June 25, 2007

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

Graduate School of International Cooperation Studies, Kobe University

Department of Economic Development and Policies

Academic Advisor : Prof. Seiichi FUKUI

Student No. : 024i063i

Name : Phoumin HAN

 Acknowledgements

I would like to express my deepest gratitude and thanks to Ministry of Education,

Culture, Sport, Science and Technology (Monbukagakusho) for generously financing my

study in the Doctoral programme, Department of Economic Development and Policies,

Graduate School of International Cooperation Studies, Kobe University, Japan.

I am greatly indebted to my academic advisor Professor Seiichi Fukui, who

encouraged me and provides valuable advice during my study and graciously supported me

in data collection, and provides me an opportunity to be a teaching assistant in

Econometrics seminars. Your advice, patient and care have greatly contributed to my

success. I would like to express very grateful thanks to you wholeheartedly.

I am very grateful to Professor Terukazu SURUGA and Dr. Keiichi OGAWA, who

generously agreed to serve as committee members of my dissertation. Your valuable

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

knowledge and facilitating many academic procedures.

I am very grateful to Mr. Shivakumar Srinavas, who provides me a continuous

advice and support in my professional career as well as my academic work such as editing

and commenting my work.

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

1.1 Background and Rationale of the Study 1

1.2 Research Questions and Objectives 7
1.2.1 Econometric Analysis on Root Causes of Child Labour in Cambodia 12
1.2.2 Child’s Wage and Trade-off between Human Capital and Working 13
Hours
1.2.3 Effects of Agricultural Child Labour on Child Health and Its 15
Determinants
1.3 Significance of the Study 18
1.4 Organization of the Study 23

Chapter II: CURRENT SITUATION OF CHILD LABOUR: Perspectives from Poverty,

Education and Health

2.1 Current Situation of Poverty and Child Labour in Cambodia 25

2.2 Educational System, Enrolment, Dropout, Literacy and Educational Expenses 30
2.3 Health System, Child Mortality, Illness, Preventive Measures and Health Care 41
Expenses

Chapter III: UNDERLYING ROOT CAUSES OF CHILD LABOUR: Empirical Evidence and

Policy Implications for Cambodia

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 Appendix III: FARM CHILD OF WEALTH PARADOX 87

Chapter Appendix III: TABLES

Annex 3.1 Tobit Estimate of Hours of Agricultural Child Labour 93

Annex.3.2 Matrix of Correlation of Independent Variables 94

Chapter IV: CHILD’S WAGE, HUMAN CAPITAL AND HOURS WORKED TRADE-OFF:

Simple Theoretical and Empirical Evidence for Policy Implications

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 Appendix IV: TABLES

Annex 4.1 Multinomial Logit_ Maximum Likelihood Estimate of Child’s 130

Activities
Annex 4.2 Matrix of Correlation of Independent Variables of Model 131
Specification 1
Annex 4.3 Matrix of Correlation of Independent Variables of Model 131
Specification 2
Annex 4.4 Simultaneous Estimates of Earning Income and Hours Worked of 132
Children in the Past 7 Days (Log Earnings in the past 7 Days)

Chapter V: HEALTH EFFECTS OF CHILD LABOUR AND ITS DETERMINANTS: Evidence and

Policy Implications Drawing from a Study of Four Villages, Cambodia

5.1 Introduction 133

5.2 Overview of Villages Surveyed 138
5.3 Theoretical and Empirical Framework 145
5.3.1 Theoretical Model 145
5.3.2 Empirical Model 147
5.3.3 Data and Variables 154
5.4 Results and Discussions 156
5.5 Conclusion 167

Chapter Appendix V: TABLES AND GRAPHS

Annex 5.1 Correlations between Parents’ Education and Media 170

Annex 5.2 Probit Coefficient Estimates, the Estimated Value of “Listening to 170
Radio”
Annex 5.3 Probit Coefficient Estimates, the Estimated Value of “Watching 171
TV”

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

Chapter VI: CONCLUSION AND POLICY RECOMMENDATIONS TOWARD POVERTY AND

CHILD LABOUR IN CAMBODIA

6.1 Conclusion of this Study 181

6.1.1 Major Findings in Chapter III 182
6.1.2 Major Findings in Chapter IV 183
6.1.3 Major Findings in Chapter V 184
6.2 Policy Recommendations and Policy Implications 186
6.2.1 Major Policy Recommendation for Investing in Children’s Education 186
6.2.2 Major Policy Recommendation for Investing in Children’s Health 188
6.2.3 Major Policy Recommendation on Job Creation in Agricultural Sector 189
6.2.4 Other Policy Implications Emerging from this Study 190

LIST OF REFERENCES 192

v
 LIST OF TABLES

Table Titles Page

No.
1.1 Percentage of Headcount Index in Cambodia by Strata and by Years 4
2.1 Current Working Children 5-17 Years Old by Strata and Sex 26
(Extrapolation)
2.2 Distribution of Poverty and Child Labour by Provinces (Percent) 28
2.3 Regression of Child Labour and Poverty 30
2.4 Number of Schools, Classes, and Teaching Staff in Cambodia 31
2.5 Net Enrolment Ratio in Lower Secondary and Primary Education by 35
Province and Sex
2.6 Dropout Rate by Grade and Provinces 37
2.7 Population 5-24 Years Currently Attending School by Level and Sex 38
(Percent)
2.8 Literacy Rate, Population 7 Years and over by Stratum and Sex (Percent) 39
2.9 Average Annual Expenses by Education Level 39
2.10 Impact of Child Labour on Net Enrolment in Primary School 40
2.11 Impact of Child Labour on Net Enrolment in Lower Secondary School 41
2.12 Child Mortality Rate and Working Child by Province 44
2.13 Source of Money Spent on Health by Province (Percent) 47
2.14 Impact of Child Labour on Health 49
3.1 Frequency, Row, and Column Percentage of the Child’s Activities by Age 74
3.2 Descriptive Statistics (Aged 5-14) 77
3.3 Maximum Likelihood Estimates of Children’s Activities (Aged 5-14) 79
3.4 Children’s Activities by Poverty Status, 1999 (Percentage of All 83
Households with at least One Child in Age Group 5-14 Years Old)
4.1 Descriptive statistics 113
4.2 Simultaneous Estimates of Earning Income and Hours Worked of 115
Children in the Past 7 Days
4.3 Hours Worked of Children in the Past 7 Days by Age, Sector and Sex 119
(Economic and Non-economic Activity)
4.4 Regression Coefficient Estimates of ‘Schooling Attainment Relative to 120

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

2.1 Enrolment Rates by Age, for 1999 and 2004 35

2.2 Household Members in Bad or Very Bad Health by Sex Evaluated by 46
Household Head or Spouse (Percent)

LIST OF MAPS

2.1 Proportion of Provincial Child Labour to Total Child Labour in 28

Cambodia (Percentage)
2.2 Distribution of Operational District or Referral Hospital 43
2.3 Distribution of Health Post 43

vii
 C
Chhaapptteerr II

IIN
NTTR
ROOD
DUUC
CTTIIO
ONN

1.1. Background and Rationale of the Study

1
Child labour is a pervasive problem throughout the world, especially in

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

participation during the Millennium declaration and, subsequently, the Millennium

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

concrete objectives for human development. The international community's efforts to

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

through the generations. It is nevertheless over-simplistic to attribute the problem solely to

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

households are very useful for breaking this cycle.

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-

consumption and better education compared to land-poor households.

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 .

Table 1.1: Percentage of Headcount Index in Cambodia by Strata and by Years

Regions 1993/94 1996/97 1999 1999 (Both 2004*

(Round 2) Rounds)

OPL FPL OPL FPL OPL FPL OPL FPL OLP

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

Total 39.0 20.0 36.1 17.9 35.9 11.5 51.1 28.9 35

Source: CSES 1993, 1996/97, 1999, 2004

Note: OLP-Overall Poverty Line; FPL-Food Poverty Line
* The 2004 Percentage of Head Count Index is Based on the Recalculation of the World
Bank in Cambodia.

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

were about one in every two children were found to be working.

In fact, the majority of children in Cambodia are found to be involved in labour and

attending school. Therefore, the above aforementioned consequences of child labour as

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

maintain consumption in the household (see Chapter V for details).

The causal antecedences of child labour are numerous and socio-economic

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

government’s Second Socio-Economic Development Plan II (SEDP-II, 2001), the National

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

medical and educational expenses.

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

a child who works 2, or 3, or 4 hours a day. Furthermore, Cambodian children tend to

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

education and health, as detailed in Chapter IV and V of this dissertation.

1.2. Research Questions and Objectives

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

important contribution to poverty alleviation and to development in general (Udry, 2006).

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

to incorporate the subsistence constraint on consumption. This extension yields several

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

wage/productivity. Second, it implies that if the relative productivity of child labour is

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

benefits of an educational investment and arrive at some conclusions as to the profitability

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

labour or other activities if the altruistic 9 parents come into play.

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

made several research questions:

(1) What are the main causes of child labour?

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

child education or schooling? Children in developing countries inevitably have

to work. However, the majority of Cambodian children combine work with

study. Does their human capital formation through their level of education

significantly determine their wage rate?

(3) Does child labour have negative impact on human capital formation through

child health and nutrition?

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

other related hypotheses of the above aforementioned.

(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

children’s education plays a significant role in their wage rates.

(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

of child labour concerning the interests of the child.

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

methodologies to be used in order to achieve the objective of this study.

1.2.1. Econometric Analysis on Root Causes of Child Labour in Cambodia

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

subsistence level of consumption, there is no child labour. Furthermore, the household

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

education; (b) household’s characteristics: poverty status (1 if household consumption is

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

for heteroskedasticity in variance of y conditioning on x variables (Wooldridge, 2003). To

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

significant role in determining the children’s activities.

1.2.2. Child’s Wage and Trade-off between Human Capital and Working Hours

To investigate the impact of child labour on human capital formation through

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

hours by children in Cambodia and the effect on their education.

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.

In the empirical specification of the Human Capital Earning function, Mincer

(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

children are represented by the simultaneous equation system.

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

learning measures such as variables of Schooling Attainment Relative to Age (SAGE

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

likely to be endogenous, thus OLS coefficient estimates may be biased. Therefore, it is

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

establish our regression model for our study.

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

child health. However, Cambodian children work on farms as a traditional agricultural

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

agricultural child labour in Cambodia.

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

community endowment factors. Furthermore, information received through media is a

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

information through the received information function.

In the econometric analysis, the study wishes to estimate the relation between

health characteristics and child health outcomes. There are, of course, many measurements

of health such as general health status (GHS), self-reported morbidity/illness/injuries,

nutritional-based indicators (nutritional intake), and nutritional-based indictors

(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.

1.3. Significance of the Study

The significance of this study is that (1) it has demonstrated the originality and its

contribution to the growing literature of child labour; (2) it provides a significant

contribution to policy formulation for poverty reduction in Cambodia through the

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

originality of this study can be summarised as follows:

(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

amongst non-poor households if the subsistence level of consumption is set at a lower

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

literatures of child labour.

(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

that a child is combining work with study, or is engaged in neither.

(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

mentioned literature, O’Donnell et al (2003) found little evidence of the contemporaneous

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

indicators of access to information such as reading newspapers, watching television, and

listening to the radio.

In contrast to the above literatures, this study observes that Cambodia children

work on farms as a traditional agricultural practice and their involvement is highly

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

importance of general self-health assessments and the anthropometrics measure of height

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

level there will be a consequence to child health.

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

losing its substance.

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

comprehensive review of the current situation of poverty, child labour, human

development through education and health in Cambodia. Chapter III consists of an

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

causes of child labour.

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.

Chapter V focuses on health effects and their determinants of agricultural child

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

capital refers to social coherences and mutual helps.

Finally is the chapter of conclusion. It composes of a summary of all the chapters of

analyses, and proposes policy recommendations, and policy implications towards child

labour in Cambodia. This chapter is brief, but comprehensive summary of results. The end

of the chapter proceeds by the bibliography.

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

2.1. Current Situation of Poverty and Child Labour in Cambodia

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

some suggestion on the relationship of poverty and child labour in Cambodia.

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

exploitation.” In January 1997, the National Assembly of the Royal Government of

25
Cambodia had adopted a new labour code. The new labour code sets the minimum age of

employment at 15 (Article 177) 12.

Table 2.1: Current Working Children 5-17 Years Old by Strata and Sex (Extrapolation)

Regions Number of Working Children and Its Percentage

(Column Percent)

Male Female Total

Phnom Penh 54,577 61,698 116,275

(4.7%) (5.5%) (5.1%)

Other Urban 119,776 120,607 240,383

(10.2%) (10.8%) (10.5%)

Rural 996,089 925,714 1,921,803

(85.1%) (83.7%) (84.4%)

Total Cambodia 1,170,441 1,108,019 2,278,460

(100%) (100%) (100%)

Source: CCLS-2001-02

The Ministry of Labour and Vocational Training (MoLVT) is mandated to lead in

actions to combat child labour. Regardless of the government institution, it is observed in

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

program utilizes government officials as teachers in its education programs to prevent

trafficking in source provinces; LICADHO provides education and awareness-raising on

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.

In order to be visually seen the distribution of child labour in Cambodia, the

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

combined with agriculture as well.

27
Map 2.1: Proportion of Provincial Child Labour to Total Child Labour in Cambodia
(Percentage)

Otdar Mean Chey

Stueng Traeng

Banteay Mean Chey

Preah Vihear

Siem Reab Rotanak Kiri

Bat Dambang
Krong Pailin Kampong Thum Kracheh
Mondol Kiri

Pousat

Kampong Chhnang
Kampong Cham

Kaoh KongKampong Spueu

Phnom Penh Prey Veaeng
Kandal

Svay Rieng

Kampot Takaev

Krong Preah Sihanouk

Krong Kaeb

Source: Author’s calculation and mapping technique, data Province

from CCLS 2001-02 Distribution of Child Labour
(Percent)

Note that the red color is the highest distribution of child 1 to 5

labour. 5 to 9

9 to 13

Table 2.2: Distribution of Poverty and Child Labour by Provinces (Percent)

Provinces Percentage of Population Percentage of working

Below Food Poverty Line children aged 5-14 years old

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.

Suggestion on the Impact of Poverty on Child Labour: As noted in the World

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.

Table 2.3: Regression of Child Labour and Poverty

Dependent Variable is “Proportion of working Coefficient P-value

children aged 5-14 years old”

Proportion of people whose income is less than food .2046809 0.090

poverty line

Constant 52.56058 0.000

Number of obs = 24 F( 1, 22) = 2.81 Prob > F = 0.0077

R-squared = 0.1730 Adj R-squared = 0.1133 Root MSE = 13.542

2.2. Educational System, Enrolment, Drop-out, Literacy and Educational Expenses

Educational System: Before 1975, the country adopted a French-based education

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

period of learning lasting 45 minutes (MoEYS, 2002).

Table 2.4: Number of Schools, Classes, and Teaching Staff in Cambodia

Particulars Number Number Classes Teaching Non-Teaching

13
of of in Staff Staff 14
Schools Classes Pagoda
Total Female Total Female
1,345 2,316 4 2,650 2,624 183 164
Pre-School (778*)
Primary School 6,180 61,648 309 50,140 20,678 10,701 2,256
Lower Secondary Level 800 10,744 28 19,082 6,302 2,903 595
Upper Secondary Level 232 3,513 12 6,025 1,662 804 191
Source: Education Management Information System, MoEYS, 2004-2005
Note: * is the private school

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

dropouts. Despite the government’s budget constraints, the Royal Government of

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

of recurrent expenditures incurred by central and provincial levels.

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

Education Sector Support Program (ESSP). The Ministry describes it as an education

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

efficiency of education as part of a decentralized, pro-poor approach, consistent with the

Government’s broader social and economic development agenda.

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-

sufficiency for the Cambodian educational system.

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

if the rates were much lower then.

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

provinces is less behind the male counterpart.

Figure 2.1: Enrolment Rates by Age, for 1999 and 2004

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

Source: CSES-2004 and CSES-1999

Table 2.5: Net Enrolment Ratio in Lower Secondary and Primary Education by Province

and Sex

Provinces Net enrollment ratio in lower Net enrollment ratio in

secondary education primary education
Female Male Total Female Male Total

Banteay Mean Chey 10.2 11.6 10.9 87.2 87.1 87.1

Bat Dambang 13.2 13.5 13.3 87.6 93.8 90.7
Kampong Cham 15.2 17.8 16.5 85.1 91.5 88.2

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

cities) to continue their education.

Table 2.6: Dropout Rate by Grade and Province

Provinces Dropout rate by grade (Grade 1-12)

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

Source: Education Management Information System, MoEYS, 2001-2002

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

attending the formal school system.

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

More importantly, the recent Cambodia Socioeconomic Survey 2004 (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

Cambodia Phnom Penh Other Urban Rural

Both Sexes 67.1 89.1 75.0 63.5

Male 74.4 93.9 80.5 71.3

Female 60.3 84.7 69.6 56.3

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

is for a student in “Undergraduate/Graduate”. Since the higher level schools are

predominantly found in Phnom Penh the average annual expenses are consequently much

higher there.

Table 2.9: Average Annual Expenses by Education Level

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

well elaborated and explored in chapter IV of this dissertation.

Table 2.10: Impact of Child Labour on Net Enrolment in Primary School

Dependent Variable is “Net enrollment ratio in Coefficient P-value

primary education”

Proportion of working children aged 5-14 years old .0648116 0.631

Constant 82.08167 0.000

Number of obs = 24 F( 1, 22) = 0.24 Prob > F = 0.6311

R-squared = 0.0107 Root MSE = 8.977

40
Table 2.11: Impact of Child Labour on Net Enrolment in Lower Secondary School

Dependent Variable is “Net Enrolment Ratio in lower Coefficient P-value

secondary education”

Proportion of working children aged 5-14 years old -.1221569 0.366

Constant 22.60304 0.001

Number of obs = 24 F( 1, 22) = 0.85 Prob > F = 0.3663

R-squared = 0.0372 Root MSE = 8.9336

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

increase in the level of government expenditure based on improved information on costs,

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

financing system that will promote improvement in quality in the delivery of an

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

availability of medical assistants and medical doctors; lack of continued education

opportunities; lack of outreach to the community health centres by higher level

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.

Map 2.2: Distribution of Operational District or Referral Hospital

#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

Map 2.3: Distribution of Health Post

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

Kiri, and Rotanak Kiri province.

Table 2.12: Child Mortality Rate and Working Child by Province

Provinces Child mortality rate Percentage of working

children age 5-14

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

Source: Cambodia Demographic and Health Survey,-2000, and CSES-1999

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

Household Head or Spouse (Percent)

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.

According to the same report of CSES-2004, at least 81 percent of the children

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

from night blindness.

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

population, who sold assets in order to finance the cost of sickness.

Table 2.13: Source of Money Spent on Health by Province (Percent)

Provinces Source of money spent on Source of money spent on

health care is borrowed with health care is from sold
interest assets

Phnom Penh 12.1 1.6

Kaoh Kong 11 1.9
Kandal 12.2 2.8
Kampong Chhnang 2.8 3.7
Kampong Cham 13.1 5.1

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

create additional financial barriers to seeking 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

unlikely to provide a solution to the level and method of funding.

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.

Table 2.14: Impact of Child Labour on Health

Dependent Variable is “Child Mortality Rate” Coefficient P-value

Proportion of working children aged 5-14 years old .1138336 0.768

Constant 54.53177 0.029

Number of obs = 24 F( 1, 22) = 0.09 Prob > F = 0.7679

R-squared = 0.0040 Root MSE = 25.7

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

enthusiastically care for their children education and other welfare.

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

children's human capital formation in response to changes in the relative wage/productivity

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

do not. Social infrastructures are shown in this study to be important characteristics to

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

significantly that they tend to employ their own children.

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,

parents’ occupation, parents’ education, gender, number of children, social infrastructure,

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

axioms of Basu and Van.

3.2. Existing Literatures and Alternative Hypotheses

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

paradox, parent’s occupation, parent’s education, gender, number of children, social

infrastructure, regional difference such as rural or urban, and other contributing

hypothesises on the determinants of child labour .

3.2.1. Poverty and Altruism

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.

Similarly to the model of economics of child labour, “bequests” constraint of

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

markets are imperfects (Baland and Robinson, 2000).

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

models of intergenerational transfers. The first model treats parental investment in

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

two sided altruism.

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

(1997), and Ray (2000).

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

by international organizations, e.g., the World Bank.

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

Initiatives toward Global Academy, Graduate School of International Cooperation and

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.

3.2.3. Parents’ Occupation

Several studies show the importance of parent employment characteristics on child

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

determining of children’s activity such that if the father employed in a vulnerable

occupation, for example, day labour or wage labour, it raises the probability that child will

work full time or combine work and study.

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

and schooling are responsive to opportunities to work, in particular to opportunities

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

it depends on the occupation of the parents.

3.2.4. Parents’ Education

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

labour’s participation) is found inconsistently based on social, cultural and political

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

household to be at work on the farm. Moreover, children of the female-headed households

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

schooling with work in Bangladesh.

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

3.2.6. Number of Children

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)

treated number of children as exogeneity.

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

rupees in Tamil Nadu and Karnataka, India.

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

investments in skills obtained in school.

62
3.2.7. Social Infrastructure

Other factors, especially social infrastructure, indeed, play significant role in

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

and hours of child labour.

3.2.8. Other Factors

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

school because of the availability of services. Furthermore, neighbouring effects,

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

continue higher education.

3.3. Empirical Framework

3.3.1. Empirical Model

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,

j=1 if the child both works and studies,

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

j=3 if the child neither work nor study (idle).

Let P j be the probabilities associated with these j=0,1,2,3 categories. By

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

Where 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 children’s activities. The vector xi denotes observed individual and

exogenous variables that include (a) child characteristics: i.e., age, age2, sex, primary

education, and secondary education; (b) household’s characteristics: poverty status (1 if

household consumption above poverty line, 0 otherwise), parents’ employment status,

24
parents’ education, and female household head ; and (c) the vector of Z i is represented by

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

into our model due to data limitation.

The model is automatically accounted for heteroskedasticity in variance of y

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

play significant role in determining the children’s activities. Alternatively, if households’

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

correlated with children’s activities.

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

equations of endogeneity of Multinomial Logit at once without having problem of larger

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

policy implication (Wooldridge 2003). By endogeneity, it means that numbers of children

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

determining children’s activities.

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

home and send the male one for higher education.

3.3.2. Data Used in this Study

The sample of household with children’s activities is drawn from the Cambodia

Socio-economic Survey 1998/99 (CSES-1998/99). The survey carried out by National

Institute of Statistics (NIS) of the Ministry of Planning is the cross-sectional multi-purpose

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

children’s activities termed as “idle”, which composed of a broad category of children

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

than “non-economic activity”.

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

education, number of children in the household, parents’ occupation, female is the

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

used as a discrete variable instead of grouping as “primary, secondary, high school, or

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

analysis is given in Table 3.2.

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.

3.3.3. Results of Empirical Estimate

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

important for policy directions to wards child labour in Cambodia.

The results suggest that Cambodia children tend to participate in school or

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-

governmental organization, and international organization have positive effect on

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

determinants of child labour.

An additional issue on child activity in Cambodia is that we have observed that

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

the category of child idle.

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

5 4 3 190 632 829

0.48 0.36 22.92 76.24 100.00
0.43 0.40 2.68 21.46 7.08

6 7 10 311 565 893

0.78 1.12 34.83 63.27 100.00
0.75 1.32 4.39 19.19 7.63

7 11 21 532 430 994

1.11 2.11 53.52 43.26 100.00
1.19 2.77 7.52 14.60 8.49

8 12 29 620 293 954

1.26 3.04 64.99 30.71 100.00
1.29 3.82 8.76 9.95 8.15

9 11 39 604 163 817

1.35 4.77 73.93 19.95 100.00
1.19 5.14 8.53 5.53 6.98

10 24 73 847 186 1,130

2.12 6.46 74.96 16.46 100.00
2.59 9.62 11.97 6.32 9.65

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

12 34 68 766 100 968

3.51 7.02 79.13 10.33 100.00
3.66 8.96 10.82 3.40 8.27

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

15 178 107 520 102 907

19.63 11.80 57.33 11.25 100.00
19.18 14.10 7.35 3.46 7.75

16 221 86 441 107 855

25.85 10.06 51.58 12.51 100.00
23.81 11.33 6.23 3.63 7.30

17 257 83 301 122 763

33.68 10.88 39.45 15.99 100.00
27.69 10.94 4.25 4.14 6.52

Total 928 759 7,078 2,945 11,710

7.92 6.48 60.44 25.15 100.00
100.00 100.00 100.00 100.00 100.00
Source: Author’s calculation from the sample CSES-1999

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

Appendix III, Table Annex 3.1).

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

the growing literature of child labour.

76
Table 3.2: Descriptive Statistics (Aged 5-14)

Variables and Related Hypotheses Definition Mean Std.Dev.

Variables Related Hypotheses

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)

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

Age of child -.5647422*** .1761312 -.3659512*** .046315*** -.0082827*** .0065671* -.0445994***

(.1731576) (.1408344) (.0992556) (.01348) (.00303) (.00368) (.01292)
Age of child^2 .0372018*** .0003363 -.003391 -.0001028 .0006049*** 5.90e-06 -.000508
(.0079225) (.0063506) (.0052273) (.00066) (.00015) (.00017) (.00063)
Child’s education .4480189*** .0626348** .9023365*** -.1130498*** .0050358*** .0021608*** .1101748***
(.0342851) (.0271544) (.0244283) (.00303) (.00065) (.00071) (.00285)
Child of HH head -.3609908** -.1798617 -.4837814*** .0732308*** -.0048456 -.0025038 -.0658814***
(.1831343) (.1619816) (.1270487) (.02027) (.00356) (.00469) (.01993)
Female child .3728089*** -.1380359 .1833281*** -.023348*** .0056662*** -.0046872* .0223692***
(.1086478) (.0889907) (.0671355) (.00875) (.00177) (.00241) (.00832)
Nb. Children in HH .0635405* .0300082 -.0161773 .0003686 .0010441* .0008577 -.0022704
(.0362754) (.0323074) (.0245004) (.0031) (.00058) (.00088) (.00301)
Mother’s education .0044885 -.080658*** -.0169437 .0037673 .000149 -.0021403*** -.001776
(.0252555) (.0216124) (.0163927) (.00214) (.0004) (.0006) (.00201)
Father’s education -.0628799*** .021046 -.051145*** .0063421*** -.0008984** .0008127* -.0062565***
(.023328) (.0182401) (.01387) (.0018) (.00038) (.00049) (.00172)
Poverty status(food) -.3659028*** -.1480021 -.2997745*** .0452594*** -.0054366** -.0026262 -.0371966**

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.

Numbers in parentheses are robust standard errors.

Multinomial logistic regression: Number of obs = 10,092 Wald chi2(87) = 106484.60
Pseudo R2 = 0.3770 Prob > chi2 = 0.0000 Log pseudo-likelihood = -5974.4556

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

parent’s income or consumption is equal to subsistent level (C= C 0 ) (Christelle, 2007).

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

suggests alternative policy, perhaps, the limited hours worked of children.

(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

needed to upgrade women’s status in decision-making.

(3) Social infrastructures are shown in this study to be important characteristics to

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

distributed in the rural areas of Cambodia.

(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

in the Cambodian Millennium Development Goals.

86
 C
Chhaapptteerr A
Appppeennddiixx IIIIII::

FFA
ARRM
MCCH
HIIL
LDDO
OFF W
WEEA
ALLT
THH PPA
ARRA
ADDO
OXX

A.3.1. Theoretical Model on a Two-Period of Child Labour

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

substitution effects arising from market imperfections.

The model is solved to give an expression for the quantity of child labour supplied in

period 1:

Lc1 = h( A0 , K 0 , wh1 , wh 2 , p r1 , p r 2 ; Z , e), Eq (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

unobservable, the demand equation for X 1 can be inverted to write K 0 as a function of X 1 .

Since X 1 is observable, it is convenient to substitute out for K 0 in equation 1 to obtain:

Lc1 = h( A0 , X 1 , wh1 , wh 2 , p r1 , p r 2 ; Z , e), Eq (2)

87
Since information on the rental price for land (p r1 , p r2 ) is lacking, the equation is conditioned

on the quantity of rented land (A r ) to get:

Lc1 = h( A0 , X 1 , wh1 , wh 2 , Ar1 , Ar 2 ; Z , e), Eq (3)

This equation forms the basis of the estimates. Consumption (X 1 ) and rented land (A r )

are treated as potentially endogenous. X is expected to capture (negative) income effects on

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.

The Role of Market Imperfections

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,

less child labour.

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

is perfect, and it is positive if both of these markets are imperfect.

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.

Consider the three possibilities:

• 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

offset one another.

• 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.

• 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.

A.3.2. Empirical Model

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

cattle on child labour, the coefficients on these variables are of interest.

Instrumental Variables

If decisions about income expenditure (X 1 ) and labour supply (for example, L c ) are made

simultaneously, then X 1 is endogenous in the equation for L c . Moreover, child labour

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

equation, for hours of work (H), be written as

H * = Zβ + Yγ + e Eq (1)

90
where hours (H) is a censored endogenous variable, X is a vector of exogenous variables, and

Y is the endogenous variable. The auxiliary equation describing Y in terms of exogenous

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 .

Substituting for e in equation 1 gives the conditional model

H * = Zβ + Yγ + uα + ε Eq (3)

where u is an estimate obtained by OLS estimation of equation 2 and equation 3 can be

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

engaged in household work on varieties of reasons depending on seasons, however, cattle

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

that most households employ their children on tending the cattle.

Land and Land Size:

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

likely to be child labour.

92
 C
Chhaapptteerr A
Appppeennddiixx IIIIII::
TTA
ABBLLEESS

Table Annex 3.1: Tobit Estimate of Hours of Agricultural Child Labour

Hours Worked of Coefficient Unconditional Conditional Probability P-value

Agricultural Child Expected on Uncensored
Value being
Uncensored
Age of the child 12.29191 5.882062 4.324401 .3815553 0.011

Age of the child^2 -.4387867 -.2099731 -.154369 -.0136205 0.051

Female Child 4.217907 2.018805 1.486072 .1303435 0.063

Fitted Income 1.66E-07 7.96E-08 5.85E-08 5.16E-09 0.897

Acres .0002184 .0001045 .0000768 6.78E-06 0.758

Acres^2 -1.63E-08 -7.78E-09 -5.72E-09 -5.05E-10 0.604

Land dummy 10.94556 3.754506 3.126332 .304337 0.080

Family size .784275 .3752999 .2759147 .0243448 0.378

Nb. Of labour -1.725527 -.8257183 -.6070553 -.0535624 0.159

Age of HH head .0220066 .0105308 .0077421 .0006831 0.881

Female HH -3.768407 -1.8033 -1.325758 -.1169758 0.003

Father’s education -.0387142 -.018526 -.01362 -.0012017 0.947

Mother’s education .27438 .1312993 .0965293 .0085171 0.624

Cattle 11.46269 5.485253 4.032674 .3558153 0.000

Irrigated land 1.69212 .8296921 .6061288 .052548 0.626

PreyChangVa 3.172932 1.58267 1.151829 .0983833 0.365

KhanDamra 21.06588 14.1226 10.38028 .5344486 0.000

KolKorm 13.00743 8.6701 3.97766 .080074 0.000

Resid-Income -2.65E-06 -1.27E-06 -9.33E-07 -8.24E-08 0.002

Constant -98.44926 -47.11102 -34.6353 -3.055979 0.000

Source: Author’s calculation from field survey 2006

Number of obs = 214 LR chi2(18) = 185.86 Prob > chi2 = 0.0000
Pseudo R2 = 0.1617 Log likelihood = -481.69771
Obs. summary: 101 left-censored observations at hours worked<=0
113 censored observations.

93
Table Annex.3.2: Matrix of Correlation of Independent Variables
(1) (2) (3) (4) (5) (6) (7) (8) (9)

(1) Age of child 1.0000

(2) Age of child^2 0.9889 1.0000

(3) Child’s education 0.3028 0.3570 1.0000

(4) Child of HH head 0.0569 0.0527 0.0186 1.0000

(5) Female child 0.0026 0.0039 0.0052 -0.0097 1.0000

(6) Nb. Children in HH 0.0008 -0.0076 -0.0479 0.1671 -0.0008 1.0000

(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

(foreign gov/ IOs)

(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)

(10) Employer of HH 1.0000

(government)

(11) Employer of HH -0.023 1.0000

(state enterprise)

(12) Employer of HH -0.032 -0.0045 1.0000

(joint venture)

(13) Employer of HH -0.124 -0.0170 -0.0235 1.0000

(private)

(14) Employer of HH -0.019 -0.0026 -0.0036 -0.0138 1.0000

(foreign gov/ IOs)

(15) Employer of HH -0.025 -0.0035 -0.0048 -0.0183 -0.0028 1.0000

(NGO)

(16) Employer of HH -0.622 -0.0853 -0.1177 -0.4468 -0.0692 -0.0919 1.0000

(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)

(19) (20) (21) (22) (23) (24) (25) (26) (27)

(19) Clean water 1.0000

(20) Electricity 0.7478 1.0000

(21) Dist. to Pri School -0.2494 -0.2593 1.0000

(22) Dist. to Sec School -0.2442 -0.2864 0.3063 1.0000

(23) Dist. to high school -0.3020 -0.3880 0.2316 0.5630 1.0000

(24) Cham -0.0467 0.0131 -0.0181 0.0700 0.0434 1.0000

(25) Chinese 0.1264 0.0885 -0.0381 -0.0210 -0.0368 -0.0113 1.0000

(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)

(28) Other urban 1.0000

(29) Rural -0.7945 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 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

world, as reflected by the UN Millennium Summit and subsequent Millennium Development

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

are poor, do in fact combine work with study.

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

positive impact on total household income.

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

have negligible effect on human capital accumulation in Pakistan and Nicaragua.

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

combining work with study, or is engaged in neither.

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

capital development of children, as well as a broader social policy in Cambodia.

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

maintain a sufficient level of household consumption. Therefore, so as to ascertain the

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

provide the following investigations:

(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,

their education is influenced by their income.

(2) Estimate the trade-off between hours worked as child labour and schooling outcomes,

and thus establish the threshold level.

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

(Log Earnings in the past 7 Days).

4.2. Theoretical Framework

4.2.1. Rate of Return to Human Capital Investment

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

extensively researched education’s importance in determining individual differences in wages

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

and are trained outside of schools, especially on jobs.

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:

n = length of working plus length of schooling of children

= length of working for children without schooling

Ys = annual earnings of individual child with s years of schooling

Vs = present value of an individual’s child earnings at start of schooling

r = discount rate

102
 t = 0, 1, 2,…, n time, in years

d = difference in the amount of schooling, in years

e = base of natural logarithms

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

And when discounting process is continuous is:

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

schooling is found by letting V s =V s-d :

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

pay, the difference between earnings of individuals is due to difference in investment of d

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

ration, K s, s-d , as follow:

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

we can write the following formula:

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

function of time spent at school.

4.2.2. Hours Worked of Children and Schooling Trade-off

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

parents’ utility function is defined by the equation (1) below:

Max U (C p , δhC , a ) Eq (1)

lC , x

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

capital function of children as:

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

budget constraint in equation (3) expressed as:

C p + x = I p + WC lC ; or C p = I p + WC lC − x Eq (3)

where I p is parents’ income and W c l c is child’s income. We assume that parents’

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

of children if and only iff both W c and δ are high.

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

4.3.1. Rate of Returns to Human Capital of Child Labour

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:

Ei = β 0 + β1t i − β 2 t i + β 3 Chi + β 4 l c + β 5 Z i + U 1i ; i ∈ {1,2,3....N }

2
Eq. (1)

li = β 0 + β1t i − β 2 t i + β 3 Chi + β 4 Ei + β 5 Z i + β 6 Bi + U 2i ; i ∈ {1,2,3....N }

2
Eq. (2)

where t i is years of experience, E s is earning capacity based on schooling, l c is the

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.

4.3.2. Human Capital Formation of Children and Hours Worked Trade-off

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:

hc = ( β 0 + β 3 AgeG1 + β 4 AgeG 2) + ( β 1 + β 5 AgeG1 + β 7 AgeG 2)l C +

( β 2 + β 6 AgeG1 + β 8 AgeG 2)l C2 + β 9 Chi + β 10 Povi + Eq.(1)
β 11 SchExi + β 12 X i + β 13 Z i + U i1

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

rearrange (1) as:

hc = β 0 + β 1l c + β 2 l c2 + β 3 AgeG1 + β 4 AgeG 2 + β 5 ( AgeG1.l C ) +

β 6 ( AgeG1.lC ) 2 + β 7 ( AgeG 2.lC ) + β 8 ( AgeG 2.lC ) 2 + β 9 Chi + Eq. (2)
β10 Povi + β11 SchExi + β12 X i + β13 Z i + U i1

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

problem. The instrumented equations are:

l c* = δ 0 + δ 1Chi + δ 2 SchExi + δ 3 X i + δ 4 IVi + δ 5 Z i + U i 2 Eq. (3)

Povi* = δ 0 + δ 1Chi + δ 2 SchExi + δ 3 X i + δ 4 IVi + δ 5 Z i + U i 5 Eq. (4)

where vector X i denotes observed individual and exogenous variables that include

household’s characteristics such as parents’ education, gender of the household head,

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

include these into our model due to data limitation.

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

CSES-1998/99 by regress on sets of child and household characteristics. Because variable of

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

simply the reduced form:

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-

off hours worked by each group of the child as follows:

( β1* + β 5* )
• ( AgeG1.l ) = −
*
is the trade-off hours worked by children in the age
2( β 2* + β 6* )
C

group 6-9, with human capital formation of the child.

( β1* + β 7* )
• ( AgeG 2.lC* ) = − is the trade-off hours worked by children in the age
2( β 2* + β 8* )

group 10-12, with human capital formation of the child.

β1*
• l c* = − is the trade-off hours worked by children in the age group 13-14, with
2 β 2*

human capital formation of the child.

4.3.3. Dataset and Variables

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

( β 1 + β 5 ) and its square is ( β 2 + β 6 ) . To

29
Note that the slope of the hours worked of the age group 6-9 is
do this, we plug AgeG1=1 into equation (5). The same process is done for the slope of the hours worked of the
age group 10-12. The base group is the age group 13-14

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

on the search of proxy of this variable.

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

are based on the recommendations of Expert Committee on Unemployment Estimates (1970)

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

example, Rosati and Rossi, 2003.

Table 4.1: Descriptive statistics

Variable Definition Mean Std. Dev.

Earning of the child in Earning of the child in the past 7 days (Riels currency) 20637.23 34419.42
the past 7 days
Years of Schooling Number of Years of Schooling of the Child 2.342079 2.131733
Child’s activities: =1 if Child Combined both Works and Studies, 0 if Child 1.688475 .4631321
Works only (Base Category)
=2 if Studies only, 0 if Child Works only (Base Category) 1.711408 .9575172
=3 if Child Idles, 0 if Child Works only (Base Category)
.5857673 .4926195
SAGE Index of Schooling Attainment Relative to Age 36.64419 36.24471
Hours worked Number of Hours Worked of the Child for Economic and 13.14201 14.24304
Non-economic Activity
Mother’s education =1 if Mother Completed Secondary Education of Higher .1692431 .3749754
School’s expenditure Expenditure on Child’s Schooling Such as “School 281.0938 243.6285
fee…etc” per Child per Year
Age of child Age of the Child 9.732331 2.783275
Female Child =1 if Female Child; 0 for Male Child .4882995 .4998749
Father’s education =1 if Father Completed Secondary Education or Higher .3054886 .4606249
Poverty status =1 if Non-poor Household; 0 if Poor (Measure against .8357207 .3705376
National Poverty Line)
Nb. Babies Number of Babies in Each Household (Age 0-4 Years .5647426 .738297
Old)
Nb. Children Number of Children in Each Household (Age 5-14 Years 2.908193 1.387854
Old)
HH size Number of People in Each Household 7.42013 3.077704
Gender of HH =1 if Male Is Household Head; 0 if Female .8746277 .3311484
Age of HH Age of the Household Head 44.07611 10.15445
Uses piped water =1 if Household Uses Piped Water, 0 Otherwise .4987472 .5000102
Electricity =1 if Household Uses Public Electricity, 0 Otherwise .3588616 .4796778
Telephone =1 if Household Access to Telephone, 0 Otherwise .3854772 .9867412
Rural =1 if Rural; 0 if Urban .3167872 .4652347
Source: Author’s calculation

Total observation: 21153

113
4.4. Results of Empirical Estimate

4.4.1. Result Estimates on Rate of Return to Human Capital of Child Labour

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

not have such problem of endogeneity.

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

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

Hours worked in the 792.5454 0.000 1229.375 0.000 --------- ---------

past 7 days
Earning in the past 7 ----------- -------- ----------- -------- .0003635 0.000
days
Experience -7891.478 0.068 -7041.981 0.002 2.844115 0.248

Experience^2 1482.185 0.037 1396.803 0.048 -.5736763 0.154

Child’s edu 2453.036 0.067 2459.712 0.065 -1.364666 0.080

Female child 7726.83 0.099 6699.239 0.151 -1.411991 0.593

Uses piped water 33172 0.036 35457.02 0.025 -15.23362 0.088

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

Electricity -4210.289 0.427 -4984.002 0.346 2.479883 0.409

Telephone 196.9774 0.913 -270.7241 0.880 -.0944786 0.927

Rural -7458.829 0.182 -8092.405 0.147 4.43117 0.162

Enrolment ----------- -------- ----------- --------- -12.01092 0.000

Nb of infants ----------- -------- ----------- --------- -1.2196 0.449

Nb of children ----------- -------- ----------- --------- 1.020219 0.255

Constant -7873.213 0.322 -21581.41 0.006 31.03527 0.000

Hausman Test: Ho: difference in coefficients not systematic

χ 2 = 109.65
P-value =0.0000
Source: Author’s calculation

(a)- Report of OLS goodness of fit:

Number of obs = 171 F( 9, 161) = 7.57 Prob > F = 0.0000
R-squared = 0.2974 Adj R-squared = 0.2581 Root MSE = 29446

(b)- Report of simultaneous equations goodness of fit:

Equation Obs Parms RMSE "R-sq" F-Stat P

Earning in the past 7 171 9 30765.57 0.2330 17.18 0.0000

days
Hours worked in the 171 12 17.52939 0.2905 14.95 0.0000
past 7 days

4.4.2. Result Estimates on Hours Worked and Its Schooling Trade-off

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

Hausman-test of endogeneity confirms that covariates of “hours worked by children” and

“poverty status” are endogenous with “schooling outcomes” both in the equation of “SAGE”

as well as “Years of Schooling”. We further employ over-identification-test for all the

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

age these children will join in labour force.

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.

Nevertheless, several variables such as number of infant in the household “Nb. Of

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

of parents who are male dominant in the household-decision making.

The community infrastructure variables are statistically significant. In general, the

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

access to telephone uses.

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)

Age Urban Rural

Male Female Total Male Female Total
5 1.985158 1.996094 1.990485 1.92803 1.882129 1.905123
6 3.120536 3.268502 3.189146 4.31962 4.016667 4.172078
7 5.428969 5.926694 5.678695 7.179641 7.1875 7.183385
8 7.941253 8.03662 7.987127 10.21008 8.964809 9.601719
9 9.699571 11.00595 10.3399 10.52632 12.47138 11.43036
10 13.08807 13.84041 13.45871 16.73421 15.60825 16.16537
11 15.08604 15.8637 15.47155 18.93036 18.26836 18.60168
12 15.97448 18.94987 17.42036 21.40056 21.73491 21.57317
13 19.74906 20.5683 20.15321 24.28192 24.02609 24.1595
14 21.01169 23.70889 22.33532 26.6447 28.45918 27.47434
Total 11.83085 12.98726 12.39593 14.77606 14.72482 14.75108
Source: Author’s calculation

119
Table 4.4: Regression Coefficient Estimates of ‘Schooling Attainment Relative to Age: SAGE
Index (Sample Consists of Children Aged 6-14 Years Old)

Dependent Variable is SAGE (OLS) SAGE (IV-Tobit)

“SAGE” (a) (b)
Coefficient P-value Coefficient P-value

Hours Worked .4176843 0.005 18.89065 0.000

Hours Worked^2 -.008298 0.000 -.3758835 0.000
Poverty Status 3.079484 0.000 71.50554 0.000
AgeGroup1 (6-9 years) 40.01021 0.000 70.73415 0.218
AgeGroup2 (10-12 years) 9.3677 0.000 128.787 0.024
Hours Worked of AgeGroup1 .794538 0.000 16.20902 0.003
Hours Worked of AgeGroup1^2 -.023327 0.000 -.5320158 0.000
Hours Worked of AgeGroup2 .1222615 0.225 18.2854 0.001
Hours Worked of AgeGroup2^2 -.0034323 0.402 -.4656784 0.000
Female Child 1.249265 0.004 1.656922 0.207
School’s expenditure .0193533 0.000 .016312 0.000
Father’s education 3.867102 0.000 5.811144 0.000
Mother’s education 3.901434 0.000 5.158382 0.006
Nb. Babies -3.32965 0.000 -4.586247 0.000
Nb. Children -1.33007 0.000 -2.274986 0.014
HH size .2582736 0.037 -.193576 0.749
Gender of HH -2.511102 0.000 -14.21748 0.000
Age of HH .1638695 0.000 -.0249581 0.779
Uses piped water 2.798912 0.000 2.738037 0.057
Electricity 10.59781 0.000 8.441322 0.028
Telephone .8118143 0.000 2.167221 0.009
Rural -.3509035 0.504 -8.082366 0.008
_cons -39.49364 0.000 -140.6441 0.007

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

(a) Report of goodness of fit of OLS

Number of obs = 19575 F( 22, 19552) = 402.91 Prob > F = 0.0000

R-squared = 0.3119 Adj R-squared = 0.3112 Root MSE = 29.955

(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)

Dependent Variable is Schooling (OLS) Schooling (IV-Tobit)

“Years of Schooling” (a) (b)
Coefficient P-value Coefficient P-value

Hours Worked .0142959 0.076 .8294997 0.001

Hours Worked^2 -.0002971 0.000 -.0171784 0.000
Poverty Status .2543471 0.000 3.359152 0.000
AgeGroup1 (6-9 years) 3.800747 0.000 1.6203 0.025
AgeGroup2 (10-12 years) 1.983296 0.000 3.940988 0.019
Hours Worked of AgeGroup1 .0391141 0.000 .7488304 0.002
Hours Worked of AgeGroup1^2 -.0011943 0.024 -.0258533 0.000
Hours Worked of AgeGroup2 .022813 0.000 .803669 0.001
Hours Worked of AgeGroup2^2 -.0004873 0.287 -.0197112 0.000
Female Child .0688087 0.001 .071003 0.224

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

(a) Report of goodness of fit of OLS

Number of obs = 19575 F( 18, 19556) = 1489.22 Prob > F = 0.0000

R-squared = 0.5782 Adj R-squared = 0.5778 Root MSE = 1.3732

(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

measured by variables of schooling outcome such as “SAGE”, “Years of Schooling” and

“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

under our estimation are summarized as follow:

(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

household’s decision-making of allowing children to carry combined work with study

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.

(2) We found parents’ education is very important in the determinant of schooling

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

and the likelihood of a child working.

(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

worked by children with respect to their age group.

(6) The presence number of babies and children in each household do have negative

impact on schooling outcomes of children. Of course, this is rather simple to understand

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

look after a smaller brothers or sisters.

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

changes to the outcome of child labour.

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

A.4.1: Empirical Model

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 works only,

j=1 if the child both works and studies,

j=2 if the child attends school only and,

j=3 if the child is idle (neither works nor studies).

Let P j be the probabilities associated with these {j=0,1,2,3} categories. By considering

“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 :

β j xi = β 0 + β1l c* + β 2 l c*2 + β 3 Chi + β 4 Povi* + β 5 SchExi + β 6 X i + β 7 Z i + U i1

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.

The model is automatically accounted for heteroskedasticity in variance of y

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

performing the multinomial logit model.

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

appropriate proxy or instrumental variable of these covariates, or (ii) there is no econometric

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

age of the child tends to explain its activity.

A.4.2. Results of Multinomial Logit Estimates:

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

Nancy H. Chau (2003), and Blunch and Dort (2000).

129
 C
Chhaapptteerr A
Appppeennddiixx IIV
V::
T
TAAB
BLLE
ESS

Table Annex 4.1: Multinomial Logit_ Maximum Likelihood Estimate of Child’s Activities

Variables Parameters Marginal effects (dy/dx)

Work & P- Study P- Idle P- Work Work & Study Idle
study value only value value only study only

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

(Outcome Child’s activities==0 ((child works only) is the comparison group)

Source: Author’s calculation

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.

Report of goodness of fit of Multinomial logistic regression

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)

(11) Nb of infants 1.0000

(12) Nb children 0.1989 1.0000

Table Annex 4.3: Matrix of Correlation of Independent Variables of Model Specification 2

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
(1) Hours 1.0000
worked
(2) Hours 0.8991 1.0000
worked ^2
(3) Poverty -0.051 -0.048 1.0000
status
(4) School’s -0.042 -0.055 0.1849 1.0000
expenditure
(5) Age of 0.4909 0.3380 0.0495 0.0553 1.0000
child
(6) Age of 0.4876 0.3431 0.0481 0.0534 0.9914 1.0000
child ^2
(7) Female 0.0271 0.0208 0.0023 0.0035 0.0030 0.0022 1.0000
Child
(8) Father’s -0.067 -0.062 0.1328 0.3890 -0.034 -0.033 -0.009 1.0000
education
(9) Mother’s -0.048 -0.045 0.0957 0.3021 -0.018 -0.017 0.0053 0.3756 1.0000
education
(10) Nb. babies -0.021 -0.007 -0.032 0.0121 -0.144 -0.144 -0.001 0.0218 -0.008 1.0000
(11) Nb. -0.006 0.0082 0.1143 0.1249 0.0034 0.0004 -0.008 0.0137 -0.021 0.1350
Children
(12) HH size -0.020 -0.013 0.1995 0.2501 0.0553 0.0530 -0.014 0.1022 0.0560 0.2883
(13) Gender of -0.024 -0.029 0.0733 0.3240 -0.038 -0.038 0.0037 0.1860 0.0018 0.1028
HH
(14) Age of HH 0.0412 0.0234 0.1370 -0.011 0.1852 0.1846 -0.015 -0.033 -0.039 -0.178
(15) Uses piped 0.0420 0.0300 0.0631 0.0792 0.0318 0.0306 -0.001 0.1051 0.0884 -0.077
water
(16) Electricity -0.137 -0.121 0.2660 0.2286 0.0444 0.0452 -0.004 0.2720 0.2343 -0.140
(17) Telephone -0.062 -0.046 0.0801 0.0898 0.0093 0.0096 -0.007 0.1144 0.1076 -0.043
(18) Rural 0.0769 0.0671 -0.286 -0.206 -0.012 -0.012 -0.001 -0.175 -0.149 -0.025
Continued:
(11) (12) (13) (14) (15) (16) (17) (18)

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)

Variable Simultaneous Equation of Earning Income and Hours Worked of Children

in the Past 7 Days
Log Earning income Hours worked
(past 7 days) (past 7 days)
Coefficient P-value Coefficient P-value

Hours worked in the past 7 .0513797 0.000 --------- ---------

days
Earning in the past 7 days ----------- -------- .0002641 0.000

Experience -.2648685 0.071 2.20501 0.371

Experience^2 .0462983 0.044 -.452735 0.260

Child’s edu .1158011 0.053 -1.164622 0.035

Female child -.0064488 0.975 -.5903644 0.823

Uses piped water .4332574 0.540 -12.16025 0.074

Electricity .0797525 0.737 2.129701 0.479

Telephone -.014821 0.855 -.3099499 0.763

Rural -.3609458 0.149 4.299707 0.175

Enrolment ----------- --------- -16.24919 0.000

Nb of infants ----------- --------- -.6766803 0.685

Nb of children ----------- --------- 1.417706 0.126

Constant 7.127606 0.000 33.95921 0.000

Report of Simultaneous equations goodness of fit:

Equation Obs Parms RMSE "R-sq" F-Stat P

Log Earnings in past 7 days 171 9 1.369463 0.2084 13.83 0.0000
Hours worked in past 7 days 171 12 16.83475 0.3456 11.77 0.0000

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

Rural Cambodian children remain largely engaged in agricultural activities and

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

Millennium Development Goals.

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

water and sanitation, etc.

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

non-enrolled children are working.

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

knowledge alone appears to be crucial in raising a healthy child in Morocco.

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

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 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

policy implications towards agricultural child labour in Cambodia:

(1) Investigate the determinants of child health, and whether the current hours worked

by agricultural child labour have any ill effect on their health.

(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

affect their growth and health in general.

(3) Undertake policy analysis based on evidence emerging from this study to come up

with policy implications for agricultural child labour in Cambodia.

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

age 2-20 years old.

5.2. Overview of Villages Surveyed

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

incidence of poverty is predominant among the male-headed households (UNDP, 1999). In an

attempt to assess the socio-economy of rural Cambodia, a study led by Professor Seiichi Fukui

of Graduate School of International Studies (GSICS), Kobe University, conducted a field

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

households, consisting of 46 households in Kan Damra village, 41 in Prey Changva, 45 in Kol

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

village’s chiefs. The research employed semi-structured interviews, qualitative, and

quantitative questions to obtain background information and specific characteristics of the

households and individuals. This is the primary field survey data 32 of the socioeconomic

conditions of each household comprising of several modules, such as farming system,

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

in exchange for other consumables.

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

than other two.

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

Damra are better off.

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

prescriptions directly from the pharmacies.

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

56 percent and males 44 percent.

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

Overview of Villages Surveyed

Kampong Speu Province Takeo Province

Village Kan Damra Prey Changva Kol Korm Trapeang

Characteristics Kralong

Total Households in the village * 92 80 168 92

Number of Households in the sample 46 41 45 36

Farming Households (Percent) 95.6 95.1 95.5 88.8

Arable Land (m2) 11,049 6,111 7,160 5,508

Irrigated Land (Percent) n.a n.a 30 n.a

Household Member (person) 4.50 4.85 4.62 5.28

Household Income (Riels/year) 2,737,375 1,896,523 3,019,313 3,188,508

Per Capita Income (Riels/year) 661,508 449,825 670,215 614,134

Households below Poverty Line (%) 63.04 80.49 57.78 61.11

Household owned cattle (Average) 1.05 0.52 0.91 0.46

Female-headed households* 9 15 33 13

The number of poorest households* 10 12 17 10

Female-headed households and 7 5 8 4

poorest *

Landless households* 1 4 4 8

Landless female-headed households* 1 3 3 3

Source: Rural Household-Survey 2006

* 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

Trapeang Kroloung, 20 in Kol Korm, and 14 in Kan Damra village.

142
Table 5.2: Frequency, Row and Column Percentage of Children Aged 5-14, by Categories of

Children Operating Machinery and Applying Pesticide

Not operate machinery Operate machinery Total

Not apply pesticide 190 0 190

100.00 0.00 100.00
90.48 0.00 88.79

Apply pesticide 20 4 24
83.33 16.67 100.00
9.52 100.00 11.21

Total 210 4 214

98.13 1.87 100.00
100.00 100.00 100.00

Source: Author’s calculation from survey data 2006

Table 5.3: Frequency, Row and Column Percentage of Children Aged 5-14, by Child Labour

and Sex

Male Female Total

Not working child 56 43 99

56.57 43.43 100.00
52.34 40.19 46.26

Working child 51 64 115

44.35 55.65 100.00
47.66 59.81 53.74

Total 107 107 214

50.00 50.00 100.00
100.00 100.00 100.00
Source: Author’s calculation from survey data 2006

143
Table 5.4: Frequency, Row and Column Percentage of Children Aged 5-14, by Sex and BMI-

Age Percentile Category

Under-weight Healthy- Risk of over- Over-weight Total

Weight weight

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

Total 100 108 4 2 214

46.73 50.47 1.87 0.93 100.00
100.00 100.00 100.00 100.00 100.00
Source: Author’s calculation from survey data 2006

Table 5.5: Frequency, Row and Column Percentage of Children Aged 5-14, by Village and
BMI-Age Percentile Category

Under-weight Healthy- Risk of over- Over-weight Total

Weight weight
Prey Changva 35 27 0 1 63
55.56 42.86 0.00 1.59 100.00
35.00 25.00 0.00 50.00 29.44

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

Total 100 108 4 2 214

46.73 50.47 1.87 0.93 100.00
100.00 100.00 100.00 100.00 100.00
Source: Author’s calculation from survey data 2006

144
5.3. Theoretical and Empirical Framework

5.3.1. Theoretical Model

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

their determinants in the context of rural agricultural child labour.

We assume that household utility is a function of household consumption, C p 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’

characteristics X, and community endowment factor Z. Furthermore, information received

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:

Max U ((C p , L p , H ( L, If , x; X , Z )) Eq (1)

L,Lp , x

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

within the threshold level hours.

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

with household consumption.

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

system of equation can be solved through the given parameters below.

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

health of the child, can be estimated as follows:

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

the parameters X , Z , SC , WC is investigated in the empirical model.

5.3.2. Empirical Model

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

health, such as General Health Status (GHS), self-reported morbidity/illness/injuries,

nutritional-based indicators measured by either nutritional intake or anthropometrics, although,

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

differs between girls and boys.

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

model can be established as follow:

Healthi = ( β 0 + β 3 AgeG1 + β 4 AgeG 2) + ( β1 + β 5 AgeG1 + β 7 AgeG 2) H i +

( β 2 + β 6 AgeG1 + β 8 AgeG 2) H i2 + β 9 Chi + β10 Incomei + β11Watch _ TVi + Eq.(1)
β12 Listen _ Radioi + β13 Re ad _ Newspaperi + β14 X i + β15 Z i + U i1

The equation (1) can be rearranged as:

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 ) H i + β 7 ( AgeG 2.H i ) + β 8 ( AgeG 2.H i ) H i + β 9 Chi +
Eq.(2)
β10 Incomei + β11Watch _ TVi + β12 Listen _ Radioi +
β13 Re ad _ Newspaperi + β14 X i + β15 Z i + U i1

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

can rearrange (2) as:

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

Square to deal with this problem. The instrumented equations are:

H i* = δ 0 + δ 1 AgriAssetsi + δ 2 HousePi + δ 3 X i + δ 4 Z i + U i 2 Eq.(4)

Incomei* = δ 0 + δ 1 AgriAssetsi + δ 2 HousePi + δ 3 X i + δ 4 Z i + U i 5 Eq.(5)

Watch _ TVi * = β 0 + + β 1Chi + β 2 X i + β 3 Z i + U i 6 Eq.(6)

Listen _ Radioi* = β 0 + + β 1Chi + β 2 X i + β 3 Z i + U i 7 Eq.(7)

Re ad _ Newspaperi* = β 0 + + β 1Chi + β 2 X i + β 3 Z i + U i 8 Eq.(8)

where the variables in equation (1) through (8) are H i (hours worked), Incomei (total

income of household), Chi (child’s characteristics), X i (household’s characteristics),

Z i (community characteristics), AgriAssetsi (agricultural assets including thresher, plough,

pump, sprayer, cart, and rake) and HousePi (imputed house price).

Finally the estimation of the structural equation (3) is simply the reduced form. The

following to estimate the group model:

150
Income Effect on Child Health Model

Healthi = 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 + Eq.(9)
β10 Incomei + β11 X i + β12 Z i + U i1

Parents’ Education Effect on Child Health Model

Healthi = β 0 + β 1 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 + Eq.(10)
β10 MEdu i + β11 FEdu i + β12 X i + β13 Z i + U i1

Media Effect on Child Health Model

Healthi = 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 +
Eq.(11)
β10Watch _ TVi* + β11 Listen _ Radioi* + β12 Re ad _ Newspaperi* +
β13 Incomei + β14 X i + β15 Z i + U i1

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* )

the age group 5-9, with child health.

( β1* + β 7* )
• ( AgeG 2.H i* ) = − is the trade-off between hours worked by children in
2( β 2* + β 8* )

the age group 10-12, with child health.

β1*
• H = − * is the trade-off between hours worked by children in the age group
*

2β 2
i

13-14, with child health.

We also perform Hausman-test of endogeneity and over-identification restriction test

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

( β 1 + β 5 ) and its square is ( β 2 + β 6 ) . To

36
Note that the slope of the hours worked of the age group 5-9 is
do this, we plug AgeG1=1 into equation (5). The same process is done for the slope of the hours worked of the
age group 10-12. The base group is the age group 13-14

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)

where xi β is the aggregated form of the explanatory variables. Therefore, the

dependent variable can be observed as follow:

Healthi ≡ 1( Healthi* > 0)

Pr( Healthi = 1 xi ) = G ( 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* ) +
Eq. (13)
β 8 ( AgeG 2.H i* ) 2 + β 9 Chi + β10 Incomei + β11Watch _ TVi +
β12 Listen _ Radioi + β13 Re ad _ Newspaperi + β14 X i + β15 Z i )
= G ( xi β ) = Φ ( z )

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

And the likelihood function is:

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

“human capital function”.

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

variances. In this case, let SC be the p × p correlation or covariance matrix to be analysed.

The eigen decomposition of SC is:

p
SC = VΛV ′ = ∑ λi vi vi'
i =1

Where vi vi' = δ ij (i.e., orthonormality)

154
 λ1 ≥ λ2 ≥ ... ≥ λ p ≥ 0

The asymptotic distribution of the eigenvectors v̂i and eigenvalues λ̂i of a covariance

matrix SC for a sample from a multivariate normal distribution N ( µ , ∑ ) was derived by

Girshick (1939). Implementing the theory of first principal component, we can generate a

predicted value of the variable called social capital “SC”.

Table 5.6: Descriptive Statistics (Sample Size of Children in Age Group 5-14 Years Old)

Variable Definition Mean Std. Dev.

Total Income Total income of household 2665966 1811022

Child’s Age Age of child 10.32243 2.677284
Female Child 1 if female child, 0 otherwise .5 .5011723
hours worked Hours worked in the past 7 days 8.483645 12.77214
Child labour 1 if working child, 0 otherwise .5373832 .4997696
Child’s education Education of child (years) 3.065421 2.386468
Father’s education Education of father (years) 4.948598 2.934684
Mother’s education Education of mother (years) 4.252336 2.53832
Face food-shortage 1 if household faces food shortage in the past year .4497354 .4987884
Meals past 7 days 1 if have enough food in the past 7 days .8738318 .3328174
Nb. Family Number of family members 5.593458 1.883497
Nb.labour Number of labour force in the family 2.5 1.157746
Nb.living apart Number of children living apart .8317757 1.182708
Breastfeeding 1 if a child is breastfed in 1 year period, 0 .2663551 .4430886
otherwise
Female household head 1 if female is a household head, 0 otherwise 1.163551 1.705109
Use toilet& boiled water 1 if household either has toilet, or use boiled water .2056075 .4050924
Immunization 1 if child is immunized, 0 otherwise .864486 .3430743
Training attended Number of trainings household attended 3.799065 3.622024
Listening to radio 1 if listening to radio, 0 otherwise .0700935 .2559031
Watching TV 1 if watching TV, 0 otherwise .2803738 .450235
Reading newspaper 1 if reading newspaper, 0 otherwise .546729 .4989788
Times of injury Number of time child injured 3.014019 5.285714

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

Source: Author’s calculation from survey data 2006

Note: Total observation 214

5.4. Results and Discussions

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

results from the 2SLS estimators.

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.

Breastfeeding: In Table 5.8, the variable “breastfeeding” is statistically significant. It

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

breastfeeding indicates the underlying socioeconomic background of that individual household

in that it is unable to provide a necessary level of nutrients for the child.

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

Development Project (RDP) implemented by Ministry of Rural Development. This finding

also supports the claim that Cambodia faces the grim reality that a significant proportion of its

children are suffering from malnutrition (CDHS-2001). In 1999, the estimation of

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

on subsistence farming, which income difficult to calculate, although we try to measure

income using several important variables such as assets of the household.

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

better child health.

Access to media, participation in health programs: Access to media (radio, television,

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

clinic located in the commune health post.

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

drinking water in the dry season.

Age of household head: The coefficient of “age of the household head” is statistically

significant while father’s and mother’s education are not.

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

clearly revealed in this study.

Household’s member living-apart: The coefficient of family member living-apart “Nb

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

members living apart.

Injury: The coefficient of variable “times of injury” is statistically significant in Table

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.

Female household head: The coefficient of “female household head” is statistically

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.

Community characteristics: A lot more has to be explored, in terms of the basic

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

Age of the Child Male Child Female Child Total

5 0 0 0
6 0 0 0
7 2.4285714 2.1428571 2.3333333
8 .27777778 1.3636364 .68965517
9 3.125 2.25 2.6875
10 6 8.4615385 7.2307692
11 9 14.166667 11.583335
12 14.666667 11.125 12.642857
13 11.714286 13.941176 12.935484
14 19.066667 21.7 20.383335
Total 7.5981308 9.3691589 8.4836449
Source: Author’s calculation from the field survey data, 2006

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

Least Squares (2SLS) and check its validity and over-identification.

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

explained by indicators of access to information- reading newspapers, watching television,

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

associated to the question “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? This

simply means that the “Social Capital” through mutual helps and solidarity contributes to the

better child health.

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 5.1: Correlations between Parents’ Education and Media

(1) (2) (3) (4) (5)

(1) Father’s Education 1.0000

(2) Mother’s Education 0.1654 1.0000

(3) Listening to Radio 0.1375 0.0792 1.0000

(4) Watching TV -0.0139 -0.0252 -0.1559 1.0000

(5) Reading Newspaper 0.1443 0.1574 -0.1606 -0.1685 1.0000

Table Annex 5.2: Probit Coefficient Estimates, the Estimated Value of “Listening to Radio”

Dependent Variable: Ratio (1 listening Standard Probit Model

to radio)
Coefficient Marginal Effect P-value
Father’s Education .1875005 .0214972 0.020
Mother’s Education .0517656 .005935 0.533
Child’s Age .0031018 .0003556 0.955
Female Child -.0828752 -.0095025 0.815
Father’ Age -.0721267 -.0082694 0.131
Mother’s Age .0366176 .0041983 0.269
Face Food Shortage -.3546563 -.0398461 0.212
Meals Past 7 Days -1.710248 -.4294686 0.000
Family Size .0464183 .0053219 0.744
Nb of Labor -.0129573 -.0014856 0.945
Nb Living Apart -.0140835 -.0016147 0.954
Health Trainings attended .0570373 .0065394 0.076
Off-Farm Income (dummy) -.1291851 -.0158954 0.737
Remittance (dummy) .7101601 .093527 0.300
Agriculture Land (dummy) -1.095103 -.2347669 0.075
Imputed House Price -3.52e-08 -4.04e-09 0.250
Agricultural Assets -2.67e-07 -3.07e-08 0.351
Consumption Goods -9.82e-08 -1.13e-08 0.032
PreyChangVa .9738844 .1537809 0.150
Khan Damra 1.282908 .2749363 0.121

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”

Dependent Variable: TV (1 if Watching Standard Probit Model

TV)

Coefficient Marginal Effect P-value

Father’s Education -.032482 -.0076341 0.575

Mother’s Education -.067615 -.0158913 0.350

Child’s Age -.0679749 -.015976 0.177

Female Child .05737 .0134857 0.820

Father’ Age -.1492797 -.0350848 0.000

Mother’s Age .1304215 .0306526 0.005

Face Food Shortage 1.76601 .4431437 0.000

Meals Past 7 Days .4884641 .0940377 0.338

Family Size -.5306923 -.124727 0.000

Nb of Labor .4723689 .1110195 0.014

Nb Living Apart .1432577 .0336694 0.413

Health Trainings attended .0160662 .003776 0.592

Off-Farm Income (dummy) .687554 .1243338 0.132

Remittance (dummy) .2844298 .0688671 0.581

Agriculture Land (dummy) -.6585218 -.1959971 0.241

Imputed House Price 5.98e-08 1.40e-08 0.084

Agricultural Assets -7.08e-07 -1.66e-07 0.018

Consumption Goods 3.51e-08 8.26e-09 0.601

PreyChangVa .1381894 .0333949 0.687

171
Dependent Variable: TV (1 if Watching Standard Probit Model

TV)

Coefficient Marginal Effect P-value

Khan Damra .5200827 .1446318 0.263

KolKorm .4719389 .1225662 0.176

Constant 1.012694 0.406

Source: Author’s Calculation

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”

Dependent Variable: Newspaper (1 if Standard Probit Model

Reading Newspaper)

Coefficient Marginal Effect P-value

Father’s Education .0319405 .012237 0.523

Mother’s Education -.0123751 -.0047411 0.833

Child’s Age .0524102 .0200793 0.238

Female Child -.0801575 -.0307041 0.725

Father’ Age .0312431 .0119698 0.206

Mother’s Age -.0089761 -.0034389 0.513

Face Food Shortage -1.603374 -.5677003 0.000

Meals Past 7 Days 1.290263 .4731316 0.000

Family Size .2934508 .1124264 0.002

Nb of Labor -.203906 -.0781201 0.174

Nb Living Apart -.2323493 -.0890173 0.197

Health Trainings attended -.0738824 -.0283057 0.013

Off-Farm Income (dummy) -.5319453 -.1883907 0.182

Remittance (dummy) .2608332 .0988787 0.583

172
Dependent Variable: Newspaper (1 if Standard Probit Model

Reading Newspaper)

Coefficient Marginal Effect P-value

Agriculture Land (dummy) .7270959 .2837629 0.116

Imputed House Price -6.03e-09 -2.31e-09 0.839

Agricultural Assets 7.65e-07 2.93e-07 0.011

Consumption Goods 1.12e-10 4.30e-11 0.999

PreyChangVa .1360177 .0516575 0.672

Khan Damra -.2553817 -.0997098 0.563

KolKorm -.5849214 -.2273585 0.111

Constant -2.57632 0.017

Source: Author’s Calculation

Number of obs = 189 Wald chi2(21) = 70.70 Prob > chi2 = 0.0000
Pseudo R2 = 0.3512 Log pseudolikelihood = -83.542712

Table Annex 5.5: Estimated Value of Total Household Income

Dependent Variable is Total Income OLS Regression with Corrected Standard of Error

Coefficient P-value

Off-farm income 1042146 0.000

Remittance 1461424 0.000

Imputed House Price .1712185 0.000

Agricultural Assets .0524898 0.282

Father Education 771831.6 0.000

Mother Education 195348.9 0.505

PreyChangVa 813727.8 0.002

KhanDamra 407852.7 0.188

KolKorm 140517.5 0.054

Constant 341436.8 0.066

Number of obs = 214 F( 9, 204) = 15.38 Prob > F = 0.0000

R-squared = 0.4421 Root MSE = 1.4e+06

173
Table Annex 5.6: Result of First Principal Component for Variable of “Social Capital”

(Principal component factors; 2 factors retained)

Factor Eigenvalue Difference Proportion Cumulative

1 1.27377 0.27307 0.4246 0.4246

2 1.00070 0.27517 0.3336 0.7582
3 0.72553 . 0.2418 1.0000

Factor Loadings

Variable 1 2 Uniqueness

Q1 0.78493 0.18379 0.35010

Q2 0.79854 -0.00658 0.36229
Q5 -0.14137 0.98330 0.01314

(Varimax rotation) Rotated Factor Loadings

Variable 1 2 Uniqueness

Q1 0.80229 0.07893 0.35010

Q2 0.79074 -0.11157 0.36229
Q5 -0.01078 0.99335 0.01314

Scoring Coefficients

Variable 1

Q1 0.63503
Q2 0.62060
Q5 0.01925

Table Annex 5.7: Matrix of Covariates Correlation Used in the Model

(1) (2) (3) (4) (5) (6) (7) (8) (9)

(1) Total Income 1.0000
(2) Child’s Age 0.0779 1.0000
(3) Female Child 0.0455 0.0631 1.0000
(4) Hours Worked -0.063 0.5065 0.0922 1.0000
(5) Hours Worked^2 -0.059 0.4179 0.0908 0.9536 1.0000
(6) Child’s Education 0.2577 0.7245 0.0375 0.4504 0.3796 1.0000

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

6.1. Conclusion of the Study

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

survey data of Cambodia (CSES-1998/99, and CCLS-2001/02) to test these hypotheses.

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

distributed in the rural areas of Cambodia.

6.1.2 Major Findings in Chapter IV

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

parents to invest more in their children’s education.

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

contribute to the policy formulation as stated in the next heading.

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

schooling compared with the rural areas.

6.1.3 Major Findings in Chapter V

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

development agencies and the Royal Government of Cambodia.

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

be established as a warning system to protect children from harmful environments.

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

mutual helps and solidarity contributes to the better child health.

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

reduce time of schooling.

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.

6.2. Policy Recommendations and Policy Implications

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,

provides herewith sets of recommendations and policy implications to the governments,

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

Cambodia. The following qualitative and quantitative policy measures are:

6.2.1 Major Policy Recommendation for Investing in Children’s Education

It is widely accepted that education plays major role in sustainable development as it

will empower men and women in society to have better knowledge for their innovation and

technology. Most developing countries encounter many constraints to improving education.

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

multiple benefits from implementing this policy.

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

Understanding that health is an important element for survival and children

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

cumbersome, which create all business environments to encounter unnecessary regulation,

licensing requirements and too many inspections. All of these can deter investors, deny job

opportunities to local communities, and impede efforts to reduce poverty.

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

Reap, and Shihanouk Ville.

6.2.4 Other Policy Implications Emerging from this Study

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

outnumbered of that female’s participation. Policies to improve and increase access to

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,

March 2007, The Asian Population and Development Association.

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”.

Geneva, International Labour Office.

Baland, J. M and J. A. Robinson, 2000, “Is Child Labour Inefficient?,” Journal of Political

Economy, 108, pp. 663-679.

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

International Labour Standard,” Journal of Economic Literature, 37, pp. 1083-1119.

Basu, K. and P. Van, 1998, “The Economics of Child Labour,” The American Economic

Review, 88, pp. 412-427.

Becker, G. S, 1960, An Economic Analysis of Fertility in Demographic and Economic

Changes in Developed Countries, NBER, Princeton University Press, Princeton, NJ.

Becker, G. S, 1962, “Investment in Human Capital: A Theoretical Analysis,” Journal of

Political Economy, 70, pp. 9-49.

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,”

Journal of Economic Literature, 14, pp. 817-826.

Becker, G. S, 1993, “Government, Human Capital, and Economic Growth,” Industry of Free

China, 79, pp. 47-56.

Bhalotra, S. and C. Heady, 2003, “Child Farm Labour: The Wealth Paradox,” The World

Bank Economic Review, 17, pp. 197-227.

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

Development 3 and Latin America and the Caribbean Region.

Blunch, Niels. H, 2004, “Maternal Literacy and Numeracy Skills and Child Health in Ghana,”

Paper Submitted for Presentation at Fourth BREAD Conference on Development

Economics, Washington DC.

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

Research WP, No.1844, World Bank, Washington, DC.

Chakraborty, S and M. Das, 2005, “Mortality, Fertility, and Child Labour,” Economics

Letters, 86, 273-278.

Chao, S and O. Alper, 1998, Accessing Basic Education in Ghana, Studies in Human

Development No.1, World Bank, Washington, DC.

193
Christelle Dumas, 2007, “Why Do Parents Make their Children Work? A Test of the Poverty

Hypothesis in Rural Areas of Burkina Faso,” Forthcoming in Oxford Economic Papers.

Cochran, Susan, Joanne Leslie, and Donald O’Hana, 1982, “Parental Education and Child

Health: Intra-country Evidence,” Health Policy and Education, 2, pp. 213-250.

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,

India,” Population and Development Review, 13, pp. 77-100.

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,

91, pp. 795-813.

Duryea, Suzanne and A. K. Mary, 2001, School Attendance, Child Labour and Local Labour

Markets in Urban Brazil, Inter-American Development Bank.

Dwyer, D. and J. Bruce, 1988, A Home Divided: Women and Income in the Third World.

Stanford, California, Stanford University Press.

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

South Africa,” Forthcoming in Journal of Development Economics.

Edmonds, E and C. Turk, 2004, “Child Labour in Transition in Vietnam,” World Bank Policy

Research Paper No.2774, Washington DC, World Bank.

194
Fallon, Peter and Tzannatos, Zafiris, 1998, Child Labour: Issues and Directions for the World

Bank, Washington, DC, World Bank.

Fan C. Simon, 2004, “Relative Wage, Child Labour, and Human Capital”, Oxford Economic

Papers, 56, pp. 687-700.

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

UNOPS, Asian and Pacific Regional Office, Kuala Lumpur.

Fuglesang, A, and D. Chandler, 1993, “Participation as Process-Process as Growth: What We

Can Learn from Grameen Bank,” mimeo, Bangladesh.

Geremek, B., 1994, Poverty: A History, Oxford: Blackwell.

Glewwe. P, 1999, “Why Does Mother’s Schooling Raise Child Health in Developing

Countries?,” Journal of Human Resources, 34, pp. 124-159.

Grootaert, C and K. Ravi, 1995, “Child Labour: An Economic Perspective,” International

Labour Review, 134, pp. 187-203.

Guarcello. L, S. Lyon and F. C. Rossati, 2004, “Impact of Working Time on Children’s

Health,” Understanding Children’s Work Discussion Paper.

HRW World Report, 2002, Children's Rights, Human Rights Watch.

Heady, Christopher, 2003, “The Effect of Child Labour on Learning Achievement,” World

Development, 31, pp. 385-398.

ILO (International Labour Organization), 1998, Child Labour: Targeting the Intolerable,

International Labour Office, Geneva.

ILO (International Labour Organization), 2002, “Every Child Counts: New Global Estimates

on Child Labour,” mimeo, International Labour Office.

195
Jensen, P and H. S. Nielsen, 1997, “Child Labour or School Attendance? Evidence from

Zambia,” Journal of Population Economics, 10, pp. 407–24.

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

Journal of Human Resources, 4, pp. 665-696.

Khanam, Rasheda, 2003, “Child Labor and School Attendance: Evidence from Bangladesh,”

mimeo, University of Sydney.

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.

Lee, B. C, L. S. Jenkins and J. D. Westaby, 1997, “Factors Influencing Exposure of Children

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

of Health Economics, 23, pp. 259-291.

Mincer, Jacob, 1974, Schooling, Experience and Earnings, Columbia University Press,

National Bureau of Economic Research, New York.

MoEYS (Ministry of Education, Youth and Sports), 2002, Education Statistics and

Indicators, Department of Planning, Phnom Penh, Cambodia.

MoEYS (Ministry of Education, Youth and Sports), 2005, Education Management

Information System (CD-Room), Phnom Penh, Cambodia.

MoEYS (Ministry of Education, Youth and Sports), 2005, Education Sector Performance

Report, Phnom Penh, Cambodia.

MoH (Ministry of Health), 2005, Master Plan for Social Health Insurances in Cambodia,

Phnom Penh, 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.

MoP (Ministry of Planning), 2003, Cambodia Millennium Development Goals Report,

Prepared with Funding from UNDP and Sida, Phnom Penh, November 2003.

NIS, 1994, Report on the Cambodia Socio-Economic Survey, National Institute of Statistics,

Ministry of Planning, Phnom Penh, Cambodia.

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,

Ministry of Planning, Phnom Penh, Cambodia.

NIS, 2002, Report on Cambodia Child Labor Survey, National Institute of Statistics, Ministry

of Planning, Phnom Penh, Cambodia.

NIS, 2004, Cambodia Inter-censal Population Survey, National Institute of Statistics,

Ministry of Planning, Phnom Penh, Cambodia.

NIS, 2004, Report on the Cambodia Socio-Economic Survey, National Institute of Statistics,

Ministry of Planning, Phnom Penh, Cambodia.

NIS, 2005, Cambodia Demographic and Health Survey, National Institute of Statistics,

Ministry of Planning, Phnom Penh, Cambodia.

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

Performance: A Review,” mimeo, March.

197
O’Donnell. O, E. Van Doorslaer and F.C. Rossati, 2002, “Child Labor and Health: Evidence

and research issues,” Understanding Children’s Work.

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

University of California at Berkeley.

Parker. D, 1997, “Child Labour: The Impact of Economic Exploitation on Health and Welfare

of Children,” Minnesota Medicine, 80, pp. 10-13; 52-55.

Patrinos, H. A and G. Psacharopoulos, 1995, “Educational Performance and Child Labour in

Paraguay,” International Journal of Educational Development, 15, pp. 47–60.

Patrinos, H. A and G. Psacharopoulos, 1997, “Family Size, Schooling and Child Labour in

Peru_ An Empirical Analysis,” Journal of Population Economics, 10, pp. 387–405.

Pranab Bardhan and Udry Christopher, 1999, Development Microeconomics, chapter 3:

Population, Oxford University Press, pp 21-31.

RGC, 2002, National Poverty Reduction Strategy, Prepared by Council for Social

Development (CSD), Royal Government of Cambodia.

RGC, 2002, Second Five-Year Socio-Economic Development Plan 2001-2005, Royal

Government of Cambodia.

RGC, 2004, Rectangular Strategy, Disseminated by Council of Ministers, Royal Government

of Cambodia.

RGC, 2005, National Poverty Reduction Strategy-Progress Report, SNEC, MEF, Phnom

Penh, Cambodia.

RGC, 2005, National Strategic Development Plan 2006-2010, Royal Government of

Cambodia.

198
Ranjan, Manash Gupta, 1998, “Wage Determination of a Child Worker: A Theoretical

Analysis,” Review of Development Economics, 2, pp. 219-228.

Raut, Lakshmi K. and Tran, Lien H., 2005, “Parental Human Capital Investment and Old-Age

Transfers from Children: Is It a Loan Contract or Reciprocity for Indonesian Families?,”

Journal of Development Economics, 77, pp. 389-414.

Ravallion, M. and Q. Wodon, 2000, “Does Child Labour Displace Schooling: Evidence from

Behavioural Responses to an Enrolment Subsidy,” Economic Journal, 110, pp. 158-176.

Ray, Ranjan and G. Lancaster, 2005, “The Impact of Children’s Work on Schooling: Multi-

Country Evidence,” International Labour Review, 144, pp. 189-210.

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

Contribution of Rural Children in India,” Econometrica, 45, pp. 1065-1079.

Satz Debra, 2003, “Child Labour: A Normative Perspective,” The World Bank Economic

Review, 17, pp. 297-309.

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

Institute, University of Wisconsin-Madison.

Schapiro, Tamar, 1999, “What is a Child?,” Ethics, 4, pp. 715-738.

199
Shah, Iqbal H and Khamma, Jitendra, 2002, “Breast-feeding, Infant Health and Child Survival

in the Asia-Pacific Context,” Asia Pacific Population Journal. 5, pp. 25-44.

Strauss. J and D. Thomas, 1998, “Health, Nutrition, and Economic Development,” Journal of

Economic Literature, 36, pp. 766-817.

Thomas. D., J. Strauss and M. H. Henriques, 1991, “How Does Mother’s Education Affect

Child Height?,” Journal of Human Resources, 26, pp. 183-211.

UCW (Understanding Children’s Work), 2007, “Child Labour and Education for All,”

Retrieved in 2007 from http://www.ucw-project.org/

UN (United Nations), 2000, UN Millennium Submit, New York.

UNDP (United Nations Development Program), 2005, Human Development Report.

US Department of Labour, 2000, By the Sweat and Toil of Children: An Economic

Consideration of Child Labour, Washington DC: US Department of Labour, Bureau of

International Labour Affairs.

USGAO (United State General Accounting Office), 1998, Child Labour in Agriculture:

Changes Needed to Better Protect Health and Educational Opportunities. Report to

Congressional Requesters.

Udry, C., 2006, “Child Labour,” Understanding Poverty, edited by Abhijit Vinayak Banerjee,

Roland Benabou and Dilip Mookherjee , Oxford University Press.

Wolfe. B and J. Behrman , 1987, “Women’s Schooling and Children’s Health,” Journal of

Health Economics, 6, pp 239-54.

Wooldridge, Jeffrey M., 2003, Introductory Econometrics, A Modern Approach, Thomson,

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