDGs).
2
female children. Hence, to stop the vicious cycle of inter-generation of poverty and child
labour, policy measures for the improvement of children’s education and health in poor
The issue of child labour is often regarded as harmful to education and health.
Concern about human capital consequences of child labour derives primarily from the
belief that work increases the children’s exposure to health hazards that could damage to
their health, and work could also force them to drop out of school due to long hours
worked and fatigue. In addition, hazards may also threaten psychological health of children
because their physiological immaturity and the biological process of growth may be more
vulnerable than adults, given that children are more prone to injury through accidents
(ILO, 1998). On the other hand, some literatures have mentioned that child labour indeed
contributes to the human capital of children by introducing income to the households and
being able to gain enough food for their growth as well as their education. Fan (2004) has
found that small increase in child labour may not be a trade-off with human capital
investment since the positive impact of increased financial resources on education may
outweigh the negative impact of reduced time of study. This is simply that children’s
labour market participation raises the financial resources, which can be spent on their
education. Bhalotra and Heady (2003) have found that children of land-rich households are
more often found in work than the children of land-poor households, which implies that
greater poverty does not lead to greater child labour. This also implies that child labour
exists mostly in land-rich households, and thus these children will have better food-
3
5
In Cambodia, many households live under the poverty line (see details in Table
6
1.1). The World Bank in Cambodia (WB, 2006) announced a significant 12 percentage
point decrease in poverty over the last decade based on household consumption data
collected in 2004. Thus, instead of announcing that poverty had decreased very modestly
from 39% to 35% in 1993-2004, the Bank announced that poverty had decreased from
47% to 35%. However, the number of the poor has not dropped in real terms because
7
population growth has made the actual number of poor people increase .
(Round 2) Rounds)
Phnom Penh 11.4 6.2 11.1 3.4 9.7 3.3 14.6 5.2 5
Other Urban 36.6 19.6 29.9 15.4 25.2 13.7 42.4 28.4 18
Areas
Rural Areas 43.1 21.9 40.1 20.0 40.1 12.1 56.1 31.5 38
5
The Cambodia Socio-Economic Survey 1993/94 (CSES, 1993/94) highlighted a high incidence of poverty,
at 39 percent of the total population. The latest CSES-2003/04 shows that the poverty rate has dropped to 35
percent
6
The Bank recalculated the national poverty rate in 1993/94 from 39% (Prescott and Pradhan, 1997 and
World Bank, 1999) to 47% using backward projection (Ear. S, 2006).
7
Based on the author calculation using independent data from CSES1993 and CSES2004, the poor
population has increased from 4,157,192 in 1993 to 4,703,697 in 2004, while the poverty ratio fell from 39 %
in 1993 to 35% in 2004.
4
This enormous amount of poverty inevitably has social consequences, including the
creation of child labour. The recent economic growth has resulted in children not only
working on family farms, but also outside of their households as wage-labour. The
Cambodia Child Labour Survey 2001-02 (CCLS-2001/02) estimated that there were about
1,516,363 children aged 5-14 who can be considered “working children”, about 44.8
percent of children in this age group. In considered the age group 5-17 years old, there
In fact, the majority of children in Cambodia are found to be involved in labour and
whether child labour is harmful or not in the context of Cambodia is a matter of empirical
investigation. Besides the worst form of child labour as defined by the ILO, most children
have, for many years, been participating in the household economy. The agrarian system in
the Kingdom has always encouraged the use of family labour in farms and livestock
management, and as a result the majority of households have tended to use children. As
school-enrolment does not commence at least until six years of age, most parents tend to
bring their children to their workplace, and so on farms, children work alongside their
parents. Furthermore, the local perceptions towards child labour among Cambodian
communities is that child labour is good because the child understands the difficult life of
parents, and thus the income deriving from child labour is regarded as important to
conditions remain crucial. The reduction and gradual elimination of child labour is vital for
the country to attain the goals set by the Royal Government in its socio-economic
development objectives and in building an equal and equitable society. The Cambodian
5
Poverty Reduction Strategy (NPRS, 2003, 2005), the 2003 Cambodian Millennium
Development Goals (CMDGR, 2003), the Rectangular Strategy (RS, 2004) and the
National Strategic Development Plan (NSDP) 2006-2010 are all important policy
documents aimed at addressing the issue of poverty reduction, better service coverage, and
improved schooling, all of which are necessary elements of a durable solution to child
labour.
However, the battle against child labour in developing countries will come at a very
heavy cost as it will reduce the total household’s income if the policy of banning child
labour is implemented. Poor households often cannot afford to send their children to school
without the income derived from their child labour. Needless to say, in Cambodia as well
as many developing countries, the government cannot afford to provide subsidies for
This study departs from the fact that there are many arguments over the issue of
child labour, especially what are the real causes of child labour. Many literatures have
proved that poverty is the main cause of child labour, while many others seem to not fully
agree. On the basis of these arguments, the first core paper of this study in Chapter III has
been formulated to find the “underlying root causes of child labour in Cambodia”.
Regardless of the causes of child labour, there are many literatures that discus the negative
impact of child labour on their health and education, while many others seem to not agree.
It is also that many studies in the past have used data limited only to “child participation in
labour force” without considering their hours worked. This could cause big confusion
among the analysis of the impact of child labour on human capital of the child. For
example, if a child works 1 hour a day, it must inevitably have a different impact than from
combine work with study. This simply means that the income derived from child labour
6
could possibly maintain child health through enough food consumption, and also enable
the child to attend school. On the basis of these facts, this study attempts to investigate the
trade-off between hours worked by children and their human capital formation through
The study of “Child Labour, Poverty and Human Capital” in Cambodia is worth
exploring as its matter is contemporary and related to individuals, households and the state
as a whole. The consequences and its connectivity of child labour to health and education
are recursive and vice-versa. Many studies, for instance, Basu and Van (1998), Basu and
Tzannatos (2003), Lee and Westaby (1997), Saupe and Bentley (1994), Kim and Zepeda
(2004), Chakraborty and Das (2004), Grootaert and Ravi (1995), Chao and Alper (1998),
Duryea and Arends (2001), Basu, Arnab K., and Nancy H. Chau (2003), and Blunch and
Dort (2000), have come to believe that the causation of child labour is poverty and the
reverse is true 8. Moreover, child labour has been believed by many economists as a cause
of future poverty, so that the direct measure to move children to school can make an
However, the reality is not simple and any policy will come at its cost. We know that
poverty and child labour are mutually reinforcing; because their parents are poor, children
must work to maintain their schooling, and in many case these children remain out of
school. Policies to reduce child labour are an overarching goal; however, each country may
take its measures and approaches differently to deal with child labour, poverty and their
human capital.
8
We provide detailed literature review in chapter 3, 4 and 5 of its relatedness of specific studies in this study.
7
There is a growing body of literatures on the study of child labour, poverty and the
human capital of children. Basu and Van (1998) have developed an economic model in
which poverty and child labour are potentially crucial components. The model asserts that
the existence or non-existence of child labour depends on the general level of productivity
in the economy. If the economy is very unproductive, child labour exists in equilibrium,
while in a very productive economy, it will not. Two assumptions were identified: first, the
“luxury axiom” stated that children are sent to work only if the household’s income from
sources other than child labour is very low. Secondly, the “substitution axiom” assumes
that child labour is a substitution for adult labour. This model has set the foundation for
many other studies. Based on this theory, policy makers now have a greater understanding
of why parents send their children to work. Fallon and Tzannatos (1998) reviewed a
variety of studies that indicated a strong negative relationship between the incidences of
child labour and household income, but the relationship is less marked in more affluent
developing countries. Ray (2000) has found a strong negative correlation between
household income and child labour, and a positive relationship between household income
and school enrolment in Pakistan, but no such relationship in the relatively wealthier Peru.
Fan (2004) has analysed child labour and children's human capital formation in response to
changes in the relative wage/productivity between child labour and adult labour. He has
demonstrated that a rise in child labour productivity may lead to an increase in both child
labour and children's human capital. So, in contrast to conventional wisdom, this model
has shown that a small increase in child labour may not adversely affect children's human
capital since the positive impact of increased financial resources for education may
outweigh the negative impact of reduced time for study. Moreover, the model is extended
interesting results. First, it shows that when the subsistence constraint is binding and
8
parental income is low, child labour will decrease with children's relative
sufficiently high, an increase in child labour will increase children's human capital
regardless of whether the subsistence constraint is binding. Third, it indicates that when
parental income is below subsistence levels, the implementation of laws that partially deter
child labour will both increase child labour and reduce children's human capital. Mincer
(1974) has conveyed to the reader the importance of investment in human capital. In his
theoretical model, all individuals attain a certain level of human capital and this level is
primarily influenced by education and training. It should also be noted that Adam Smith
(1776) was the first to suggest that an educated worker could be likened to an expensive
machine, meaning investment in human capital increases productivity. The skills embodied
in a person can be "rented out" to employers. The higher the level of skills a person has the
higher this "rent" is likely to be. Thus, the expected returns on investment in human capital
are that a higher level of earnings and greater job satisfaction over one's working lifetime.
This basic concept can be found in existing literatures, which compare the costs and
of investing in education.
In the Basu and Van model two axiom assumptions seem plausible, and several
studies have supported the validity of the hypotheses. However, we have also observed,
particularly in the studies of child labour, that there is a lack of theoretical studies which
can explain the empirical findings. For example, if the subsistence level of each household,
considered as the poverty line, is set at a higher level, it is possible that there will be a
larger number of households under the poverty line who send their children to school. On
the other hand, if the poverty line is set at a lower level, there will be a large number of
household who are non-poor, but send their children to work. If this is the case, it turns out
9
to be a contradictory view to the theoretical model of child labour, for example, altruistic
parents who are poor and still send their children to school. Taking these points into
consideration, it seems that the model of child labour could not predict the result of child
In addition, the trade-off between child labour and schooling has been assumed, for
instance, Ravallion and Wodon (2000), Patrinos and Psacharopoulos (1995, 1997), Basu
(1999), Jensen and Nielsen (1997). However, their analyses mostly are based on children’s
participation rates rather than hours worked. The only exceptions are only Rossati and
Rossi (2003), and Ray and Lancaster (2005) where hours worked of children are accounted
for. This, perhaps, indicates that data on hours worked were unavailable in many countries,
thus limited the analysis only in the domain of child’s participation in the labour force.
On the basis of the general situation related to child labour, some literatures have
shown the harmful consequences of child labour; while many others the positive impact of
child labour. These literatures of positive and negative impacts of child labour have fed
into the growing literatures, and the different opinion on the child labour issue. In addition,
many studies have shown that poverty is the cause of child labour, for in stance Basu and
Van (1998). However, there are many arguments that greater poverty does not lead to
greater child labour (Bhalotra and Heady, 2003). Following these arguments, this study has
9
The altruistic model by Becker (1976) has been wisely used in connection to household decision-making
towards their household’s members because altruists itself is willing to reduce her/his own consumption in
order to increase the consumption of others. Two assumptions were made in the model of altruism: First, the
own consumption of egoistic persons would exceed that of equally able altruistic persons; secondly, the own
consumption of egoists would be greater if the wealth of egoists and altruists were equal because altruists
give away some of their wealth to be consumed by others. From this model, it is also implicated that, if
parents are altruistic, their children will be allocated in school, or both school and work, rather than working
only.
10
(2) Does child labour have a negative impact on human capital formation through
study. Does their human capital formation through their level of education
(3) Does child labour have negative impact on human capital formation through
In order to answer the above aforementioned questions, this study aims to make the
investigation as follows:
(A) To test the Basu and Van model (1998) on the assumption that poverty is the
cause of child labour, the “wealth paradox” of Bhalotra and Heady (2003), and
(B) To investigate the impact of child labour on human capital formation through
education and health. Furthermore, this study also investigates if the level of
(C) To investigate the other factors which have positive effects on human capital
formation of the child, and provide some effective policy measures to the issue
In order to achieve the objectives of this study, the author employs both qualitative
and quantitative methods by using national dataset as well as the field survey 10 data
conducted in 2006 and observations to come up with concrete results as shown in details
10
The field survey led by Professor Seiichi Fukui, was conducted in August-September, 2006. During the
field survey, the author was one of the enumerator under the supervision of Professor Seiichi Fukui. Thus, the
opportunity to be in the field allows the author analyze deeply about the agricultural child labour.
11
from Chapter III to Chapter V of this dissertation. The next sub-headings entail about the
To analyze the underlying root causes of child labour in Cambodia, this study uses
the national dataset of Cambodia Socio-economic Survey 1998/99 (CSES-1998/99) for the
empirical analysis. Based on the existing theoretical model of child labour, the study tries
to test the existing pioneering study of Basu and Van (1998) whose assumption is that
existence or non-existence of child labour largely depends on parents’ wage rates (adult’s
wage rates). The model has shown that if the wage rates of parents are above the
prefers to send the child to work, if and only if, in the absence of income from the child,
each individual’s consumption falls below a certain exogenously fixed subsistence level of
consumption. Thus, each individual household would make a supply of labour. Christelle
Dumas (2007) has shown that child labour is subject to (i) when subsistence constraint not
binding, the adults’ income is greater than the subsistence level, but children still work;
and (ii) when subsistence constraint binding, an adult’s income is equal to the subsistence
level, but children still work. Therefore, the primary testing hypothesis in this study is that
when adults’ income is less than the subsistence level, but children are still enrolled in
school, which means that the poor households also send their children to school though in
less proportion compared to non-poor households. Furthermore, this study also tries to
prove that the assumption of Basu and Van (1998) seems plausible; however, it may not be
true if the subsistence level of consumption (poverty line) is set for different levels. The
fact is that if the subsistence level of consumption is set at lower level, this study always
observes that children of the non-poor households are found in the categories of work only,
12
combining work and study, study only, and idle. However, if the subsistence level of
consumption is set higher, this study always observes that children in poor households are
found in the categories of work only, combining work and study, study only, and idle.
To test the above hypothesis, this study employs the Multinomial Logit model to
capture four possible outcomes of the child’s activities (works only, combining work with
study, study only, and idle). In this model, there are sets of exogenous variables that
include (a) child characteristics: i.e., age, age2, sex, primary education, and secondary
above the poverty line, 0 otherwise), region of the resident (1 if urban, 0 otherwise),
parents’ employment status, parents’ education, and gender of the household head; and (c)
community characteristics such as access to clean water and sanitation and distance to
school. By employing the Multinomial Logit function, this model automatically accounts
make sure the estimation is valid and accurate, this study uses parental household’s income
by excluding the children’s income to avoid an endogeneity problem where it could be that
the households are able to sustain a relatively high households’ income per capita because
they send their children to work. This study further constructs the poverty status of each
individual out of parental income and treats it as a threshold level for the minimum need.
Using parental income allows this study to test the parental altruistic model by Basu and
Van (1998) for the luxurious axiom and substitution axiom in which parents’ wages play a
1.2.2. Child’s Wage and Trade-off between Human Capital and Working Hours
education, this study has formulated objectives to achieve the above mentioned hypotheses.
13
Since income derived from child labour will help to maintain their household’s
consumption and maintain them at school, therefore, it is worth exploring as whether the
level of education of the child labourer may help them to earn more or not. To investigate
the determinant of child’s earnings, this study aims to estimate the rate of return to child
labour from their education and experience in the labour force. Thereafter, this study also
tests the above hypotheses as whether hours worked by children have a negative impact on
their human capital formation or not. In order to achieve this objective, this study develops
both theoretical and empirical models to find the contemporary facts of current working
The study starts with the standard economic assumption that individuals are
rational utility-maximizers (see, e.g., Becker, 1965). It assumes that individuals’ parent
allocate their children’ time between working and schooling through their maximization of
household utility. The human capital function of the child is assumed to be a function of
time spent at school and school’s expenditure for which an individual household spent on
school fee, textbooks, and other extra costs of child-schooling during the year. All
decisions are made by altruistic parents, and children are treated as recipients. With the
appropriate derivation and assumptions, the model reaches its solution and propositions.
(1974) noticed that there is less of a relationship between experience and schooling, than
between age and schooling. Experience profiles of logarithm earnings are much more
nearly parallel than age profiles. If this is so, in an earning function in which earnings are
logarithmic, years of work experience shall be entered additively and in arithmetical form.
The experience term is not linear but concave. Furthermore, a child’s earnings and hours
worked are jointly determined. Therefore, the earning function and hours worked by
14
This empirical model on human capital formation and hours worked- schooling
trade-off is derived from the conceptual framework that expresses the negative relationship
between human capital formation and hours worked- schooling trade-off. This study uses
index), Years of Schooling, and Child’s Activities as dependent variables. This study
models the learning measure of the child with a set of explanatory variables such as age,
age^2, education of parents, gender of the child, number of children in the household (age
5-14), number of babies in the household (aged 0-4), poverty status of each household (1 if
above the poverty line), school expenditures, and hours worked by the child and its square.
The inclusion of child labour hours, H i , and its square is designed to capture the trade-off
point between labour hours and the learning measure of the child. According to Orazem
and Gunnarsson (2003), the child labour hours, as well as the poverty status of the child are
necessary to use appropriate instrumental variables in the dataset that have a high
correlation with hours worked by the child and poverty status, but that are not correlated
with the disturbance term of the structural equation. With these notions in mind, we can
1.2.3. Effects of Agricultural Child Labour on Child Health and Its Determinants
To investigate the impact of child labour on child health in the rural agriculture of
Cambodia, this study has formulated objectives in the above mentioned hypothesis as
whether child labour has a negative impact on their health or not. Furthermore, many
literatures have debated the goodness and badness of child labour and its consequences on
practice and their involvement is highly regarded as usefulness for their family economy.
15
Because of the unknown reality of child health effects as the results of their participation
and hours worked, therefore, this study, through the theoretical and empirical models, aims
to: (1) investigate the determinants of child health, and that if the current hours worked of
agricultural child labour have an effect on their health or not; (2) and if the current child
labour participation in agriculture has negative effect on health, this study also aims to
establish average threshold level of hours for agricultural child labour, so as their
involvement in agricultural activity will not interrupt on their growths and health in
general; and (3) provide policy implications based on the emerging evidence regarding
This study has developed a household model to understand factors affecting child
health and its determinants in the context of rural agricultural child labour. The model
assumes that household utility is a function of household consumption, and the health of
children. In this context, health features directly as an argument in the utility function of
the food of children and non-food consumption such as toilet uses, boiled water and other
sanitary facilities, hours worked by children, information received through media (access
to TV, Radio, and Newspapers), inputs to collect information, parents’ characteristics, and
function of inputs to collect information, social capital, and public service. Simply we can
express household utility and the demand function of child health. At any given income
level, households must choose where their resources will be spent on household’s
consumption expenditures and child health. The maximization utility of each household is
subjected to budget constraints in each household. From here, the first order condition
(FOC) with respect hours worked by children is simply seen as decisions made by each
parents as to how much to invest on child health along with all other given constraints in
the households. The results from FOC allow this study to interpret that (i) children’s health
16
is a trade-off with hours worked by children depending on how many hours that the child
has worked. This is interesting that, if children are working, it will contribute directly to
the household consumption, and the equilibrium could be that children health is good as
well as children are observed to be working. However, in some case, children’s health is
bad and children are working. Therefore, the best equilibrium of our interest is Pareto-
efficient which means that children’s health is good as long as they are working in the
threshold level hour; and (ii) we also can interpret that the maximization of children’s
health with respect to received information is simply the trade-off between input efforts to
collect that information through the function of child health and the input to collect that
In the econometric analysis, the study wishes to estimate the relation between
health characteristics and child health outcomes. There are, of course, many measurements
(anthropometrics); however, each measurement has its own drawbacks (Strauss and
Thomas, 1998), and among which, general health status is probably the most widely used
indicator in empirical literature in the United States. The model of this study employs two
main health indicators: general health status and BMI for age percentile. Among our
explanatory variables, we suspect that variables “Total Income” and “Hours Worked by
Children” are endogenous with child health outcomes. Therefore, we need to seek the
appropriate instrumental variables in the dataset that have a high correlation with these
endogenous variables, but they are not correlated with the disturbance term of the
structural equation. Finally, the study also performs the Hausman-test of endogeneity and
the over-identification restriction test to check whether the suspected endogenous variables
and their instrumented variables are valid or not. We are also cautious about Media effects
17
(watching TV, listening to the radio, and reading newspapers) and the father’s and
mother’s education. This is simply the case of omitted variable bias where we may omit a
variable that actually belongs to the population model. In this case, this study doubts that
the effects of mother’s and father’s education on child health may operate through the
access to media such as watching TV, listening to the radio or reading newspapers.
The significance of this study is that (1) it has demonstrated the originality and its
understanding the dilemma of child labour. On the basis of these findings, appropriate
measures could be redefined for the best interest of the child. A critical contribution and its
(1) In literatures, the argument on the causes of child labour has been assumed that
poverty is the only cause of child labour. For example, Basu and Van (1998) have shown
that the existence and non-existence of child labour depends on the parents’ wage rates. If
the parents’ wage is above subsistence level of consumption, there will be no supply of
child labour. On the other hand, parents will send their children to work if the parents’
wages stay below that subsistence level of consumption. These striking results have been
supported by many other studies as in the above mentioned literatures. However, many
studies have argued that poverty does not lead to child labour. For example, the study of
wealth paradox by Bhalotra and Heady (2003) has shown that children of the land-rich
households tend to work more than the children of the land-poor households. This finding
has implied that poverty does not lead to child labour. Although, there are debates on the
causes of child labour, however, those studies have not precisely formulated a theoretical
18
model which can explain the empirical findings of the child labour. This means that, for
example, if the subsistence level of consumption of each household (poverty line) is set at
a higher level, it is possible that there will be a larger number of households under the
poverty line who send their children to school. On the other hand, if the poverty line is set
at a lower level, there will be a large number of household who are non-poor, but send
their children to work. If this is the case, then the model of child labour developed by Basu
and Van (1998) could not predict the outcome of child labour. Furthermore, altruistic
parents will exist more amongst poor-households if the subsistence level of consumption is
set at a higher level. In contrast to the earlier case, altruistic parents will exist more
level. Taking these points into consideration, it seems that the model of child labour and
11
altruistic parents could not predict the outcome of child labour or other activities if
subsistence levels of consumption are set at different level. Therefore, this study attempts
to test these hypotheses and its findings will contribute significantly to the growing
(2) In past few decades, the analysis of the impact of child labour on their human
capital formation (education) has been vaguely defined. One of the reasons to explain that
limitation is the unavailability of data on child labour. In fact, many literatures have
analyzed the impact of child labour and its consequences on education and health, by
utilizing the variable of “participation of children in the labour force” (whether the child is
working or not). Recently, there have been several studies on the trade-off between child
11
The altruistic model by Becker (1976) has been wisely used in connection to household decision-making
towards their household’s members because altruists itself is willing to reduce her/his own consumption in
order to increase the consumption of others. Two assumptions were made in the model of altruism: First, the
own consumption of egoistic persons would exceed that of equally able altruistic persons; secondly, the own
consumption of egoists would be greater if the wealth of egoists and altruists were equal because altruists
give away some of their wealth to be consumed by others. From this model, it is also implicated that, if
parents are altruistic, their children will be allocate in school, or both school and work, rather than working
only.
19
labour and schooling, for instance, Ravallion and Wodon (2000), Patrinos and
Psacharopoulos (1995, 1997), Basu (1999), Jensen and Nielsen (1997). However, their
analyses, as mentioned earlier, were based on whether the child was working or not. The
only exceptions are Rossati and Rossi (2003), and Ray and Lancaster (2005), where hours
worked by children were accounted for. Although these two studies have used hours
worked by children in their analysis, there are discrepancies in their empirical models; for
instance, the lack of consideration of endogenous household’s income and hours worked to
the schooling outcomes of the child. Therefore, in contrast to the aforementioned studies,
this study will be the first attempt to: (i) investigate the rate of return to education from
child labour, which is crucial because the income from the child’s labour provides the
resources necessary for their schooling; (ii) the earlier studies on the trade-off between
child labour and schooling, none of them have formulated a theoretical model to
understand this trade-off. Thus this study is also the first study, which has developed both
theoretical and empirical model. A lot of improvement has been made to the empirical
model because this study take into account endogeneity, and it has added significant
covariates into the model such as “school expenditure”, because the non-inclusion of
omitted variables can cause serious problems on parameter estimates (Wooldridge, 2003);
(iii) the sample covers children aged 5-14 rather than the more limited 12-14 target group
since children are assumed to be actively engaged in domestic work from a young age, and
this age group is in accordance with the definition of child labour in Cambodia; (iv) an
estimation of the Multinomial Logit function is made to take into account the possibility
(3) The literatures of agricultural child labour, especially the effects of child labour
on child health, discusses negative impacts rather than positive ones (Guarcello et.al.,
2004). In the ILO reports in 1998, it was shown that children worldwide are being exposed
20
to hazards in their working environments resulting from the risks of having contact with
toxic pesticides, lifting heavy loads, operating machinery without appropriate training,
exposure to strong sunlight, having a lack of water and sanitation, and of other ill-related
agricultural occupations. Furthermore, agricultural child labour may involve long work
hours from a young age; for example, the bonded child labour, in India, as young as
eleven, often work sixteen or seventeen hours a day (HRW, 2002). In contrast to the above
impact of child work on health in the case of Vietnam’s agricultural child labour. Fentiman
et al (2001) also found no growth differences in rural Ghana between children’s enrolled
and not enrolled by assuming that the non-enrolled children are more likely to be working.
Beside the arguments on the negative impact of agricultural child labour on their health,
several studies have searched for the determinants of child health. Cochrane and O’Hara,
(1982), and Behrman and Deolalikar (1988) found a positive impact of parental education
on child height. Glewwe (1999) found that a mother’s health knowledge alone appears to
be crucial in raising child health in Morocco. Other studies argue that education has no
direct effect on height, but is a proxy for unobserved background variables such as
underlying maternal health (Behrman and Wolfe, 1987). More explicitly, Thomas et al
(1991) found that almost all the impact of maternal education can be explained by
In contrast to the above literatures, this study observes that Cambodia children
regarded as essential for their family economy. If children are withdrawn from agricultural
labour, their household will not be able to afford enough food and other necessities, thus it
ultimately will affect child health. Therefore, this study has emerged to address shortfalls
21
and discrepancies in the above literatures. First, there is no theoretical model so far that can
explain the consequences of child labour on child health. This reason could be that most
literatures presumed that child labour automatically has a negative impact on child health.
Secondly, the earlier studies have yet to consider the hours worked by children. One of the
reasons to explain that limitation is the unavailability of data on agricultural child labour,
especially the hours worked by children. Thirdly, Guarcello et.al (2004) recognizes the
and weight standardized for age and sex (BMI-age); however, the limitation of data has
constrained their analysis to only indicators built on self-reported illness and injuries.
Fourthly, while the above literatures have found parents’ education to have a positive
impact on child health, this study has not found any correlation between the two. The
reason to explain this finding could be that most parents’ education is low in rural
agriculture areas, thus this low level of education may not have a significant impact yet.
However, this study has found that the access to media such as listening to radio, watching
television, and reading newspapers has significant impact on child health. In short, this
study aims to test the existing hypotheses on the determinant of child health. Most
importantly, this study becomes the first study that has tried to address the above
mentioned gaps. The contribution of this research will be significantly fed into the growing
literature of agricultural child labour, and help policy makers to establish rules and
regulations on the working hours standard of child labour, in which agricultural children
will be allowed to work within the threshold level hours for which working beyond this
22
1.4. Organization of the Study
This dissertation is organized into six main chapters. Chapter I introduces the
background of the study, research subject with its brief narrative of objectives and
methodologies, and its organization of the whole dissertation. The way this dissertation is
organised, has an advantages to readers as one can select relevant chapters to read without
Chapter II describes the current situation of child labour, the perspectives from
poverty, education and health. This chapter consists of three main reviews related to the
topic of this study. The reviews include the current situation of poverty and child labour in
Cambodia, the educational system and current literacy, enrolment, and educational
expenses, and the health system, current illness and other health prevention. This chapter,
through the analysis of disaggregated data by province, has recommended for further
research into the chapter III, IV, and V. The snapshot of Chapter II is a brief, but
introduction to the study, the hypotheses, the empirical model and its estimation. The
findings of this chapter reinforce existing theories of child labour. In addition, this study
has found that non-poor households also send their children to school and combined school
and study. The findings from the simple simulation by using different levels of subsistent
levels of consumption (Poverty lines); this study is seen as the first contribution to existing
literature in terms of the counter argument of the assumption that poverty is the only
Chapter IV studies on children’s wage rates, human capital and schooling trade-off.
Like any other paper of analysis, chapter IV provides a building block and cascade from
the above studies. This study analyzes the rate of return to education among child labour in
23
Cambodia. With the notion and hypothesis that child labour increases human capital of the
child, thus it is important to estimate the rate of return to education before addressing
question of how much time children need to work in order to maintain their schooling
status. It further provides an analysis of schooling outcomes and hours worked trade-off.
labour. Its hypotheses came with challenges the ILO assumptions that child labour
damages the health of children if works are exposed to hazardous conditions. However,
child labour in Cambodia remains a traditional practice, where children have less exposure
to pesticides and other dangerous environments. Thus this chapter tries to reveal the
reasoning behind child work in agricultural areas in Cambodia. The findings suggest that
child labour is yet to post a threat on health. Surprisingly, the outcome of child health is
operated through the exposure and uses of media, where parents’ education is implicitly
revealed through access to radio, television, and reading newspapers. In addition, the role
of social capital is vital for child’s health in the context of rural Cambodia. This social
analyses, and proposes policy recommendations, and policy implications towards child
labour in Cambodia. This chapter is brief, but comprehensive summary of results. The end
24
C
Chhaapptteerr IIII
C
CUUR
RRRE
ENNT
T SSIIT
TUUA
ATTIIO
ONNO
OFF C
CHHIIL
LDDL
LAAB
BOOU
URR::
PPeerrssppeeccttiivveess ffrroom
m PPoovveerrttyy,, E
Edduuccaattiioonn aanndd H
Heeaalltthh
The results of Cambodia Child Labour Survey 2001-02 (CCLS, 2001-02) estimated
that about one in every two children was found to be working for the 5-17 year age group.
Working children approximate 84.4 percent are found to be in rural areas; 5.1 percent in
Phnom Penh and about 10.5 percent in other urban area (See, Table 2.1). In real terms,
rural agriculture harbors the largest share of the child labour accounted almost two millions
children working. This figures implied that the degree of development of Cambodia
economy may have an impact on the number and characteristics of child labour, and the
removing child labour is believed to have impact on economy as well. Thus, this
contemporary issue of poverty and child labour has to be well understood. This section will
review briefly the related regulations and some conventions on child labour, and propose
Cambodia signed the UN Convention on the Rights of the Child in September 1992
and confirmed it in July 1993, which has been reflected in Article 48 of the Constitution of
Cambodia and it explicitly states “The State shall: protect the rights of children as
stipulated in the UN Convention on the Rights of the Child, in particular the right to life,
education, protection during wartime, and protection from economic and sexual
25
Cambodia had adopted a new labour code. The new labour code sets the minimum age of
Table 2.1: Current Working Children 5-17 Years Old by Strata and Sex (Extrapolation)
(Column Percent)
Source: CCLS-2001-02
Cambodia that ILO/IPEC is the only organization working directly to eliminate the worst
forms of child labour in Cambodia. However, a significant number of NGOs are currently
working to address the dire health, economic, social and legal situations of children in
12
Further, this article stipulated that children aged between 12 and 15 years of age can be engaged in light
work provided that: i) the work is not hazardous to their health and psychological development; and ii) the
work will not affect their school attendance or their participation in vocational training program approved by
the competent authorities. It is further stated in Article 181 that minors, whatever their sex, below the age of
18 and still under the responsibility of their parents or guardians cannot engage in any type of work contract
without the prior approval of their parents or guardians.
26
Cambodia. Some of the significant organizations are: World Vision Cambodia, CARE
International, World Education, Save the Children Norway, and International Organization
for Migration (IOM), Mith Samlanh/Friends, VCAO, Cambodian League for the
Promotion and Defence of Human Rights (LICADHO), Ptea Teung Dong, and Operation
Enfants de Battambang (OEB). Briefly these organizations work to serve the needs of
children in one way or another. For example, IOM assists in repatriation and re-
integration of child beggars who have been trafficked to Thailand; World Education
children’s rights, receives and investigates children’s rights violations, provides advocacy
and has child protection networks at the grassroots level; Mith Samlanh/Friends has a
transitional home, safe migration program in Phnom Penh and Kampong Cham, support
and outreach for children, and child’s rights programs; CARE International utilizes
educational programs to keep children away from labour; Ptea Teung Dong in Battambang
provides shelter, vocational training and re-integration assistance to homeless families with
children; and OEB assists disabled children with education and life skills training.
mapping technique gives us more easily to find the highest percentage of child labour in
Cambodia disaggregated by provinces. The Map 2.1 has significantly shown that provinces
bordered with Viet Nam and Thailand contained the high incidence of child labour, thus
attention need to be focussed on these regions. These provinces are Kampong Cham,
Kandal, Prey Veng, Svay Rieng, and the Phnom Penh city. The high percentage of child
labour along the bordered provinces also implied that children have many job opportunities
more than children in other provinces, because bordered provinces has trade activity
27
Map 2.1: Proportion of Provincial Child Labour to Total Child Labour in Cambodia
(Percentage)
Bat Dambang
Krong Pailin Kampong Thum Kracheh
Mondol Kiri
Pousat
Kampong Chhnang
Kampong Cham
Svay Rieng
Kampot Takaev
9 to 13
Banteay Meanchey 30 69
Battambang 32 71
Kampong Cham 28 71
28
Provinces Percentage of Population Percentage of working
Below Food Poverty Line children aged 5-14 years old
Kampong Chhnang 39 62
Kampong Speu 12 60
Kampong Thom 12 62
Kampot 27 68
Kandal 29 75
Kep 42 64
Koh Kong 3 61
Kratie 15 49
Mondulkiri 76 44
Otdar Meanchey 81 86
Pailin 32 44
Phnom Penh 3 23
Preah Vihear 98 63
Prey Veng 37 50
Pursat 45 68
Ratanakiri 41 57
Siem Reap 45 75
Sihanouk ville 21 33
Steung Treng 19 52
Svay Rieng 32 66
Takeo 44 61
Source: CCLS-2001/02, and CSES-1998/99.
Bank study (Fallon and Tzannatos, 1998) that the incidence of child labour and children as
a proportion of the total labour force decline with per capita GDP. This basically means
that child labours has negative relationship with income. In another word, child labour has
positive relationship with poverty. However, many studies also seem to reject the
hypothesis that poverty is the only cause of child labour (see, the details in Chapter III).
29
Since data is available to carry-out a test of this hypothesis by using the data in Table 2.2.
The result of the regression in Table 2.3 has shown the positive relationship between
poverty and the incidence of child labour. This basic result seems to support many earlier
literatures on the study of poverty and child labour. However, this kind of study needs to
be further explored and further searched for the determinants of child labour. Therefore,
the details study on this subject is well elaborated and tested in Chapter III of this
dissertation.
poverty line
system that required 13 years of education. Unfortunately from 1975-79, most human
resource and infrastructure were abolished by destructive force of Khmer Rough. Schools
were reopened in early 1979 after the collapse of that killing regime. Because of the urgent
need of the country, the Ministry of Education executed a 10-year education system from
1979-86 and then expanded it to an 11-year education system from 1986 to 1996. The
Ministry has continued to improve the education system through reformed curriculum, new
30
textbooks developed and new teaching skills provided to teachers to prepare ground for the
introduction of a new 12-year education system in the 1996-97 school years. The new
system has increased the number of learning hours for every grade in the cycle of primary
education. In the framework of this new 12-year education system, one school year lasts 38
learning weeks, with 5 learning days per week, 6 periods of learning per day, and each
In the 1997-98 school year, the country has a total of 5,026 primary schools, 350
lower-secondary schools and 125 upper-secondary schools. Children who live far away
from a school find very hard to continue learning and some of them inevitably become
13
Teaching staff is the staff who are in charge of teaching everyday, librarians, operational activity teachers,
art-home economic teachers, lab staff, computer trainers, Primary School principals in school with 6 classes
or less and Vice principals with 10 classes or less; General Secondary School principals in school with 4
classes or less and Vice principals with 7 classes or less.
14
Non-teaching staff is the staff who are working everyday in school office, secretaries, accountants,
dormitory manager, health staff, drivers, cooks, Primary School principals in schools with 7 classes or more
and Vice principals with 11 classes or more; General Secondary School principals in schools with 5 classes
or more and Vice principals with 8 classes or more.
31
Cambodia has tried to ensure the equal opportunity for every child to receive a 9-year basic
education. The progress can be seen from the increasing school infrastructure in each level
of education. In 2004-05, there were 2,123 pre-schools, which included 1,345 public and
778 private pre-schools (see, Table 2.4). Among the public pre-schools, only 129 are
independent while the others are preschool classrooms within primary schools. The total
staff at pre-schools is 2,833 (2,788 female) with a teaching staff of 2,650 (2,624 female).
The Ministry encourages all provinces to promote pre-school activities as they believe,
based on research, that children who have attended pre-school are more likely and able to
continue their studies further than the ones who have not.
As for the primary school infrastructure, the total number of school also has
increased to 6,180. The number of primary school staff is 60,841 (22,934 female) of which
the teaching staffs are 50,140 (20,678 female or 41.24%). By similar policies and strategies
as in primary education, the lower secondary schools also have increased to 578. Table 2.4
also showed the increase of upper secondary schools to 232 in year 2004-05.
Problems are still found in education network, such as poor services of education
delivery and a large gap in education quality between urban schools and rural or remote
schools (MoEYS, 2005). In history, Pagoda has been playing a critical role in keeping
schools functioning well. Presently, community also have active role, through “Parents’
Associations” or “School Supporting Committees”, to raise money to cover not only the
recurrent expenses, but also to sponsor all sport and art activities. The national budget for
education can cover staff salaries (teachers and education administrators) and a small part
The most important change in the education system in recent years has been the
introduction of comprehensive planning under the Education Strategic Plan (ESP) and the
32
reform process undertaken with a sector wide approach (SWAP) to planning. The ESP is a
rolling five year plan first developed by the MoEYS for the period 2001-2005. The ESSP
is an accompanying operational plan which sets out in detail how the government will
work in partnership with donors and NGOs to achieve the aims of the ESP. The Ministry's
program is organized into a number of discrete components which are collectively known
as the Priority Action Program (PAP). The PAP is actually a funding mechanism for these
components, delivering funds for interventions designed to promote equity, quality, and
In the connection with Education for All policy, many reforms have been made
including the provision of operating budgets to schools, the abolition of registration fees,
and special incentives to the poor to attend school (e.g. scholarships). Although there have
been significant achievements, especially a surge in enrolment, the impact of the reform
process has been limited in many areas. It has had little success in resolving chronic
problems such as late and partial disbursement of budgeted funds and the quality of
provincial and local management. Its heavy emphasis on achieving quantitative impacts
and demonstrating efficiencies and cost effectiveness has also been criticized for pushing
attention to issues of quality to one side. In spite of these flaws, the ESP-ESSP initiative
represents a break with the past, a willingness to acknowledge the realities of an under-
resourced education system, and the inequities which accompany it, and an acceptance of
the need for greater accountability. It also represents a real step towards greater self-
School Enrolment: The primary school enrolment increased from 2.4 million in
2000 to a peak enrolment of 2.75 million in 2004 (MoEYS, 2005). This trend not only
reflects normal population growth but actual improvements in access, partly through higher
33
levels of participation in areas where schools are already accessible and partly through the
building of new schools in areas never served before. Ministry figures also show that Net
Enrolment Rate (NER 15 ) has risen from 83.8% to 90.1% since the year 2000. The
expansion in enrolment at primary level has also had significant implications for teacher
supply, pupil class ratio (PCR) and intake requirements at teacher training institutions, all
of which are yet to be fully addressed. In Figure 2.1, it has reported that the enrolment rates
increase until the age of 11 and then start to decline, indicating that although the minimum
age for admission is 6 years, late admissions are common. Beyond the age of 14 years, the
enrolment rates decline rapidly confirming that the drop out rates increase steadily with the
transition from primary to secondary and tertiary level of education and training. Up to the
age of 12, girls and boys have almost the same enrolment rates, then the girls’ enrolment
rates drop and are lower than the boys’. This is the same pattern as in the CSES 1999 even
It is very important to have the disaggregated data on the Net Enrolment Ratio by
province as shown in Table 2.4. The data availability enables this study to carry out a test
the relationship of child labour and child enrolment ratio in primary school and lower
16
secondary school. Furthermore, Table 2.5 has embraced the concept of Basic Education
policy in Cambodia. It is clearly indicated that the province of Rotanak Kiri, Mondol Kiri,
Otdar Mean Chey, Preah Vihear, and Kaoh Kong need immediate intention because the
Net Enrollment Rate in these provinces still stay below 80 percent. It is further that all
female Net Enrollment Ratio in these provinces are less than male counterpart. As for the
Net Enrollment Ratio in lower secondary school, the data has shown that the province of
15
Net enrollment rate: Number of pupils of the official school age enrolled in school expressed as a
percentage of the total population of the same age group
16
Basic education: The complete cycle of nine years of formal and non-formal education, starting with
grade 1 and ending with grade 9.
34
Rotanak Kiri, Otdar Mean Chey, Mondol Kiri, Preah Vihear, Stueng Traeng, Kaoh Kong,
and Krong Pailin stay below 10 percent, which indicated immediate policy intervention.
Furthermore, it is observed that female Enrolment Ratio in lower secondary school in these
100.0
90.0
80.0
70.0
Enrolment Rate
60.0
2004
50.0
1999
40.0
30.0
20.0
10.0
0.0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Table 2.5: Net Enrolment Ratio in Lower Secondary and Primary Education by Province
and Sex
35
Provinces Net enrollment ratio in lower Net enrollment ratio in
secondary education primary education
Female Male Total Female Male Total
Kampong Chhnang 13.9 18.9 16.4 91.8 95 93.4
Kampong Speu 13.3 18.4 15.9 92.1 95.2 93.7
Kampong Thum 14.2 14.7 14.5 87.2 86.9 87
Kampot 28.8 37.9 33.4 87.5 91.8 89.8
Kandal 21.2 25.1 23.2 90.1 92.8 91.5
Kaoh Kong 6.3 9.2 7.8 74.9 82.9 78.9
Kracheh 16 16.6 16.3 86.3 91.3 88.9
Krong Kaeb 18.9 17.2 18 85 94 89.7
Krong Pailin 8.8 7.5 8.1 96.4 85.7 90.6
Krong Preah Sihanouk 13.5 17.1 15.3 85.1 93.3 89.2
Mondol Kiri 5.5 4.3 4.9 62.5 78.8 70.7
Otdar Mean Chey 3.3 4.2 3.8 79 76 77.4
Phnom Penh 34.7 39.8 37.2 82.5 88.7 85.6
Pousat 11.8 14.4 13.1 86.9 97.1 91.9
Preah Vihear 5.7 5.6 5.6 76.5 79 77.8
Prey Veaeng 13.9 22.7 18.1 87.8 89.4 88.6
Rotanak Kiri 3.2 4.2 3.7 46.1 63 54.6
Siem Reab 12.4 15.4 13.9 88.9 92.5 90.8
Stueng Traeng 5.7 6.8 6.2 76.5 90.5 83.3
Svay Rieng 18 24.8 21.6 89 96 92.4
Takaev 25.4 33.6 29.6 90.1 92 91.1
Source: Education Management Information System, MoEYS, 2002-2004
Dropout Rate: It is hard to find the dropout rate data in Cambodia. However, the
improvement of data collection of the Ministry of Education Youth and Sport (MoEYS)
has enable for this study to generate the table 2.6 for details of dropout rates by grade and
province. Based on the trends of dropout rates from grade 1-12, it has indicated that there
is a high rate of dropout rate in the grade 1 and the upper secondary and high school (see
36
Table 2.6). This basically means that if children manage to survive from grade 1, they are
likely that they will at least manage to finish primary school (grade1-6). For the lower
secondary school (grade 7-9) children are likely to continue from the completion of
primary school, because of the increasing availability of school in the communes and
villages. However, the dropout rate increases in the high school because of unavailability
of school in the areas, which require those children to move to urban centres (provinces or
1 2 3 4 5 6 7 8 9 10 11 12
Banteay Mean Chey 14.4 10.9 10 11.8 14.9 17.8 19.8 19.6 23.8 12.1 9 38.9
Kampong Chhnang 9.1 9.1 6.8 8.6 9.1 9.2 15.8 15 22.1 5.5 4.5 23.1
Kampong Speu 20 12.2 9.4 9.5 8.5 9.8 21 19.6 31.5 18.2 7.6 40.4
Kracheh 14.4 12.6 10.4 11.8 30.2 15.9 14.2 8.7 21.7 21.8 16.4 43.1
Krong Kaeb 17.7 18.5 10.7 12.5 13.7 7.8 25.4 20.4 23.5 15.3 6.9 72.9
Krong Pailin 24 22.9 16.6 15.9 10.7 12.7 16 9.2 21.7 14.3 4.5 50
Mondol Kiri 40.5 41 18.4 13.7 17 9 3.9 8.9 10.4 19.4 6.7 50
Otdar Mean Chey 20.7 22.6 10 13.6 14.1 16.4 14.3 10.6 19.8 20.4 23.6
Phnom Penh 11.3 7.7 5.7 6.1 8.4 7.7 13.8 5.9 17.9 5.3 5.1 15.1
Pousat 20.6 11.9 10.1 11.8 11 11.1 18.7 11.1 22.5 10.4 6.9 37.8
Preah Vihear 21.8 18.6 15.1 16.1 19.3 14.8 21.8 10.4 25.8 25.6 14.1 5.8
37
School Attendance 17 and Literacy: According to the latest CSES-2004, some 3.7
million (55 percent) of the population aged 5-24 years, comprising 2 million males and 1.7
million females, were attending the formal school system in 2004 . Of this number 2.8
million or 75 percent were in primary schools (see Table 2.7). The number of persons in
higher education is very low. The share of the population aged 5-24 years is considerably
higher than in the survey from 1999, where 46 percent of the population (2.7 million) was
Table 2.7: Population 5-24 Years Currently Attending School by Level and Sex (Percent)
Level Sex
Male Female Both Sexes
Pre-primary 1.2 1.3 1.3
Primary 73.3 77.0 75.0
Lower Secondary 16.1 14.7 15.5
High school 6.8 5.2 6.0
Technical/Vocational 0.6 0.4 0.5
Under Graduate/ Graduate 1.7 1.0 1.4
Other 0.3 0.4 0.3
Total 100.0 100.0 100.0
Source: CSES-2004
has shown that literacy 18 is 67.1 percent for the population 7 years and over. The general
literacy rate is highest in Phnom Penh and much lower in rural areas. The literacy rate is
higher for males than females in both urban and rural areas (see Table 2.8).
17
School attendance is defined as attendance at a kindergarten, primary, lower or upper secondary school,
technical or professional school, college or university. Even when on holiday the person is considered as
being in the school system.
18
Literacy is the ability to read and write a simple message in any language.
38
Table 2.8: Literacy Rate, Population 7 Years and over by Stratum and Sex (Percent)
Stratum
Source: CSES-2004
19
Educational Expenses: The average educational expenses are estimated to about
130,000 riels per school year; 114,000 riels for female and 144,000 for male (see, Table
2.9). In 1999 the corresponding figure was 66,000 riels. Note that inflation is not taken into
account in these figures. There are large differences by educational level. The highest cost
predominantly found in Phnom Penh the average annual expenses are consequently much
higher there.
Level Riels
Pre-school 47 000
Primary 42 000
Lower Secondary 168 000
High school 393 000
Technical/Vocational 1 141 000
Under Graduate/ Graduate 2 129 000
Total, all levels 130 000
Source: CSES-2004
19
Educational expenses include school fees, tuition, text books, other school supplies, allowances for
children studying away from home, transport cost, gift to teachers, building fund etc.
39
Suggestion on the Impact of Child Labour on Education: It has been long
recognised that child labour is the direct conflict with education of the child. However, a
small increase in child labour does not necessarily result in a trade-off with human capital
investment (Fan, 2004), and increases in schooling does not necessarily translate into a
decline in child labour (Edmonds, 2005), since the positive impact of increased financial
resources for education may outweigh the negative impact of reduced time for study. In
Cambodia, this study has observed that most children like to combine both work and study.
Thus, this could mean that child labour helps their parents financial, and thus they are
maintained in school. With data availability of net enrolment ratio and the proportion of
working children age 5-14, it has enabled this study to carry-out a test of this relationship.
However, the result of the regressions (see Table 2.10 and Table 2.11) confirmed that we
do not have enough evidence to conclude this relationship. Therefore, a details study is
primary education”
40
Table 2.11: Impact of Child Labour on Net Enrolment in Lower Secondary School
secondary education”
2.3. Health System, Child Mortality, Illness, Preventive Measures and Health Care
Expenses
Health System: Similarly to all other public sectors, the public health system has
suffered from war and chronic in meeting the health needs of the population (SEDP-II,
2001-05). After the Khmer Rouge regime in 1979, only 50 medical doctors survived from
the brutal genocide. During those times, Cambodia started from ground zero as a
nationwide from rehabilitation to construction of its basic needs of public services. The
period 1980 to 1989, many health workers were trained through accelerated training
courses of varying quality. The health service delivery system was designated as a publicly
financed, staffed, and managed services based on a socialist model of health service
delivery. The period 1989 to 1995 was a time of strengthening and development with
substantial investment from the government and donor. The period 1995-1998, various
provisions on health service and management were passed into law, which opened the path
to private practice. Throughout these periods of health sector reform in Cambodia, serious
efforts have been made to improve health systems and health care financing, through an
improved allocation of resources to priority areas and efforts to increase formal revenues at
41
the provider level. According to Second-Socio Economic Development Plan-II (SEDP-II,
2001-05), the current pressure is to reach a more adequate, stable and efficient health care
appropriate volume and mix of health services and remove financial barriers to seeking
health care.
Currently, the Health Coverage Plan of the Ministry of Health has been dividing the
country into 73 operational districts (ODs) in Cambodia's 24 provinces (see Map 2.2).
Each Operational District covers a population of 100-200 000 and includes a referral
hospital and 10-20 health centres with catchment areas of approximately 10 000. Map 2.3
showed that health post has been established to cover the areas where health centres are not
available. For example, the areas of north-eastern parts and along the borders of Vietnam
and Thai, there are many health posts to help servicing the population in those remote
areas.
While significant improvement has been achieved in the public health care delivery
system, several factors related to the health workers staffing these facilities constitute a
major hazard in the development of social health insurance. These factors are: very low
salary levels, leading to preference to work in areas with opportunities to increase income
from other sources; imbalances in deployment, with some overstaffing in urban areas and
severe understaffing in remote rural areas; inadequate skills levels, as most health centres
are staffed by primary and secondary nurses and in some cases, midwives, with very low
professionals; for the health workers in health centres, isolation and lack of supervision of
higher level professionals and weak management capacity. These factors are interlinked,
42
and strongly related to the quality of care provided. If the ability to generate additional
income from patients is taken away, motivation of these staff may decrease further.
#Ñ
Oddar Meanchey
Ñ Preah Vihear
#Ñ Ratanak Kiri
#
Ñ
Ñ Banteay Meanchey
#Ñ Ñ Stung Treng
Ñ #
Ñ
Ñ Siemreap
#
Ñ
Ñ Ñ
Battambang
Ñ
#
Pailin Ñ
#
Ñ
Ñ Kampong Thom
#
Ñ
ÑPursat
#
Ñ Kratie Mondul Kiri
Ñ #
Ñ #Ñ
#Ñ Ñ Ñ Ñ
Kampong Chhnang
ÑKampong
Ñ Cham
#
Ñ
Ñ Ñ
Ñ Ñ Ñ Ñ Ñ
Koh Kong Ñ
#
Ñ Phnom Penh Ñ
#
Ñ Prey Veng
Ñ
#Ñ ÑÑ
# #
Ñ
Kampong Speu Kandal Ñ
Ñ Ñ Ñ Ñ
Ñ
ÑÑ Svay Rieng
Ñ Ñ # Ñ Ñ #
Ñ Ñ
Ñ
Takeo
Ñ
Sihanoukville Ñ
#Ñ Kampot Ñ
#
Ñ Ñ
Kep#Ñ
Ñ #
Oddar Meanchey Ñ
Preah Vihear Ñ Ñ
Ñ Ñ
# Ñ Ratanak
Ñ Kiri
# Ñ
Banteay Meanchey Ñ ÑÑ
# Ñ Stung Treng
Ñ # Ñ
Siemreap Ñ
#
Battambang Ñ
Ñ # Ñ
Pailin Ñ
#
Kampong Thom
#
Pursat ÑÑ
Ñ # Kratie Mondul Kiri Ñ
Ñ # #Ñ
Ñ Ñ
Ñ Ñ
Ñ Ñ # ÑÑ
Ñ Kampong Chhnang
Ñ Kampong Cham Ñ
# Ñ Ñ
Ñ
Koh Kong Ñ
# Phnom Penh
Ñ
Ñ #
ÑÑ Prey Veng
Ñ # # #
Ñ
Kampong Speu Kandal
Svay Rieng
#
#
Ñ Takeo
Sihanoukville Kampot
# #
Kep#
43
Child Mortality: Despite the major differences in child mortality estimates between
the Census-1998, CDHS-2000 and NHS-1998 surveys, whereas the CDHS 2000 produces
an under-five mortality rate of 124, and NHS 1998 produces an under-five rate of 115, the
1998 Census produces an under-five rate of 130 per 1,000 live births, all data provide an
high child mortality rate which draw a concerns for policy intervention. The disaggregated
data of child mortality by provinces (see, Table 2.12), it has indicated that child mortality
rate is high in the highland areas such as Kracheh, Preah Vihear, Stueng Traeng, Mondol
Phnom Penh 22 23
Takaev 40 61
Kaoh Kong 41 61
Svay Rieng 42 66
Kandal 43 75
Krong Preah Sihanouk 44 33
Kampot 47 68
Prey Veaeng 48 50
Siem Reab 49 75
Bat Dambang 52 71
Kampong Chhnang 52 62
Krong Pailin 52 44
Kampong Speu 53 60
Kampong Thum 57 62
Krong Kaeb 62 64
Kampong Cham 63 71
Banteay Mean Chey 64 69
44
Provinces Child mortality rate Percentage of working
children age 5-14
Kracheh 68 49
Otdar Mean Chey 70 86
Pousat 75 68
Preah Vihear 88 63
Stueng Traeng 94 52
Mondol Kiri 106 44
Rotanak Kiri 140 57
Illness: In recent report of CSES-2004, men are in somewhat better health than
women, as evaluated by the household heads. The proportion with bad or very bad health
is higher among women than among men, 11 percent compared with 8 percent for men.
For the Cambodian population as a whole, 10 percent or 1.3 million people considered
themselves or their spouse of being in bad or very bad health. More than 10 percent of
children under the age of 5 years are reported to be in bad health, boys more than girls (see,
Figure 2.2). Both men and women seem to have their best period from 10 to 25 years of
age. After age 25 the proportion in bad health is about the same and increasing for women
and men until age 45. After that age the proportion in bad health is clearly higher among
women.
45
Figure 2.2: Household Members in Bad or Very Bad Health by Sex Evaluated by
22
20
18
16
14 Men Women
Percent
12
10
0
Age
0 - 4 5 - 9 10 - 14 15 - 19 20 - 24 25 - 34 35 - 44 45 - 54 55 - 64 65 +
Source: CSES-2004
Preventive Measure: Breast milk is the primary source of nutrients for infants and
also transfers immunities from mother to child. The WHO recommends exclusive
breastfeeding during the first six months of life. According to CSES-2004, about 97
percent of Cambodian children under age 2, have been breastfed during some period.
However, only 30 percent of children were given breast milk as the first thing after birth.
Almost two out of three children are first given water or sugar water. There is a difference
between Phnom Penh and the rest of the country. In Phnom Penh, more than 45 percent are
given breast milk and only about 7 percent are given sugar water.
under age 2 were fully vaccinated against tuberculosis; three doses of DPT vaccine to
prevent diphtheria, pertussis, and tetanus; at least three doses of polio vaccine; and one
46
dose of measles vaccine. For 6 percent of the children their mothers could not show a
health card but the child was said to be fully vaccinated. Vitamin A was given to 83
percent of the children. 3 percent of the Cambodian children under age 2 have suffered
Health Care Expense: A recently study on the health insurance (MoH, 2005)
concluded that paying for health care, particularly secondary or tertiary care is still a major
cause of destitution among the poorest sections of the community. High user fees in
referral hospitals have often resulted in the poor having to sell productive assets (e.g. land,
livestock), or going into debt to pay for treatment, thus increasing their vulnerability. Table
2.13 has shown the source of money spent on health care by province. The data indicated
that more than 10 percent of the population in Banteay Mean Chey, Kampong Cham,
Kandal, Phnom Penh, and Prey Veaeng province, have used the mechanism of borrowing
to finance the health expenses. Besides the borrowing, people have used their assets to
finance the cost of medical treatment. The data showed that Svay Rieng, Prey Veaeng,
Pousat, and Kampong Speu province has the high proportion of more than 10 percent of
47
Provinces Source of money spent on Source of money spent on
health care is borrowed with health care is from sold
interest assets
Kampong Thum 6.4 8.4
Banteay Mean Chey 27.5 8.7
Takaev 8.1 8.7
Svay Rieng 3.9 11.3
Prey Veaeng 15.7 11.8
Pousat 7.6 16.5
Kampong Speu 7 17.9
Source: Cambodia Demographic and Health Survey,-2000
The data from the Cambodia Demographic and Health Survey 2000 (CDHS, 2000)
has shown that household reported spending an average of $20.7 to receive health care
following injury or illness in the past month. For that episode, 94% of the amount was
spent on health services while 6% was spent on transport. For the most part, the household
expenditure is in the form of user fees to public and private providers at the time of use or
the direct purchase of medicines from private pharmacies or drug vendors. In addition to
the costs of health care, the costs of transport and of borrowing money to pay for care
The impact of the excessive burden on the family imposed by health care
expenditures in the public and private sectors is increasingly recognized. Health care
financing has therefore remained a major stumbling block in balanced health system
development to meet multiple health needs, provide reasonable remuneration for health
workers and facilitate efficient planning and operation. According to the same study of
health insurance (MoH, 2005), the government still provides only 9% of the current health
expenditures; development partners provide around 19% while the rest now around 72%
comes through household out-of-pocket payment for care, at the time of use. Given the low
48
implementation rate of the government budget for health care, and the competing needs of
other government agencies dealing with social services, it is unlikely that government
spending on health will increase. More effective collection and use of the user fees are also
Suggestion on the Impact of Child Labour on Health: It has been wisely believed
that child labour posts threat on health. This belief come from the fact that most children
exposed to unaware hazard condition such as exposing long hours worked in the factory, or
other sectors. However, it is observed Cambodian children in rural areas interact with their
environment where agriculture work is very common. Given data availability of child
mortality by province, it allows this study to test as whether working children may have
affect on the child mortality. This relationship is not direct because working children is in
the age group 5-14, while the child mortality rate is the child under 5 years old. However,
the child mortality could be best proxy of child health, as it may require the elder child to
work longer hours to support the family, which could explain the general child mortality.
The postulation of the relationship is shown in Table 2.14. The result of the regression has
indicated that we do not have enough evidence to conclude that working child has negative
relationship with child mortality rate. Therefore, it must be best to study in detail through
the field survey data, which is elaborated well in Chapter V of this dissertation.
49
C
Chhaapptteerr IIIIII
U
UNND
DEER
RLLY
YIIN
NGGR
ROOO
OTTC
CAAU
USSE
ESS O
OFF C
CHHIIL
LDDL
LAAB
BOOU
URR::
E
Emmppiirriiccaall E
Evviiddeennccee aanndd PPoolliiccyy IIm
mpplliiccaattiioonnss ffoorr C
Caam
mbbooddiiaa
3.1. Introduction
Cambodia has signed and ratified ILO Convention No. 138, concerning the
Minimum Age for Admission to Employment. The Convention specifies a minimum age
for employment of not less than 15 years and not below the age of compulsory schooling
(Article 2.3). The Current State Plan of Action has been concentrating on the demand for
child labour by employers. However, this study believed that the push factor of children
into labour force (the reasons why children enter the labour market) is far more important
than the pull factor (the demand for child worker). This research attempts to address the
root causes of child labour as this would ensure that we focus on the systemic and basic
causes that push children into the labour market. This would lead to more permanent and
sustainable solutions. If children do not come to the labour market, the question of their
being employed will not arise. Dealing with the supply side will entail a comprehensive
approach that addresses the root causes of child labour in Cambodia as well as other
developing countries.
In the following chapter IV of this dissertation, one of the research findings is that
there are many children who go to school because they work and combine both work and
education. Thus this implies that removing children from labour force is not the best
solution for them since the state does not have ability to solve the poverty. All children,
and more so children who work, are living thinking, feeling human beings who also show
their altruistic to their parents by providing labour to generate some income for the solution
50
of poverty. Their families, vice versa, love them no less that we do our children and would
In recent decades, the literature on child labour and poverty has been documented.
Basu and Van (1998) have developed an economic model in which poverty and child
labour is potentially crucial components. The model asserts that the existence or non-
existence of child labour depends on the general level of productivity in the economy. If
the economy is very unproductive, child labour exists in equilibrium, while in a very
productive economy, it will not. Two assumptions were identified: first, the “luxury
axiom” stated that children are sent to work only if the household’s income from sources
other than child labour is very low. Secondly, the “substitution axiom” assumes that child
labour is a substitution for adult labour. This model set the foundation for many other
studies, and policy makers now have a greater understanding of why parents send their
children to work. Fallon and Tzannatos (1998) reviewed a variety of studies that indicated
a strong negative relationship between the incidence of child labour and household income,
but the relationship is less marked in more affluent developing country. Ray (2000) finds a
strong negative correlation between household income and child labour, and a positive
relationship between household income and school enrolment in Pakistan, but no such
relationship in the relatively wealthier Peru. Fan (2004) analyses child labour and
between child labour and adult labour. He demonstrates that a rise in child labour
productivity may lead to an increase in both child labour and children's human capital. So,
in contrast to conventional wisdom, this model shows that a small increase in child labour
may not adversely affect children's human capital since the positive impact of increased
financial resources for education may outweigh the negative impact of reduced time for
study.
51
Basically, this study argues that the two axiom assumptions of Basu and Van
(1998) seem plausible, and several studies have supported the validity of the hypotheses.
However, this study has observed that, particularly in the studies of child labour, there is
lack of theoretical studies, which can explain the empirical findings. For example, if the
subsistence level of each household, considered as the poverty line, is set at a higher level,
it is possible that there will be a large number of households under the poverty line, who
send their children to school. On the other hand, if the poverty line is set at a lower level,
there will be a large number of household who are non-poor, but send their children to
work. If this is the case, it turns to be a contradictory view to the theoretical model of child
labour, for example, altruistic parents who are poor and still send their children to school.
Taking these points into consideration, it seems that the model of child labour could not
predict the result of child labour or other activities if the altruistic 20 parents come into play.
In contrast to the aforementioned studies, this study finds two striking results (i)
many poor household send their children to school or combine school and work. This has
indicated that altruistic parents, though poor, still care for the welfare of their children in
terms of educational investment; (ii) the non-poor households also send their children to
work and combine work with school. This study is unlike the Basus and Van (1998) whose
results of theoretical model implied that poor households will not afford the schooling of
their children and thus children will be located only at labour force, and while the rich will
afford the luxurious good, meaning that children be located at school only. This study
contributes to the new rethinking on “causes of child labour” and the results make this
20
The altruistic model by Becker (1976) has been wisely used in connection to household decision-making
towards their household’s members because altruists itself is willing to reduce her/his own consumption in
order to increase the consumption of others. Two assumptions were made in the model of altruism: First, the
own consumption of egoistic persons would exceed that of equally able altruistic persons; secondly, the own
consumption of egoists would be greater if the wealth of egoists and altruists were equal because altruists
give away some of their wealth to be consumed by others. From this model, it is also implicated that, if
parents are altruistic, their children will be allocate in school, or both school and work, rather than working
only.
52
study different from the earlier by providing alternative outcome of child labour.
Furthermore, this study finds that fathers’ education have significant role in human capital
accumulation of their children, for which they are in line with the previous work by Ray
(2000), Deb and Rosati (2004), Blunch et.al (2002), Bhalotra and Heady, C (2003), and
Khanam (2003); But what make this finding different from other works is father’s
education play a major role in the determinant of child labour, where mothers’ education
reduce child labour and increase children’s enrollment at school. Female children tend to
be child labour rather than schooling. Households possessed of livestock are proved
Finally, this study contributes to the contemporary issue of child labour on two
important features. First, this study feed into the growing literature of child labour and
poverty through the finding that the parents of the poor households have high altruistic for
their children by sending their children to school. At the same time, parents of the non-
poor households possess less altruistic by sending their children to work. However, the
status of poor or non-poor is largely depending on the poverty line, in which this study has
used the simple simulation by allowing the changes of poverty line to observe the outcome
of child activity. Secondly, this study draws some policy implications toward poverty
reduction through the understanding of root causes of child labour for Cambodia.
The organisation of this study is as follows. Section 3.2 is the review on existing
literature and alternative hypotheses describing about the determinants of child labour,
which provides a critical review including the poverty and altruism, wealth paradox,
and other contributing factors on determinants of child labour. Based on this review on the
hypotheses, the study conducts empirical test to find out the root causes of child labour in
53
Cambodia. Section 3.3 presents the empirical framework. It consists of empirical model,
data used, and the results of empirical estimate. The study proposes a method to compute
multiple outcomes by using Multinomial Logit function to estimates the results of parents’
decision on child activity, in which it composes of four categories (work only, both work
and study, school only, and idle whose activity neither attends school nor work). By using
large data of socio-economic survey of Cambodia 1999, this makes the random sample
robustness and the sample population falls under asymptotically normal distribution.
Section 3.4 is the conclusion of the study. It provides the new rethinking on the causes of
child labour, and it challenges to existing literatures on the strong assumption of the two
As mentioned in the above introduction and previous chapters, this study attempts
to understand the root causes of child labour by using empirical evidence from dataset of
Cambodia Socioeconomic Survey 1999, and test against some critical evidences. Thus the
following hypotheses has been proposed to explain about poverty and altruism, wealth
There are several previous studies focusing almost exclusively on poverty as the
main determinant of child labor, for example, Basu and Van (1998), Basu and Tzannatos
(2003), Lee and Westaby (1997), Saupe and Bentley (1994), Kim and Zepeda (2004),
Chakraborty and Das (2004), Grootaert and Ravi (1995), Chao and Alper (1998), Duryea
and Arends (2001), Basu, Arnab K., and Nancy H. Chau (2003), and Blunch and Dort
54
(2000). Some of those studies are purely empirical, theoretical or both and they tend to
reinforce the positive link between poverty and child labour. Among these studies, Basu
and Van (1998) developed a model of child labour that has been subsequently used in a
wide range of analyses, in which the economy exhibits multiple equilibrium of demand and
supply of child labour. The model provides that existence or non-existence of the child
labour depends on the general level of productivity of the economy. If economy is very
unproductive, child labour exists in equilibrium, while very productive, it does not. The
main finding of the work of Basu and Van (1998) is that: if the wage rate of parents is high
above subsistent level, all children will be sent to school and at the same time, all families
prefer to remain small; and on the other hand, if the wage rate of parent is lower than
subsistent level, all children will be sent to work, and each family decides to have many
children.
parents and “capital market imperfections” are used to conclude the rationale decision of
parents about the trade-off between child labour and the accumulation of human capital,
even if parents are altruistic and child labor is socially inefficient because parents fail to
internalize its negative effects of child labour when bequests are zero or when capital
However, small increase in child labour may not be trade-off with human capital
investment (Fan, 2004) since the positive impact of increased financial resources on
education may outweigh the negative impact of reduced time of study. This is simply that
children’s labor market participation raises the financial resources and spent on their
education.
The above altruism assumptions seem plausible and several studies have support
the validity of the hypotheses. However, we also observe, particularly in the studies of
55
child labour that we are lack of theoretical studies which can explain the empirical fact
findings, for example, if subsistence level of each household considered as poverty line,
has been set at higher, it is possible that there will be a large number of households under
poverty line who send their children to school. On the other hand, we also observe that if
the poverty line is set at the lower level, there will be a large number of household who are
non-poor send their children to work. If this is the case, it turns to be a contradictory view
to the theoretical model of child labour, for example, altruistic parents who are poor and
still send their children to school. Taking these points into consideration, it seems that the
model of child labour could not predict the result of child labour or other activities if the
altruistic 21 parents come into play. Such that a behavior of altruism and the investment in
human capital hypothesis are plausible, however, they have been debated by Raut and Tran
(2005) that parents and their children are mutually altruistic. They have considered two
children’s education as a loan with the terms of repayment set by parents. In this
framework they have also considered the loan contracts both with and without two-sided
altruism and carried out specification tests to choose between the two. The specification
tests rejected the no-altruism case in favor of the two-sided altruism case. In the second
model, parents decide how much they want to invest in children, and children decide how
much to pay back. In this model, the two-way transfers are determined by reciprocity with
21
The altruistic model by Becker (1976) has been wisely used in connection to household decision-making
towards their household’s members because altruists itself is willing to reduce her/his own consumption in
order to increase the consumption of others. Two assumptions were made in the model of altruism: First, the
own consumption of egoistic persons would exceed that of equally able altruistic persons; secondly, the own
consumption of egoists would be greater if the wealth of egoists and altruists were equal because altruists
give away some of their wealth to be consumed by others. From this model, it is also implicated that, if
parents are altruistic, their children will be allocate in school, or both school and work, rather than working
only.
56
3.2.2. Wealth Paradox
Since land is the most important store of wealth in agrarian societies and a
substantial fraction of households do not own land (Bhalotra and Heady, 2003), this
challenges the commonly held presumption that child labour emerges from the poorest
households, for example, US Department of Labour (2000); and Basu and Van (1998).
There are, of course, growing empirical studies that have failed to find the positive
relationship between poverty and child labour, for example, Nielsen (1998), Canagarajah
The study by Bhalotra and Heady (2003) on wealth paradox using data from
Pakistan and Ghana found that greater poverty does not lead to greater child labour. They
brought an interesting analysis of the “wealth paradox” simply the effects of farm size on
child labour. Their hypothesis was based on their remarkable observation that children of
the land-rich households are often found in work than the children of land-poor
households. The phenomenon is termed here as the wealth paradox. The basic theoretical
model is derived from a model of peasant household in an economy with imperfect market
for labour, land, and credit, which allowing two periods to capture the impact of child work
in period 1 on productivity in period 2. This model simply means that the ownership of
productive assets, especially land can affect child labour in various ways: (i) the effect
arrives through both the gain in work experience and the possible lowering educational
attainment because this can have negative wealth effects, with large land holdings
generating higher income, making it easier for households to forgo the income that child
work brings, and (ii) capital market imperfection that results in lower interest rates for
households that can offer land as collateral, reinforce the wealth effect allowing large
landowners to borrow more to meet insurance needs or finance their children’s education.
57
Another study seems to convincingly raise doubts on the poverty and child labour
is Fallon and Tzannatos (1998). Their study notes in the World Bank issues paper that
there is negative association between income and the incidence of child labour for low-
income country ($500 US or less). But this association becomes less marked in the more
affluent developing countries (in the $1,000 to $4,000 income range). It is not clear
whether the apparent lack of association at higher incomes is due to statistical or other
reasons, although the relation between child labour incidence and per capita income across
countries is likely to be affected by cultural differences. This view is now increasingly held
Since the data of Cambodia Socio-economic Survey 1998/99 did not have enough
information for us to test the hypothesis of wealth paradox of child farm, we employ
another field survey dataset in 2006 conducted by author together with a group of students
directed by Professor Seiichi Fukui, under the financial support from the project of New
Studies. The result of the empirical testing is attached in Chapter Appendix III.
There are several possibility of the outcomes of this testing. According to Bhalotra
and Heady (2003), they consider imposing the assumption of perfect land markets as well.
The positive incentive effect of land disappears even if the labour market is imperfect,
because land for which hired labour cannot be found can be rented out. The coefficient on
land is zero if the credit market is perfect and negative if the credit market is imperfect.
The case of perfect labour markets is analogous to that of perfect land markets. The
positive incentive effect of land disappears because labour can be hired out. This holds
whether or not land markets are perfect. The coefficient on land is therefore zero or
negative, depending on whether the credit market is perfect or not. Land has a positive
incentive effect if and only if both land and labour markets are imperfect. In this case the
58
land coefficient is zero if either the land market or the labour market is perfect, and it is
positive if both of these markets are imperfect. The study also suggests that the relative
size of credit versus land and labour market imperfections can be discerned by observing
the estimated coefficient on land. Consider the three possibilities (1) If the estimated
coefficient on land size is zero, either all markets are perfect or the credit market is perfect
and either the land or labour markets is perfect. Alternatively, it is possible that all three
markets are imperfect and the positive and negative land effects offset one another; (2) If
the estimated coefficient on land is positive, either all three markets are imperfect or the
credit market is perfect but both land and labour markets are imperfect. In this case both
the land and labour markets can be inferred to be imperfect; and (3) If the coefficient on
land is negative, either all three markets are imperfect or credit markets are imperfect and
either the land or the labour market is perfect. In this case the credit market can be inferred
to be imperfect.
labour, for example, Bhalotra and Heady (2003) note that children from larger land-owning
households work more than children from smaller land-owning homes, implying that child
labour does not decrease with wealth. Similarly, Edmonds and Turk (2004) find that
Vietnamese households that own their own businesses tend to make their children work
more. Khanam (2003), beside the indication of a positive education of parents and the
probability of child schooling, also notes that parents’ occupation is very important in
occupation, for example, day labour or wage labour, it raises the probability that child will
59
Similarly, Parikh and Sadoulet (2005), in their studies on the effect of parents’
occupation on child labour and school attendance in Brazil, investigates how child labour
provided by children’s own parents. They found that after controlling for household,
parental, regional, and child characteristics, children whose parents are self-employed or
employers are more likely to work than children of employees, irrespective of the sector of
parent activity. Furthermore, the paper also confirms a recent finding that children from
areas with high average adult employment rates are more likely to work than children from
areas with low average adult employment rates. Finally, since twice as many children of
the self-employed and employers both work and go to school as those of employees, the
paper suggests that child labour does not necessarily represent a trade off with schooling as
Parents’ characteristics, especially the education of mother and father have shown
the impact on human capital accumulation of the child, for instance, Ray (2000), Deb and
Rosati (2004), Blunch et.al (2002), Bhalotra and Heady (2003), and Khanam (2003). Those
studies confirmed that there exists positive link between parent’s education and the
likelihood of a child attending school, and similarly a negative link between parent’s
education and the likelihood of a child working. Among these studies, some tend to
emphasize the crucial role of mothers’ education on the child schooling, i.e., Deb and
Rosati (2004), and Ray (2000). However, the evidence by Kim and Zepeda (2004) found
rather contradiction that the higher the parents’ education level, the higher the probability
children will work but the fewer hours they will work.
60
3.2.5. Gender
In most empirical work, gender of the child labor (male and/or female child
environment. Bhalotra and Heady, C (2003), besides the wealth effect, also found that
child age in Pakistan has positive effect on hours worked, which is much larger for boy
than for girls, and children of the household head are more likely than other children in the
in Pakistan work significantly more and the effect is bigger for boys and girls. Khanam
(2003), in the study of child labour and school attendance by using Bangladeshee data,
indicated a positive gender coefficient that girls are more likely than boys to combine
In Ghana case study, Blunch and Dort (2000) besides confirming result of positive
link between poverty and child labour, also indicates the evidence of a gender gap in child
labour linked to poverty, since girls as a group as well as across urban, rural and poverty
sub-sample consistently are found to be more likely to engage in harmful child labour than
boys, and there exist structural differences in the process underlying harmful child labor in
Ghana across gender, across rural and urban location as well as across poverty quintiles of
household
Child labour and number of children are endogenous in the household decision-
making model (Becker 1960; Deb and Rosati, 2004). However, this problem exists in the
empirical work as obviously seen in previous studies on the model of children’s activities,
for example, Ray and Lancaster (2004), Ray (2000), Blunch et.al (2002), Khanam (2003)
61
Numbers of children in each household naturally have direct impact on food
consumption and human capital accumulation of the child, for example, the trade-off
between child quantity and quality within a family. The models date back to Becker
(1960), who was motivated to explain the empirical regularity that families with higher
income have fewer children. He theorized that, as income rises, individuals may choose to
increase the average quality and reduce quantity. A key element of the model is an
interaction between quantity and quality in the budget constraint that leads to rising
marginal costs of quality with respect to family size; this generates a trade-off between
quality and quantity. With regard to the household consumption, the recent empirical study
by Nagaraj (2002) found that the number of children aged 5-14 attending school rise
monotonically as monthly household expenditure rise from less than 120 rupees to 455-560
Basu and Van (1998) also implicated the interaction between fertility decisions and
labour market outcomes. If the economy is characterized by small families, labour is scare,
adult wage is high, families can afford to keep children out of the labour force, and all
families prefer to remain small. Another possible outcome is that adults wage are low,
families are so poor that all children must work, and each family decides to have many
children. Another study that helps explain the relationship between child labour and
fertility is Rosenzweig and Evenson (1977). Their study is on the child farm labour, which
explicitly takes into account the economic contribution of children in agricultural areas in
rural population of India. The findings supported the hypothesis that one of the basic
conditions motivating Indian families to bear relatively large numbers of children in the
late 1950’s was the high return to the use of raw labour power of children compared to
62
3.2.7. Social Infrastructure
increasing the decision of children’s schooling, for instance Chao and Alper (1998)
analyze the access to basic education in Ghana for children between 10-14 years of age.
They found that school participation rate is closely related to distance to primary school,
pupil-teacher ratio at the primary level, access to drinking water and roads. Deb and Rosati
(2004) also notes about the importance of social infrastructure, especially the increasing
primary and secondary school in rural areas does increase school attendance and reduces
the probability that a child works or is idle. Ray (2000) also points out in the result of his
study of empirical evidence for Peru and Pakistan that the provision of good schools can
do a lot to reduce child labour in South Asia and to break the strong link between poverty
Note that the below alternative hypotheses lack of theoretical and empirical
findings. However, this study will also test these hypotheses empirically. The regional
differences between rural or urban areas will naturally impact on the activities of the child.
For instance, it is believed that children living in urban areas are easily being located at
especially parents in urban areas are more aware of the value of education since they are
exposed to urban media and advertisement of school. There is relationship between social
infrastructure and the probability of schooling of the child. In the context of Cambodia,
there are beliefs that Chinese people tend to train their children for businesses, while other
ethnic group like to send their children to public school. Therefore, this study will also test
this hypothesis as whether this pattern of thinking still exits in reality. Female household
63
head has high degree of altruistic, and thus will tend to locate children at school. Female
labour force has bargaining power in the household, and thus leads to play significant role
in the child’s welfare such as their education. Child of the household head tends to be in
school, while child not belong to household heads tend to work. Age of the child has strong
correlation to child labour. The older age of the child tend to work than the younger age.
The absence of policy of “education for all” in 1999 made many children to stop school
and cannot continue to junior high school. The fact is that during those time, most areas of
rural part have bad infrastructure and junior high school exist only in urban centre of
districts or provinces only, thus it because an big obstacle for many children who want to
To test hypotheses in this study which are related to existing theoretical structures
above, the author, instead of using Bivariate Probit to capture just two possible outcomes
of the children’s activities, applies Multinomial Logit that allows us to use for four
possible outcomes to reflect full picture of parents’ decision towards their children’s
activities. We model the child’s activity, which is categorized into four categories denoting
by:
22
j=0 if the child works only,
22
In order to capture the entire population of child worker, any duration of work in any economic activity,
during the reference period made children eligible for enumeration as child worker. Furthermore, Work is
defined as an economic activity that a child performs for pay, profit or family gain. It includes paid
employment; operating a farm or business; working for a household economic activity (like food processing
or raising of livestock) without pay; working as an apprentice in order to learn a skill or a craft; without
necessarily receiving wages; and production of paddy or vegetables, say, solely for home consumption. Also,
include is the holding of a job, even if the person is temporarily absent because of vacation, strike or illness.
Production of fixed assets for the use of the own household, such as building or repairing the house is also
consider as work.
64
j=2 if the child attends school 23 only, and
considering “j=2 if the child attends school only” as the base/reference ( j-1) category, so
we can express the probabilities and the likelihood function of for the Multinomial Logit
model as follows:
exp( β j xi + zi )
P( yij = j xi ) = Pij = for j≠2 (3.1)
1 + ∑ j =0 exp( β j xi + zi )
j −1
1
P ( y 2 = 2 xi ) = Pi 2 = (3.2)
1 + ∑ j =0 exp( β j xi + z i )
j −1
n j −1
log L = ∑∑ yij log Pij (3.3)
i =1 j =0
falls in the jth category, and y ij =0 otherwise). β j is the covariate effects of response
exogenous variables that include (a) child characteristics: i.e., age, age2, sex, primary
community characteristics such as access to clean water and sanitation, distance to school,
and the regional dummy such as Phnom Penh, other urban and rural areas. Although there
are more community characteristics that may affect the utility of the parents such as
23
The term schooling includes attendance at a kindergarten, primary, lower or upper secondary school,
technical or professional school, college or university.
24
Female household head is the female adult member of the household who is accepted and recognized by
the other household members as head.
65
pagoda, water irrigation, road, access to market and others, but we may not include these
conditioning on x variables (Wooldridge, 2003). Besides that, the multi-co linearity has
been checked by the correlation matrix of covariates and its results confirmed that there
was no any independent variable that might cause suffers to our model. However, we know
that the problem of endogeneity is the most important to our model if we are not carefully
to select and understand the nature of these covariates. One of the possible endogeneity is
the household income that one may imply that the wealthier households could be expected
to induce the households not to send their children to work. However, the problem is that
the households are able to sustain a relatively high households’ income per capita because
they send their children to work. In fact, this is the case that leads to biased estimates and
also on the estimates of the other parameters of the model. Knowing this potential problem,
I use parental household’s income by excluding the children’s income to avoid such
endogeneity with children’s activities. I also further construct the poverty status of each
individual out of parental income and treat it as a threshold level for the minimum need.
By using parental income, it allows this study to test the parental altruistic model by Basu
and Van (1998) for the luxurious axiom and substitution axiom in which parents’ wage
income may have been used, there is one possible solution to avoid endogeneity, that is
instrumental variable, but it is rather hard to come up with instrumental variable that is not
Based on previous studies on the model of children’s activities such as Ray and
Lancaster (2004), Ray (2000), Blunch, et. al (2002), Khanam (2003) treated number of
children as exogeneity and they neither discussed on this possible endogeneity. This is
66
perhaps (i) due to the lack of available data for appropriate proxy or instrumental variable
of this covariate, (ii) there is no econometric program that can command the simultaneous
standard error, however, we can solve it by two stage process, but it may cause the
downward biases due to larger standard error occurs, and (iii) it may not necessarily to
have exact partial effect since Multinomial Logit results place emphasis on the direction of
its magnitude rather than the coefficients, therefore, knowing the direction is sufficient for
are determined by parental decision. However, the proximate determinants of fertility are
the use of contraceptive, the timing of breast-feeding, the frequency of intercourse, the
infant mortality rate, low age of marriage, household economy, and other environmental
influence such as neighboring and cultural context (Das Gupta, 1987; Pranab Bardhan and
Christopher Udry, 1999). Given complexity and difficulties to find appropriate instrument
variable. Thus, I presume, they tend to relax this assumption. It is the same way that
Christopher Heady (2000), in his study on Ghanaian data, recognized the endogeneity of
child labour’s hours on the child’s educational outcome, but does not tackle it in the
estimation.
The model regards the covariate of poverty status as important to understanding the
major part of the cause of child labour, and many studies did support this hypotheses
including the recent study by Nagaraj (2002) who found that the number of children aged
5-14 attending school rises monotonically as monthly household expenditure rise from less
than 120 rupees to 455-560 rupees in Tamil Nadu and Karnataka, India. Other covariates
also help to explain the phenomena. Age of the child is included to capture parental
behavior of allocating their children’s activities. It was motivated by the premier believe
that age of the child tends to explain its activity. Another important covariate is the
67
parental occupational status. The study by Bhalotra and Heady, C (2003) using data from
Pakistan and Ghana, i.e., found that the larger amounts of land tend to make their children
work more because of labour market imperfection. Edmonds and Turk (2004) found
something similar in Viet Name. The self-employed household or private businesses are
more likely to send their children to work. Thus parental occupations do play parts in
To be broader, a simple economic poverty need not be the sole reason for child
labour. Of course, we have cultural and social factors to look at. Therefore, this model
include community characteristics to capture those dimension, even though they are not
enough to represent all, but rather help us to understand more of the dimension and the
interaction of children’s activities. More importantly, the sex of the child is also
contributing factor to the parents’ decision beside the consumption alone. As mentioned,
cultural practice may vary from region to regions based on sex, religion, and security. For
Cambodia, perhaps, security tends to make mind set of parents to keep female children at
The sample of household with children’s activities is drawn from the Cambodia
to supplement the data base generated through CSES-1996/97, fill critical data gaps in a
number of topics, and meet the data needs for analyzing and monitoring poverty, and
support the anti-poverty program. The total sample size of this survey is 6,000 households,
which were divided into two equal rounds of 3,000 households in each round. This is in
order to capture seasonal change in Cambodia context. The first round was conducted in
68
March 1999 and the second round was in August 1999. More importantly, the CSESE-
1998/99 included more features of employment status, child labour, per capita expenditure
of the households, health and education expenditure. The sampling design of CSES-
1998/99 involved the stratification of the country into five domains based on ecological
zones into which the country is divided and also treating Phnom Penh as a separate
domain, and the cross cutting rural and urban sectors being treated as separate strata. Thus
ten strata were created (0-Phnom Penh, 1-Plain, 2-Tonle Sap, 3-Coastal and 4-Plateau and
mountain). Each domain was classified into rural and urban. And Phnom Penh was treated
as a separate zone.
Based on the working definition of the child labour 25, which covered the age group
5-14 years, it allows us to extract from the micro-data in the employment module and the
economic activity status. This survey population had excluded children engaged as workers
and living in boarding house, we have, in this study, include this characteristics into the
either work nor study. In deed, we consider the children whose activities fall under this
category termed as the “subsistence work” or the work at the “household chores” rather
Therefore, the sample in this study consider household with children’s activities
aged 5-14 years old to be included because child labour and schooling and other categories
are all relevant in that group. We disaggregate “sector” into three areas as “Phnom Penh”,
“Other Urban” and “Rural”. The reason to treat Phnom Penh as separated sector is due to
the special nature of fast development from the rest of other urban in the country. From the
total sample, we draw a sample of 10,092 individuals that each child spent their time in one
25
In order to capture the entire population of child workers, any duration of work in any economic activity,
during the reference period, made children eligible for enumeration as child workers. This extended
definition used in child labour module of the CSES 1999, as information on labour force status, covering
aged 5-14 years who living in the household.
69
or both the following activities: work only termed as child labour, both work and studies,
studies only, neither work nor studies termed as idle. The model of Multinomial Logit is
used to capture the nature and phenomena of children’s activities. The explanatory
variables consist of a set of children characteristics, for instance, the child’ age, child’s
education, child of the household head and child is male or female; the household’s
characteristics included the age of parents (mother and father separately), parent’s
household head, female labour force, and status of poverty in each household. The poverty
line to identify the household whose living standard below or above the threshold are
drawn separately based on the poverty line of Phnom Penh (2470 Riel equivalence of 0.64
US dollar), other urban (2093 Riel equivalence of 0.55 US dollar) and the rural (1777 Riel
equivalence of 0.46 US dollar). For those who income and consumption fall below the
poverty are considered as poor coded as “zero”, and the opposite is the non-poor which is
coded “one”. We use the dummy variable to represent the poverty status, instead of the
continuous measure of income because it is well known that the measure of income in
developing countries have significant measurement error at the lower end of income
distribution. Furthermore, we did avoid from the endogenous income by subtracting the
household total income from the child’s income. This will reflect the purpose of our
economic concept where parent’s income or consumption plays the main role of
determinant of children’s activities. We also assume the exogeneity of the fertility that
reflected in the number of children in each household. The level of education of parents is
advanced level” because we want to reflect the origin of the question and the grouping
became difficult for the advanced level. The summary statistics for the variable used in the
70
Children’s activities are obtained from the “Module of Income and Employment in
page 6 of 23” in question No.3, column 4, 6 and 9. Children were grouped into 4 groups
with a sufficient condition to extract these groups. By the origin of the question No.3,
column 4, children were grouped into three groups only, but by combining the question
No.3, column 6 and 9, we are able to draw a full picture of children’s activities.
As needed to test additional hypothesis of wealth paradox of the child farm, we use
field survey dataset in 2006 (see chapter V) to complement in this study. The empirical
result of this testing is given in Chapter Appendix III, Table Annex 3.1.
Table 3.2 presents descriptive statistics of variables used and its related hypothesis
in this study. Table 3.3 presents the results of Multinomial Logit of the Maximum
Likelihood estimate. Table 3.4 presents simple simulation by using the cross-tabulation,
allowing the changes of poverty lines to observe the outcome of child activity. The
regressors used in the child activity model include the following variables: age of child and
its squared, child’s education, child of the household head, female child, number of
children in the household, mother’s education, father’s education, poverty status, types of
employer of household head (government, state enterprise, joint venture, private, foreign
government/ IOs, NGO, self-employed), female labour force, female household head,
clean water, electricity, distance to primary school, distance to secondary school, distance
to high school, ethnicity (Cham, Chinese, Vietnamese, local group), other urban, and rural
areas. The structure of the model allows the estimations of marginal effect which are
combined school and work at the younger age, and remains in the combined activity at
71
later age of childhood. The variable of “child’s education” has showed the positive impact
for the category of “child work only”, “both work and study”, and “child idle”. These
results have implied that children had studied up to one level and shift their activities to
either work or combined work and study as well as doing neither (idle) rather than
continuation of their study. This is very strange but it was true that those children tend to
stop schooling and be located in other activities because of the absence of policy of
“education for all” in year 1999 which had made children to stop school when they had
reached junior high school. Furthermore, many parts of Cambodia, especially in rural
areas, junior high school were not available, thus need those children to move to urban
centers, which could be the reason to block their continuation of higher study. The
presence numbers of children in the household tend to increase the probability of working.
The coefficient of poverty status “dummy variable_ code 1 if above poverty line, 0
otherwise” indicates that children of the poor households tend to be child labour, combined
work and schooling, or idle. Children of non-poor households tend to be found in school
meaning that there exist positive association between poverty and child labour and
negative association between poverty and human capital accumulation of the child.
Therefore, we confirm this validity. However, how strong of this link is subject to more
debate because we have found the results in Table 3.4 that majority of the poor also tend to
send their children to school and combined work and study. In contrast, this study also
found that many non-poor have sent their children to work and combined work and study.
Hence, the model of child labour developed by Basu and Van (1998) which used income of
adult as a threshold to predict the existence or non-existence of child labour may not be
truly applicable when the “altruistic parents” come into play. And of course, several
surveys especially the survey of Raut and Tran (2005), in the case of Indonesian families,
found that parent and their children are mutually altruistic. The parents’ occupation tends
72
to determine the child’s activities. Employment status as government employee, non-
probability of child’s schooling, and negative effective on child labour. And conversely is
the employment status as private enterprise, private, joint venture, and self-employed.
Father’ education has positive impact on human capital accumulation of the child.
This study also finds that there is preferences bias of parents toward gender of the
child in terms of schooling, child labour, and other activities. The coefficient of the female
child (dummy variable_ 1 if female child, and 0 otherwise) indicates that female child is
found to specialize in working, both work and study. This phenomenon clearly indicates
the preferences bias of parents towards gender of the child. Moreover, this phenomena is
observed across not only gender structure, but also across rural vis a vis urban and poor vis
a vis non-poor. The higher demand for female child workers are closely linked to the
feminization works in the context of Cambodia, for instance, the work related to traditional
textile, garment manufacturing, farming, and retail trade and restaurants. These results
show the female children play important economic actors in order to improve the economic
status of their family. Lastly, social infrastructure is very important factor to increase the
probability of child’s schooling, and other variable such as ethnicity seems not be the
many children as high as 27 percent of all children are classified in the category of “child
idle”. This result takes to mean that many children are neither engage in economic-activity
nor study at school. This high percentage of child idle raise issues as who are these
children and why they are idle? The fact is explained in the national report of Cambodia
Socio-economic Survey 1999 that “the inclusion of children aged 5 years where as the
minimum age for admission to school is 6 years”. This is why it accounts the high
73
percentage of child idle. It should also be noted that this study have inclusively include the
child of doing household-chores in the idle category as well. The separation of this
category will make the data more complicated as we had already have four category of
children’s activity such as “child work only, child study only, child combined both work
and study, and child idle”. To have precise understanding on this issue, Table 3.1 gives
more details of the child activity broken down by age category from 5-17 years old. We
found that many children at the age 5 - 6 years old are out of school, which accounts into
Table 3.1: Frequency, Row, and Column Percentage of the Child’s Activities by Age
Age of the Child Child works Child studies Child idle Total
child works only and studies only
11 15 48 622 88 773
1.94 6.21 80.47 11.38 100.00
1.62 6.32 8.79 2.99 6.60
74
Age of the Child Child works Child studies Child idle Total
child works only and studies only
13 64 93 676 71 904
7.08 10.29 74.78 7.85 100.00
6.90 12.25 9.55 2.41 7.72
14 90 99 648 86 923
9.75 10.73 70.21 9.32 100.00
9.70 13.04 9.16 2.92 7.88
Wealth paradox of the child farm labour, inter-alias is proven to have significantly
true that households with possession of farm land tend to employ their own children on the
farm. However, the size of farm itself is not statistically significant, meaning that most
farmers owned relatively equal size of land in the case of Cambodia (see Chapter
Table 3.4 has provided details on the counter-argument on the earlier hypothesis of
the two axioms of Basu and Van. The simple simulation by using the cross-tabulation (see,
Table 3.4) has allowed the changes of poverty lines to observe the outcome of child
activity. This basically means that, for example, if the subsistence level of consumption of
each household (poverty line) is set at a higher level, it is possible that there will be a large
number of households under the poverty line who send their children to school. On the
75
other hand, if the poverty line is set at a lower level, there will be a large number of
household who are non-poor, but send their children to work. If these are the case, then the
model of child labour developed by Basu and Van (1998) could not predict the outcome of
child labour. Furthermore the altruistic parents will exist more amongst the poor-household
if the subsistent level of consumption is set at higher level. In contrast to the earlier, the
altruistic parents will exist more amongst the non-poor households if the subsistent level of
consumption is set at lower level. Taking these points into consideration, it seems that the
model of child labour and altruistic parents could not predict the outcome of child labour
or other activities if subsistent levels of consumption are set at different level. Therefore,
this study attempts to test these hypotheses and its findings will contribute significantly to
76
Table 3.2: Descriptive Statistics (Aged 5-14)
Age of child Hypothesis under sub-section 3.2.8 (Elder child tends to Age in year 10.0222 3.129216
work than the younger)
Age of child^2 Hypothesis under sub-section 3.2.8 Age in year^2 110.2354 63.41253
Child’s education Hypothesis 3.2.8 Education of the child 3.320848 2.155787
Child of the HH head Hypothesis 3.2.8 (Child of the household head tends to =1 if child of the household head .8899128 .3130139
be in school)
Female child Hypothesis 3.2.5 (preferences bias of parents toward =1 if child is female .4841459 .4997733
the gender of the child)
Nb. Of child in HH Hypothesis 3.2.6 (Number of children and child labour) Number of children in the household (1-14 Y) 3.6739 1.468522
Mother’s education Hypothesis 3.2.4 (relationship between parents’ Education of mother 5.432521 2.211425
education and human capital of the child)
Father’s education Hypothesis 3.2.4 Education of father 6.726615 2.891023
Poverty status Hypothesis 3.2.1 (poverty and child labour). Testing =1 if above poverty line .5274475 .4992708
hypothesis of Basu and Van (1998)
Employer of HH Hypothesis 3.2.3 (Parents’ occupation tends to =1 if civil servant .1462545 .3533786
(government) determine the child’s activities)
Employer of HH Hypothesis under sub-section 3.2.3 =1 if works for state enterprise .003369 .0579481
(state enterprise)
Employer of HH Hypothesis under sub-section 3.2.3 =1 if employer is joint venture .0058462 .0762405
(joint venture)
Employer of HH Hypothesis under sub-section 3.2.3 =1 if works for private company .0826397 .2753505
(private)
Employer of HH Hypothesis under sub-section 3.2.3 =1 if works for foreign government or .0021799 .0466413
(foreign gov/ IOs) International organization
Employer of HH Hypothesis under sub-section 3.2.3 =1 if works for non-governmental organization .0034681 .0587912
77
Variables and Related Hypotheses Definition Mean Std.Dev.
Variables Related Hypotheses
(NGO)
Employer of HH Hypothesis under sub-section 3.2.3 =1 if self-employed .6957987 .4600911
(self-employed)
Female L Force Hypothesis 3.2.8 (Female labour force has bargaining =1 if female is employed .1229687 .3284175
power, thus be altruistic for child’s education)
Female household head Hypothesis 3.2.8 (Female household head has high =1 if female is the household head .1486326 .3557436
degree of altruistic for child’s education)
Clean water Hypothesis 3.2.7 (social infrastructure and the Access to clean water (percent) 15.56342 31.26297
probability of schooling)
Electricity Hypothesis under sub-section 3.2.7 Access to electricity (percent) 29.40815 39.14817
Dist. to pri school Hypothesis under sub-section 3.2.7 Distance to primary school (kilometer) 14.41677 9.030921
Dist. to sec school Hypothesis under sub-section 3.2.7 Distance to secondary school (kilometer) 53.01467 71.11883
Dist. to high school Hypothesis under sub-section 3.2.7 Distance to high school (kilometer) 109.8806 142.788
Cham Hypothesis 3.2.8 (Different ethnic group tends to effect =1 if Cham .025763 .1584353
the preference of child’ activities)
Chinese Hypothesis under sub-section 3.2.8 =1 if Chinese .0046572 .0680876
Vietnames Hypothesis under sub-section 3.2.8 =1 if Vietnamese .010107 .1000293
Localgroup* Hypothesis under sub-section 3.2.8 =1 if Local group .0035672 .0596222
Other urban Hypothesis 3.2.7 (regional differences and child =1 if other urban .295977 .4565033
labour)
Rural Hypothesis under sub-section 3.2.7 =1 if rural .6059255 .4886752
Source: Author’s calculation
Total observation: 10,092
Note: Local group is referred to local ethnic minority such as Charai, Pnong, Kavet, Kreung, Thampuen and other fews. These ethnic group mostly are the hill tribe living in
northeastern part of Cambodia.
78
Table 3.3: Maximum Likelihood Estimates of Children’s Activities (Aged 5-14)
79
Variables Parameters Marginal effects (dy/dx)
Child works Both works & Child idle Child Studies Child works Both works & Child idle
only studies only only studies
(.1180257) (.0996607) (.0773066) (.01085) (.00221) (.00283) (.01045)
Employer of HH -1.039086*** .2779762 -.6710889*** .0695455*** -.0114995*** .0115842 -.0696302***
(government) (.3357055) (.3207689) (.2202095) (.02211) (.00294) (.01118) (.01919)
Employer of HH -34.79871*** -35.7100*** .1049717 .0283351 -.0182437*** -.0315805*** .0214891
(state enterprise) (.4565152) (.3815718) (.940546) (.12963) (.00169) (.00224) (.1296)
Employer of HH -.6403945 .6646311 1.401699*** -.2550556*** -.0104689 .00936 .2561645***
(joint venture) (1.201961) (.6753667) (.4206732) (.09589) (.00704) (.02462) (.09981)
Employer of HH .1587997 .4739852 .167018 -.0352125 .0019611 .0145838 .0186675
(private) (.3214914) (.3405993) (.227736) (.03229) (.0056) (.01318) (.0304)
Employer of HH -.8005417 .6656655 -1.723474** .0915215* -.0081714 .0327506 -.1161007***
(foreign gov/ IOs) (1.30717) (.8006083) (.8234916) (.05211) (.01057) (.04555) (.02349)
Employer of HH -35.56373*** .3056567 -2.297048* .1303801*** -.0183976*** .0161607 -.1281432***
(NGO) (.4311103) (.7940289) (1.177662) (.03889) (.0017) (.03347) (.02093)
Employer of HH -.3399582 .4874517 .0750087 -.0141171 -.00628 .0121725* .0082246
(self-employed) (.2756557) (.2991789) (.2009583) (.02536) (.00506) (.00695) (.0242)
Female L force .6414577 .0894803 .6405866** -.0994033** .0102211 .0010926 .0902748*
(.4396256) (.4545225) (.3013604) (.05064) (.01032) (.01199) (.05016)
Female head HH .320144 .0631127 -.4228315 .037868 .0068222 .0031893 -.0478794*
(.4222668) (.4355598) (.2857545) (.03177) (.00868) (.01275) (.02812)
Clean water -.0079062* -.011405*** -.0055796*** .0010186*** -.0001085 -.0002858*** -.0006244**
80
Variables Parameters Marginal effects (dy/dx)
Child works Both works & Child idle Child Studies Child works Both works & Child idle
only studies only only studies
(.0047156) (.0032143) (.0021541) (.00029) (.00008) (.00008) (.00026)
Electricity -.0125266*** .0019761 -.008584*** .0011261*** -.0001817*** .0000947* -.0010391***
(.0028642) (.0018349) (.001501) (.0002) (.00005) (.00005) (.00018)
Dist. to Pri School .0129662*** -.0143916** .0154984*** -.0016582*** .0001782** -.0004628** .0019428***
(.0048208) (.0068149) (.0033231) (.00046) (.00008) (.00019) (.00041)
Dist. to Sec School .0016829** .0017431** .0022539*** -.0003261*** .0000209* .0000378* .0002674***
(.0007166) (.0007251) (.0006269) (.00008) (.00001) (.00002) (.00008)
Dist. to high school .0025402*** .0007619* .0012675*** -.0001995*** .0000374*** .0000145 .0001475***
(.0003817) (.0004453) (.000289) (.00004) (.00001) (.00001) (.00004)
Cham ✣ -.3529869 -.4199686 .3659926 -.0376314 -.005438 -.0106375** .0537069
(.3743743) (.296557) (.2260922) (.03507) (.00418) (.00529) (.03553)
Chinese ✣ -35.76386*** -35.9113*** -1.830899** .1683942*** -.019095*** -.0329754*** -.1163237***
(.5755359) (.3169971) (.8977876) (.02494) (.00176) (.00234) (.02482)
Vietnamese ✣ 1.050284* -35.6746*** 2.300957*** -.4461192*** .0043623 -.0400314*** .4817884***
(.5832957) (.2037901) (.3125098) (.0703) (.01102) (.00281) (.07068)
Local group ✣ 2.61835*** .2692113 1.070362 -.2670246* .132676** .0034248 .1377734
(.5873322) (1.028767) (.7440332) (.15046) (.06186) (.02414) (.13582)
Other urban ✝ .4609731 1.808241*** .4694422** -.1238507*** .0051697 .071548*** .0471329*
(.4562226) (.4040072) (.1898801) (.0321) (.00809) (.02349) (.02573)
Rural ✝ .6852924 1.807558*** .4476921** -.0979578*** .0086816 .0440404*** .0452358**
81
Variables Parameters Marginal effects (dy/dx)
Child works Both works & Child idle Child Studies Child works Both works & Child idle
only studies only only studies
(.4668131) (.4010913) (.194527) (.02498) (.00671) (.0097) (.02242)
Constant -3.226138*** -6.11681*** -.1858545
(1.141452) (.9495765) (.5772007)
(Outcome Child’s activities==2 ((child attends school only) is the comparison group)
Note: ***; **; * are statistically significant at the 1, 5 and 10 percent level respectively.
The sign “✣” indicates the ethnic dummy in which the comparison group is Khmer ethnic majority.
The sign “✝” indicates the regional dummy in which the comparison group is Phnom Penh capital city.
82
Table 3.4: Children’s Activities by Poverty Status, 1999 (Percentage of All Households with at least One Child Considering Age Group 5-14
Years Old)
Child’s activities Age group 5-14 years old: using different poverty line
using overall poverty line using food poverty line using Basu and Van’s hypotheses
(1) (2) (3)
Poor Non-poor Total Poor Non-poor Total Poor Non-poor Total
Works only: Persons 278 172 450 170 280 450 3 447 450
Works only: Row % 61.78 38.22 100.00 37.78 62.22 100.00 0.67 99.33 100.00
Works only: Column % 5.83 3.23 4.46 6.29 3.79 4.46 6.25 4.45 4.46
Both work & study: Persons 348 242 590 177 413 590 6 584 590
Both work & study: Row % 58.98 41.02 100.00 30.00 70.00 100.00 1.02 98.98 100.00
Both work & study: Column % 7.30 4.55 5.85 6.55 5.59 5.85 12.50 5.81 5.85
Study only: Persons 2,582 3,754 6,336 1,401 4,935 6,336 17 6,319 6,336
Study only: Row % 40.75 59.25 100.00 22.11 77.89 100.00 0.27 99.73 100.00
Study only: Column % 54.14 70.52 62.78 51.81 66.80 62.78 35.42 62.91 62.78
Child idles: Persons 1,561 1,155 2,716 956 1,760 2,716 22 2,694 2,716
Child idles: Row % 57.47 42.53 100.00 35.20 64.80 100.00 0.81 99.19 100.00
Child idles: Column % 32.73 21.70 26.91 35.36 23.82 26.91 45.83 26.82 26.91
Total: Persons 4,769 5,323 10,092 2,704 7,388 10,092 48 10,044 10,092
Total: Row % 47.26 52.74 100.00 26.79 73.21 100.00 0.48 99.52 100.00
Total: Column % 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
(1) The 1999 Overall Poverty Line takes the value 2305 per person per day in Phnom Penh, 1926 in Other Urban and 1687 in Rural areas.
(2) The 1999 Food Poverty Line takes the value 1737 per person per day in Phnom Penh, 1583 in Other Urban and 1379 in Rural areas.
(3) Using Basu and Van’s hypotheses “the lowest income among the group of child study only” to construct the poverty line, which take the value of 843 Riels per persons
per day in Phnom Penh, 417 in Other Urban and 364 in Rural areas.
Note: These poverty lines refer to the cost of achieving the standard of living provided by the reference bundle of food constructed by Prescott and Pradhan (1997), with an
allowance for non-food, non-rent and non-medical consumption based on typical value of non-food spending by households whose total expenditure just equals the cost of the
food poverty line.
Source: Author’ calculation
83
3.4. Conclusion
This study has estimated the model of parents’ decision to allocate their child activity
by using the maximum likelihood of Multinomial Logit function. The structure of estimation
allows us to analyze broader picture of dynamics of child activity. Our findings are
summarized as follows:
(1) Our finding gives some idea of how this study is different from the earlier studies
and our empirical results gave a different approach to observe the outcome of child labour.
The earlier study tries to test several hypotheses including that if child labour are subject to (i)
subsistence constraint not binding_ meaning parent’s income or consumption is greater than
subsistent level (C> C 0 ) and children work; and (ii) subsistence constraint binding_ meaning
Referring to Altruism model of Becker and the two axioms of Basu and Van 26, a household
prefers to send the child to work if and only if the absence of income from the child, each
individual’s consumption falls below certain exogenously fixed subsistence level, C 0 . Thus,
each individual household would make supply of labour. However, this study observed that
non-poor households also send their children to school. Therefore, the primary striking result
is that C< C 0 and children enrolled in school as well. Furthermore, the non-poor households
also send their children to work and combined work and study. Therefore, we argue the strong
assumptions of modern economics of child labour (Basu and Van, 1998) due to the fact that
“subsistent level” used to draw the parents’ decision as whether the child is observed in “child
works” or “child schooling” is hard to say in the real world since this empirical estimates
proves that the poor also send their children to school even though in lesser proportion, and at
the same time, the non-poor also send their children to work. This finding implies that on the
one hand, the strong existence of altruistic parents in the context of Cambodia, and on the
26
Basu and Van (1998) based the child labor model on the (a) The Luxury Axiom: A family will send the
children to the labor market only if the family’s income from non-child labor sources drop very low, and (b) The
Substitution Axiom: From the firm’s point of view, adult labor and child labor are substitutes.
84
other hand, the needs of child labour in Cambodia in the household economy. Therefore,
policy to absolutely ban on child labour may upset Cambodian population at large. This
(2) We find that fathers’ education have significant role on human capital
accumulation of the child, for which they are in line with the previous work by Ray (2000),
Deb and Rosati (2004), Blunch et.al (2002), Bhallotra and Heady (2003), and Khanam (2003);
But what make this finding different from other works is father’s education play a major role
on the determinant of child labour, where mothers’ education do not. This reason can only be
explained by the cultural practice where male is the dominant decision-maker in the household,
thus they tend to fall under this circumstance. Therefore, gender mainstreaming into society is
reduce child labour and increase children’s enrolment at school. This proved to be valid
elsewhere as well, for example, the study of Chao and Alper (1998) found in the case of
Ghana that school participation rate is closely related to distance to primary school, access to
drinking water and roads. Deb and Rosati (2004) also notes about the importance of social
infrastructure, especially the increasing primary and secondary school in rural areas does
increase school attendance and reduces the probability that a child works or is idle. Ray (2000)
also points out in the result of his study of empirical evidence for Peru and Pakistan that the
provision of good schools can do a lot to reduce child labour in South Asia and to break the
strong link between poverty and hours of child labour. Therefore, the policy implication is to
increase access to basic infrastructure through the rehabilitation of schools, roads, clean water
and electricity.
(4) Female children tend to be child labour rather than schooling. This implies there are
many other barriers, besides economic term, that lead to hindering access of female’s
participation in education. The policy implication is to give priority for female’s education
85
through special policy for girl education, for example, providing girl dormitory, school
boarding program and toilet facilities and other needs. Female need assessment needs to be
conducted, so that an appropriate policy could be drawn towards female education policy.
(5) Households with possession of cattle tend to employ their own children. Therefore,
the wealth paradox has been proved valid, however, through the ownership of number of
cattle, but not through the land size. This is because land size is equally small and equally
(6) Because of an absence of policy of “educational for all” in 1999 and the lack of
junior high school in many areas of rural Cambodia, thus this study have found that many
children stopped school when they reach to one level “junior high school” in which it require
those children to travel or move to stay in urban centre in order to continue their education.
This result contributes to policy and evaluate against the progress that have been made so far
by the Royal Government of Cambodia to improve the access for all children to education by
year 2015 for at least their basic education “the completion of 9 years of schooling” as stated
86
C
Chhaapptteerr A
Appppeennddiixx IIIIII::
FFA
ARRM
MCCH
HIIL
LDDO
OFF W
WEEA
ALLT
THH PPA
ARRA
ADDO
OXX
This chapter appendix sets out a model of the peasant household in an economy with
imperfect markets for labour, land and credit. Allowing two periods enables us to capture the
impact of child work in period 1 on productivity in period 2. This effect arises through both
the gain in work experience and the possible lowering of educational attainment. The model
specifies the effects of farm size on child labour, which, in addition to a wealth effect, includes
The model is solved to give an expression for the quantity of child labour supplied in
period 1:
where subscripts 1 and 2 refer to periods 1 and 2, L c is child labour, A 0 is owned land, K 0 is
initial financial wealth, w h is the wage of hired labour, p r is the price of rented land, Z refers to
exogenous taste shifters, and e refers to unobservable characteristics and optimization errors.
The first-order conditions can similarly be solved to describe other endogenous variables, such
as period 1 consumption, X 1 , in terms of the exogenous variables. Like equation 1 these will
be demand equations that depend on prices and initial wealth (denoted by land [A 0 ] and other
financial wealth [K 0 ]). Since estimation of equation 1 faces the problem that K 0 is
87
Since information on the rental price for land (p r1 , p r2 ) is lacking, the equation is conditioned
This equation forms the basis of the estimates. Consumption (X 1 ) and rented land (A r )
child labour associated with both land and other financial capital. Land owned will generate
the incentive and collateral effects described. As these are of opposite signs, the sign of the
coefficient on land is ambiguous a priori. The following Section describes what can be learnt
about imperfections in different markets based upon the estimated coefficient on land in
equation 3.
According to the literature of Bhallotra and Heady, 2003, they try to explain the role
of market imperfection as follow: The model allows imperfections in each of the labour, land,
and credit markets. This section explores the role played by each in determining the level of
child labour. In every case considered, the negative wealth effect of land is taken to be
captured by X 1 . Conditioning on X 1 , the expected sign of the land coefficient will depend on
which market imperfections dominate. If both land and labour markets are imperfect,
households with land to farm will have an incentive to employ child labour. As this incentive
is stronger the larger the plot of land, land size will have a positive effect on child labour. At
the same time, land size will be associated with weaker credit constraints and, to that extent,
Consider imposing the assumption of perfect land markets. The positive incentive
effect of land disappears even if the labour market is imperfect, because land for which hired
labour cannot be found can be rented out. The negative collateral effect of land persists as long
as the credit market is imperfect. Hence the coefficient on land is zero if the credit market is
88
perfect and negative if the credit market is imperfect. The case of perfect labour markets is
analogous to that of perfect land markets. The positive incentive effect of land disappears
because labour can be hired out. This holds whether or not land markets are perfect. The
coefficient on land is therefore zero or negative, depending on whether the credit market is
perfect or not. If credit markets are assumed to be perfect, there is no negative collateral effect
of land. Land has a positive incentive effect if and only if both land and labour markets are
imperfect. In this case the land coefficient is zero if either the land market or the labour market
The preceding discussion suggests that the relative size of credit versus land and
labour market imperfections can be discerned by observing the estimated coefficient on land.
• If the estimated coefficient on land size is zero, either all markets are perfect or the credit
market is perfect and either the land or labour markets are perfect. Alternatively, it is
possible that all three markets are imperfect and the positive and negative land effects
• If the estimated coefficient on land is positive, either all three markets are imperfect or the
credit market is perfect but both land and labour markets are imperfect. In this case both
• If the coefficient on land is negative, either all three markets are imperfect or credit
markets are imperfect and either the land or the labour market is perfect. In this case the
It is unusual to have data that span an individual childhood (period 1) and adulthood (period
2), making it difficult to obtain structural estimates of a dynamic model. The analysis here is
limited by the single cross-section of data. The dependent variable is hours of child work on
89
the family farm. As there is considerable variation in hours, this measure is preferable to the
participation measure used in most previous research. Since many children do not participate
in farm work, the Tobit estimator is used. This model is a extent modification to Bhalotra and
Heady (2003) that limit the wealth paradox to only land. Since Cambodia agricultural child
labour include time of caring cattle, then it is important to include the cattle as one of the
wealth paradox.
Main Variables
The key regressor, land size (At), is defined as acres of farm land owned or operated by the
household. A quadratic term is included to allow the sizes of the wealth and substitution
effects to vary with land size. Since the land size are not vary much amongst household, we
prefer to use a dummy of having land or not to capture the tendency of child labour employed
on owned land. As no previous study has considered the effects of the mode of operation of
Instrumental Variables
If decisions about income expenditure (X 1 ) and labour supply (for example, L c ) are made
contributes to resources available for consumption. Most previous studies ignore this problem.
As the data do not offer a relevant natural experiment, X1 is instrumented. It is difficult to find
a valid instrument for income in a model of (child) labour supply. In the wider literature,
commonly used instruments for income are education or occupation. Now let the main
H * = Zβ + Yγ + e Eq (1)
90
where hours (H) is a censored endogenous variable, X is a vector of exogenous variables, and
variables Z 1 (Z 1 includes Z) is
Y = Z1π + u Eq (2)
The error terms e and u are assumed to be jointly normally distributed. Let e = uα + ε .
H * = Zβ + Yγ + uα + ε Eq (3)
estimated by the standard Tobit procedure. A test of α = 0 is a test of the null hypothesis that
Y is exogenous.
A.3.3. Results of Empirical Estimate (using field survey data 2006, considering cattle as
one of the most inserting wealth paradox)
Table Annex 3.1 is the results of Tobit estimation, since hours worked of agricultural
children are censored at zero hour. The result of our estimation is interesting in the sense that
health paradox hypothesis is valid, however, in term of number of cattle in each household. As
a matter of reality, Cambodia is like many countries in Asia and Africa where children
raising is considered as one of the wealth paradox which intensively employed their owned
children on this activity since it is considered by most family as adult labour saving.
Therefore, Cambodia is one of the special cases taking into account the number of cattle in the
family to test the child labour of wealth paradox. We also run the Tobit regression of hours
worked of children conditional on child schooling, however, it was confirmed from our
regression that our data do not have such conditioning, thus the whole regression could be run
by Tobit alone or OLS. We also knew that our shape of data is fitted with Tobit as many
children have hours worked censored at zero. We prefer to interpret the results for the
following:
91
Number of Cattle:
The coefficient on the number of “cattle” is statistically significant and positive. This
basically means that child labour hour is strongly associated with the number of cattle in each
household. It also indicates that labour market in rural Cambodia is imperfection both labour
hours of children. This result confirms that there exists the strong evidence on wealth paradox
in the context of Cambodia. It is also evidence with our observation during the field interview
The coefficient on dummy variable having land or not, is statistically significant, while
the coefficient on the size of the land is not. This also very interesting results which can tell us
that land in Cambodia are equally small and distributed, thus the size of the land does not
matter at all.
Other variables:
The coefficient on the “age” of children and its square are both statistically significant,
though opposite sign. This takes to mean that children in Cambodia tend to work as child
labour when they are young, and beyond certain age they will be located in other activity
which is not a child labour any more. Household income, parents’ education, family size,
number of adult labour and irrigated land are not statistically significant. These basically mean
that all these variables are yet to play a direct role in the determinants of child labour in the
context of rural Cambodia. Another word is that we do not have enough evidence to conclude
that these variables have an effect on rural child labour. The coefficient on “femaleHH”
female household head is statistically significant, which means that female household head has
higher degree of altruistic on children. Nevertheless, there are unobserved endowment effects
on the child labour, but it only tell us that to be in Khan Damra village and Kol Korm are
92
C
Chhaapptteerr A
Appppeennddiixx IIIIII::
TTA
ABBLLEESS
93
Table Annex.3.2: Matrix of Correlation of Independent Variables
(1) (2) (3) (4) (5) (6) (7) (8) (9)
(7) Mother’s education 0.0128 0.0141 0.0862 -0.0588 0.0054 -0.1122 1.0000
(8) Father’s education 0.0425 0.0439 0.1202 -0.0909 -0.0016 -0.1123 0.3839 1.0000
(9) Poverty status 0.0671 0.0667 0.0692 -0.0389 0.0254 -0.1783 0.1194 0.1444 1.0000
(10) Employer of HH 0.0276 0.0255 0.1023 0.0449 0.0116 -0.0185 0.1785 0.2598 0.1681
(government)
(11) Employer of HH -0.0015 -0.0022 0.0018 -0.0045 -0.0043 0.0195 -0.0087 0.0128 0.0108
(state enterprise)
(12) Employer of HH 0.0093 0.0091 0.0103 0.0096 -0.0064 -0.0030 0.0124 0.0045 0.0358
(joint venture)
(13) Employer of HH -0.0057 -0.0069 0.0044 0.0311 -0.0000 -0.0179 0.0137 0.0239 0.0407
(private)
(14) Employer of HH 0.0016 0.0004 0.0004 0.0160 -0.0005 -0.0129 0.0234 -0.0026 0.0355
(15) Employer of HH 0.0022 0.0035 -0.0077 0.0168 0.0130 -0.0101 0.0231 0.0360 0.0039
(NGO)
(16) Employer of HH -0.0315 -0.0312 -0.0954 0.1199 -0.0163 0.0937 -0.1667 -0.2438 -0.1973
(self-employed)
(17) Female L force 0.0364 0.0349 -0.0039 -0.1593 -0.0069 -0.1383 0.0544 0.1342 -0.0168
(18) Female head HH 0.0353 0.0353 -0.0014 -0.2887 -0.0022 -0.1630 0.0647 0.1543 0.0052
(19) Clean water 0.0757 0.0751 0.1531 -0.0697 0.0142 -0.1109 0.1839 0.2391 0.3416
(20) Electricity 0.0778 0.0778 0.1399 -0.0602 0.0261 -0.1104 0.1876 0.2344 0.3973
(21) Dist. to Pri School -0.0207 -0.0182 0.0044 0.0213 0.0005 0.0638 -0.0746 -0.0880 -0.1279
(22) Dist. to Sec School -0.0428 -0.0424 -0.0174 0.0329 -0.0079 0.0918 -0.0831 -0.1053 -0.1491
(23) Dist. to high school -0.0531 -0.0525 -0.0239 0.0444 0.0081 0.0927 -0.1079 -0.1367 -0.1654
(24) Cham -0.0019 -0.0015 -0.0143 0.0302 0.0044 0.0494 -0.0211 -0.0476 -0.0277
(25) Chinese 0.0175 0.0165 0.0214 -0.0160 0.0021 0.0171 0.0175 0.0153 0.0644
94
(26) Vietnamese -0.0018 -0.0023 -0.0053 0.0076 0.0192 -0.0096 0.0178 -0.0122 0.0566
(27) Local group -0.0252 -0.0235 0.0183 0.0097 -0.0090 0.0019 0.0114 -0.0371 -0.0576
(28) Other urban -0.0065 -0.0060 -0.0171 -0.0086 -0.0109 0.0516 0.0199 0.0044 0.0364
(29) Rural -0.0376 -0.0374 -0.0799 0.0416 -0.0065 0.0265 -0.1133 -0.1315 -0.206
(Continued)
(10) (11) (12) (13) (14) (15) (16) (17) (18)
(government)
(state enterprise)
(joint venture)
(private)
(NGO)
(self-employed)
(17) Female L force -0.102 -0.0035 -0.0133 0.0524 -0.0175 -0.0148 0.0947 1.0000
(18) Female head HH -0.125 -0.0074 -0.0181 0.0261 -0.0196 -0.0182 -0.0221 0.8943 1.0000
(19) Clean water 0.2564 0.0621 0.0744 0.1218 -0.0055 0.0012 -0.3472 0.0066 0.0496
(20) Electricity 0.2965 0.0821 0.0679 0.1602 0.0196 0.0130 -0.4135 0.0011 0.0464
(21) Dist. to Pri School -0.0649 -0.0080 -0.0148 -0.0715 -0.0052 -0.0136 0.1228 -0.0316 -0.0468
(22) Dist. to Sec School -0.0884 -0.0195 -0.0131 -0.0867 -0.0025 -0.0088 0.1520 -0.0371 -0.0533
(23) Dist. to high school -0.1434 -0.0259 -0.0362 -0.0790 0.0207 -0.0241 0.1939 -0.0132 -0.0305
(24) Cham -0.0266 -0.0093 -0.0128 -0.0190 -0.0075 -0.0100 0.0432 -0.0143 -0.0162
(25) Chinese -0.0149 -0.0040 0.1212 0.0335 -0.0032 -0.0043 -0.0447 -0.0188 0.0120
(26) Vietnamese -0.0374 -0.0058 -0.0080 0.0412 -0.0047 0.0077 -0.0023 -0.0152 -0.0073
(27) Local group 0.0023 -0.0032 -0.0044 -0.0113 -0.0026 -0.0035 0.0146 0.0108 0.0058
(28) Other urban 0.0904 0.0025 0.0065 0.0098 -0.0056 0.0153 -0.0892 -0.0268 -0.0243
(29) Rural -0.2098 0.0036 -0.0318 -0.0724 0.0151 -0.0183 0.2529 0.0026 -0.0211
95
(Continued)
(26) Vietnamese 0.0432 0.0803 -0.0417 0.0328 -0.0009 -0.0165 -0.0070 1.0000
(27) Local group -0.0279 -0.0419 -0.0023 0.0261 -0.0087 -0.0091 -0.0039 -0.0057 1.0000
(28) Other urban -0.0678 0.0678 -0.0062 -0.0288 -0.1238 0.0485 -0.0069 -0.0227 0.0535
(29) Rural -0.3906 -0.4154 0.1396 0.1474 0.2534 -0.0107 -0.0597 0.0305 -0.0380
(Continued)
(28) (29)
Note: The coefficients of the matrix correlations indicate that the model is secured from the multi-co
linearity. Technically, if coefficient is greater than four and smaller than eight, one can draw
assumption that there is weak correlation, but it does not suffer the mode. However, if the coefficient is
greater than eight, one shall omit that variable or combine both variables into one.
96
C
Chhaapptteerr IIV
V
C
CHHIIL
LDD’’SS W
WAAG
GEE,, H
HUUM
MAAN
NCCA
APPIIT
TAAL
LAAN
NDDH
HOOU
URRSS W
WOOR
RKKE
EDDT
TRRA
ADDE
E--O
OFFFF::
SSiim
mppllee T
Thheeoorreettiiccaall aanndd E
Emmppiirriiccaall E
Evviiddeennccee ffoorr PPoolliiccyy IIm
mpplliiccaattiioonnss
4.1. Introduction
Child labour has long been recognized as being in direct conflict to the human capital
formation of children, and the common explanation is that schooling competes with the labour
intensive jobs that children do, whether it be wage labour, employment in family enterprises,
or collective/community labour activities. By this view, the low current income of families
keeps their children out of school and thus perpetuates poverty in the next generation
(Ravallion and Wodon, 2000). Some studies have found that poverty is the cause of child
labour (Basu and Van, 1998; Edmonds and Turk, 2004), and thus it concluded that eradicating
poverty will also eradicate child labour. Perhaps most, if not all, would agree with this strong
association, however, the most important question relates to the transitional mechanism to get
out of poverty. This is the most difficult task and has been a challenge for all leaders in the
Goals (UN, 2000). Human capital formation through schooling and training is among the most
important factor in the fight against poverty (Becker, 1993). This binding correlation between
poverty, child labour and human capital presents a trade-off, and this is the trap that is the
vicious cycle of poverty. The most pressing question is whether these children will have an
opportunity for education if their households remain poor or living just above the poverty line.
Of course, the basic reality uncovered by this research is that many children whose households
97
Given the observation of children combining work with study, a hypothesis was
formulated postulating that child labour does indeed lead to the human capital development of
children in Cambodia as well in other developing countries. While this study has proved that
child labour has a negative relationship with child schooling, this general relationship has
caused some considerable misunderstanding of the correlation between child labour and
schooling. In reality the most important issue is not whether the child works or not, but how
many hours they are participating in labour. For example, this study has found that working
just a few hours per day, not only does not negatively affect schooling, but does also in many
cases positively impact schooling, by providing the household with the income necessary to
allow the child to attend school. Like any theory based on a production function or the rate of
return to human capital (Mincer, 1974; Becker, 1975), the number of hours worked by
children will become a trade-off with their schooling if the child works more than the
threshold hours. If the assumption (child labour benefits their human capital) is true, meaning
that children’s education level provides bargaining power for their wage rate, then this will
encourage parents to choose both labour and schooling for their child because of the increased
In recent decades, various studies have assumed that a trade-off between child labour
and human capital formation exists (Basu 1999, Baland and Robinson, 2000; Fan, 2004;
Rosati and Tzannatos, 2000; Basu and Tzannatos 2003; Ray and Lancaster, 2005; Patrinos and
Psacharopoulos, 1995; Rosati and Rossi, 2004; Heady. C, 2003), but there is very little data
that analyses the trade-off between the hours worked by children and schooling outcome. This
section will review the current body of literature and highlight some key results relevant to this
study.
Basu and Van (1998) made an important contribution to the policy analysis on the
trade-off between child labour and their schooling status by using the parents’ wage rate as the
only important determinant of their child activities. In the same setting, Baland and Robinson
98
(2000) have used a “bequest constraint” of parents and “capital market imperfections” to
determine the rationale of parents concerning the trade-off between child labour and the
accumulation of human capital. The “bequest constraint” refers to the parents’ understanding
of possible future financial benefit which impact on their present decision concerning their
child’s schooling. “Capital market imperfections” refers to the absence of formal financial
institutions, which implies the existence of only informal “money lenders” as the source of
capital. Patrinos and Psacharopoulos (1995) found that the same factors that predict an
increase in child labour also predict reduced school attendance and an increased chance of
repetition. Similarly, Heady. C (2003) has used the direct measures of reading and
mathematics ability to conclude a negative relationship between child labour and educational
achievement in Ghana. However, a small increase in child labour does not necessarily result in
a trade-off with human capital investment (Fan, 2004), and increases in schooling does not
necessarily translate into a decline in child labour (Edmonds, 2005), since the positive impact
of increased financial resources for education may outweigh the negative impact of reduced
time for study. The outcome of child labour has been argued about over many decades, and
findings are varied depending on historical, political, social and economical factors.
In contrast to the aforementioned studies on the trade-off between child labour and
schooling that focus on whether the child participates in labour or not, rather than analyzing
the connection between the number of hours worked by children and their schooling,
Akabayashi and Psacharopoulos (1999) have used time-log data from a 1993 survey in the
United Republic of Tanzania to investigate the relationship between child work and human
capital development. This study found that factors that increase the children’s working hours
also decrease their hours of study, and so hours of work are negatively correlated with
studying ability. Ray and Lancaster (2004) conclude in their case study of evidence from
Belize, Cambodia, Namibia, Panama, Philippines, Portugal, and Sri Lanka that children’s
work, even in limited amounts, adversely affects the child’s learning, indicated by a reduction
99
in the school attendance rate. However, the paper suggests that if some light work is permitted
for children aged 12-13, as suggested in the ILO Convention 138, Article. 7, then it should be
accompanied by a campaign to improve adult education levels. Better educated parents will,
by ensuring that their children make more efficient use of non labour time for study, help to
reduce the damage done to the child’s learning. In contrast, a similar study on child labour
supply (Rossati and Rossi, 2003) seems to reject the assumption that a few hours of work only
In contrast to the aforementioned studies, this research will be the first attempt to: (i)
investigate the rate of return to education from child labour, which is crucial the income from
the child’s labour provides the resources necessary for their schooling; (ii) in the model of the
child labour and schooling trade-off, this study allows omitted variables to enter directly into
the explanatory variables because the non-inclusion of omitted variables into the model can
cause serious problems on parameter estimates (Wooldridge, 2003); (iii) the sample covers
children age 5-14 rather than the more limited 12-14 target group, since children are assumed
to be actively engaged in domestic work since a young age, and this age group is in
accordance with the definition of child labour in Cambodia; (iv) and estimation of the
Multinomial Logit function is made to take into account the possibility that a child is
Finally, the results show that children’s education plays a significant role in
determining their wage rate, which implicitly explains the logic behind the household’s
decision to allow the child to undertake both work and study, thus further explaining parent’s
decision to keep investing in their child’s education. This study also found that being non-poor
as well as the father’s and mother’s level of education have statistically significant effects on
child schooling outcomes. In addition, a further important finding of this study is that if the
average hours worked by children stays below the threshold level (19, 22, and 25 hours per
week for the age group of 6-9, 10-12, and 13-14 respectively), then there is no negative effect
100
on schooling. These research findings have implications for government policy of human
With the fact that majority of children in Cambodia are found in the combined work
and study, thus this research assumes that child labour increase human capital of children if
children work under the threshold level of hours. Furthermore, children are able to guarantee
their schooling because of their child labour. In other words, children’s incomes are needed to
threshold level of working hours below which there is no negative impact on schooling, it is
first necessary to measure the rate of return of child labour. With this view, this study aims to
(1) Estimate the rate of return to child labour. In the literature, the rate of return begins
only after education is completed and work begins, and this means that the investment
in education only pays off when education is finished and work begins. However, this
study measures the mutual impact of both child labour on education, and education on
child earnings. Because some of the children are in the category of “both work and
study”, their earning capacity is influenced by their education and, at the same time,
(2) Estimate the trade-off between hours worked as child labour and schooling outcomes,
This paper is organized as follows. Section 4.2 presents two simple theoretical
frameworks on (i) the rate of return to human capital of the child, and (ii) the hours worked by
children and schooling trade-off. Section 4.3 presents the empirical frameworks for the two
above mentioned theoretical frameworks, and explains the data and variables used in this
study. Section 4.4 discusses the results, and section 4.5 presents the conclusion. Finally,
Chapter Appendix IV presents the empirical model and its results of the estimate of
Multinomial Logit function. Tables Annex 4.2 and Annex 4.3 contain the tables of correlation
101
among covariates used in our specification function. Table Annex 4.4 presents the results of
Simultaneous Estimates of Earning Income and Hours Worked of Children in the Past 7 Days
Since the beginning of the industrial revolution, literacy and knowledge have become
increasingly valuable relative to basic manual skills. This increasing value has led to wage
premiums for educated workers. Not surprisingly, an educated workforce is the dominant
factor in explaining differences in regional growth and prosperity. As a result economists have
and regional differences in economic growth. The paper on investment in human capital by
Becker (1962, 1975) emphasizes on education and training, the most important investments.
Of course, formal education is not the only way to invest in human capital. Workers also learn
A number of studies use Mincer’s human capital earnings function because this model
is the most commonly employed method in labour economics. Mincer (1974) uses a simple
regression model “with a linear schooling term and a low-order polynomial in potential
experience”. Using Mincer’s model, we can easily adopt this method for the estimation of rate
of return to human capital of child labour. We simply brief the method as follow:
Supposed that:
r = discount rate
102
t = 0, 1, 2,…, n time, in years
Then, the rate of returns to human capital of the child when discounting process is
discrete is:
n
Vs = Ys ∑ (1 + r )
t = s +1
−t
n Ys (e − rs − e − rn )
Vs = Ys ∫ e −rt dt =
s r
Similarly the present value of earnings of an individual child who engages in s-d years
of schooling is:
n Ys −d (e − r ( s −d ) − e − rn )
Vs − d = Ys ∫ e dt = − rt
( s −d ) r
The ratio, K s, s-d , of annual earnings after s years to earnings after s-d years of
Ys e − r ( s −d ) − e − rn e r ( n+ d − s ) − 1
K s ,s −d = = −rs = r ( n−s )
Ys −d e − e −rn e −1
The ration, K s, s-d , is clearly larger than unity, and a positive function of r, and a
negative function of n. This take to mean that people with more schooling have higher annual
years of schooling, and difference is larger the shorter the general span of working life.
Since we are interested in the n as a fixed span of earning life, then we can write the
103
n+ s Ys −rs
Vs = Ys ∫ e −rt dt = e (1 − e −rn );
s r
n+ s −d Y
Vs −d = Ys −d ∫ e −rt dt = s −d (1 − e −rn )e −r ( s −d ) ;
s −d r
−r ( s −d )
Y e
K s ,s −d = s = −rs = e rd
Ys −d e
Ys
Now supposed that s − d = 0, then we obtain: K s , 0 = = e rs . Taking the logarithms,
Y0
ln Ys = ln Y0 + rs.
This equation shows that percentage increments in earnings are strictly proportional to
the absolute differences in the time spent at school, with the rate of return as the coefficient of
proportionality. Precisely the equation shows the logarithm of earnings as a strict linear
The conceptual framework of child labour supply and human capital formation trade-
off is based on the standard economic assumption that individuals are rational utility-
maximizers (see, e.g., Becker, 1965). In this simple theoretical model, we assumed that
individual parents will allocate their children’ time between working and schooling through
their maximization of household utility. The human capital function of the child is assumed to
be a function of time spent at school and school’s expenditure for which individual household
spent on school fee, textbooks, and other extra cost of child-schooling during the a year. All
decisions are made by altruistic parents, and children are treated as recipients. We assume that
104
where C p is the consumption of parents, h c is human capital function of the child, δ is
the proxy of altruistic parents, and a, is the household’s characteristics. As mentioned, the
human capital formation of children is the function of school’s expenditures, x, total available
for study denoting (1-l c ), and child characteristics, Ch i . Thus we can express the human
hC = hC ( x, 1 − lC , Chi ) Eq (2)
x ≥ 0 ; 0 ≤ 1 − lC ≤ 1 ; Chi > 0
In this model, each household maximize its utility in equation (1) with subjected to the
C p + x = I p + WC lC ; or C p = I p + WC lC − x Eq (3)
income function I p (z) is a function of household assets A, education, and other household’
characteristics. For simplicity we can write the household budget constraint as:
C p = A + I p ( z ) + WC lC − x Eq (4)
The First Order Condition (FOC) is made with respect to l c and x. We also assume that our
model will produce the interior solution after the FOC, so that we can find our static
equilibrium if any known initial values are provided for. Therefore the FOC are:
∂U ∂U ∂U ∂hC
= WC + δ . . (−1) = 0 Eq (5)
∂lC ∂C p ∂hC ∂ (1 − lC )
∂U ∂U ∂U dhC
= (−1) + δ . . =0 Eq (6)
∂x ∂C p ∂hC dx
The results of FOC in equation (5) and equation (6) can be rearranged as follows:
∂U ∂U
δ. = WC . Eq (7)
∂ (1 − lC ) ∂C p
dhC ∂U ∂U
δ. = / Eq (8)
dx ∂C p ∂hC
105
∂U
From equation (7) we can interpret that is the household utility maximization
∂ (1 − lC )
with respect to time of schooling. The static equilibrium for the interior solution of the time of
∂U
schooling is depending on which is the utility maximization with respect to consumption
∂C p
of the household. This household consumption could be high or low is pretty much depend on
WC
which is child wage rate and the degree of parents’ altruistic. So it is clear that if W c is
δ
high, it will leads to high consumption of the household, and thus time of schooling is high as
well. However, if parents’ altruistic δ is high, it will be that the household will reduce its
consumption in order to maximize the schooling hour of children. Therefore, in equation (7),
the household will enjoy both high household consumption as well as high hours of schooling
From equation (8) we can also interpret that the maximization of human capital of
children with respect to school expenditure is directly depending on the degree of parents’
altruistic, and this maximization is a trade-off between household’s consumption itself and the
human capital of the child. This basically means that the more the expenditure on child
schooling, it will reduce household’s consumption and increase the human capital of children.
Simply enough to solve the problems of our model, we can just solve the system of
equations as following:
L* = L* ( z , WC , a ) ; 1 − lC* = L* Eq (9)
x * = x * ( z ,WC , a ) Eq (10)
Given that any known initial value of lC* and x* , we can find the interior solution with
the provided parameters z , WC , a . The sign of the parameters is investigated in the empirical
model.
106
4.3. Empirical Framework
In the empirical specification of the human capital earning function, Mincer (1974)
noticed that there is less of an interaction, if any, between experience and schooling and than
between age and schooling. Experience profiles of earning are much more nearly parallel than
age profiles. If so, in an earning function in which earnings are logarithmic, years of work
experience shall be entered additively and in arithmetical form. The experience term is not
linear but concave. Because child’s earnings and its hours worked are jointly determined
Therefore, the earning function and its hours worked is simply the simultaneous equation
system as follows:
hours worked of the child, and Ch i and Z are characteristics of the child and community
respectively, and U is a disturbance term. In the equation (2), the variable B i is exogenous.
Mincer (1974) had noted that the use of age alone instead of experience in the earning function
results in the omission of some variables. Therefore, our model includes the experience of the
child labourer.
This empirical model on human capital formation and hours worked trade-off is
derived from the conceptual framework that expresses the negative relationship between
human capital formation and hours worked trade-off. We use the learning measures such as
variable of Schooling Attainment Relative to Age (SAGE index), years of schooling, and
107
child’s activities (child works and child idle, child combined works and studies, and child
studies only) as dependent variables. We model the learning measure of the child on set of
explanatory variables such as age, age^2, education of parents, gender of the child, number of
children in the household (age 5-14), number of babies in the household (aged 0-4), poverty
status of each household (1 if above poverty line), school’s expenditure, and hours worked by
children and its square. The inclusion of child labour hours, l c , and its square is designed to
capture the trade-off point between labour hours and learning measure of the child.
Furthermore, this study is designed to analyse the trade-off hours worked by children in
different age group. To do so, according to Wooldridge (2003) 27, the model must be written
with and interaction between hours worked by children in different group 28 (However, one
group must report the full sample of the hours worked by all children, and the rest just report
the hours worked by children in that particularly age group, by setting the rest to zero) and the
dummy of respective age groups. Obtaining this interaction is easy by including the variable
hours worked by respective groups along with respective dummies. In this case, we have
classified three age groups with the based interaction group of age group 13-14. These groups
are 6-9 years, 10-12, and 13-14 years, along with their respective hours worked by children,
such as variables ( AgeG1.lC ) and ( AgeG 2.lC ) (hours worked by children in the age group 6-9,
and 10-12 respectively) and its squares. According to Orazem and Gunnarsson (2003), the
child labour hours as well as the poverty status of the child are likely to be endogenous, thus
OLS coefficient estimates may yield biased. Therefore, we need to seek for appropriate
27
There are also occasions for interacting dummy variables to allow for a difference in slopes. The design to
capture the different slopes is given in details in Wooldridge (pp.233, 2003). He also explains how to do so in the
econometric model as explained in page 234, in the equation (7.17). However, a clear example is given in page
235, example 7.10.
28
According to Wooldridge (pp, 234, 2003), using the same techniques, we must write the model with an
interaction between hours worked by children and AgeG1, and AgeG2, by producing a other two new variable, in
this case, is hours worked by children in the AgeG1, and AgeG2 respectively, called this variables as
( AgeG1.lC ) and ( AgeG 2.lC ) . Obtaining the interaction is easy and we shall not be daunted by the variables
( AgeG1.lC ) and ( AgeG 2.lC ) , which is zero for any other children outside these age groups in the sample, and
equal to the level of hours worked by children in each respective group.
108
instrumental variables in the dataset that have high correlation with hour worked of the child
and poverty status, but they are not correlated with the disturbance term of the structural
equation. With these notions in mind, we can establish our regression model as follows:
Because AgeG1 and AgeG 2 are the dummy variables corresponding to different age groups of
the child, and its value is strictly belongs to {0,1}, thus the square of itself will remain the
same. This notion can be written as: AgeG1 = ( AgeG1) 2 , and AgeG 2 = ( AgeG 2) 2 . So, we also
can write that: ( AgeG1.lC )lC = ( AgeG1.lC ) 2 , ( AgeG 2.l C )l C = ( AgeG 2.lC ) 2 Therefore, we can
Because variables lC , and Povi are assumed to be endogenous to the model, which means that
lC , and Povi are correlated with U i1 . Thus, the estimation of equation (2) will be biased.
Therefore, we use instrumental models of the Two Stage Least Square to deal with this
where vector X i denotes observed individual and exogenous variables that include
109
household size, number of babies in the household (age 0-4 years old), number of children
(age 5-14 years old), and age of the household head. The vector Ch i is child characteristics
such as age of the child, and female child. The vector Z i is represented by community
characteristics such as access to clean water and sanitation, hygienic latrine, telephone and
rural or urban. Although there are more community characteristics that may affect the utility of
the parents such as pagoda, water irrigation, road, access to market and others, but we may not
From the structural equation (2), we have variable hours worked by children and its
squares which are endogenous with learning measures. Avoiding this endogeneity, we have
used instrumental variable IV i containing variable such as “assets of TV, radios, cars, boats,
motor-boats, tractors, bicycle, and other household assets”. Note that the variable of school’s
expenditure is not available in the dataset of CCLS-2001/02; however, since this variable is
very important for our structural equation, we have obtained the coefficient estimates from
school’s expenditure in the structural equation (2) is predicted value from other regression as
above mentioned; therefore this variable is considered as independent and uncorrelated with
the disturbance term. If the variable of school’s expenditure is available from the dataset
directly, we believe that this variable would be endogenous with learning measure, and thus
will be instrumented as well, in which the estimation of structural equation is obtained through
the search for coefficients estimate from set simultaneous equations either through Seemingly
Unrelated Regression (SUR) or Three Stages Least Square (3SLS). Additional important
household characteristic was added to the model such as poverty status variable rather than
household’s income. This potential endogenous variable has been accounted for in the above
model by the instrumented variables. Finally the estimation of the structural equation (2) is
110
hc = β 0 + β1l c* + β 2 l c*2 + β 3 AgeG1 + β 4 AgeG 2 + β 5 ( AgeG1.lC* ) +
β 6 ( AgeG1.lC* ) 2 + β 7 ( AgeG 2.lC* ) + β 8 ( AgeG 2.lC* ) 2 + β 9 Chi + Eq. (5)
β10 Povi + β11 SchExi + β12 X i + β13 Z i + U i1
The First Order Condition (FOC) on (5), with respect to hours worked is to check the
turning point of the hours worked of the child that beyond this turning point “hours worked
threshold” of the child will trade-off with human capital formation. However, since we allow
for different slopes 29 for difference age group of the child, therefore, we can derive the trade-
( β1* + β 5* )
• ( AgeG1.l ) = −
*
is the trade-off hours worked by children in the age
2( β 2* + β 6* )
C
( β1* + β 7* )
• ( AgeG 2.lC* ) = − is the trade-off hours worked by children in the age
2( β 2* + β 8* )
β1*
• l c* = − is the trade-off hours worked by children in the age group 13-14, with
2 β 2*
In this study, we have used dataset of Cambodia Child Labour Survey, 2001/02 for our
data analysis on child labour hours and other salient characteristics of the child. The sample
size of this dataset covered 12,000 households which were interviewed on the nature of child
labour in both economic and non-economic activity. Although the survey covers detailed
information of child labour in the age group 5-17 years old for the comparative purpose of
111
child labour in the regions as well as a compliance to ILO convention No.138, but we only
consider working children in the age group of 5-14 years to be our sample in this study
because the adopted new labour code in 1997 by Cambodia National Assembly sets the
minimum age of employment at 15 only (Article 177). We have, therefore, 21153 children
breaking down into 10824 male and 10329 female contain in the sample.
Table 5.1 presents descriptive statistics used in the empirical model. The variable
SAGE meaning Index of Schooling Attainment Relative to Age is constructed to be one among
the alternative measure of “schooling outcomes”. The index is simply derived from the
Years of schooling
equation that: SAGE = ×100 . The higher of the index is the
Age − Age of School Entry
better of schooling outcomes of children. We also notice that the variable of school’s
expenditure is not available in the CCLS-2001/02 dataset. However, we have obtained the
predicted value of this variable by using the coefficients estimates from another regression of
Cambodia Socio-economic Survey 1998/99. This technique is very smart and eases our burden
For the earning of the child, this study uses the indicator of “current status of working
children” as the measurement of all economic activity of the child. The CCLS-2001-02 has
adopted two approaches (Usual Economic Activity and Current Status Economic Activity)
among the three different approaches to measure employment and unemployment. The three
approaches are: (1) Usual status approach with a reference period of 365 days preceding the
date of survey; (2) Current weekly status approach with a reference period of seven days
preceding the date of survey; (3) Current daily status approach with each day of the seven days
preceding date of survey as the reference period. The NSS classifications of economic activity
set up by the Planning Commission. Since child labour is believed to involve in “Current Staus
of Economic Activity”, thus this study has decided to employ this variable in the model of
112
estimations. It is worth to note that the past body of literature also employed this variable, for
113
4.4. Results of Empirical Estimate
Table 4.2 presents coefficient estimates of return to human capital of child labour in
real amounts (Riels). In table 4.2, column (a) presents the coefficients estimates performed by
Ordinary Least Square (OLS), and column (b) presents the coefficients estimates performed by
simultaneous equation system under assumption that child’s earnings and its hours worked are
jointly determined. Using Hausman-test, we confirmed that we have enough evidence to reject
our null hypothesis of OLS, and accept the results of simultaneous estimate. The recent body
of literature places concern on the endogeneity of “education” with earning function (Duflo,
2001). However, the endogeneity comes from the concern that different locations under
different educational programmes (treatment and non-treatment areas) may have an effect on
the quality of education, and thus it will also affect the earning capacity. The instrumented
variable is just “location of treatment and non-treatment”. Therefore, our study takes into
account this concern. However, the result 30 of the endogeneity test shows that our data does
The results are interesting to our hypothesis of child labour and child schooling at the
same time. Because most Cambodian children are likely to combine work with school,
therefore, their wage rate or earnings are most important to maintaining their household
consumption and other security. As the result of child labour, then they are able to be located
in school by their parents. Let us find the significance of these results if the education and their
experience do have role in determining their earning income or not. We found that education
among child labourer give higher return around 12 percent to their income equivalence of 0.62
USD per week (see, also Table Annex 4.4). This rate of return is high and can be comparable
30
The result of the Instrumental variable regression (IV-regression) of the possibility of endogeneity of “child’s
education” with child earnings shows that our model and data is free from the endogeneity problem. Using the
Hausman Test, we cannot reject our hull hypothesis.
114
to the rate of return among adult labour of non-farm men in the United Sate of America in
1959 (see Mincer, 1974). The coefficient of experience and its squared indicate that all
working children with experience less than 2.6 years have negative impact on their earnings.
After 2.6 years, this experience becomes positive relationship with child’s earnings with the
return of 0.35 USD per week. This finding is somehow the first finding of return to child
labour in terms of their education and experience. In literature, adult workers- once completed
schooling- the first year of experience does give return to their income in smaller amounts than
another year of experience. However, our case is different because children at the 1st year till
the 2.6 years of experience still have negative impact on their earning power. Unlike adult
labour, child laborers need at least 3 years of experience in order to have positive impact on
their earnings.
Furthermore, we also found that children of households with access to public piped
water make more income than the non-piped water. This result reflects urban characteristics
where job availabilities are at child’s disposure, which also could give higher wage than the
rural one.
Table 4.2: Simultaneous Estimates of Earning Income and Hours Worked of Children in the
Past 7 Days
115
Variable Earning income of Simultaneous Equation of Earning Income and Hours
children in the past 7 Worked of Children in the Past 7 Days
days (OLS) (b)
(a) Earning income Hours worked
(past 7 days) (past 7 days)
Coefficient P-value Coefficient P-value Coefficient P-value
Table 4.4 and Table 4.5 show the results of schooling outcomes by using Ordinary
Least Square (OLS) and Two Stage Least Square Method of Tobit estimates (IV-Tobit). The
“poverty status” are endogenous with “schooling outcomes” both in the equation of “SAGE”
instrumented variables if they are valid or not. The Sargon’s statistics suggest that all
instrumented variables passed the over-identification restriction test. Provided that all
116
coefficients in the regressions are standard robust errors to account for heteroskedasticity, we
therefore are sure the results from the IV-Tobit are consistent and unbiasedness. The results in
Table 4.4, Table 4.5 and Table Annex 4.1 show that schooling outcomes are nonlinear
function of age. The coefficients estimate of “Age of the child and its squared” are both
statistically significant, although are in opposite signs. This means that Cambodia children
tend to participate in school or combined school and work at younger age, but beyond a certain
Interestingly, in Table 4.4 and Table 4.5, the variables of “Poverty status”, “School
expenditure”, “Father’s education” and “Mother’s education” are positive and statistically
significant. These findings imply that children of non-poor household tend to be in school
rather than working. Additional amount of school expenditures (school fee, extra-fee, text
book, books, pencils, and other school’s expenses) tend to increase human capital formation of
the child further. Both father’s and mother’s completion of primary or higher education plays
significant role in human capital formation of their children. While Table Annex 4.1 shares
most of the common results as in Table 4.4 and Table 4.5, however, it provides more precise
results in terms of child combining work with study or doing neither. The marginal effects
suggest that a non-poor household is associated with a 37 percent point increase in the
probability that child is doing a combined work and study, a 14 percent point increase in the
probability that child attends school only and a 15 percent point decrease in the probability
that child is likely to be idle. A 10 unit increase in school’s expenditures on the child is
associated with a 0.1 percent point increase in the probability of child “combining work and
study”, a 4 percent point increase in the probability of “child study only” and a 0.1 percent
point decrease in the probability of “child idle”. Father’s and mother’s completion of primary
education or higher are strongly associated with 4 and 5 percent point increase in the
probability that child is likely to combine work and study. Mother’s completion of primary
117
education or higher is associated with a 1 percent increase in the probability that child is likely
to attend school only, and a 2 percent point decrease in the probability that child is idle.
baby”, number of children in the household “Nb. Of children” and male household head
“Gender HH head” are negative and statistically significant. These negative relationships with
schooling outcomes imply that additional number of infant and number of children in the
household tend to reduce time of schooling of the child. The marginal effects in Table Annex
4.1 show that the present number of infants in the household reduce the probability that child
doing combined “work and study” by 1 percentage point, child doing “study only” by 3
percentage points and child doing nothing “idle” by 1 percentage point. Moreover, current
number of children in the household “Nb.of children” reduce the probability that child
combines “work and study” by 2.6 percentage points, child doing “study only” by 0.3
percentage points and reduce the probability that child is doing nothing “idle” by 1 percentage
point. Male household head tends to reduce the probability of child’s schooling. This indicates
that Cambodia’s situation of increasing children’s education has constrained by old paradigm
results show that children tend to be in school provided that their household are well
connected to urban infrastructure such as using piped water, having public electricity, and
Finally we come to the most important ceteris paribus of hours worked of children. The
variables “hours worked” and its square are statistically significant, though are of opposite
signs. This clearly indicates that hours worked of children have a U-shape relationship with
schooling’s outcomes. Thus, finding a turning or trade-off point between hours worked of
child labour and their human capital formation is rather simple as given in the empirical
model. The first order condition with respect to hours worked suggests that the trade-off hours
118
is about 19 hours per week for the age group 6-9, about 22 hours per week for age group 10-
12, and about 25 hours per week for age group 13-14. These findings essentially have implied
that Cambodian children could possibly work within these threshold level hours per week
without having negative impact on their human capital formation. Nevertheless, working
beyond theses threshold hours, child’s hours worked will have negative effects on their
schooling. For example, the finding of this research gives policy implication particularly for
targeting working child at the age of 14 years old in rural area whose working hours exceed
the threshold level of 25 hours per week (see Table 4.3). The results also imply that the
compulsory education, especially the junior high school, will be effected if these rural children
of the age group 14 years old continue working more than a threshold level.
Table 4.3: Hours Worked of Children in the Past 7 Days by Age, Sector and Sex (Economic
and Non-economic Activity)
119
Table 4.4: Regression Coefficient Estimates of ‘Schooling Attainment Relative to Age: SAGE
Index (Sample Consists of Children Aged 6-14 Years Old)
Instrumented: Hours worked, Hours worked^2, Hours Worked by Other Age Group, Poverty status
Instruments: Cart, Motorboat, Boat, Bicycle, Cyclo, Motorbike, Tractor Truck,
Car, Fridge, TV, and all exogenous regressors in the model.
Wald test of exogeneity: Prob (Chi) = 0.000
Obs. summary: 7318 left-censored observations at SAGE<=0
120
Dependent Variable is SAGE (OLS) SAGE (IV-Tobit)
“SAGE” (a) (b)
Coefficient P-value Coefficient P-value
12257 uncensored observations
Tests of endogeneity of: Hours worked, Hours worked^2, Poverty status
H0: Regressors are exogenous
Wu-Hausman F test: 87.35255 F(3,19549) P-value = 0.00000
Durbin-Wu-Hausman chi-sq test: Prob (Chi) = 0.00000
Sargan statistic (overidentification test of all instruments): Prob (Chi) = 0.1988
Source: Author’s calculation
(b) Report of goodness of fit of IV-Tobit: Two-step tobit with endogenous regressors
Number of obs = 19575 Wald chi2(22) = 2777.58 Prob > chi2 = 0.0000
Table 4.5: Regression Coefficient Estimates of Years of Child’s Schooling (Sample Consists
of Children Aged 6-14 Years Old)
121
Dependent Variable is Schooling (OLS) Schooling (IV-Tobit)
“Years of Schooling” (a) (b)
Coefficient P-value Coefficient P-value
School’s expenditure .0012642 0.000 .0008782 0.000
Father’s education .2249461 0.000 .275552 0.000
Mother’s education .1888585 0.000 .2199084 0.009
Nb. Babies -.2066663 0.000 -.2089925 0.000
Nb. Children -.0775353 0.000 -.095823 0.020
HH size .0113688 0.056 -.0168188 0.531
Gender of HH -.1905025 0.000 -.691658 0.000
Age of HH .0086746 0.000 -.002371 0.549
Uses piped water .1410249 0.000 .1034266 0.106
Electricity .6145574 0.000 .3550592 0.038
Telephone .0630423 0.000 .110706 0.003
Rural -.0197443 0.433 -.3567143 0.008
_cons -3.772771 0.000 -3.637413 0.117
Instrumented: Hours worked, Hours worked^2, Hours Worked by Other Age Group, Poverty status
Instruments: Cart, Motorboat, Boat, Bicycle, Cyclo, Motorbike, Tractor Truck,
Car, Fridge, TV, and all exogenous regressors in the model.
Wald test of exogeneity: Prob (Chi) = 0.0000
Obs. summary: 3639 left-censored observations at ChSchool<=0
15936 uncensored observations
Tests of endogeneity of: Hours worked, Hours worked^2, Poverty status
H0: Regressors are exogenous
Wu-Hausman F test: 138.12831 F(3,19549) P-value = 0.00000
Durbin-Wu-Hausman chi-sq test: Prob (Chi) = 0.00000
Sargan statistic (overidentification test of all instruments): Prob (Chi) = 0.2151
Source: Author’s calculation
(b) Report of goodness of fit of IV-Tobit: Two-step tobit with endogenous regressors
Number of obs = 19575 Wald chi2(18) = 6576.62 Prob > chi2= 0.0000
122
4.5. Conclusion
This study uses dataset of Cambodian Child Labour Survey 2001/02 (CCLS-2001/02)
to investigate the trade-off between child labour and their human capital formation, as
“Child’s activities” included the possibilities of child combining work with school or doing
neither. It also investigates the determinants of child schooling and that of the income earnings
from child labour. The estimation of child labour’s earnings turn out to be important if child’s
earnings will endogenously help smoothening households’ consumptions, and thus human
capital of the child could be maintained if child doing combined work with study. The results
(1) Even though they are child labour, their education play significant role in their
wage rate determinant. We found that, in general, the rate of return to education of child
labourers is about 12 percent (equivalence of 0.62 USD per week) considerably high. This rate
of return is comparable to the Mincer’s model estimation of American adult labour of non-
farm men in 1954 (see Mincer, 1974). This result indicates the importance of logic behind the
because child’s education even though among the child labourers are proved to be significant
in generating incomes, thus help parents to invest more on their children’s education.
outcomes of their children. However, this education has a positive impact only for parents
whose education had at least completed primary school or higher. This finding provides
consistency in the earlier works though the yardstick of measurement may have different
techniques such as allowing a unit change in parents’ schooling rather that a completion of
primary, or secondary school. The similarity to Ray and Lancaster (2005), Rossati and Rossi
(2003), Ray (2000), Deb and Rosati (2004), Blunch et.al (2002), Bhallotra and Heady (2003),
123
and Khanam (2003), is that there exists positive link between parent’s education and the
likelihood of a child attending school, and similarly a negative link between parent’s education
(4) The average hours worked of Cambodian children stay below threshold level hours.
This implies that child labour is rather increase human capital formation of children as long as
their hours worked will not exceed the threshold level. The specific thresholds for the age
group 6-9 and 10-12 are 19 and 21 hours per week respectively. All working children in this
age group have their average working hours below the threshold level. This finding tends to
reinforce the theory of Fan (2004), which states that a small increase in child labour may not
be trade-off with human capital investment, since the positive impact of increased financial
resources on education may outweigh the negative impact of reduced time of study. This is
simply that children’s labour market participation raises the financial resources and spent on
their education.
(5) The older age group 13-14 has the threshold hours of 25 hours per week. One of the
warrant attentions is that children in rural areas in the age group of 14 years old tend to work
more hours exceeding the threshold level. This basically means that this study contributed to
the policy implications, and thus helps children to achieve their compulsory education by
looking their each specific constraint and find the way to solve it from its roots. By
establishing the threshold level hours, it enables involved agency to set laws on the hours
(6) The presence number of babies and children in each household do have negative
because Cambodian has a large family size (see MoP 1999) and the older child will look after
the younger one which ultimately will reduce time of schooling. As also observed during the
field survey in 2006 (see, chapter V), it is very natural that older children spent much time to
124
(7) Better social infrastructures are shown in this study to be important characteristics
to reduce child labour and increase the probability that children attend school or combine work
with school. This proved to be valid elsewhere as well, for example, the study of Chao and
Alper (1998) found in the case of Ghana that school participation rate is closely related to
distance to primary school, access to drinking water and roads. More obvious is that the
improved school infrastructure has impact on the school enrolment rate (see, Chapter II).
Regardless of the above mentioned contributions to the determinants of child labour and child
schooling, there are many other factors that could also explain the outcomes of child labour in
the context of Cambodia. Those factors are social dimensions, which related to concepts,
environment, cultural and norms which stay out of our control. However, this study captures
only major factors which give possibility for the policy intervention, and could possibly give
125
C
Chhaapptteerr A
Appppeennddiixx IIV
V::
M
MUULLTTIIN
NOOM
MIIA
ALL LLO
OGGIITT M
MOOD
DEELL O
OFF C
CHHIILLD
DAAC
CTTIIV
VIITTY
Y
Among the three types of dependent variables of learning measure, there is only
variable “child’s activities” is a categorical variable consists of “child works only, child
combine work with study, child studies only, and child idles). Therefore, an appropriate model
for the estimation is either Multinomial Probit or Logit function. The estimates of multinomial
logit model takes into account of the possibilities of the child combining work with school or
doing neither. Therefore, child’s activities here consist of four categories denoting by:
“j=0 if the child work only” as the base/reference ( j-1) category, so we can express the
probabilities and the likelihood function of the Multinomial Logit model as follow:
exp(β j xi + zi )
P( yij = j xi ) = Pij = for j≠0 Eq. (5)
1 + ∑ j =0 exp(β j xi + zi )
j −1
1
P( y0 = 0 xi ) = Pi 0 = Eq. (6)
1 + ∑ j =0 exp( β j xi + zi )
j −1
n j −1
log L = ∑∑ yij log Pij Eq. (7)
i =1 j =0
where :
126
And y ij is the polytomous or set of dummy variables (y ij =1 if the ith individual falls in
the jth category, and y ij =0 otherwise). β j is the covariate effects of response categories of the
child’s activities. The vector xi denotes observed individual and exogenous variables that
include (a) child characteristics such as age and female child (b) household’s characteristics
such as poverty status and hours worked had been instrumented on set of household assets as
in the above procedure in equation (2&3), parents’ education, and gender of the household
head; and (c) the vector of Z i is represented by community characteristics such as access to
clean water and sanitation, hygienic latrine, telephone and rural or urban. Although there are
more community characteristics that may affect the utility of the parents such as pagoda, water
irrigation, road, access to market and others, but we may not include these into our model due
to data limitation.
conditioning on x variables (Wooldridge, 2003). Beside that, the multi-co linearity has been
checked by the correlation matrix of covariates and its results confirmed that there was no any
independent variable that might cause suffers to our model. However, we know that the
problem of endogeneity is the most important to our model if we are not carefully to select and
understand the nature of these covariates. One of the possible endogeneity is the poverty status
that one may imply that the wealthier households could be expected to induce the households
not to send their children to work. However, the problem is that the households are able to
sustain a relatively high households’ income per capita because they send their children to
work. In fact, this is the case that leads to biased estimates and also on the estimates of the
other parameters of the model. Therefore, the instrumental variable on poverty status and
hours worked were done simultaneously to get the fitted value of these variable before
Based on previous studies on the model of child’s activities such as Ray and Lancaster
(2004), Ray (2000), Blunch, Canagarajah, and Goyal (2002), Khanam (2003), they neither
127
discussed on this possible endogeneity. This is perhaps (i) due to the lack of available data for
program that can command the simultaneous equations of endogeneity of multinomial logit at
once without having problem of larger standard error, however, we can solve it by two stage
processes, but it may cause the downward biases due to larger standard error occurs, and (iii) it
may not necessarily to have exact partial effect since Multinomial Logit results place emphasis
on the direction of its magnitude rather than the coefficients, therefore, knowing the direction
is sufficient for policy implication (Wooldridge 2003). Given the complexity and difficulties
to find appropriate instrument variable. Thus, I presume, they tend to relax this assumption. It
is the same way that Heady, C (2000), in his study on Ghanaian data, recognized the
endogeneity of child labor’s hours on the child’s educational outcome, but does not tackle it in
the estimation. Our model regards the covariate of poverty status and hours worked as crucial
to understanding the major part of the cause of child labor, and many studies did support this
hypotheses including the recent study by Nagaraj (2002) who found that the number of
children aged 5-14 attending school rises monotonically as monthly household expenditure
rise from less than 120 rupees to 455-560 rupees in Tamil Nadu and Karnataka, India. Other
covariates also help to explain the phenomena. Age of the child is included to capture parental
behavior of allocating their children’s activities. It was motivated by the premier believe that
Table Annex 4.1 shows result of Multinomial Logit estimates of child’s activities in
order to take into account of child combining work with school or doing either (coefficients
with robust standard errors, and marginal effects). In Table Annex 4.1, the variable “Female
child” is negative and statistically significant in child’s category of “work only”, “combine
work and study”, and “idle”. This essentially implied that being a female child tend to be
128
working rather than schooling. The marginal effects show that being a female child is
associated with almost 1 percent point decrease in the probability that child is doing “combine
work and study”, a 2 percent point decrease in the probability that child is doing “study only”,
and a 1 percent point decrease in the probability that child is “idle”. The marginal effects in
Table Annex 4.1 indicate that being in rural area increase the probability that child is likely to
work only by 6 percentage points. The probability that children are likely to attend school only
or combine work with schooling is higher among the non-poor households. This does not
mean that the poor households did not send their children to school. However, the evidence by
in Chapter III and the Child Labour Survey Report (2002) show that most children in
Cambodia like to combine both work and study, which indicated that majority of Cambodia
population live around poverty line. This finding is also consistent with the previous
theoretical and empirical works such as the work of Basu and Van (1998), Rosati and Rossi
(2003), Ray and Lancaster (2005), Basu and Tzannatos (2003), Lee and Westaby (1997),
Saupe and Bentley (1994), Kim and Zepeda (2004), Chakraborty and Das (2004), Grootaert
and Ravi (1995), Chao and Alper (1998), Duryea and Mary (2001), Basu, Arnab K., and
129
C
Chhaapptteerr A
Appppeennddiixx IIV
V::
T
TAAB
BLLE
ESS
Table Annex 4.1: Multinomial Logit_ Maximum Likelihood Estimate of Child’s Activities
Hours worked* .3830596 0.000 .2769782 0.002 -.083125 0.364 -.062138 .0811754 .0053853 -.024422
Hours worked ^2* -.0073907 0.000 -.0058916 0.001 .001363 0.453 .0012213 -.0015215 -.000158 .000459
Poverty status* 2.01048 0.000 2.475953 0.000 -.726529 0.065 -.352566 .373302 .1376232 -.158358
School’s expend .0006806 0.000 .0006892 0.000 -.000399 0.003 -.000111 .0001422 .0000325 -.000063
Age of child 1.291402 0.000 .6681355 0.000 -.886815 0.000 -.188404 .3157251 .003994 -.123326
Age of child ^2 -.0590459 0.000 -.0522285 0.000 .0157566 0.001 .0098198 -.0120305 -.001852 .004063
Female Child -.0660707 0.071 -.2400699 0.000 -.212327 0.000 .0215472 .0075823 -.018891 -.010237
Father’s education .1556989 0.001 -.026093 0.691 -.036983 0.589 -.020823 .0417756 -.012164 -.008787
Mother’s edu .2794888 0.000 .2682861 0.000 -.196931 0.025 -.044140 .0582648 .0114643 -.025588
Nb. Babies -.0797959 0.005 -.3582614 0.000 -.242817 0.000 .0279159 -.0129614 -.030101 -.010776
Nb. Children -.1116291 0.000 -.0866333 0.005 .1072337 0.001 .0165663 -.0267315 -.002949 .0131151
HH size -.0111494 0.333 .0608551 0.000 .0751641 0.000 -.005102 -.00421 .0049538 .0043591
Gender of HH -.1275475 0.030 -.0958298 0.268 -.103377 0.234 .0230515 -.0210725 -.000838 -.001140
Age of HH .0075415 0.001 .0109649 0.000 -.000262 0.934 -.001427 .0011916 .0006687 -.000433
Uses piped water .1230068 0.001 .259 0.000 -.244154 0.000 -.030993 .0040234 .016985 -.009984
Electricity .4925001 0.000 .9701896 0.000 -.315162 0.000 -.104885 .0412805 .0712119 -.007606
Telephone .0235106 0.354 .0093739 0.769 -.060149 0.101 -.002424 .0077542 .0000405 -.005370
Rural -.3754365 0.000 -.1626368 0.010 .103996 0.085 .0623894 -.0790758 -.007242 .009444
Constance -9.959484 0.000 -6.147937 0.000 6.658385 0.000
Note: * Variables hours worked and its squared and poverty status are instrumented as above. The residual of
the 3 assumed endogenous variables are plugged into the multinomial logit, and they are all significant at 1
percent level in the two categories (see Wooldridge, 2003, pp. 506). Note also that there is no test of endogeneity
for the multinomial probit or logit function provided by the statistical software. This is done by manually.
Number of obs = 21153 Wald chi2(54) = 6649.92 Prob > chi2 = 0.0000
Pseudo R2 = 0.2983 Log pseudolikelihood = -20915.219
130
Table Annex 4.2: Matrix of Correlation of Independent Variables of Model Specification 1
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
(1) Earning 1.0000
(2) Experience 0.0933 1.0000
(3) Experience^2 0.0728 0.9344 1.0000
(4) Child’s edu 0.0387 0.2234 0.1788 1.0000
(5) Female child -0.006 -0.004 -0.004 0.0312 1.0000
(6) Piped water -0.015 0.0088 -0.003 0.0928 -0.002 1.0000
(7) Electricity -0.072 -0.027 -0.042 0.2715 0.0043 0.1533 1.0000
(8) Telephone -0.055 0.0092 0.0205 0.0500 0.0147 0.0418 0.1894 1.0000
(9) Rural -0.004 0.0308 0.0280 -0.116 -0.000 -0.062 -0.390 -0.122 1.0000
(10) Enrolment -0.265 -0.011 -0.012 -0.048 -0.055 -0.035 0.0117 -0.009 0.0347 1.0000
(11) Nb of infants 0.0277 -0.071 -0.059 -0.161 -0.000 -0.026 -0.088 -0.000 -0.052 -0.024
(12) Nb children 0.0684 -0.016 -0.017 -0.009 0.0030 -0.036 -0.010 0.0747 -0.119 -0.040
Continued:
(11) (12)
131
(11) Nb. 1.0000
children
(12) HH size 0.7039 1.0000
(13) Gender of 0.1389 0.1677 1.0000
HH
(14) Age of HH 0.0608 0.3399 -0.089 1.0000
(15) Uses piped -0.089 -0.057 -0.014 0.0113 1.0000
water
(16) Electricity -0.099 -0.008 -0.043 0.0934 0.2082 1.0000
(17) Telephone 0.0793 0.0645 0.0148 0.0235 0.0576 0.2147 1.0000
(18) Rural -0.095 -0.186 0.0053 -0.117 -0.093 -0.396 -0.130 1.0000
Note: The coefficients of the matrix correlations indicate that the model is secured from the multi-co linearity.
Technically, if coefficient is greater than four and smaller than eight, one can draw assumption that there is weak
correlation, but it does not suffer the model. However, if the coefficient is greater than eight, one shall omit that
variable or combine both variables into one.
Table Annex 4.4: Simultaneous Estimates of Earning Income and Hours Worked of Children
in the Past 7 Days (Log Earnings in the past 7 Days)
132
C
Chhaapptteerr V
V
H
HEEA
ALLT
THHE
EFFFFE
ECCT
TSS O
OFF C
CHHIIL
LDDL
LAAB
BOOU
URRA
ANND
D IIT
TSS D
DEET
TEER
RMMIIN
NAAN
NTTSS::
E
Evviiddeennccee aanndd PPoolliiccyy IIm
mpplliiccaattiioonnss D
Drraaw
wiinngg ffrroom
m aa SSttuuddyy ooff FFoouurr V
Viillllaaggeess,, C
Caam
mbbooddiiaa
5.1. Introduction
domestic household chores. Their daily life is simple, helping their parents with various
activities ranging from raising seedlings to harvesting during the agricultural season. Besides
that, children also spend time tending to animals after school, collecting firewood, and
catching fish and other animals in the local area. At a glance, children have a significant role
in the household economy since their labour has a direct impact on household welfare.
Agricultural child labourers are viewed by their community as good children who understand
the difficulties faced by their parents in day to day life. Given the current rural livelihood in
Cambodia where subsistence farming is prevalent, it is usual for children to interact with the
agricultural environment in which they live. Therefore, determining the effects of such
agricultural activities on children’s health is an important factor for policy makers to integrate
into the overall framework of poverty reduction in Cambodia, as well as to achieve the
In recent decades, there have been an increasing number of studies on the impact of
child labour on human capital formation. However, only a few of the existing studies connect
child labour with child health issues. Agricultural child labour, even though it is widely agreed
to be one of the most dangerous, employs a much larger number of children than other sectors
(USGAO, 1998). This activity is consistently ranked amongst the most hazardous industry for
mortality and morbidity (Cooper and Rothstein, 1995). Agricultural child labour is not only a
133
problem in developing countries, but is also very evident in developed countries like the
United States of America, where millions of adolescents work to earn money, gain
responsibility and independence, and help their parents financially. There is a rich body of
literature on agricultural child labour in developed countries, and this reflects the centuries of
child labour in Europe from the seventeenth to the twentieth century (Rahikainen, 2004), and
in American history (Hindman, 2002). The impact of child labour in developing countries,
especially the “rural agricultural child labour” that seems to have been ignored in recent laws,
on the long term health of children is underestimated due to the neglect and acceptance that by
working in agriculture, children will inherit know-how from their parents and thus prepare
themselves for adulthood (Satz. D, 2003). In fact, most studies investigating the impact of
child labour on health assume that such impacts are only negative (Guarcello et.al., 2004). In
the ILO reports in 1998, it was shown that worldwide, children are being exposed to hazards
in their work environment, such as contact with toxic pesticide, lifting heavy loads, operating
machinery without appropriate training, being exposed to strong sunlight, having a lack of
As is often the case, agricultural child labour may involve longer hours from a younger
age. In India, bonded child labourers as young as eleven often work sixteen or seventeen hours
a day (HRW, 2002). In the recent study of Guarcello et.al., (2004) in the case of Cambodia,
Bangladesh and Brazil, the causal link between hours of work and ill-health shows that
number of hours worked exerts a significant effect on the probability of negative health
outcomes. Although agricultural child labour remains widespread in the developing world, and
even though some children are being exposed to hazards at work, there seems to be a
consensus that such work can have positive effects on children through developing discipline,
responsibility, self-confidence and independence, and by teaching children how to inherit the
agricultural skills from their parents. On the other hand, such child labour can also have a
negative impact on health and development of children (Parker, 1997; ILO 1998). While
134
health and well-being of all age groups is important, the developmental nature of childhood
and adolescence provides the foundation for adulthood. Rapid growth and development, such
as changes in body proportions, height and weight are indicators to monitor growth patterns
(DHHS, 2006). O’Donnell et al (2003) have found little evidence of the contemporaneous
impact of child work on health in the case of Vietnam’s agricultural child labour, but work
undertaken during childhood raises the risk of illness up to five years later and the risk
increases with the duration of work. However, Fentiman et al (2002) have found no growth
differences in rural Ghana between child enrolled and not enrolled in school, assuming that the
Several studies have documented the positive impact of parental education on child
height (Cochrane et.al, 1982; Jere and Anil, 1990). Many other studies have argued that
education has no direct effect on height, but is a proxy for unobserved background variables
such as underlying maternal health (Wolfe and Behrman, 1987). More explicitly, Thomas et al
(1991) found that almost all the impact of maternal education can be explained by indicators
of access to information- reading papers, watching television, and listen to radio. In Ghana,
Blunch (2004) has found that parental literacy and numeracy skills have positive effect on
intermediate (pre- and postnatal care and vaccinations) and final child health outcomes (illness
and mortality). Since mothers often spend more time with their children than fathers, the
mother’s education level will have a greater effect on the child health. Empirically, Maitra
(2004) has found that a woman’s education has a stronger effect on health care usage relative
to that of her husband and a woman’s control over household resources (ability to keep money
aside) has a significant effect on health care usage. Glewwe (1999) found that mother’s health
Unlike the previously mentioned studies, this study is the first to investigate the health
effects of agricultural child labour, a subject hitherto virtually untouched in the literature on
135
child labour. This study is also special in the sense that a field survey was conducted in 2006 31
to fill some gaps mentioned by earlier studies. For instance., Guarcello et.al (2004) although
height and weight standardized for age and sex (BMI-age), they did not have such data, and so
used data on self-reported illness and injuries as a proxy. Despite the common understanding
on the existence of the trade-off between child health and hours worked by children, this study
has established the trade-off hours (threshold level hours), which shall facilitate policy-makers
when drawing up new laws or regulations for child labour. This study finds that agricultural
child labour does have an effect on health, if working hours exceed the threshold hours (15,
17, and 21 hours per week for the age group 5-9, 10-12, and 13-14 respectively). Currently,
the results suggest that Cambodia agricultural child labour has yet to be a concern since the
average child’s working hours is just 8.5, which is well below the threshold level. However,
there is some small proportion of children whose hours worked is above the threshold level,
drawing attention to the policy implications of protecting children from working longer hours.
Other salient determinants suggest that child health is strongly affected by access to media
such as listening to radio, watching television, and reading newspaper. It is surprised that the
role of parents’ education does not have effect on child health. The study also suggests several
other determinants of child health including the role of social capital and public facilities.
Finally this study contributes to the growing literature on child labour and health effects, and
sheds light on policy implications toward protecting agricultural child labour in Cambodia.
This study aims to provide evidence of agricultural child labour in rural Cambodia and
its consequences on health. It is accepted that agricultural child labour has a negative impact
on health, if children are exposed to hazard condition, as outlined by the ILO (1998). However,
agricultural child labour in the context of Cambodia remains a traditional practice where child
31
The field survey conducted in 2006, is led by professor Seiichi Fukui. under the financial support from the
project of New Initiative toward Global Academic Collaboration, Graduate School of International Cooperation
Studies, Kobe University.
136
involvement in the household economy and their daily labour is viewed by villagers as light
work, and necessary substitute for adult labour. Because of the reality of the effects of work on
child health is basically unknown; this study will provide the following investigations, with
(1) Investigate the determinants of child health, and whether the current hours worked
(2) If the current child labour participation in agriculture has negative effect on health,
this study aims to establish an average threshold level of hours for agricultural
child labour so that their involvements in agricultural activities will not adversely
(3) Undertake policy analysis based on evidence emerging from this study to come up
This paper is organized as follows. Section 5.2 presents the overview of the villages
surveyed. Section 5.3 presents the conceptual and empirical framework. The conceptual model
is built from the household utility model to understand the allocation of resources to child
health within a household that faces budget constraints while maximizing their utility. Based
on the theoretical model, the empirical framework is developed, so that results can be
observed empirically. Section 5.4 discusses the results of estimation. In this section each
salient determinants of child health is interpreted according to the level of significance and the
direction of each variable. Section 5.5 is the conclusion. Finally, Chapter Appendix V consists
of Tables and Graphs. Tables Annex 5.1 and Annex 5.7 contain tables of correlation among
covariates used in the specification function. Tables Annex 5.2 to Annex 5.5 provide the
estimated values of variables “watching TV”, “listening radio”, “reading newspaper”, and
“total household income”. Table Annex 5.6 gives the results of the first component analysis
for the variable “social capital” used in the model. However, the structure of estimation
137
accounts automatically for the instrumental variables without making a two-step estimation of
IV. The estimated value is merely shown to indicate the suitability of the estimated values
used in the structural model. Graphs 5.1 to 5.4 provide BMI-age and Stature-age for children
Cambodia has a total of 13,406 villages, 1609 communes, and 183 districts in 24
provinces (MoI, 2007). One important demographic characteristic of the communes and
villages of Cambodia is the large proportion of female-headed households, even though the
attempt to assess the socio-economy of rural Cambodia, a study led by Professor Seiichi Fukui
survey of rural households in four villages in Kampong Spueu and Takeo Provinces. These
four villages were selected based on the recommendation of the JICA supported Rural
Development and Resettlement Project (RDRP). It is worth explaining the background of the
projects in these areas of study. The beginnings of this project back in time to the first
UNTAC- run election in 1992, during which many returnees were to be reintegrated into
society. The project started from 1992 to March 2004, with the support from many Asian
experts from Japan, Indonesia, Malaysia, Philippines and Thailand. The provinces of
Kompong Speu and Takeo were chosen as project areas because of relatively short distance
(50 to 100 km) and easy access from National Road No.3 and 4 (RDP, 2004).
Given the above background information of the research sites, the village socio-
economic survey was conducted to explore and understand the principal causes of rural
problems such as poverty, vulnerability and child health. The survey employed a random
sampling method to select the sample from the village household list. The total sample is 168
138
Korm, and 36 in Trapeang Kroloung village. Prior to the village survey, informal discussions
were held with staff from the Ministry of Rural Development, UNDP’s staff and the four
households and individuals. This is the primary field survey data 32 of the socioeconomic
household consumption, migrations, agricultural child labour, child health and social capital.
The 168 households interviewed consisted of 214 children aged 5-14 years old, broken down
into 107 female and 107 male. The results of the survey show that the majority of children in
our study neither use pesticide nor operate machinery. Out of the total children in our sample,
99 children are non-working child and 115 working child. The study adopts the definition of
working child according to the ILO’s convention No. 138. In this context, any children
involved in agricultural activities even if for only 1 hour in the past 7 days would be regarded
as a working child.
Table 5.1 presents background information of the villages surveyed. In general, people
in rural areas, including the people in this sample of the survey, make their living from
agricultural cultivations; in which rice is the main crop and vegetables are a supplementary
source of income. Vegetables grown include pumpkin, sponge gourd, green pepper, cabbage,
tomato, cucumber and taro. Cows and bullocks are considered to be the big assets of the
household as these draught animals provide a variety of purposes for farmers to cope with
seasonal shock. Farmers will only sell their cow or bullock when there is an unexpected need
for a large expenditure such as sickness, debts, and school expenses for children who need to
move to an urban area, a marriage, a death, or other traditional ceremony. Poultry such as
32
The field survey was conducted from September 15th to October 4th 2006, under the financial support of the
project of New Initiative toward Global Academic Collaboration, Graduate School of International Cooperation
Studies, Kobe University. The survey team consists of several doctoral students including the author of this
paper, under the direction and guidance of Professor Seiichi Fukui.
139
chickens and ducks have traditionally been raised for domestic food consumption or for selling
A large family size, low income and poor infrastructure are all characteristics of the
life of the rural poor. However, the socio-economy of Kol Korm and Kan Damra villages are
better compared to the villages of Prey Changva and Trapeang Kroloung. About 30 percent of
arable land in Kol Kom village is irrigated by a reservoir which has the capacity to provide
water for farmers to cultivate in both seasons, and thus it is the area where vegetables and
fruits are grown. The village of Kan Damra is observed to have better land quality and a larger
population of cattle compared to other areas in the survey. Regardless of these reasons, human
factors are also important in explaining why Kol Korm and Kan Damra villages are better-off
Table 5.1 has highlighted some general situation of the four villages in the past three
years (note that the star “*” indicates the source of information when the situation of the four
villages were being evaluated in 2004). The figures show that Kol Korm and Kan Damra were
not any better-off at all compared to the Trapeang Kralong and Prey Changva villages.
However, the recent field socio-economic survey in 2006 indicates that Kol Korm and Kan
Being rural children, various tasks are assigned from a young age both at home and in
the fields. Their parents are too busy with their day-to-day activities to ensure that children go
to school (RDP, 2004). The very low salary of teachers forces them to supplement their
income so as to support their families. Generally, the health condition has improved over the
past decade. The commune health posts are located around 2-5 km from each village. There
are four possible health care options available to the villagers in the survey areas. The first
option is Kru Khmer who is a traditional Khmer medicine practitioner using herbs and
spiritual rites. The second option is a private clinic run by a person who may or may not have
a basic degree in medicine. The third option is the commune health centres or the government
140
hospitals in the two provincial capitals. The fourth option is seeking medical advices and
Village development is important for villagers. However, the village chief plays a
crucial role in the process of public works and other development activities. It is also worth
explaining the role of the village chief in development. Given that “he and very rarely she” is
the village chief who has been the village chief for decades. The complexity and the rapid
change in the planning system from the central command to decentralization have made many
of the village chiefs confused about their role. Usually, the village chief controls everything in
the village and his decision is made without any discourse with ordinary villagers. However,
this is not always the case. For example, in Kol Kom village, the chief who is the most literate
among those in the survey areas is also the most energetic in terms of participatory village
development. Her commitment to the village has made the entire village better off compared
to other villages. Therefore, the quality of village is also attributed to the quality of the village
chief.
Table 5.2 shows the prevalence of children that are involved in operating machinery
and/or applying pesticide. In the sample of this survey, 88.79 percent of children are neither
using pesticide nor operating machinery. Only 11.21 percent of children are in the high risk
group, whereby they are involved in operating machine or applying pesticide, or both. Table
5.3 presents the number of working children by sex. The figure shows that females account for
Table 5.4 presents Body Mass Index (BMI) of children by sex. The figure shows that
46.73 percent of the children are considered as malnutrition, which draws a concern of their
health, and amongst which 52.34 percent is male and 41.12 is female child.
141
Table 5.1: Basic Characteristics of the Villages Surveyed
Characteristics Kralong
Female-headed households* 9 15 33 13
poorest *
Landless households* 1 4 4 8
* UNOPS: the project evaluation report on the Rural Development Project, 2004
Table 5.5 presents BMI for the four villages of the study. The figures show that 35
percent of children in Prey Changva village are in the category of malnutrition, 31 percent in
142
Table 5.2: Frequency, Row and Column Percentage of Children Aged 5-14, by Categories of
Apply pesticide 20 4 24
83.33 16.67 100.00
9.52 100.00 11.21
Table 5.3: Frequency, Row and Column Percentage of Children Aged 5-14, by Child Labour
and Sex
143
Table 5.4: Frequency, Row and Column Percentage of Children Aged 5-14, by Sex and BMI-
Male 56 47 4 0 107
52.34 43.93 3.74 0.00 100.00
56.00 43.52 100.00 0.00 50.00
Female 44 61 0 2 107
41.12 57.01 0.00 1.87 100.00
44.00 56.48 0.00 100.00 50.00
Table 5.5: Frequency, Row and Column Percentage of Children Aged 5-14, by Village and
BMI-Age Percentile Category
Khan Damra 14 22 0 0 36
38.89 61.11 0.00 0.00 100.00
14.00 20.37 0.00 0.00 16.82
Kol Korm 20 36 2 1 59
33.90 61.02 3.39 1.69 100.00
20.00 33.33 50.00 50.00 27.57
Trapeang 31 23 2 0 56
Kralong 55.36 41.07 3.57 0.00 100.00
31.00 21.30 50.00 0.00 26.17
144
5.3. Theoretical and Empirical Framework
The theoretical model of this paper focuses on the health effects of agricultural child
labour and its determinants. As mentioned in the literature review, the outcomes of child
health result from multi-dimensional factors (Strauss and Thomas, 1998). In the context of
Cambodia, agricultural child labour is very common in rural areas. Most rural farms are
agricultural-based families. Unlike the mechanised agriculture of the West, children are
helping their parents with daily chores and learning their future job (Satz, 2003). Most
children of agricultural families are expected to help their parents in farm activities. Therefore,
this study develops a household model to understand the factors affecting child health and
health of children, H(.). In this context, health features directly as an argument in the utility
function of hours worked by children L, information received through media If, parents’
through media If is a function of inputs to collect information L p , social capital SC, and public
service PS. Simply, we can express the household utility and the demand function of child
health as follows:
If * = If * ( L p , SC , PS ) Eq (2)
At any given income level, households must choose how their resources will be spent
on household’s consumption expenditure and child health. Therefore the household budget
constraint is:
C p + x = I p + WC L ; or C p = I p + WC L − x Eq (3)
145
where I p is parents’ income and WC L is child’s income. The maximization utility of
each household in equation (1) is subject to the budget constraint in equation (3). Therefore,
the First Order Condition (FOC) with respect to L, L p , x is simply seen as decisions made by
each parent concerning how much to invest in child health, considering other given constraints
in the households. The FOC gives the following interior solution to our problem.
∂U ∂U ∂U dH
= .WC + . =0 Eq (4)
∂L ∂C p ∂H dL
∂U ∂U ∂H dIf
+ . . =0 Eq (5)
∂L p ∂H ∂If dL p
∂U ∂U ∂U dH
= (−1) + . =0 Eq (6)
∂x ∂C p ∂H dx
Rearranging equations (4), (5) and (6), can express the following results for
interpretation.
∂U dH ∂U ∂U ∂U dH
= −WC . ; or = −WC . / Eq (7)
∂H dL ∂C p ∂H ∂C p dL
∂U ∂H dIf ∂U
. . > 0 ; and <0 Eq (8)
∂H ∂If dL p ∂L p
∂U ∂U dH ∂U
= . = Eq (9)
∂x ∂H dx ∂C p
∂U
From equation (7), it can be said that is the household utility maximization with
∂H
respect to child’s health. The static equilibrium for the interior solution of child’s health is a
trade-off between the utility maximization with respect to consumption and the utility
maximization with respect to hours worked. In this respect, we can interpret that the longer
hours worked by children will trade-off with child health. Therefore, the best equilibrium is
the Pareto-efficient which means that children’s health is good as long as they are working
146
∂U ∂H dIf
From equation (8) it can be seen that . . is the utility maximization with
∂H ∂If dL p
∂U ∂H dIf
respect to input of collecting information. In this respect, . . > 0 if parents get
∂H ∂If dL p
∂U
access to media; otherwise < 0 if parents do not have access to media.
∂L p
∂U dH
From equation (9) we can interpret that . is utility maximization with respect
∂H dx
to health’s expenditure in order to increase child’s health. This maximization is the trade-off
Finally, we can solve the problem of our objective function through the following
system of equation (10), (11) and (12), in that if any known initial value is given, then the
L* = L* ( X , Z , SC , WC ) Eq (10)
L p = L p ( X , Z , SC , WC )
* *
Eq (11)
x * = x * ( X , Z , SC , WC ) Eq (12)
Therefore, the objective function of the human capital of the child, particularly the
H * = H * ( L* , If * , x * ; X , Z ) Eq (13)
Solving the equation (13) will provide us the solution of the child outcome. The sign of
In this section we briefly discuss the objectives of the econometric analysis and
equations, and then discuss the parameters of interest along with the identification and
estimation strategy. The purpose of this study is to estimate the relationship between health
147
care characteristics and child health outcomes. There are, of course, many measurement of
each measurement has its own drawbacks (Strauss and Thomas, 1998). General Health Status
is probably the most widely used indicator in empirical literature of the United States.
Guarcello et.al (2004) however, recognized the importance of general self-health assessment
and anthropometrics measure of height and weight standardized for age and sex (BMI-age),
although, data constraints limited their analysis to indicators of self-reported illness and
injuries.
This study’s model of agricultural child labour and health effects employs two main
health indicators: General Health Status and BMI for age percentile. The General Health
Status indicator is the self-health evaluation where respondents provide an assessment on their
own health. The anthropometric indicator used in this study is the Body Mass Index for age
and sex (BMI-age). Although the BMI number is calculated the same way for children and
adults, the criteria used to interpret the meaning of the BMI number for children and teens are
different from those used for adults (DHHS, 2006). For example, the BMI age-and sex-
specific percentiles (see Chapter Appendix V, Graphs 5.1 to 5.4) are used for children and
teens to deal with the amount that body fat changes with age and the amount of body fat
The conceptual framework laid the foundations for the estimation of child health
outcomes based on their nutritional intakes, and personal and family characteristics. Among
our explanatory variables, it is suspected that the variables “Total Income” and “Hours
Worked” are endogenous with child health outcomes. Therefore, it is necessary to determine
for appropriate instrumental variables in the dataset that have a high correlation with these
endogenous variables, but that are not correlated with the disturbance term of the structural
equation.
148
This research is also concerned with Media effects (watching TV, listening to radio,
and reading newspaper) and father’s and mother’s education. This is simply because of
omitted variable bias where an important variable may inadvertently omitted. In this case, it is
doubtful that mother’s and father’s education have a direct effect on child health, but these
variables may operate through the access to media 33. With these notions in mind, regression
Because AgeG1 and AgeG 2 are the dummy variables corresponding to different age groups of
the child, and its value is strictly belongs to {0,1}, thus the square of itself will remain the
same. This notion can be written as: AgeG1 = ( AgeG1) 2 , and AgeG 2 = ( AgeG 2) 2 . So, we also
can write that: ( AgeG1.H i ) H i = ( AgeG1.H i ) 2 , ( AgeG 2.H i ) H i = ( AgeG 2.H i ) 2 Therefore, we
33
We estimate the probit maximum likelihood of the equation “access to radio, TV and reading newspapers”,
and get these fitted value to insert into our main equation of child health outcomes.
149
Healthi = β 0 + β1 H i + β 2 H i2 + β 3 AgeG1 + β 4 AgeG 2 + β 5 ( AgeG1.H i ) +
β 6 ( AgeG1.H i ) 2 + β 7 ( AgeG 2.H i ) + β 8 ( AgeG 2.H i ) 2 + β 9 Chi +
Eq.(3)
β10 Incomei + β11Watch _ TVi + β12 Listen _ Radioi +
β13 Re ad _ Newspaperi + β14 X i + β15 Z i + U i1
Because variables H i , Incomei , and “access to media such as watching television, reading
newspapers, and listening to the radio” are assumed to be endogenous to the model, which
means that H i , Incomei , and “access to media” are correlated with U i1 . Thus, the estimation
of equation (3) will be biased. Thus, we use instrumental models of the Two Stage Least
where the variables in equation (1) through (8) are H i (hours worked), Incomei (total
pump, sprayer, cart, and rake) and HousePi (imputed house price).
Finally the estimation of the structural equation (3) is simply the reduced form. The
150
Income Effect on Child Health Model
The inclusion of child labour hours, H i , and its square is designed to capture the trade-
off between labour hours and health of the child. Furthermore, this study is designed to
analyse the trade-off between hours worked and child health for children in different age
groups. To do so, as done by Wooldridge (2003) 34 , the model must be written with and
interaction between hours worked by children in different groups 35 and the dummy of
respective age groups. Obtaining this interaction by including the variable hours worked for
34
There are also occasions for interacting dummy variables to allow for a difference in slopes. The design to
capture the different slopes is given in details in Wooldridge (pp.233, 2003). He also explains how to do so in the
econometric model as explained in page 234, in the equation (7.17). However, a clear example is given in page
235, example 7.10.
35
According to Wooldridge (pp, 234, 2003), using the same techniques, we must write the model with an
interaction between hours worked by children and AgeG1, and AgeG2, by producing a other two new variable, in
this case, is hours worked by children in the AgeG1, and AgeG2 respectively, which are written as
( AgeG1.H i ) and ( AgeG 2.H i ) . Obtaining the interaction is easy and we shall not be daunted by the
variables ( AgeG1.H i ) and ( AgeG 2.H i ) , which is zero for any other children outside these age groups in the
sample, and equal to the level of hours worked by children in each respective group.
151
respective groups along with respective dummies is a relatively simple operation. In this case,
three age groups are classified with the reference interaction group of age group 13-14. These
groups are 5-9, 10-12, and 13-14 years old, and the respective hours worked such as variables
( AgeG1.H i ) and ( AgeG 2.H i ) (hours worked by children in the age group 5-9, and 10-12
respectively) and their squares are also used in the equations. If the coefficient of hours
worked and its squared are statistically significant with U-shape function, then the First Order
Condition (FOC) on (9), (10) and (11), with respect to hours worked, is to check the turning
point of the hours worked of children against child health outcome. However, since we allow
for different slopes 36 for different age groups. The trade-off hours worked by each group can
be derived as follows:
( β1* + β 5* )
• ( AgeG1.H i* ) = − is the trade-off between hours worked by children in
2( β 2* + β 6* )
( β1* + β 7* )
• ( AgeG 2.H i* ) = − is the trade-off between hours worked by children in
2( β 2* + β 8* )
β1*
• H = − * is the trade-off between hours worked by children in the age group
*
2β 2
i
to check as whether the suspected endogenous variables and their instrumented variables are
valid or not. In case, this model cannot pass these tests, the model can simply estimate our
structural equation (3) by standard probit model for the equation of “General Health
152
Assessment” and by OLS for the equation of “BMI for Age percentile”. The standard Probit
model of “General Health Assessment” is expressed through the latent variable as follow:
Healthi* = xi β + ε i
xi β = β 0 + β1 H i* + β 2 H i*2 + β 3 AgeG1 + β 4 AgeG 2 + β 5 ( AgeG1.H i* ) +
β 6 ( AgeG1.H i* ) 2 + β 7 ( AgeG 2.H i* ) + β 8 ( AgeG 2.H i* ) 2 + β 9 Chi +
β10 Incomei + β11Watch _ TVi + β12 Listen _ Radioi + Eq. (13)
β13 Re ad _ Newspaperi + β14 X i + β15 Z i
ε i xi ~ N (0,1)
Then the standard normal cumulative distribution function and the standard normal
density are:
(z)
1 ( z) 2
Φ ( z ) = ∫ φ (v)dv ; φ ( z ) = exp[− ]
−∞ 2π 2
s n
ln L = ∑ ln Φ{1 − ( xi β )} + ∑ ln Φ ( xi β ) ;
i s +1
N = 1..., s, ( s + 1)......n
153
5.3.3. Data and Variables
The variables used for the empirical analysis are given in Table 5.6. The variable
“Income” and “Hours worked” are suspected to have error terms correlated with “child health
outcomes”. Therefore these variables are instrumented based on sets of exogenous variables
such as “house price”, “livestock value”, “off-farm employment”, “farm land size” and
Lastly, the variable “Social Capital” is the explanatory variable. However, this variable
is generated through the first component analysis (PC1) based on the three questions asked
during the field survey. The first question is: “How many percentage of villagers participated
in cooperative works such as building public house, pagoda, or road in the village in the last
one year? And rate as follow: 3 if everyone (almost everyone), 2 if about half, and 1 if no one
or few”. The second question is: “If there is water supply problem or natural calamity in this
village, how many percentages of villagers will cooperate to solve the problem? And rate as
follow: 3 if everyone (almost everyone), 2 if about half, and 1 if no one or few”. The third
question is: “Please tell me whether in general you agree or disagree. If you lose a pig or a
cow, someone in the village would help look for it and would return it to you, and rate as
follow: 3 if agree, 2 if neither agree nor disagree and 1 if disagree”. The First Principal
Component analysis technique is used for data reduction (see, Stata9 Multivariate Statistics,
page 369). This technique helps to reduce the number of variables in the analysis by
describing a series of uncorrelated linear combinations of the variables that contain most of the
p
SC = VΛV ′ = ∑ λi vi vi'
i =1
154
λ1 ≥ λ2 ≥ ... ≥ λ p ≥ 0
The asymptotic distribution of the eigenvectors v̂i and eigenvalues λ̂i of a covariance
Girshick (1939). Implementing the theory of first principal component, we can generate a
Table 5.6: Descriptive Statistics (Sample Size of Children in Age Group 5-14 Years Old)
155
Variable Definition Mean Std. Dev.
Machine operation 1 if child operates machinery, 0 otherwise .0186916 .135751
Apply pesticide 1 if child apply pesticide, 0 otherwise .1121495 .3162902
Experience Number of year child in labor force 1.17757 1.895451
PreyChangVa Prey ChangVa village .2943925 .4568378
KhanDamra Khan Damra village .1682243 .3749424
KolKorm Kol Korm village .2757009 .4479145
Table 5.8 presents coefficients estimate of “General Health Assessment”. Table 5.8,
column (A), (B), and (C) presents the IV-Probit coefficients. It is assumed that total
household’s income and hours worked by children are endogenous to the model of child health
outcomes. In addition, we also assumed that the access to media is endogenous with the model
as well (for the estimated value, see Table Annexes 5.2, 5.3, and 5.4). Using Hausman-test, we
have enough evidence to reject our null hypothesis of exogeneity, and thus we tend to interpret
the result from IV-Probit coefficients estimate. Similarly, Table 5.9 shows the coefficients
estimate of “Standardized BMI for Age Percentile”. Table 5.9, column (A), (B) and (C)
present coefficients estimated by Two-Stage Least Squares (2SLS). The “Standardized BMI
for Age Percentile” equation passes the test of endogeneity. We, therefore, tend to interpret the
The results of empirical estimation are interesting for our hypothesis of child labour
and health effects. In Table 5.8, the coefficients of “hours worked by children and it’s
squared” are statistically significant. These findings essentially imply that agricultural child
labour, especially hours worked have an effect on child health if they work more than the
threshold level of hours (15 hours for children in age group 5-9, about 17 hours for children in
156
age group 10-12, and about 21 hours for children in the age group 13-14). However, in general,
the average working hours of children is about 8.5 hours per week, which suggests that the
present situation of child labour in rural Cambodia does not have a negative effect on health.
Of course, there is some small proportions of children whose working hours exceed their
respective threshold level, which could be important for policy intervention for that particular
age group.
shows that child breastfeeding is associated with an 11 percent point increase in the
probability that the child is healthy. However, breastfeeding here is defined as a child whose
mother provided breastfeeding from the first day of birth up to 12 months old. Breastfeeding
for longer period does not give any effect on child health in the context of rural Cambodia.
These findings are consistent with the observations of the field interviews with parents, i.e., a
quote from many mothers was “because of insufficient milk and nutrition, thus child was/is
given longer periods of breastfeeding, and at extreme case child was breastfed up to five
years”. Shah, Iqbal H & Khamma Jitendra (2002) indicated that breast milk alone does not
provide all nutrients needed by an infant older than 6 months of age. The behavioural research
suggests that by six months of age, the developmental readiness in infants to take semi-solids
as well as physiological evidence is mature enough to handle food. Therefore, the longer
Immunization: In Table 5.8 columns (C) and Table 5.9 columns (B), the variable
“immunization” is statistically significant. The marginal effect shows that a child immunized
against seven preventable diseases has a 21 percent point increase in the probability of being
healthy. This finding supports the evidence claimed by PATH (2004) that immunization has
been a relatively successful public health program in Cambodia contributing to improved child
health, which in recent years, has seen polio eradicated, and the incidence of measles sharply
157
reduced. Based on the field interview, all the four villages in the study have commune health
centres equipped with trained health staff and nurses. The CSES-1998/99 village data clearly
shows that child immunization coverage is much higher in villages that have public health
clinics than in those that do not have. There is other evidence such as that given by village
health volunteers that availability of health infrastructure has been gradually improving over
the past 10 years, and thus child health has improved significantly.
Age of the child: In Table 5.8, and Table 5.9, the coefficients of child age groups are
statistically significant. The negative sign on the children age group “AgeGroup1, and
AgeGroup2” simply mean that children in rural Cambodia do not have better health when they
are young, but health gradually improves as they grow up. This evidence is consistent with our
observation that during the past 10 years, there are lots of health and nutrition programme
implemented in the villages by the World Food Programme (WFP) as well as the Rural
also supports the claim that Cambodia faces the grim reality that a significant proportion of its
malnourished children aged less than 5 years in Cambodia was 56 percent and 44 percent in
this age group had moderate to severe stunted growth (MoP, 1999). Now, these children have
inevitably joined in the agricultural labour force that this study has analysed. This study found
that many of these same children continue to suffer from malnourishment (see Table 5.5).
Income of Household: In both Table 5.8, and Table 5.9, the variable “Income” is not
statistically significant. This result implies that people in rural Cambodia are highly dependent
Social Capital: The variable “Social Capital” is significant and positive (see, Tables
5.8, and 5.9). This variable is generated through the first component analysis (PC1) based on
the three questions asked during the field survey. The result of the first component (PC1)
158
shows that the positive and significant effect of “Social Capital” has contributed to child
health. Among the three questions used to form the variable “Social Capital”, we have found
that the impact is strongly associated to the questions of (1) “How many percentage of
villagers participated in cooperative works such as building public house, pagoda, or road in
the village in the last one year?”; and (2) “If there is water supply problem or natural calamity
in this village, how many percentages of villagers will cooperate to solve the problem?”. This
simply means that the “Social Capital” through mutual helps and solidarity contributes to the
and newspapers) has been found to have a positive relationship with the improvement of child
health (see Table 5.8 and Table 5.9). This finding support the earlier study by Thomas et al
(1991) who found that the entire impact of maternal education can be explained by indicators
of access to information- reading papers, watching television, and listening to radio. We also
find that a household’s participation in health programs has a positive relationship with child
health. Based on the field interview, health training programs have been conducted since 1999
with support from the project of “Rural Development Project” of the Ministry of Rural
Development. Most villagers tend to agree that there has been an improvement in health since
the establishment of the project. Furthermore, more facilities have been equipped by the
Ministry of Health, such as the establishment of monthly health checks done by the mobile
Public health practice “water and sanitation”: The results of the empirical finding are
consistent with our field observations in that the improvement of general health perception and
body mass of children are closely linked to the health facilities and the understanding of
general health knowledge. For example, the application and use of latrine and boiled water
consumption leads to improved general health and body mass of children (see, Table 5.9).
Access to safe drinking water and sanitation is crucial in Cambodia because of the prevalence
159
of water-borne diseases. According to CSES-1998/99 only 23 per cent of villages obtain water
from public or private taps; the vast majority of villages (nearly one-half) rely on wells for
Age of household head: The coefficient of “age of the household head” is statistically
Parents’ education: Both coefficients on mother’s and father’s education are not
significant in all the models. As mentioned in the conceptualized framework, the theory
generally does give any unambiguous conclusions concerning the impact of education on the
demand for health inputs, thus pointing at the need for empirical analysis. Furthermore, the
role of education has been ignored when a illness arises, because the attitude in the context of
Cambodia is that elders provide substantial advice to the primary caregivers, meaning that
health consultation is a home-based decision (CDHS-2001). This basically means that we may
not expect a direct effect of parent’s education on child health, while the seniority in age is
living apart” is statistically significant (see, Table 5.8). The direction of coefficients suggests
that the more household members living-apart is associated with a 19 percent increase in the
probability that a child is in good health. This basically reflects the situation in Cambodia
whereby many households finance their consumption or debts with remittances from family
6.8. This basically means that a child’s health has a negative relationship with the experience
of being injured while working. The negative impact is an 8 percent point reduction in child
health.
Hours worked and its square: The results here emphases the hours worked by children
and related health effects. The estimated equations of child health outcomes imply that hours
160
worked by children has a diminishing effect. The coefficient on “hours worked in the past 7
days” is positive, and the coefficients on its squares is negative (see Table 5.8, column (C)).
Therefore the quadratic has a parabolic shape, and we can therefore calculate the turning
points. We found that agricultural hours worked by children would not have a negative effect
on “child normal growth” if he/she works within the threshold of 15 hours for children in the
age group 5-9, about 17 hours for children in the age group 10-12, and about 21 hours for
children in the age group 13-14. These threshold levels imply that children shall be permitted
to work within each respective threshold, although exceeding this limit would harm the child’s
health. Although most children worked within the threshold level, this study observed (Table.
5.7) that children in the age group of 14 years old, especial female children, worked an
average of 21 hours thus almost exceeding the threshold level for that particular group.
Therefore, appropriate measures shall be taken to prevent any negative health effects for those
children.
significant (see Table 5.9). This means that we cannot ignore the role of gender in intra-
household resource allocation. Small scale qualitative studies document that households in
which women have more power devote a greater proportion of resources to child-centred
expenditures (Dwyer and Bruce, 1988), although there is little quantitative validation of
differential spending patterns. This also indicates that female household head sacrifice for the
consumption of children, or in other words, she has a high degree of altruism towards her
children.
infrastructure, facilities, location, and history of each village, because we have found simply
that child health, in general, is better in Khan Damra and Kol Korm villages compared to the
Trapeang Kraloung village. Through our observation, Khan Damra and Kol Kom village are
the better-off villages in terms of health understanding as well as child health. The community
161
characteristics can possibly explain that there are existing practices that lead to better health
outcomes emerging from day-to-day health enhancing behaviours, such as better personal
hygiene, regular access to preventive treatments such as timely vaccination, understanding and
practicing home-gardening which provides nutrition for children. Many of these practices
occur unconsciously and are often related to fundamental rules that households live by, rather
than being conscious decisions regarding the allocation of time and money. Despites the above
factors that explain the better health in Kol Korm and Khan Damra, the following explanation
could also be relevant: The study observed that about 30 percent of arable land in Kol Kom
village is irrigated from the reservoir which can provide water for farmers to plant main crops
in both seasons, and thus it is the areas where vegetables and fruits are grown. The village of
Kan Damra is observed to have better land quality and a larger population of cattle than other
areas in the study. Furthermore, human factors are regarded as part of the reasons to explain
why Kol Korm and Kan Damra villages are better-off than the other two.
Table 5.7: Hours Worked by Children in the Past 7 Days by Age and Sex
162
Table 5.8: Probit Coefficient Estimates of “Self-Health Assessment”
Dependent Variable is Model 1: Income Effect Model 2: Parents’ Education Effect Model 3: Media Effect
Self-Health Assessment (A) (B) (C)
Coefficient Marginal P-value Coefficient Marginal P-value Coefficient Marginal P-value
Effect Effect Effect
Female Child -.1282028 -.0381935 0.620 -.1284704 -.0382222 0.215 -.0558432 -.0160099 0.829
Hours Worked .0750146 .0223485 0.079 .0776668 .0231078 0.096 .0509598 .0146087 0.040
Hours Worked^2 -.0018439 -.0004914 0.091 -.0019442 -.0005409 0.037 -.0011667 -.0003338 0.068
AgeGroup1(5-9 years) -.1296214 -.0390184 0.083 -.0631051 -.0188722 0.198 -.2900621 -.0809308 0.082
AgeGroup2(10-12 years) -.9037728 -.2964019 0.098 -.8698805 -.2844307 0.019 -.850445 -.2701443 0.034
Hours Worked AgeGroup1 .1479783 .044086 0.103 .1263494 .037592 0.083 .1905293 .0546194 0.011
Hours Worked AgeGroup1^2 -.0049846 -.0016767 0.053 -.0041427 -.0013251 0.004 -.0068558 -.0016254 0.075
Hours Worked AgeGroup2 .0505324 .0150547 0.142 .048853 .014535 0.143 .0305644 .0087619 0.055
Hours Worked AgeGroup2^2 -.0017175 -.0004946 0.086 -.0017786 -.0004829 0.015 -.0011523 -.000415 0.068
Food Shortage -.8035321 -.2435406 0.001 -.6963945 -.2106949 0.007 .6369931 .1769535 0.105
Meals Past 7 Days .7538164 .2628358 0.050 .5259977 .1770263 0.167 .494988 .1610971 0.178
Family Size -.0443033 -.0131989 0.604 -.0601404 -.0178932 0.494 -.2877715 -.0824959 0.057
Nb Labour -.088956 -.026502 0.562 -.1074345 -.0319644 0.470 .2737682 .0784816 0.149
Nb Living Apart .5879378 .1751597 0.009 .6794677 .2021583 0.003 .6750387 .1935144 0.003
Breastfeed .3544117 .0985739 0.339 .3561478 .0988762 0.314 .4349573 .1139801 0.050
Female HH -.1392123 -.0414744 0.189 -.1194448 -.0355377 0.361 -.1463034 -.041941 0.253
Latrine & Water .1134201 .0329008 0.530 .0634054 .0185882 0.846 .1912935 .0522423 0.578
Vaccine .6470899 .2215787 0.116 .551285 .185765 0.172 .6324832 .2104753 0.035
Health Training Attended .0191804 .0057143 0.327 .0180192 .0053612 0.655 .0390936 .011207 0.421
Times of Injury -.2739504 -.0816159 0.000 -.2719401 -.0809089 0.000 -.2942753 -.0843604 0.000
163
Operate Machine -.44261 -.1507468 0.676 -.6115349 -.2155253 0.557 -.6273188 -.2168249 0.542
Apply Pesticide .1166202 .0335359 0.800 .1329907 .0379929 0.776 .2785602 .0727576 0.598
Experience .1557901 .0464133 0.187 .1850082 .0550445 0.109 .1952883 .0559836 0.013
PreyChangVa -.2048761 -.0628585 0.660 -.3337315 -.1039836 0.414 -1.053499 -.3401382 0.108
KhanDamra .6325665 .1549987 0.086 .4174288 .1096555 0.071 .192383 .0580976 0.055
KolKorm .5750683 .1538809 0.027 .5457258 .1466681 0.035 .0032768 .0009388 0.095
Social Capital .0889625 .0265039 0.009 .1127742 .0335531 0.081 .0872938 .0250246 0.054
Father’s Education ------------ ------------ -------- .0165464 .004923 0.752 ------------ ------------- --------
Mother’s Education ------------ ---------- -------- .0753077 .0224059 0.254 ------------ ------------- --------
Fitted Income 1.53e-07 4.56e-08 0.361 ------------- ---------- -------- -2.77e-08 -7.94e-09 0.675
Fitted Radio ------------ ---------- ------- ----------- ---------- ------- .6014409 .172416 0.004
Fitted TV ------------ ---------- ------- ----------- ---------- ------- -.05286 -.0151535 0.167
Fitted Newspaper ---------- ---------- ------- ----------- ---------- ------- .7359069 .2109636 0.073
Constant .1087932 ---------- 0.032 .2268703 ---------- 0.859 2.008076 ----------- 0.082
For (A) column: Sargan statistic (overidentification test of all instruments): Prob (Chi) = 0.1035; Wald test of exogeneity Prob (Chi) = 0.0463
For (B) column: Sargan statistic (overidentification test of all instruments): Prob (Chi) = 0.208; Wald test of exogeneity Prob (Chi)= 0.0115
For (C) column: Sargan statistic (overidentification test of all instruments): Prob (Chi) = 0.1933 Wald test of exogeneity Prob (Chi) = 0.0186
Instrumented: Hours worked, Hours worked ^2, Other Hours Worked, Income
Instruments: Off farm income, Remitance, House’s Price, Agricultural Assets, PreyChangVa, KhanDamra, and all set of exogenous variable in the structural
equation (1).
Source: Author’s Calculation from the Field Data Survey 2006
Model 1: Probit regression ; Number of obs = 189 Wald chi2 = 71.43 Prob > chi2 = 0.0000 Log pseudolikelihood = -65.167848 Pseudo R2 = 0.3326
Model 2: Probit regression ; Number of obs = 189 Wald chi2 = 73.80 Prob > chi2 = 0.0000 Log pseudolikelihood = -64.787041 Pseudo R2 = 0.3359
Model 3: Probit regression; Number of obs = 189 Wald chi2 = 73.44 Prob > chi2 = 0.0000 Log pseudolikelihood = -61.733515 Pseudo R2 = 0.3625
164
Table 5.9: Regression Coefficient Estimates of “Standardized MBI for Age”
Dependent variable: Model 1: Income Effect Model 2: Parents’ Education Effect Model 3: Media Effect
Standardized BMI for age (A) (B) (C)
Coefficient P-value Coefficient P-value Coefficient P-value
Female Child 3.965239 0.000 3.928882 0.000 3.909478 0.000
Hours Worked .1053447 0.130 .0985389 0.254 .0866252 0.310
Hours Worked^2 -.0024236 0.219 -.0021728 0.201 -.0018596 0.191
AgeGroup1(5-9 years) -3.281799 0.000 -3.15687 0.000 -2.70053 0.000
AgeGroup2(10-12 years) .024269 0.075 -.171223 0.032 -.284048 0.070
Hours Worked AgeGroup1 -.4266631 0.122 -.4475925 0.080 -.3442021 0.169
Hours Worked AgeGroup1^2 .0176913 0.333 .0210319 0.223 .0137227 0.310
Hours Worked AgeGroup2 .0041501 0.168 .0263447 0.201 .0236861 0.215
Hours Worked AgeGroup2^2 .0004753 0.207 .0001619 0.135 .0001986 0.118
Food Shortage -.0894498 0.127 -.330069 0.425 -2.120924 0.073
Meals Past 7 Days .533729 0.107 1.034251 0.221 .9706179 0.404
Family Size -.2798665 0.023 -.2517831 0.040 -.7733987 0.005
Nb Labour -.1294373 0.528 -.1544481 0.456 -.6598867 0.172
Nb Living Apart .2648841 0.313 .1789082 0.461 .2232999 0.391
Breastfeed .5329946 0.191 .4530749 0.149 .4390329 0.171
Female HH .5550894 0.015 .4958094 0.030 .567759 0.018
Latrine & Water 1.482578 0.004 1.195278 0.017 1.357195 0.010
Vaccine 1.105777 0.167 1.393922 0.025 1.220614 0.147
Health Training Attended .08489 0.086 -.0799322 0.127 -.1164206 0.079
Times of Injury .0198258 0.222 .0233227 0.253 .0128551 0.284
Operate Machine -.116591 0.323 .4294896 0.534 -.0493766 0.470
165
Apply Pesticide -.1633346 0.223 -.1636784 0.116 -.0839388 0.204
Experience .1363467 0.034 -.1886293 0.286 -.1774865 0.327
PreyChangVa 1.023142 0.154 1.179842 0.076 1.792504 0.108
KhanDamra 1.233058 0.034 -.776674 0.002 -.6030075 0.042
KolKorm 1.495397 0.045 1.711783 0.024 2.135123 0.013
Social Capital .5553333 0.055 .4246287 0.127 .5731378 0.058
Father’s Education ------------- ---------- -.1058423 0.327 ------------ ---------
Mother’s Education ------------- ---------- -.1358837 0.184 ------------ ---------
Fitted Income -2.39e-07 0.328 --------------- ---------- -1.84e-07 0.364
Fitted Radio ------------ ----------- ------------- --------- -.6005243 0.083
Fitted TV ------------ ----------- ------------- --------- .2078035 0.068
Fitted Newspaper ------------- ----------- ------------- ---------- -.9362405 0.093
Constant 8.430487 0.000 8.644417 0.000 6.836952 0.001
For (A) column: Hansen J statistic (overidentification test of all instruments): Prob (Chi) = 0.1084 ; Wald test of exogeneity Prob (Chi) = 0.0323
For (B) column: Hansen J statistic (overidentification test of all instruments): Prob (Chi) = 0.1351; Wald test of exogeneity Prob (Chi) = 0.0500
For (C) column: Hansen J statistic (overidentification test of all instruments): Prob (Chi) = 0.1319 ; Wald test of exogeneity Prob (Chi) = 0.0267
Instrumented: Hours worked, Hours worked ^2, Other Hours Worked, Income
Instruments: Off farm income, Remitance, House’s Price, Agricultural Assets, PreyChangVa, KhanDamra, and all set of exogenous variable in the structural
equation (1).
Source: Author’s Calculation from the Field Data Survey 2006
Model 1: Number of obs = 189 F( 28, 160) = 8.36 Prob > F = 0.0000 R-squared = 0.3013 Root MSE = 2.6653
Model 2: Number of obs = 189 F( 29, 159) = 8.2 Prob > F = 0.0000 R-squared = 0.3116 Root MSE = 2.6435
Model 3: Number of obs = 189 F( 31, 157) = 8.77 Prob > F = 0.0000 R-squared = 0.3396 Root MSE = 2.6366
166
5.5. Conclusion
This study is the first to investigate the health effects of agricultural child labour, a
subject hitherto virtually untouched in the literature on child labour. This study is also special
in the sense that a field survey was conducted under the financial support from the project of
New Initiative towards Global Academic Collaboration to fill some gaps mentioned by earlier
studies. For instance, Guarcello et.al (2004) though recognizing the importance of general
self-health assessment and anthropometrics measures of height and weight standardized for
age and sex (BMI-age), they did not have such data, and so used data on self-reported illness
and injuries an a proxy. In addition, this study observes that most parents’ education is low in
rural agricultural areas, which draws concerns that the impact of parents’ education on child’s
health may not be observed directly, however, its impact on child’s health could be observed
through access to media such as watching TV, listening to the radio, and reading newspapers.
We employ several techniques to guarantee goodness of the estimation by using Two Stage
The findings of this study contribute to the growing literature on child labour and
health effects. The three striking results in this study are: (1) the current situation of
agricultural child labour has yet to be a concern on health as the average working hours of
children is about 8.5 hours per week, while the threshold level is 15 hours for the children in
the age group 5-9, about 17 hours for the children in the age group 10-12, and about 21 hours
for children in the age group 13-14. However, there are small proportions of children whose
hours worked exceeding this threshold level which draw attention for policy makers for the
target group of children. The result also implies that child could work within threshold hours
of respective group without having negative effect on child health; (2) Accessing to media has
significant impact on child health, while parents’ education is yet to be evidence of child
health in the context of rural Cambodia. This finding is consistent with the earlier work by
167
Thomas et al (1991) who found that almost all the impact of maternal education can be
and listen to the radio; and (3) the result of the first component (PC1) shows a positive and
significant effect of “Social Capital” contributing to child health. Among the three questions
used to form the variable “Social Capital”, we have found that the impact is strongly
works such as building public house, pagoda, or road in the village in the last one year? This
simply means that the “Social Capital” through mutual helps and solidarity contributes to the
Finally there are contributing factors that explain in agricultural child health. Those
factors are: “breastfeeding” is associated with a 11 percent point increase in the probability
that child is healthy; child immunized against seven preventable diseases is associated with a
21 percent point increase in the probability that child is healthy; the negative coefficients on
age group “AgeGroup1, and AgeGroup2” simply mean that children in rural Cambodia do not
have better health when they were young, but health gradually has improved over time, when
they grow up. The improvement of general health perceptions and body mass of children are
well associated with the health facilities and the understanding of general health knowledge,
for example, the application and use of latrine and boiled water consumption have led to the
improved general health and body mass of children. The more the household members living
apart is associated with a 19 percent increase in the probability that child is in good health.
The times of injury has negative impact on child’s health around 8 percent points reduction.
The female household heads sacrifice for the consumption of children, or in other words, she
has a high degree of altruism towards her children. Finally, community characteristics can
possibly explain that there are existing practices that lead to better health outcomes emerging
from day-to-day health enhancing behaviours, such as better personal hygiene, regular access
168
to preventive treatments such as timely vaccination, understanding and practicing home-
gardening which provide nutrition to children. Many of these practices occur unconsciously
and are often related to fundamental rules that households live by, rather than conscious
decisions regarding allocation of time and money. This also means that the village of KolKom
and KanDamra have stood better-off in terms of general health compared to the village of
Trapeang Kraloung. Regardless of the above mentioned, human factor is parts of the
contributing factor that could explains the regional difference that Kol Korm and Kan Damra
village are better-off in terms of health and other socio-economic conditions. In addition, it is
observed that the irrigated areas and fertile soil could be the most important factors to add to
the reasons of regional differences. Moreover, the natural endowment factors such as
irrigation and land quality also important factors to make the regional difference. In this
study, it is observed that about 30 percent of arable land in Kol Kom village is irrigated from
the reservoir which can provide water to farmers to do main crop for both seasons, and thus it
is the areas where vegetables and fruits are largely grown. The village of Kan Damra is
observed to have a better land quality and a larger population of cattle than the other areas in
the survey.
169
C
Chhaapptteerr A
Appppeennddiixx V
V:
T
TAAB
BLLE
ESS A
ANND
DGGR
RAAPPH
HSS
Table Annex 5.2: Probit Coefficient Estimates, the Estimated Value of “Listening to Radio”
170
Dependent Variable: Ratio (1 listening Standard Probit Model
to radio)
Coefficient Marginal Effect P-value
KolKorm 1.645202 .3256475 0.039
Constant .1691039 0.924
Source: Author’s Calculation
Number of obs = 103 Wald chi2(18) = 34.12 Prob > chi2 = 0.0122
Pseudo R2 = 0.6598 Log pseudolikelihood = -12.610214
Table Annex 5.3: Probit Coefficient Estimates, the Estimated Value of “Watching TV”
TV)
171
Dependent Variable: TV (1 if Watching Standard Probit Model
TV)
Number of obs = 189 Wald chi2(21) = 49.88 Prob > chi2 = 0.0004
Log pseudolikelihood = -47.405941 Pseudo R2 = 0.3580
Table Annex 5.4: Probit Coefficient Estimates, the Estimated Value of “Reading
Newspaper”
Reading Newspaper)
172
Dependent Variable: Newspaper (1 if Standard Probit Model
Reading Newspaper)
Number of obs = 189 Wald chi2(21) = 70.70 Prob > chi2 = 0.0000
Pseudo R2 = 0.3512 Log pseudolikelihood = -83.542712
Dependent Variable is Total Income OLS Regression with Corrected Standard of Error
Coefficient P-value
173
Table Annex 5.6: Result of First Principal Component for Variable of “Social Capital”
Factor Loadings
Variable 1 2 Uniqueness
Variable 1 2 Uniqueness
Scoring Coefficients
Variable 1
Q1 0.63503
Q2 0.62060
Q5 0.01925
174
(7) Face Food Short -0.178 0.0483 -0.027 -0.076 -0.115 0.0113 1.0000
(8) Meals Past 7 Days 0.1294 0.0374 -0.160 0.1861 0.1895 0.2018 -0.102 1.0000
(9) Family Size 0.0667 0.1251 -0.049 -0.065 -0.042 -0.039 -0.117 -0.094 1.0000
(10) Nb of Labor -0.051 0.0117 -0.089 -0.066 -0.029 -0.067 -0.022 -0.066 0.4600
(11) Nb Living Apart 0.298 0.1075 -0.027 -0.038 -0.031 0.087 -0.004 0.0157 0.3337
(12) Breastfeeding 0.171 -0.037 -0.116 -0.041 -0.043 -0.057 -0.156 0.1567 0.2744
(13) Female 0.0228 0.0497 -0.000 0.0268 0.0591 0.1205 -0.100 0.0094 -0.201
Household
(14) Latrine & Boiled -0.266 -0.074 -0.050 -0.046 -0.044 -0.094 -0.028 0.0724 -0.043
(15) Immunization -0.011 -0.112 0.1208 -0.045 -0.053 -0.051 0.083 -0.106 -0.177
(16) Trainings attended 0.024 -0.065 -0.127 -0.026 -0.080 -0.031 -0.093 0.0784 -0.022
(17) Times of Injure 0.0397 -0.094 -0.040 -0.145 -0.113 -0.142 -0.062 -0.023 -0.007
(18) Machine Operate -0.107 0.1302 -0.072 0.2408 0.2356 0.1759 -0.059 0.0561 -0.023
(19) Apply Pesticide -0.01 0.2758 0.0019 0.3222 0.2781 0.1609 -0.062 0.0393 0.1118
(20) Experience 0.0315 0.5495 0.0218 0.6197 0.5364 0.4165 0.0711 0.0981 0.0750
(21) PreyChangVa -0.354 -0.033 0.0842 0.0124 -0.028 -0.249 0.1012 -0.338 0.0671
(22) KhanDamra -0.083 0.0617 -0.027 0.1333 0.1517 0.1288 -0.167 0.1143 -0.106
(23) KolKorm Village 0.155 0.0543 0.0622 -0.002 0.0069 0.2736 0.0512 0.1673 -0.210
(24) Father’s Edu 0.3487 -0.103 -0.044 -0.062 -0.046 0.0972 -0.206 0.2906 -0.160
(25) Mother’s Edu 0.0705 -0.048 0.0009 0.042 0.0474 0.0544 -0.268 0.3360 -0.143
(26) Access to TV -0.069 -0.096 0.0802 -0.126 -0.143 -0.127 -0.101 -0.273 0.0689
(27) Access to Radio 0.039 -0.119 -0.055 -0.080 -0.086 0.0005 0.4064 0.0258 -0.269
(28) Access 0.0745 0.1301 -0.004 0.1667 0.1900 0.1281 -0.474 0.2522 0.2054
Newspaper
Continued
(10) (11) (12) (13) (14) (15) (16) (17) (18) (19)
(10) Nb of Labor 1.0000
(11) Nb Living Apart 0.5316 1.0000
(12) Breastfeeding -0.006 -0.000 1.0000
(13) Female -0.131 -0.044 -0.127 1.0000
Household
(14) Latrine & Boiled 0.0517 0.0113 -0.061 -0.039 1.0000
(15) Immunization -0.034 -0.205 -0.248 0.1363 0.1237 1.0000
(16) Trainings attended -0.045 -0.050 0.2510 -0.140 -0.157 0.0024 1.0000
(17) Times of Injure 0.1562 0.1499 0.0672 -0.142 -0.130 -0.272 -0.004 1.0000
(18) Machine Operate -0.011 -0.102 -0.088 0.087 0.1097 0.0587 -0.076 -0.076 1.0000
(19) Apply Pesticide -0.072 -0.116 0.0441 -0.009 0.0237 0.0013 0.0544 -0.052 0.4051 1.0000
(20) Experience -0.105 -0.060 0.0170 -0.038 0.0136 0.1057 -0.020 -0.127 0.1182 0.3585
175
(21) PreyChangVa 0.1130 0.0005 -0.022 -0.375 -0.013 -0.003 0.2506 0.0398 -0.096 0.1929
(22) KhanDamra -0.072 -0.093 -0.186 0.7313 0.0138 0.1233 -0.038 -0.087 0.1456 0.0808
(23) KolKorm Village -0.189 -0.126 -0.104 -0.041 -0.048 0.0100 -0.189 -0.124 0.0719 -0.116
(24) Father’s Edu -0.204 -0.177 0.0125 0.0409 0.1461 0.1584 0.0475 -0.127 0.0933 -0.049
(25) Mother’s Edu -0.132 -0.100 -0.081 0.1307 0.1632 0.1905 0.0264 -0.062 0.0958 0.0456
(26) Access to TV -0.038 -0.026 -0.027 -0.092 0.0379 0.0653 0.0964 -0.005 -0.057 -0.093
(27) Access to Radio 0.0473 -0.025 -0.168 -0.013 0.0395 0.0441 -0.019 0.0353 -0.091 -0.189
(28) Access -0.043 -0.006 0.1496 0.0570 -0.104 -0.094 -0.055 0.0053 0.1260 0.2108
Newspaper
Continued
(20) (21) (22) (23) (24) (25) (26) (27) (28)
(20) Experience 1.0000
(21) PreyChangVa 0.1445 1.0000
(22) KhanDamra 0.0335 -0.274 1.0000
(23) KolKorm Village -0.012 -0.410 -0.260 1.0000
(24) Father’s Edu -0.084 -0.351 0.1281 0.2425 1.0000
(25) Mother’s Edu -0.017 -0.253 0.2472 0.0612 0.5810 1.0000
(26) Access to TV -0.079 0.1178 -0.030 0.0692 0.1596 0.0866 1.0000
(27) Access to Radio -0.060 -0.050 0.0049 0.0560 -0.098 -0.119 -0.139 1.0000
(28) Access 0.0860 -0.090 0.1161 -0.037 0.1250 0.1499 -0.171 -0.728 1.0000
Newspaper
Note: The coefficients of the matrix correlations indicate that the model is secured from the multi-co linearity.
Technically, if coefficient is greater than four and smaller than eight, one can draw assumption that there is weak
correlation, but it does not suffer the model. However, if the coefficient is greater than eight, one shall omit that
variable or combine both variables into one.
176
Graph 5.1: Body Mass Index-for-Age Percentiles (Girls, 2 to 20 Years)
177
Graph 6.2: Body Mass Index-for-Age Percentiles (Boys, 2 to 20 Years)
178
Graph 5.3: Stature-for-Age Percentiles (Girls, 2 to 20 Years)
179
Graph 5.4: Stature-for-Age Percentiles (Boys, 2 to 20 Years)
180
C
Chhaapptteerr V
VII
C
COON
NCCL
LUUSSIIO
ONNA
ANND
D PPO
OLLIIC
CYYR
REEO
OMMM
MEEN
NDDA
ATTIIO
ONNSS T
TOOW
WAAR
RDD
PPO
OVVE
ERRT
TYYA
ANND
DCCH
HIIL
LDDL
LAAB
BOOU
URR IIN
NCCA
AMMB
BOOD
DIIA
A
In Cambodia, many households live under the poverty line. The World Bank (WB,
2006) announced that poverty had decreased from 47% to 35%. However, the number of the
poor has not dropped in real terms because population growth has made the actual number of
poor people increase from 4,157,192 in 1993 to 4,703,697 in 2004. The recent Cambodia
Child Labour Survey 2001 estimated that there were about 1,516,363 children aged 5-14 who
can be considered “working children”, about 44.8 percent of children in this age group. In
considered the age group 5-17 years old, there were about one in every two children were
found to be working. The issue of child labour is often being regarded as harmful to education
and health. For the past decades, the relationship between poverty and the incidence of child
labour has been researched. Most of the findings confirmed the positive relationship between
the two. In addition, most body of literatures also found the negative relationship between
child labour and child’s education as well as child labour and child’s health. In Chapter II of
this study, the results of regressions by using Cambodia’s data on the disaggregated provinces
have shown that there exists positive relationship between poverty and child labour; however,
it does not find any significant relationship between child labour and child’s education as well
as child labour and child’s health. Therefore, this study tries to use more detailed national
181
6.1.1 Major Findings in Chapter III
This study has rejected the strong assumption of the two axioms of Basu and Van that
used parents’ wage rate as the only determinant factor for the supply of child labour, or child
schooling. The simple simulation by allowing the changes of the subsistence level of
consumption (poverty line), this study has observed that poor households have high degree of
altruism for their children, when we set the poverty line at higher level. If the poverty line is
set at a lower level, the parents of non-poor households have a higher degree of altruism.
Therefore, the primary striking result is that when parents’ income is less than subsistence
level (C< C 0 ), children are still enrolled in school. Furthermore, non-poor households also
send their children to work and combining work and study. Therefore, we disagree with the
strong assumptions of modern economics on child labour (Basu and Van, 1998) due to the
fact that the “subsistence level” used to draw the parents’ decisions as whether the child is
observed in “child works” or “child schooling” is hard to apply to the real world since this
study’s empirical estimates prove that the poor also send their children to school even though
in lesser proportion, and at the same time, the non-poor also send their children to work. This
finding implies that on the one hand, there is a strong existence of altruistic parents in the
context of Cambodia, and on the other hand, needs for child labour in Cambodia in the
household economy.
This study, however, did not reject the hypothesis that poverty is the cause of child
labour. We found that the probability of children to enroll in school is higher among non-poor
households. However, this does not mean that poor households did not send their children to
school. This finding is also consistent with the previous theoretical and empirical works such
as the work of Basu and Van (1998), Rosati and Rossi (2003), Ray and Lancaster (2004),
Basu and Tzannatos (2003), Lee and Westaby (1997), Saupe and Bentley (1994), Kim and
Zepeda (2004), Chakraborty and Das (2004), Christiaan and Ravi (1995), Chao and Alper
182
(1998), Duryea and Arends (2001), Basu, Arnab K., and Nancy H. Chau (2003), and Blunch
and Verner (2000). In connection to the hypothesis of child labour and poverty, this study also
found that households with possession of cattle tend to employ their own children. Therefore,
the wealth paradox has been proved valid, however, through the ownership of number of
cattle, but not through land size. This is because land size is equally small and equally
This study has found that the rate of return to education of child labourer is about 12
percent (equivalence of 0.62 USD per week) considerably high. This rate of return is
comparable to the Mincer’s model estimation of American adult labour of non-farm men in
1954 (see Mincer, 1974). This result indicates the importance of logic behind the household’s
decision-making of allowing children to work combined with study because child’s education,
even among child labourer, is proved to be significant in generating income, thus helping
Most importantly, for the human capital through education and hours worked trade-
off, this study has found that the average hours worked of Cambodian children stays below
threshold level hours. The specific thresholds for the age groups 6-9, 10-12, and 13-14 are 19,
21, and 25 hours per week respectively. All working children in this age group have their
average working hours below the threshold level. This implies that child labour rather
increases human capital formation of children as long as their hours worked does not exceed
the threshold level. This finding tends to reinforce the theory of Fan (2004), which states that
a small increase in child labour may not be a trade-off with human capital investment, since
the positive impact of increased financial resources on education may outweigh the negative
impact of a reduced time of study. This is simply that children’s labour market participation
183
raises the financial resources spent on their education. Although the average working hours of
all children stays below the threshold level for human capital formation through education, we
observed that the older age group 13-14 tends to work exceeding the threshold level. This will
In addition to the above aforementioned findings, this study also has found that
parents’ education is very important in determining the schooling outcomes of their children.
However, this education has a positive impact only for parents who had at least completed
primary school or higher. As far as the gender equality and empowerment are concerned, this
study also found that female children tend to be working rather than attending school. This
implies there are many other barriers besides economic terms that lead to hindering access of
female’s participation in education. This finding also contributes to the important policy for
gender equality in Cambodia. Furthermore, this study has also found that social
infrastructures are shown in this study to be important characteristics to reduce child labour
and increase children’s enrollment at school. Being in urban areas has more advantages to
This study has found that media plays a significant role in child health. Any household
has gained access to radio, TV and newspapers also evidently shown the increase in child
health. This evidence feeds into the growing literatures on the role of media in health (see
Strauss and Thomas, 1998). This is one of the important determinants of health of agricultural
child labour. This also indicated the increasing role of expanding media in Cambodia in the
last decades following the general improvement of public health through active support from
184
Most importantly, for the human capital through health and hours worked trade-off,
this study has found that the average working hours of agricultural child labour stays below
the threshold level. The specific thresholds for the age groups 5-9, 10-12, and 13-14 are 15,
17, and 21 hours per week respectively. In general, this study has found that agricultural child
labour has yet to have negative health effects, but rather it benefit the household economy,
and teach children to inherit skills from parents and predecessors. The gradual elimination of
child labour will be favored as long as it does not disgrace the interests of children’s welfare
to education and health. This study also suggests a close monitoring on agricultural child
labour as it is stated by the ILO that the negative effects on child health vary according to
their operating environments. Therefore, this type of study has to be repeated often and it shall
The result of the first component (PC1) shows that the positive and significant effect
of “Social Capital” has contributed to child health. Among the three questions used to form
the variable “Social Capital”, we have found that the impact is strongly associated to the
questions of (1) “How many percentage of villagers participated in cooperative works such as
building public house, pagoda, or road in the village in the last one year?”, and (2) “If there is
water supply problem or natural calamity in this village, how many percentages of villagers
will cooperate to solve the problem?”. This simply means that the “Social Capital” through
For fertility and family planning, this study has found that the presence number of
babies and children in each household do have negative impact on schooling outcomes of the
child. Of course, this is rather simple to understand because Cambodian has a large family
size (see MoP, 1999) and a younger child is taken-care by the older childe who ultimately will
185
Other preventive measures are also important for child health. This study has found
that child breastfeeding is associated with a 11 percent point increase in the probability that
child is healthy. Child immunized against seven preventable diseases is associated with a 21
percent point increase in the probability that child is healthy. The improvement of general
health perceptions and body mass of children are well associated with health facilities and an
understanding of general health knowledge. For example, the application and use of latrine
and boiled water consumption leads to improved general health and body mass of children.
Derived from the main fact-findings in this study, a set of economic, social and
political implications are recommended for pertaining to have the best options to maintain
human capital of children, especially their education and health, and the job creation policy in
rural agriculture for sustainable development and poverty alleviation. This study, hitherto,
international organizations, national and local agencies, and the private sectors whose work
share concerns on children and human capital of long-term development for the kingdom of
will empower men and women in society to have better knowledge for their innovation and
186
While Cambodia has been doing quite well in increasing number of primary schools
through out the country 37, however, there are concerns on disproportional increases in the
lower secondary school buildings and the increase number of children. The Education Sector
Performance Report 2005 shows that performance in the education sector has improved
during academic year 2004-05 and that significant improvements have been made over the
medium term since 1999-2000 and the beginning of the education sector reform process, as
outlined in the Education Strategic Plan (ESP) 2001-05. However, these improvements have
not been sufficient to meet the MoEYS's targets in some key areas, particularly lower
secondary net enrolment rates and promotion rates in primary education. These two areas are
closely linked as improvements in primary promotion will lead to greater student numbers
able to gain access to the lower secondary level (details, see Chapter II). This is clearly a
priority area for further strengthening as present rates of improvement are not sufficient to
reach the Cambodian Millennium Development Goal (CMDG) in 2010 nor the Education For
All (EFA) targets for 2015 38 . Although in the recent years, school facilities have been
expanded for grades 7-9, it is not enough to respond to the demand. The data of the MoEYS
(MoEYS, 2004-2005) shows that there are only 749 communes from a total of 1,621, with
lower secondary schools. In remote areas NER is lowest at 3.9%, i.e. more than 95% of
children aged 12-14 are not enrolled in lower secondary schools, which implies that there still
exists an acute shortage of secondary school in these areas. Therefore, this study recommends
for the Royal Government of Cambodia and stakeholders in the development to pool
resources for the construction of lower secondary schools throughout the country. There are
37
Statistical data for the 2004-05 school year shows the total number of primary school as 6,180 with 2,682,129
students. This includes 1,266,420 girls which is 47% of the total indicating that at this level the enrolment of
boys and girls is similar with effectively no gender gap at primary school.
38
The National Plan for Education for All (EFA) set its base year at 2003 and its target at year 2015.
187
6.2.2 Major Policy Recommendation for Investing in Children’s Health
development, the RGC states in its National Health Policy that "The RGC recognizes that
there are two systems of health care services, one run by RGC and the other run by the private
sector. The RGC focuses mainly on awareness of sanitation, vaccination, treatment and other
public health related services. This study notices that number of health centres have been
increased the last decades due to commitment from the RGC and the support from the
international development agencies. Many maternity wards and health centres are being built
by the RGC at the community levels (see details in Chapter II). Besides the above
aforementioned progresses, this study has found that the population has to travel more than 5
kilometres to reach an operational health clinic facility. The difficulties are found in
transportation and high cost of treatment. In case of serious sickness, patients are sent to the
hospitals in Phnom Penh and this process can sometimes be difficult for children because of
slow transportation and slow services after their arrival at the hospitals. Furthermore, poor
people find it impossible to send their children from rural areas because of high travel costs
and fees charged for treatment. Most of the government hospitals or health centres have
inadequate equipment, facilities and medicine and do not have enough professional staff.
Apart from this there are only a few general hospitals or private inadequately trained medical
practitioners who cannot provide a satisfactory level of health services for children and the
population. Therefore, this study recommends for the RGC to improve the basic health
services at local levels. There are multiple benefits from implementing this policy.
188
6.2.3 Major Policy Recommendation on Job Creation in Agricultural Sector
The population growth rate remains high 1.81 percent from 1998-2004 (APDA 39,
2007; NIS, 2004) with an average family size of 5.2 (CDHS, 2000). Throughout the country,
only 47 percent of children continue to lower secondary education. In the remote areas, the
NER is quite low at 3.9 percent. Therefore, the concern is that the current mechanism and
policy cannot absorb rural labour surplus as large proportion of children will not continue the
lower secondary school, unless the government will implement large scale infrastructure of
secondary school facilities and school buildings, and develop job opportunities in agricultural
sector. We also notice that the situation of private sector is exceptionally narrowly-based in
the sense that the economy depends largely on the growth of tourism and garment industry.
Experience from countries of sustainable growth indicates that the diversification of growth is
a road to equitable growth, which creates jobs and reduces poverty. From a cost and resources
standpoint, Cambodia may have a comparative advantage in agro-industry. If value added can
be raised, agro-industry can contribute to diversifying the economy by shifting more rural
employment to manufacturing. To shed some light, this study recommends for the RGC to
create business environment focusing on job creation in rural agricultural sector. The specific
recommendations are: (1) Small scale farmers increasingly have to adapt their farm
businesses to market changes, improve efficiency and increase profitability to take advantages
of new opportunities that arise. Most farmers have small land size in average 0.35 hectares,
which reached a stage where self-sufficiency in the production of rice is not possible (APDA,
2007). The report of the Asian Population and Development Association (APDA) also
mentioned that farmers consider various solutions for their survival in the current situation
they are in, with labour migration being one answer. Therefore, farmers have to be compelled
39
In 2007, the Asian Population and Development Association (APDA) conducted a based study on impact of
population issue on agriculture and rural development for the Kingdom of Cambodia focusing on Phnom Penh
and Svay Rieng province.
189
to become better managers in order to survive. In this case, the integrated farming system are
highly recommended; (2) Throughout Cambodia, the introduction of new farm method (Green
Revolution) is not easy as it require a reliable water supply. In order to increase the rice and
other agricultural productivity, therefore, the government shall place high priority on the
construction of irrigation facilities; (3) The government and private sectors and development
partners have to encourage and help farmers to organise and work in small groups as
corporative, in order to build their capacity to produce for market and respond to market
demands; (4) Policy to support agro-industry, and post-harvest management for making a
range of value-added; (5) Policy to develop supply chain will benefit farmers from increased
earnings, and helps to stimulate the development of pre-processing workshops, which will
ultimately create new jobs; (6) Finally, the government has to commit to policy reforms on
the ease of doing agribusiness in Cambodia. The current procedure of doing agribusiness is
licensing requirements and too many inspections. All of these can deter investors, deny job
The Implementation of policy to create jobs in agricultural sector will solve many
problems, especially the poverty reduction through job creation. Besides the above mentioned
benefits, this policy will also have effect in reducing labour migration (seasonal migration for
alternative income) which helps reduce urban slum in main city such as Phnom Penh, Siem
The rate of return from education among child labourers is high. Therefore, the
increased educational opportunities for children, particularly child labourers are likely to have
significant effects on maintaining their schooling. The policy implication is that, there are
190
several economic benefits (private return) as well as social benefits from increasing human
capital of children. The study has found that the average hours worked of children in
Cambodia stay below the threshold level hours. Therefore, this study has recommended that
children shall be allowed to work within the threshold hours (19 hours, 21 hours, and 25 hours
per week for the age group 6-9, 10-12, and 13-14 respectively). However, the immediate
policy intervention shall be addressed to the children in the age group 13-14 because their
working hours exceeded the threshold level. As far as the gender inequality is concerned, the
study has found that male children’s participation in human capital accumulation is
female’s education is needed to deal separately from the male counterpart because female
children have different needs to allow them to join in school, for instance., providing girl
dormitory and school boarding program, toilet facilities, changing the biasness of parents’
decision towards male’s education, improving social security would cerate favourable
condition for more girls’ participation in school. Implementing the gender equality
opportunity will benefit to women whose status in social, political and economical are under
represented as a results of socio-political and cultural context imbedded to the society from
the past. This study has also found that media has a positive impact on child health. Therefore
the policy to increase household’s access to media is important to increase child health. There
are many private and social benefits in increasing household’s access to media.
191
REFERENCES
APDA, 2007, Base Study on Impact of Population Issue on Agriculture and Rural
Development, Kingdom of Cambodia, Focus on Phnom Penh and Svay Rieng Province,
Akabayashi, H. and G. Psacharopoulos, 1999, “The Trade-off between Child Labour and
Human Capital Formation: A Tanzanian Case Study”, Journal of Development Studies, 35,
pp. 120-140.
Ashagrie, K, 1998, “Statistics on Child Labour and Hazardous Child Labour in Brief”.
Baland, J. M and J. A. Robinson, 2000, “Is Child Labour Inefficient?,” Journal of Political
Basu, Arnab K., and Nancy, H. Chau, 2003, “Targeting Child Labour in Debt Bondage:
Evidence, Theory, and Policy Implications,” The World Bank Economic Review, 17, pp.
255-281.
Basu, K and Z. Tzannatos, 2003, “The Global Child Labour Problem: What Do We Know
and What Can We Do?,” The World Bank Economic Review, 17, pp. 147-173.
Basu, K, 1999, “Child Labour: Cause, Consequences, and Cure, with Remarks on
Basu, K. and P. Van, 1998, “The Economics of Child Labour,” The American Economic
192
Becker, G. S, 1965, “A Theory of the Allocation of Time,” Economic Journal, 75, pp. 493-
517.
Becker, G. S, 1975, Human Capital: A Theoretical and Empirical Analysis, 2nd Ed.,
Columbia University Press (for National Bureau of Economic Research), New York.
Becker, G. S, 1976, “Altruism, Egoism, and Genetic Fitness: Economic and Sociobiology,”
Becker, G. S, 1993, “Government, Human Capital, and Economic Growth,” Industry of Free
Bhalotra, S. and C. Heady, 2003, “Child Farm Labour: The Wealth Paradox,” The World
Blunch, Niels. H and V. Dort, 2000, “Revisiting the Link between Poverty and Child Labour,
the Ghanaian Experience,” The World Bank, Africa Technical Families, Human
Blunch, Niels. H, 2004, “Maternal Literacy and Numeracy Skills and Child Health in Ghana,”
Blunch, Niels. H,, C. Sudharshan and G. Sangeeta, 2002, “Short and Long-term Impacts of
Economic Policies on Child Labour and Schooling in Ghana,” Social Protection Discussion
Paper, No.0212.
Canagarajah, S. and H. Coulombe, 1997, “Child Labour and Schooling in Ghana,” Policy
Chakraborty, S and M. Das, 2005, “Mortality, Fertility, and Child Labour,” Economics
Chao, S and O. Alper, 1998, Accessing Basic Education in Ghana, Studies in Human
193
Christelle Dumas, 2007, “Why Do Parents Make their Children Work? A Test of the Poverty
Cochran, Susan, Joanne Leslie, and Donald O’Hana, 1982, “Parental Education and Child
DHHS, 2006, “BMI for Children and Teens,” Department of Health and Human Services,
Centers for Disease Control and Prevention (CDC), Retrieved November 30, 2006,
from http://www.cdc.gov/nccdphp/dnpa/bmi/childrens_BMI/about_childrens_BMI.htm
Das, Gupta Monica, 1987, “Selective Discrimination Against Female Children in Rural Punja,
Deb, Partha and F. Rosati, 2004, Estimating the Effect of Fertility Decisions on Child Labour
and Schooling, A Joint Research Project of the ILO, World Bank and Unicef.
Duflo, Esther, 2001, “Schooling and Labour Market Consequences of School Construction in
Indonesia: Evidence from an Unusual Policy Experiment,” The American Economic Review,
Duryea, Suzanne and A. K. Mary, 2001, School Attendance, Child Labour and Local Labour
Dwyer, D. and J. Bruce, 1988, A Home Divided: Women and Income in the Third World.
Ear, S., 2006, The Political Economy of Aid, Governance, and Policy-Making: Cambodia As
Seen from Global, National, and Sectoral Perspectives, Doctoral Dissertation, University
of California, Berkeley.
Edmonds V. Eric, 2005, “Child Labour and Schooling Response to Anticipated Income in
Edmonds, E and C. Turk, 2004, “Child Labour in Transition in Vietnam,” World Bank Policy
194
Fallon, Peter and Tzannatos, Zafiris, 1998, Child Labour: Issues and Directions for the World
Fan C. Simon, 2004, “Relative Wage, Child Labour, and Human Capital”, Oxford Economic
Fentiman, A, A. Hall and D. Bundy, 2002, “Health and Cultural Factors Associated with
Enrolment in Basic Education: A Study in Rural Ghana,” Social Science and Medicine, 52:
pp. 429-439.
Foo Francis, 2004, Rural Development Project: Project Evaluation Report, Supported by
Glewwe. P, 1999, “Why Does Mother’s Schooling Raise Child Health in Developing
Heady, Christopher, 2003, “The Effect of Child Labour on Learning Achievement,” World
ILO (International Labour Organization), 1998, Child Labour: Targeting the Intolerable,
ILO (International Labour Organization), 2002, “Every Child Counts: New Global Estimates
195
Jensen, P and H. S. Nielsen, 1997, “Child Labour or School Attendance? Evidence from
Jere R. Behrman, Anil B. Deolalikar, 1990, “The Intra-household Demand for Nutrients in
Rural South India: Individual Estimates, Fixed Effects, and Permanent Income,” The
Khanam, Rasheda, 2003, “Child Labor and School Attendance: Evidence from Bangladesh,”
Kim, Jongsoog and Zepeda, Lydia, 2004, “Factor Affecting Children’s Participation and
Amount of Labour on Family Farms,” Journal of Safety Research, 35, pp. 391-401.
to Major Hazards on Family Farms,” Journal of Rural Health, 13, pp. 206–215.
Maitra. P, 2004, “Parental Bargaining, Health Inputs, and Child Mortality in India,” Journal
Mincer, Jacob, 1974, Schooling, Experience and Earnings, Columbia University Press,
MoEYS (Ministry of Education, Youth and Sports), 2002, Education Statistics and
MoEYS (Ministry of Education, Youth and Sports), 2005, Education Sector Performance
MoH (Ministry of Health), 2005, Master Plan for Social Health Insurances in Cambodia,
MoI (Ministry of Interior), 2007, Related Laws and Administrations, Retrieved in 2007, from
http://www.interior.gov.kh/
196
MoP (Ministry of Planning), 1999, A Poverty Profile of Cambodia, Phnom Penh, Cambodia.
Prepared with Funding from UNDP and Sida, Phnom Penh, November 2003.
NIS, 1994, Report on the Cambodia Socio-Economic Survey, National Institute of Statistics,
NIS, 1998, Cambodian Census, National Institute of Statistics, Ministry of Planning, Phnom
Penh, Cambodia.
NIS, 1999, Report on the Cambodia Socio-Economic Survey, National Institute of Statistics,
NIS, 2002, Report on Cambodia Child Labor Survey, National Institute of Statistics, Ministry
NIS, 2004, Report on the Cambodia Socio-Economic Survey, National Institute of Statistics,
NIS, 2005, Cambodia Demographic and Health Survey, National Institute of Statistics,
Nagaraj, K., 2002, “Female and Child Workers in a Household Industry: A Case Study of
Beedi Industry in Karnataka and Tamil Nadu,” WP, Madras Institute of Development
Studies, Gandhinagar.
Nielsen, Helena Skyt, 1998, “Child Labour and School Attendance: Two Joint Decisions,”
CLC-WP 98-15, Centre for Labour Market and Social Research, Aarhus, Denmark.
Orazem, P. and V. Gunnarsson, 2003, “Child Labour, School Attendance and Academic
197
O’Donnell. O, E. Van Doorslaer and F.C. Rossati, 2002, “Child Labor and Health: Evidence
PATH, 2004, Cambodia Child Health Strategy Development: Immunization, Phnom Penh,
Cambodia.
Parikh, Anokhi and Sadoulet, Elisabeth, 2005, “The Effect of Parents’ Occupation on Child
Labour and School Attendance in Brazil,” Department of Economics, Oxford University and
Parker. D, 1997, “Child Labour: The Impact of Economic Exploitation on Health and Welfare
Patrinos, H. A and G. Psacharopoulos, 1997, “Family Size, Schooling and Child Labour in
RGC, 2002, National Poverty Reduction Strategy, Prepared by Council for Social
Government of Cambodia.
of Cambodia.
RGC, 2005, National Poverty Reduction Strategy-Progress Report, SNEC, MEF, Phnom
Penh, Cambodia.
Cambodia.
198
Ranjan, Manash Gupta, 1998, “Wage Determination of a Child Worker: A Theoretical
Raut, Lakshmi K. and Tran, Lien H., 2005, “Parental Human Capital Investment and Old-Age
Ravallion, M. and Q. Wodon, 2000, “Does Child Labour Displace Schooling: Evidence from
Ray, Ranjan and G. Lancaster, 2005, “The Impact of Children’s Work on Schooling: Multi-
Ray, Ranjan, 2000, “Child Labour, Child Schooling, and Their Interaction with Adult Labor:
Empirical Evidence for Peru and Pakistan,” The World Bank Economic Review, 14, pp. 347-
367.
Rosati, F. C and Rossi, 2003, “Children’s Working Hours and School Enrollment: Evidence
from Pakistan and Nicaragua,” The World Bank Economic Review, 17, pp. 283-295.
Rosati, F. C and Z. Tzannatos, 2000, “Child Labour in Vietnam,” mimeo, The World Bank,
Rosenzweig, M. R., and R. Evenson, 1977, “Fertility, Schooling, and the Economic
Satz Debra, 2003, “Child Labour: A Normative Perspective,” The World Bank Economic
Saupe, W. E and S. Bentley, 1994, “Introduction to the Survey, Status of Wisconsin Farming-
Special Edition: The 1993 ATFFI Family Farm Survey,” Madison, WI7 Department of
Agricultural and Applied Economics and Agricultural Technology and Family Farm
199
Shah, Iqbal H and Khamma, Jitendra, 2002, “Breast-feeding, Infant Health and Child Survival
Strauss. J and D. Thomas, 1998, “Health, Nutrition, and Economic Development,” Journal of
Thomas. D., J. Strauss and M. H. Henriques, 1991, “How Does Mother’s Education Affect
UCW (Understanding Children’s Work), 2007, “Child Labour and Education for All,”
USGAO (United State General Accounting Office), 1998, Child Labour in Agriculture:
Congressional Requesters.
Udry, C., 2006, “Child Labour,” Understanding Poverty, edited by Abhijit Vinayak Banerjee,
Wolfe. B and J. Behrman , 1987, “Women’s Schooling and Children’s Health,” Journal of
South-Western, USA.
World Bank, 2006, Cambodia Halving Poverty by 2015?, Poverty Assessment 2006, Prepared
by the World Bank for the Consultative Group Meeting, Phnom Penh, February 2006.
200