ESSAYS ON EDUCATION AND INTERGENERATIONAL TRANSFERS …

~h\J 2D1~:4(P),7

ESSAYS ON EDUCATION AND INTERGENERATIONAL TRANSFERS ININDONESIA

A DISSERTATION SUBMITTED TO THE GRADUATE DIVISION OF THEUNIVERSITY OF HAWAI'IIN PARTIAL FULFILLMENT OF THE

REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

IN

ECONOMICS

AUGUST 2005

By

Maliki

Dissertation Committee:

Andrew Mason, ChairpersonTimothy J. Halliday

Sang-Hyop LeeGerard Russo

Robert D. Retherford

ACKNOWLEDGMENT

lowe a great debt of gratitude to a number of people for their support and

guidance during my studies and in the preparation and completion of this

dissertation. First of all, I would like to express my deepest gratitude to Andrew

Mason, who advised me through all the stages of this dissertation. His comments

and suggestions were very nurturing in the development of my research. I really

appreciate for his encouragement, which made me believe that I could complete

this dissertation. Gerard Russo and Robert Retherford provided me with

constructive comments and encouragement throughout the writing of this

dissertation. My gratitude is to Sang-Hyop Lee for motivating me to achieve my

highest quality. I received significant guidance and advice along the way from

Timothy Halliday. I am also grateful to Francois Wolff for his comments, as they

improved the quality of this dissertation.

Ronald Lee of the University of California at Berkeley provided me with access to

Susenas data, which was necessary for this research. Carl Boe assisted me with

utilizing Berkeley's resources, which greatly enriched my writing. My gratitude for

Walter W. McMahon of the University of Illinois at Urbana-Champaign, Gustavo

de Santis, Robert Sparrow, and Edward Norton all of whom provided me with

important additional data or helpful programs. I would like to express my thanks

to the participants in the Global Conference in Education Research Results in

Prague. It was a pleasure for me to present one of my papers at this conference,

111

where I received valuable comments. My appreciation is for Professor Peter

Orazem who made a lot of important comments during the conference. East~

West Center Fellowships provided me with financial support as well as Research

Grants, which made this dissertation possible. Joy Sakurai of the USA Embassy

in Jakarta has allowed me to use their facilities for Remote Dissertation Defense.

Comfort Sumida gave her valuable time for editing this dissertation. Jaida

Samudra, Kimberly Burnett, Jenny Garmendia and David Lusan also sacrificed

their time to check over this dissertation and made it worth reading. Benny Azwir,

Muhammad Ikbal, and Erna Rosita have worked hard to provide me with

important data on the education public budget. Nicole and Turro Wongkaren are

wonderful friends who kept the graduate school atmosphere livelier. My

appreciation is given to Suchart's family and Beet for being our friends and

family. I know that you both will follow very soon. Keep your spirits up. Someday

we will meet again either in Jakarta, Chiang-Mai, Bangkok, or some other part of

the world.

My mother and parents-in~law provided me with a lot of support. This step of my

life is incomplete without your blessing. Wini, my wife, has stood behind me at

every step along the way. She has shown limitless patience and endurance

throughout the entire period of my study. Last, but not least, my daughter, Zafia,

who makes everywhere home with her presence. Everything we have been doing

is only for you.

IV

ABSTRACT

The objective of this dissertation is to investigate how private intergenerational

transfers respond to public policy changes in Indonesia. In particular, this paper

investigates how private education and non-education transfers respond to newly

implemented education policies. This dissertation contributes to the existing

literature on intergenerational transfers where there exist few investigations

regarding empirical relationships between household decisions on school

demand and child labor supply, private intergenerational transfers, and public

policy.

This dissertation consists of three essays. The first essay develops a new

method of estimating familial education and non-education transfers and public

education transfers, using Indonesian Socio Economic Survey (Susenas) and

Indonesian government budgeting data. The purpose of the second essay is to

investigate how the introduction of nine-year compulsory education affects school

enrollment and child labor supply. The third essay examines how this same

education policy has influenced familial educational investment decisions and

non-educational transfers in Indonesia.

Using the distance to the nearest school as an approximation of education

policies implemented between 1993 and 1996, difference-in-differences results

indicate that the policies led to a decline in child labor supply and an increase in

v

school enrollment of 2% to 4% among children aged 11 to 15. However, child

labor did not decline proportionally. For an average of 6 km change in the school

distance, non-educational transfers increased by as much as 5%. On the other

hand, educational transfers increased by 10%. The non-educational transfer

changes were due to both declining child labor income and increasing non

educational consumption. Thus, parents bear the higher education cost and lost

opportunity cost by sending their children to school. In addition, non-education

transfers are complementary with education expenditures. It is concluded that

parents are still bound by the compulsory education laws. Households require

their children to work in order for them to fulfill the higher expenditures. A

subsidy is necessary in order for them to send their children to school and to

reduce the child labor supply.

VI

TABLE OF CONTENT

ACKNOWLEDGMENT 111

ABSTRACT V

LIST OF TABLES IX

LIST OF FIGURES XI

ESSAY 1: ESTIMATION OF PRIVATE EDUCATION AND NON-EDUCATIONTRANSFERS AND PUBLIC EDUCATION TRANSFERS 1

1. BACKGROUND AND OBJECTIVE 22. LITERATURE REVIEW 43. EDUCATION IN INDONESIA 11

3. 1 Education System and Policy 113.2 Background on Indonesian Education Financing 17

4. DATA DESCRiPTION 225. METHODOLOGY 31

5.1 Estimation ofPrivate Education Transfers 325.2 Estimation of Public Education Transfers .43

6. RESULTS AND DiSCUSSiON .466.1 Estimation ofPrivate Education Transfers Results 466.2 Estimation ofNon-education Transfers Results 596.3 Estimation ofPublic Education Transfers 63

7. CONCLUSiONS 76

ESSAY 2: EDUCATION POLICY, CHILDREN'S SCHOOLING, AND LABORDECISIONS .....................................................................•.................................79

1. BACKGROUND AND OBJECTiVE 802. LITERATURE REVIEW 833. CONCEPTUAL FRAMEWORK 874. DATA AND EMPIRICAL STRATEGy 97

4.1 Data Description 974.2 Empirical Strategy 1074.3 Simple Differences 112

5. EMPIRICAL RESULTS 1165. 1 The Effect of School Distance on Children's Activities 1165.2 The Effect of Education Policies on Parental Labor Decisions 133

6. CONCLUSiONS 137

Vll

ESSAY 3: THE EFFECT OF EDUCATION POLICY ONINTERGENERATIONAL TRANSFERS 140

1. MOTIVATIONS AND OSJECTIVES 1412. LITERATURE REVIEW 1453. CONCEPTUAL FRAMEWORK 149

4. 1 Data Sources 1544.2 Empirical Analysis 158

5. CONCLUSIONS 191

APPENDiX 194

REFERENCES 196

YIn

LIST OF TABLES

1.1 Type of Budget and Responsible Ministry 18

1.2 Total Annual Expenditures on Education by Government Agencies, bySource of Funds, and Levelof Schooling, 1995 - 1996 23

1.3 Mean Value by Education of Household Head 27

1.4 Goodness-of-fit Average Cost (Coefficients of Regression) Over Non-Parametric Average Cost .49

1.5 Regression Results of Estimated Data on Real Data 50

1.6 Average Age of Transfers' Recipients and Providers 57

1.7 Equivalence Scale Comparison 59

1.8 Education Financing by Ministries and School Level (in Billion Rupiah) .....65

1.9 Average Age of Public Education Transfers 69

1.10 Average and Accumulated Private and Public Education Transfers 75

2.1 Variable Means on School Enrollment and Employment 100

2.2 Variable Mean on Employment of Household Heads and Spouses 102

2.3 Regression Results on Determinants of School Enrollment and EmploymentDecisions, Susenas 1993 106

2.4 Interpretation of Difference-in-Differences of Equation (2.9) 111

2.5 Non-parametric Difference-in-Differences Tabulation on Child Labor andEnrollment Between 1993 and 1996 114

2.6 OLS Regression Results for Difference-in-Differences with Four DifferentDependent Variables: Coefficients of Interaction 118

IX

2.7 OLS Regression Results with Empl9yment, School, and Hours Worked asthe Dependent Variables, Clustered by Sub-district Level: Coefficients ofInteraction 123

2.8 Non-parametric DID: Effect of a Change in Distance on Parental LaborSupply, 1993 to 1996 135

2.9 Coefficients of Interaction Difference-in-Differences: the Effect ofEducationPolicy on Parental Labor Supply 136

3.1 Descriptive Statistics 157

3.2 Interpretation of DID model 161

3.3 Non-parametric Difference-in-Differences Tabulation on Schooling Demandand Employment Decisions 164

3.4 Regression (OLS) Results for Difference-in-Differences Between TwoTypes of Cohort (Treatment Group 12 - 15 and Control Group 20 - 25) ....168

3.5 Estimates the Effects of Education Policy on Non-Education Transfers:Coefficients of Interaction Between Age Variable Dummy at 1993 or 1996and Distance to the Nearest School at 1993 or 1996 175

3.6 Estimates of the Effects of Education Policy on Education Transfers:Coefficients of Interaction Between Age Variable Dummy at 1993 or 1996and Distance to the Nearest School at 1993 or 1996 179

Appendix Table 1 Education Policy Milestone in Indonesia 194

Appendix Table 2 Summary of Comparative Static 195

x

LIST OF FIGURES

Figure

1.1 General Education and Islamic Education System in the 1950's 12

1.2 Formal School System Based on Law No.2 1989 13

1.3 Enrollment Rate and Education Financing Over Time 18

1.4 Private Education Transfers Resources 30

1.5 Illustration of The Engel Method 35

1.6 Illustration of The Rothbarth Method .40

1.7 Regression Results for Education Expenditures on Enrolled Age Groups:Estimated Coefficients jJ by Age Group .47

1.8 Comparison Between Actual Data and Predicted Individual EducationExpenditure 48

1.9 Regression Results of Education Expenditures on Enrolled Age Groups:Susenas 1993, 1996, 1999, and 2002 52

1.10 Private Education Transfers Profiles 1993, 1996, and 1999 53

1.11 Monthly Education Transfers Outflow by Household Head as PrincipalEarners 55

1.12 Net Education Transfers Flow with Household Head as Principal Agents ..56

1.13 Private Education Transfers Flow 58

1.14 Consumption Allocation Using Split Method 61

1.15 Consumption Allocation Profile Using the Split Method and LinearProportion Allocation (0.2 - 0.8) for Children 62

1.16 Average Public and Private Expenditure Per Capita Per School Level. ......68

1.17 Per Capita Public Education Transfers Outflow and Inflow 68

Xl

1.18 Per Capita Public and Private Education Transfers by Age of Recipient1993, 1996, 1999 71

2.1 The Effect of Education Subsidy and Compulsory Education on SchoolDemand 92

2.2 The Welfare Analysis of the Effects of Education Subsidy on School andChild Labor Demand 96

2.3 Non-parametric Difference-in-Differences Results Using 1993/1996 and1993/2002 115

2.4 Effect of the Junior High School Distance Changes on School Enrollmentand Employment Decisions by Age for All Years (1993,1996,1999,2002)...................................................................................................................125

2.5 Effect of the Junior High School Distance Changes on Number of HoursWorked by Age 126

2.6 DID of School and Work Decisions Using 1993, 1996, 1999, and 2002Survey Data: Effect of the Program by Age Comparing Boys vs. Girls andUrban vs. Rural 129

2.7 DID of School and Work Decisions Using 1993, 1996, 1999, and 2002Survey Data: Effect of the Program by Age Siblings' Effect on HouseholdDecisions 131

3.1 School Demand and Working Decision 166

3.2 Estimates of the Effects of Education Policy on Non-Education Transfers:Coefficients of Interaction Between Age Variables Dummy and Distance tothe Nearest Junior High School 177

3.3 Estimates of the Effects of Education Policy on Education Transfers:Coefficients of Interaction Between Age Variables Dummy and Distance tothe Nearest Junior High School 180

3.4 Estimates of the Effects of Education Policy on Labor income: Coefficientsof Interaction Between Age Variable Dummy and Distance to the NearestSchool 183

3.5 Estimates of the Effects of Education Policy on Non-EducationConsumption: Coefficients of Interaction Between Age Variable Dummy andDistance to the Nearest School. 184

xu

3.6 Estimates of the Effects of Education Policy on Non-Education TransfersUsing Restricted Sample: Coefficients of Interaction Between Age VariableDummy and Distance to the Nearest School 187

3.7 Estimates of the Effects of Education Policy on Education Transfers UsingRestricted Sample: Coefficients of Interaction Between Age Variable Dummyand Distance to the Nearest School 188

3.8 Estimates of the Effects of Education Policy on Labor Income UsingRestricted Sample: Coefficients of Interaction Between Age Variable Dummyand Distance to the Nearest School. 189

X111

ESSAY 1: ESTIMATION OF PRIVATE EDUCATION AND NON-EDUCATIONTRANSFERS AND PUBLIC EDUCATION TRANSFERS

1

1. Background and Objective

Education can be perceived either as an investment or as consumption (Schultz

1960). As an investment, education 'enables' children and allows them to

contribute to a productive economy. Education stimulates and increases human

potential. In a society where retired parents commonly depend on their children

for support, the education of children is an investment for both parents and

children. Retired parents can expect returns on the wealth they have invested

into children's education. In other societies with a more accessible capital

market, parents may educate children for their own satisfaction.

Children's education and earnings enable parents to preserve or even enhance

their social status without concern for future monetary consequences. Within

these societies in general and among the parents who receive higher utility from

a child's education, education is perceived primarily as consumption, rather than

investment. However, there is no definite distinction between parents who

perceive education as consumption and those who perceive it as investment.

Parental attitudes often lie between the two perceptions. In addition to receiving

satisfaction from their children's educational achievements, parents expect some

financial return in the future.

The purpose of this paper is to develop a new method of estimating both familial

education and non-education transfers and public education transfers using the

Indonesian Socio Economic Survey (Susenas) and Indonesian government

2

budgeting data. The estimation of familial and public education transfers is an

attempt to construct an estimate of education transfers at the national level.

National education transfers are an element of the National Transfers Account

(NTA) projecti.

Deficiencies in investigations of intergenerational transfers have stemmed

primarily from the unavailability of sufficient data on individual transfers. This

paper is the first attempt to develop a new methodology for estimating individual

intergenerational transfers. Constructing a model of familial and public transfers

based on four years of national survey and fiscal data provides a significant

advance to the literature on intergenerational transfers with implications for

application to further economic analysis and policy formation.

Decomposition of household level educational expenditures, into data at the

individual level, enables us to analyze the age profile of private educational

transfers and comparison with public educational transfers. Transfers flow from

private sources such as parents or household heads to school age groups.

Public sources originate from taxation of the productive age group and are then

allocated by the government for education. This paper will analyze both private

and public educational transfers from a macroeconomic perspective. Based on

concepts presented in Becker and Tomes and Becker and Murphy, this paper

further investigates the relationship between government and parental transfers

for education. The question to be addressed is how private transfers for

education respond to government transfers and policy enforcement.

3

In the following section, I provide a brief review the literature to better illustrate

the importance of education transfers to human capital development. I describe

briefly the education financing system in Indonesia in section 3. In Section 4, the

data used for both private and public education account estimates are described.

The methodology of account estimation, results and analysis are covered in

Sections 5 and 6 respectively.

2. Literature Review

Whether perceived as an investment, consumption, or a combination of the two,

education involves intergenerational transfers. Quality of children can be

measured by their education or health. Education as an indicator of children's

quality is a means of human capital transmission, as intergenerational education

transfers are sustained from generation to generation. Parents transmit 'value' to

their children in the form of an education, and these children do the same for

their own in the future. Despite the lack of an explicit contract between parents

and children, this mechanism works in general. The 'value' brought by parents is

strong enough to sustain the mechanism into the future.

Previous attempts have beeh made to explain the flow of resources across

generations in both the familial system and public system. This paper follows the

conceptual framework developed by Lee for analyzing the intergenerational

transfer system. A synthesis of Lee's theoretical framework is applied on

4

intergenerational transfers as constructed by the National Transfers Account

Team (framework details provided in the NTA proposal submitted to the National

Institutes of Health (NIH) by Lee and Mason, 2004). Lee (1994) applies the

framework to transfers in the United States using household level data,

distinguishing between education, health, and social security. Lee and Edwards

forecast public transfers in the United States based on the contingencies of

current government policy. Luth similarly estimates intergenerational transfers in

Germany. Mason and Miller develop a model of familial transfers for Taiwan.

Mason and Ogawa also build on the model of the familial system by examining

the effect of bequests and living arrangements on savings in Japan. This paper

fills the gap in comprehensive estimation of individual private transfers and

estimation of public education transfers at the individual level, rather than the

household.

Familial education transfers have not been explicitly and comprehensively

examined. Lee (1994) finds, based on household level analysis, that in the

United States the direction of both private and public educational transfers is

downward, from older to younger age groups. This contrasts with the direction of

health and social security transfers, which flow from younger to older age groups.

The flow of educational transfers differs between developed countries and

developing countries in that the average age of recipients in developing countries

tends to be younger than those in developed countries. Lee also completes a

comprehensive construction of the direction of individual educational transfers.

5

Becker and Tomes discuss the trade-off between the quantity and quality of

children. Parents decide child quality by spending on education for their children.

This investment complements the inherited ability of children, which cannot be

controlled, as it is largely genetic. Hence, intergenerational transfers consist of

human capital and non-human capital transfers. Parents, regardless of wealth

status, transmit ability to their children, invest in their education, and provide gifts

and/or bequests. Altruistic parents react to a child's ability endowment by

adjusting educational investment, gifts, and/or bequests. Becker and Tomes

argue that parents invest in education for their children based on their perception

of children's initial 'endowments'. Parents are assumed to carry perfect

information on their children's abilities and use this information to determine the

amount they will transfer to their children. To compensate less able children,

parents bless them with non-human capital transfers, e.g. bequests.

Parental decisions to contribute more to the quality of their children by investing

in their education lead to a sacrifice of control over the quantity of children. The

shadow price of increasing quality is linearly related to the shadow price of

increasing quantity. Thus, for the same level of child quality, increasing the

number of children will increase costs. On the other hand, an exogenous

increase in the demand for quality will increase the shadow price of quantity

making quantity relatively more expensive.

Becker and Tomes (1976) also recognize the role of government in enhancing

children's quality. In addition to parental transfers, the government generates

6

educational transfers by providing public schools, supplies, and other support.

Becker and Tomes argue that government intervention in children's education

through public compensatory programs positively redistributes wealth. Hanushek,

Leung, and Yilmaz assess the relationship between educational subsidies,

negative income tax (NIT), and wage subsidies as government redistribution of

wealth. A government has to balance the bUdget by allocating taxes earned from

workers to the groups with which it is intervening. The transfer mechanism is

understood to move from the productive age group to the unproductive school-

age group. The interaction between family and government transfers may

depend on how far the government supports education and how supportive the

family is towards educating children, which may in turn depend on family

background, family income, and other unobserved factors.

There are at least three parties that have a direct relationship with education:

parents, government, and children. Becker and Tomes (1976) initiate the

analysis into the relationship among parents, children, and government. In an

extension to Becker and Tomes (1976), Becker and Murphy further discuss the

existence of government involvement in the family decision. Government

intervention in the parent-child relationship is aimed at ensuring that children

receive enough education. Market failure due to credit market constraints and

imperfect information are legitimate reasons for government intervention in family

I

decisions. Therefore, the government attempts to ensure parents send their

children to obtain enough education. Government intervention through increases

in the development budget for the education sector can reduce the inefficiency

7

that may result when relying solely on parents to invest in children's education

(Becker and Murphy 1988).

Previous literature investigating the effect of government intervention on private

education decisions is limited. There exist few studies on compulsory education.

Investigations of the effect of government subsidies and tuition policies on

enrollment rates have been conducted. Peltzman (1973) examines the

interaction between government subsidies and private education expenditures at

the college level. Schultz (2004) evaluates the effect of school subsidies, through

the Mexican Progressa poverty program, on enrollment. Duflo evaluates the

effect of constructing new elementary schools on the years of education and

earnings in Indonesia. In general, these studies find a positive effect of

government intervention on enrollment rates especially on basic education, years

of education, and future earnings.

Public education programs decrease social inequality. Benabou assesses the

effects of employing progressive income taxes and re-distributive educational

financing on macroeconomic indicators such as income, inequality, inter

temporal welfare, etc. in a dynamic heterogeneous-agent economy. He

concludes that education finance policies have relatively undistorted policy

implications compared to taxes and transfers. Educational policy also improves

income growth more than taxation and transfers.

8

Little investigation into the interaction between private decisions and government

intervention on human capital investment has been conducted. Zilcha and

Viaene contribute to this literature by modeling the effect of the form of

intergenerational transfer on income distribution and capital accumulation. A two

period overlapping generations model is used. Altruistic parents have two

transfer options, improving their offspring's earnings by investing in their

education or through a transfer of physical capital. Zilcha and Viaene conclude

that the parents' choice of transfer generates a less equal income distribution if it

produces more output. Under the presence of total public education intervention,

less equal intergenerational income distributions occur due to altruism and

providing bequest that result in higher aggregate output.

Cremer and Pestieau investigate the effects of an income tax structure and

optimal tax or subsidies on private education, as well as its effects on public

education provision. The model assumes that parents receive a 'warm glow'

effect, or personal satisfaction by investing their wealth in their children's

education. This is, to some extent, limited altruism in parents' behavior. Cremer

and Pestieau use a successive generation model with each of the generations

differentiated by their working ability. Each generation works and invests for their

children's education. Educated children vary by their probability of enhancing

their productivity in the future. It is found that if public and private education

investments are perfect substitutes, the public education provision will crowd out

private investment, but is still desirable. Through a non-linear tax or subsidy on

9

private investment, the "public education provision will contribute to relaxing the

self-selection constraint" (Cremer and Pestieau p.17).

Credit market limitations, redistribution of income, and the positive externalities of

education are justifications used by De Fraja for constructing an education policy

to be imposed by utilitarian government that intervenes in the household

decisions on education investments. His model provides non-uniform education

provisions, which depends on parental income and children's abilities to influence

parental contributions to the education funds. It is concluded that education

provision with the objective of income re-distribution will conflict with equity and

efficiency issues, as investing in the more able children improves efficiency more

than investing in the less able children.

Government intervention in education varies by country and may also vary at the

district level for decentralized systems. Some countries affect the direct cost of

education through vouchers or scholarships. Most governments regulate their

citizens through compulsory education laws. Direct interventions, such as

compulsory education, are not only popular in developing countries, but also in

developed countries during their early stages of development. Governments may

also tax income and re-distribute it to citizens through education provisions.

Human capital investment is important as an engine of growth and in improving

income distribution. Household reaction towards government intervention,

however, will vary depending on how a government interferes. Compulsory

10

education will require households to maintain children's education levels.

Voucher systems or support systems will change household income constraints.

Effects of the latter policy will depend on a family's preferences for child

education.

3. Education in Indonesia

3.1 Education System and Policy

The Indonesian education system is regulated by Law No.2 Year 1989 on

National Education System. The system is based on Indonesia's early national

education system from the 1950's that segregated schools into two divisions,

forming the general and Islamic systems. Figure 1.1 and Figure 1.2 provide a

visual comparison between the school systems over two generations. A stratified

school system of the 1950's is also presented in Figure 1.1. The formal education

system consists of general education and religion-based education that extends

from kindergarten through the higher education system. Religion-based

education follows the same levels as general education. The basic education

program consists of a 6-year elementary school and 3-year junior high school.

Senior secondary schools are divided into general and vocational schools. The

kindergarten and preschool levels are gaining popularity, and many families are

sending their children to school at an earlier age.

11

HigherEducation

Senior High SChool

Junior High SChool

Primary School

Islamic Law TeacherTraining SChool

Islam SeniorHigh School

Training School for

Islam Junior High SChoolTeacher of the Islam

religion

Religion Based Primary SChool

22

21

20

19

18

17

16

15

14

13

12

11

10

9

8

7

6

AGE

Sources: Compulsory Education in Indonesia

Figure 1.1 General Education and Islamic Education System in the 1950's

12

22

21

20

19

18

17

16

15

14

13

12

11

10

9

8

7

6

AGE

HigherEducation

SecondarySchool

Basic Education

Preschool

IslamicDoctorate Professional

DoctorateProgram

Program Program

IslamicMaster Professional

MagisterProgram Program

Program

Islamic UndergraduatUndergraduat e Degree Diploma 4

Diploma 3e Program ProgramDiploma 2

Diplomat

tslamicSS General SS Vocational SS

IslamicJSS General JSS

Islamic PrimaryPrimary SchoolSchool

Islamic Kindegarten Kindegarten

Sources: Indonesia Educational Statistics in Brief 2000/2001

Figure 1.2 Formal School System Based on Law NO.2 1989

13

Compulsory education first appeared on the development agenda in 1950, five

years after Independence Day, August 1945. The main objectives were to reduce

the large educational gaps between aristocrats and non-aristocrats, men and

women, and different ethnic groups. Indonesian aristocrats benefited from

education access during the Dutch colonial days, while the non-aristocrats

usually did not have open access to education. Further, the unequal access in

education between men and women was of primary concern.

Compulsory education helped to stimulate equality among different ethnic

groups, where some had previously benefited from the Dutch system. There

were initially 5 million students in elementary school (Sekolah Rakyat) with

another 5 million school-age children that needed schooling . As a result, the

implementation of the compulsory program required additional school buildings,

teachers, and school supplies. Despite the limited budget, the program started in

specifically appointed districts in all provinces, and by 1959, nearly 60% of the

districts had enforced compulsory education. Table A in the appendix

summarizes the milestones of education policies imposed in Indonesia since the

Colonized Era.

Education development has been the priority in all long-term development plans

during the Soeharto era. The first priority was to reach the quantity of education

as provided for in the 1973 President Instruction Program (SD INPRES), when

the Indonesian government received a windfall from the oil price shock. As the

14

biggest oil producers and OPEC members, Indonesia received a large surplus

from oil exports, as the oil price nearly tripled. The government built around 150

thousand new school units, 166 thousand new classrooms, and rehabilitated

nearly 380 thousand school units from fiscal year 1973/1974 to fiscal year

1993/1994. The government also formally abolished tuition fees at the

elementary level in 1973 and financed increased book supplies and teacher

development programs.

The SO INPRES program was one of the most extensive projects that a

developing country had ever implemented. In 1984, as a part ofthis program, the

Indonesian government formally started its 6-year compulsory basic education

requirements. In 1994, ten years later, the government extended the program to

9-years of compulsory basic education.

The goal of the compulsory education program, to achieve universal basic

education, was threatened by the financial crisis. The social safety net

constructed soon after the 1997 financial crisis was meant to protect the poor

from its impact. The government started a scholarship program to assist the poor

in the primary and secondary school-aged groups in overcoming the impact of

the crisis.

The most recent development in the education policy is the effort to decentralize

the education system at the district level. According to Law number 22/1999 on

district governments and Law number 25/1999 on decentralization, the central

15

government has to hand authority over education policies to the district

governments. Prior to the decentralized system, education policy in Indonesia

was fully centralized. The central government had complete control over the

budget allocations for education to all schools. The Ministry of National Education

was the main executor for the education program with assistance from the

Ministry of Religion Affairs, the Ministry of Finance, the National Development

and Planning Agency, and the Ministry of Home Affairs. Under this centralized

system, most budgeting decisions were made by the central government through

the Ministry of National Education, leaving public schools with almost no

budgeting authority.

The decentralization policy has had major implications for education financing.

The districts were granted have the opportunity to decide whether education is

their development priority or not. Investment in education after decentralization

may become more complex and diverse. The policy does not guarantee that

there will be more investment in education. To overcome the possibility of lower

education investment, there is extensive research being conducted to find

alternative education financing with more in depth involvement of the community

in funding efforts. By not depending on the government, and involving the

community, it is expected that schools will improve their accountability and

efficiency. In this case, the role of private contributors will become more

significant with the governments, particularly the central government, playing a

supervisory role.

16

3.2 Background on Indonesian Education Financing

The government of Indonesia has committed itself to making education a priority

for Indonesian development over time. Figure 1.3 presents the time series of

enrollment rates for every school level with the scale indicated on the left vertical

axis. The figure also presents percentage of education financing out of the total

development budget, and the percentage increment of aggregate private

consumption for comparison. Scale for these series is located on the right vertical

axis. As shown the universal 6-year basic education program was achieved in

the middle of 1980's. Enrollment rates of school levels higher than elementary

gradually increased over time.

The percentage of the educational budget out of the total national development

budget has fluctuated over time. On average, the educational development

budget has been about 11 % of the national development budget. Change of

private consumption over time also has fluctuated. Historically, this does not

relate to the rise of enrollment rates. The proportion of education expense in

GNP was approximately 1.9% in 1975 increasing to 2.2% in 1995 (Tilaar 1995).

This proportion was relatively smaller than the education investments of Thailand

and Singapore.

17

120.00% 60.00%

20.00%

10.00%

30.00%

CI)

f40.00% 5i

~CI)0...o

~a:::

leC)

50.00%

80.00%

20.00%

40.00%

60.00%

100.00%

SftI

a:::....c

~ecw

1973/1974 1978/1979 1983/1984 1988/1989 1993/1994 1998/1999

_primary_higher

_junior-Education Portion

__senior

-private consumption

Figure 1.3 Enrollment Rate and Education Financing Over Time

Table 1.1 Type of Budget and Responsible Ministry

Responsible Ministry Type of Budget School Level

Ministry of National Education (MONE) Recurrent Junior High SchoolDevelopment Senior High SchoolOperational Maintenance (OPF) Higher EducationQuality Improvement

Ministry of Home Affairs (MOHA)President' Primary School Instruction

Elementary School(SD-INPRES)Teacher Salary

Ministry of Religion Affairs (MORA) Recurrent All levels of religion based schoolDevelopment

MONE, MOF, and MORA Social Safety Net (JPS or PKM-BBM) All levels

MOFand MOHA Primary School Subdisies Elementary School

18

Four ministries are responsible for managing national education finances as

shown in Table 1.1. Including the four ministries, the National Development and

Planning Agency (Bappenas) coordinates the financing at the macro level for all

the levels of education, and indirectly coordinates the five ministries in the

execution of education program planning. The Ministry of National Education

(MONE) and Ministry of Finance (MOF) coordinate to finance junior and senior

high schools, as well as higher education. While the Ministry of Home Affairs

(MOHA) and MOF take care of most of the education financing at the primary

school level, MONE's obligation is to ensure efficiency and quality by organizing

the primary level curriculum. MONE is also responsible for curriculum

development for other school levels. The Ministry of Religious Affairs (MORA) is

involved at all levels of religion-based schools' financing. The four ministries,

MONE, MOF, MOHA, and MORA are direct executive agencies for their

respective school levels. Recently, district governments have more authority in

allocating their budget due to the decentralization policy.

Schools receive several types of financing that are included in recurrent and

development budgets. Primary schools receive Presidential Instruction

Development Funds (SO INPRES; started 1972/1973), the Education

Development Subsidy (Subsidi Bantuan Pembangunan Pendidikan or SBPP;

started in 1992), the Operational Subsidy (Bantuan Operasional Pendidikan or

BOP), and funds from the Social Safety Net program (started in 1998 after the

financial crisis). MOHA and MOF jointly manage elementary school teachers'

19

salaries. The Education Development Subsidy (SBPP) is part of MOF's recurrent

budget. Three ministries, MOHA, MOF, and Bappenas, administer SD-INPRES.

The objective of SD-INPRES is to enhance primary school development in rural

areas by constructing more school buildings and classrooms, rehabilitate older

schools, train more teachers, and develop new curricula. Some operational

block grants such as the Social Safety Net project are coordinated by MONE.

Block grants are allocated directly to schools, thereby giving them wider authority

to manage operating funds used to buy school supplies and textbooks.

Data on the SD-INPRES fund managed by MOHA, MOF, and Bappenas are

available from the Ministry of Finance (MOF) and Bappenas. Recently, following

the decentralization policy, the SD-INPRES scheme has changed. Funds are

allocated to budgets at the district level. This paper focuses on three fiscal years,

1992/1993, 1995/1996, and 1998/1999. During these years, SD-INPRES was

centrally managed. Data on primary school teachers' salaries are the most

difficult to obtain. Recent data is available from Bappenas, but prior to fiscal year

1995/1996 average salaries have to be estimated. ii

Secondary schools receive tuition fees from students, which are managed by

MONE at the district level, Operational and Maintenance Facility Funds

(Operasional dan Pemeliharaan Fasilitas or OPF), the block grant for school

operations (BOP), and Social Safety Net funds. MONE and MOF are responsible

for financing junior and senior high schools, including the management of

teachers' salaries for the secondary school and higher education teachers'

20

salaries. MONE and MOF are also responsible for managing the operational

funds for public and private higher education institutions. Higher education

institutions fund their activities with tuition received from students and recurrent

and development budgets from MONE.

In addition to the formal system of education, informal educational programs also

receive serious attention from the government. For example, MONE manages

several programs focused on eradicating illiteracy among school dropouts.

Besides their curriculum, MONE administers their finances and coordinates with

district level governments. This budget is very limited when compared to the

formal education budget.

Table 1.2 shows the government funding profile for 1995/1996 as presented in

Bray and Thomas . The Ministry of National Education acts as the principal

executor and manages almost 51 % of the total education spending. Due to

existence of religion-based education or 'Madrasah', the Ministry of Religion

Affairs carries around 4% of the total budget. Finally, the Ministry of Home

Affairs, which mainly administers the primary level of education, accounts for

nearly 38% of the total budget. Most of the government funds are designated for

recurrent budget, with the majority for teacher salaries. Table 1.2 indicates that

even though the government incurs a big portion of spending at the primary level,

it actually pays more per student in secondary and higher education levels than

per primary level student.

21

Public transfers for educational expenditures are estimated using governmental

data for the relevant fiscal years. Some of the bUdget sources are not available

per school level. Therefore, they are obtained from the methods and/or results of

previous studies . I estimateiii funds allocated by MORA and Operational Funds

(OPF) and Quality Improvement funds allocated for secondary schools and

higher education. I took the proportion of existing allocation per education level

available in Bray and Thomas (1998), multiplied by the total budget of the

respected funds to find per education level budget allocation. The allocation of

operational funds (BOP) is estimated using the proportion of operational funds for

each level of school disbursed through the other type of operational funds such

as the Social Safety Net ProgramiV.

4. Data Description

Indonesian Socio-Economic Survey Data (Susenas) is used. Susenas is an

annual nationally representative socio-economic survey, which collects detailed

socio-economic information on households and individuals that live in the same

households. Data collected includes age, gender, relation to household head,

highest level of education attended, education institution attending (private or

public), and other socio-economic data for each member of the household. The

annual socio-economic data collected produces the core-Susenas. A brief

consumption survey is conducted with the annual Susenas survey.

22

Table 1.2 Total Annual Expenditures on Education by Government Agencies, by

Source of Funds, and Level of Schooling, 1995 - 1996

Ministry ofNational MinistryofReligion AffairEducation President's

Nunberof Prim1ryPrinury

MinistryofStudents School

SchoolHmIr Total

Type ofScOOls (1000) Recurrent Developnmt SubsidiesInstruction

Affairs Recurrent Developnmt

PRIMARY 29,448Public 24,057 77 110 742 4,387 5,316Private 1,892 161 161MJdrasah, Public 193 7 21 4 32MJdrasah, Private 3,306

JR SECONDARYPublic 4,684 1,130 630 1,760Private 2,262 20 9 29MJdrasah, Public 355 44 61 105MJdrasah, Private 1,103 7 23 30

SR SECONDARYPublic, gereraI 1,429 497 535 1,032Public, vocatiornl 500 261 186 447Private, gereral 1,148 7 7Ptivate, vocational 1,148 -MJdrasah, public 192 194 20 214l\Ihdrasah, jrivate 259 14 14

-TERTIARY -

Public 853 612 759 1,371Private 1,450 59 14 73MJdrasah, Public 279 58 36 94MJ.draslh, Private 68 -

-Other Education 60 290 350

-Adrrinistration 923 6 929

-TOTAL-AlL -SCHOOLLEVEIB 45,178 3,569 2,513 110 749 4,569 337 117 11,%4

DISIRIBUTION 30"1< 210/. 10/. 60/. 380/. 30/. 10/. 100"1<

Sources: Bray and Thomas (1998). Note: Madrasah is Islam-based school. All monetary values

are in Billions of Rupiah.

23

Every three years, in addition to the core-Susenas, a detailed sUNey of socio

economic aspects of the households, including health, education, expenditure

and income is conducted. This detailed socio-economic household data set is

called the module-Susenas. Core-Susenas and module-Susenas data are

compatible and easily merged. Module-Susenas data from 1993, 1996 and 1999

are used for my analysis. These Susenas data sets cover detailed expenditures

(food and non-food items) and income of the households.

Education is one of the non-food household expenditure items included in the

data. In addition, I also use module-Susenas data, which provides detailed

education expenditures for the years 1992, 1995, and 1998. These Susenas data

sets have information regarding access to education facilities, principle agents

who pay for education costs and study activities after school time. A detailed

education expenditure sUNey is needed to re-evaluate the estimation method

conducted for household expenditures and income modules (Susenas 1993,

1996, and 1999). Finally, by matching core-Susenas, which covers individual

characteristics and the same year of module-Susenas covering household

education expenditures, I construct the individual level education expenditures.

School-age enrollment profiles are also used to construct the public education

expenditures

Table 1.3 provides descriptive data on household and individual data for three

years. Panel A presents household characteristics and panel B displays

24

individual characteristics. Each year of data is divided based on level of

education of the household head: those who only complete primary education

and those who have higher education than primary level. In the third column, all

samples are included: thus all categories of households are included. The

monetary data are in monthly bases with Rupiah currency, which was exchanged

at Rp. 2,500 for one US Dollar in 1995. Non-educated parents or those with

lower than primary school education, accounted for about 15% of total samples,

are indicated by zero years of education and are excluded because it is

suspected that these observations may be missing.

The number of children is slightly higher for lower educated parents compared to

higher educated parents. This figure declines slightly over time to 1.86 in 1999

from around 2.08 in 1993. Household heads with lower education are older, while

higher educated parents are relatively younger. Over time, the average age of

lower educated parents is rising, while the average age of higher educated

parents is relatively stable. Including all samples, the average age of the

household head is around 44 to 45 years.

In general, household investment in education accounts for only 2% of total

expenditures, or approximately of Rp. 5,248 or USD 2.00 per month in 1993 and

higher in nominal value over the following years. This proportion is relatively

stable and does not change over time. The share of education expenses is

higher for higher educated parents; nearly triple that of lower educated parents.

Higher educated parents spend more on education, while lower educated

25

parents spend slightly less. Food share in average is about 59% for lower

educated parents in 1993, and lower for higher educated parents during the

same year. In 1999, the food share is around 63% when including all samples.

Panel B of Table 1.3 presents individual characteristicsv. Included in the data are

all members of households in the sample. The average age is between 23 and

26 years. Individuals from households with lower educated household heads

tend to be older than those coming from households with higher educated

household heads. In general, if we include all samples, the average age is about

26 years. School enrollment varies from 24% to 30%. Almost 50% of the sample

is male. Parents' education relates positively to the higher enrollment of

individuals, as those with higher educated parents completed higher degrees of

education.

26

Table 1.3 Mean Value by Education of Household Head

1993Household Head Education

Higher than

Primary Primary All

Panel A: Household Characteristics

Number of Children 2.26 2.21 2.08(1.63) (1.62) (1.66)

Household head year of Education 6.15 11.44 5.54(0.58) (2.14) (4.06)

Age of Household head 41.66 39.41 44.99(12.25) (11.17) (13.89)

Education share expenses** 0.020 0.031 0.02(0.04) (0.05) (0.04)

Food Share expenses** 0.59 0.52 0.59(0.13) (0.14) (0.13)

Total Expenditures 177,757.50 322,748.30 190,469.20(284,093.40) (332,379.00) (251,864.70)

Education Expenditures 4,515.10 12,313.18 5,248.94(23,740.12) (34,992.11 ) (22,319.11 )

Labor Income 173,174.30 204,491.30 154,521.80(207,878.40) (327,167.30) (229,939.30)

Number of Observation' 15,928 15,852 50,740

Panel B: Individual Characteristics

Age 24.49 23.91 26.13(17.69) (16.62) (18.94)

Male** 0.51 0.49 0.50(0.50) (0.50) (0.50)

Only Completed Primary** 0.37 0.07 0.18(0.48) (0.26) (0.38)

Only Completed Junior level** 0.06 0.21 0.09(0.24) (0.40) (0.28)

Completed higher than Junior level** 0.24 0.46 0.36(0.43) (0.50) (0.48)

Enroll at School** 0.25 0.30 0.24(0.43) (0.46) (0.43)

Labor Income 60,652.08 66,233.12 58,153.11(140,948.30) (205,920.30) (150,053.00)


*AII Monetary values are in Rupiah/month (1.00 USD = 2,500 Rupiah, 1996 exchange rate).

Standard Deviations are in parentheses. ** in percentage terms.

27

Table 1.3 (Continued) Mean Value by Education of Household Head


Higher than

Primarv Primary All


Number of Children 2.17 2.07 1.99(1.59) (1.51 ) (1.59)

Household head year of Education 6.14 11.56 6.05(0.56) (2.20) (4.20)



Food Share expenses** 0.57 0.50 0.56(0.12) (0.14) (0.13)


Education Expenditures 6,686.03 19,589.86 9,006.79(17,296.34) (50,275.17) (29,922.00)

Labor Income 248,701.30 278,726.70 226,151.80(320,425.40) (397,816.10) (339,006.10)



Age 25.25 24.65 26.63(17.96) (16.93) (18.98)

Male** 0.50 0.49 0.50(0.50) (0.50) (0.50)


Only Completed Junior level** 0.11 0.19 0.12(0.32) (0.39) (0.32)


Enroll at School** 0.25 0.29 0.24(0.43) (0.45) (0.43)

Labor Income 90,562.53 102,455.10 88,834.92(220,266.40) (276,689.50) (230,524.40)




28

Table 1.3 (Continued) Mean Value by Education of Household Head


Higher than

PrimaN Primary All


Number of Children 2.01 1.88 1.86(1.48) (1.44) (1.51 )

Household head year of Education



Food Share expenses** 0.64 0.58 0.63(0.11 ) (0.13) (0.12)


Education Expenditures 9,293.39 24,388.54 12,683.98(25,768.70) (58,535.69) (38,258.09)

Labor Income 402,688.50 601,148.10 452,093.30(385,130.80) (648,095.80) (498,724.40)

Number of ObseNation' 18,124 21,525 61,228


Age 26.44 25.34 27.59(18.30) (17.16) (19.19)

Male** 0.51 0.50 0.50(0.50) (0.50) (0.50)


Only Completed Junior level** 0.13 0.21 0.14(0.34) (0.41 ) (0.35)


Enroll at School** 0.23 0.27 0.23(0.42) (0.44) (0.42)

Labor Income 138,733.60 183,854.00 134,742.90(306,256.80) (447,788.40) (342,356.80)

Number of ObseNation' 77,958 89,849 254,016



29

Figure 1.4 provides an age profile of private education expenditures by source for

1992 and 1995. Both years display a similar pattern: from ages 5 to around 25

the main source of private education expenditures are parents. Other primary

sources of education expenditures are other relatives or individuals themselves.

Individuals become self-sufficient from the age of 20 years. Other sources of

expenditures are institutional, governmental, or non-relative sources. These

sources start to contribute to education expenditures from around the age of 20.

In 1995, others sources are slightly higher than in 1992 at the age of 30. This

information on sources of education expenditures is important when formulating

the private education transfers account to construct transfers outflow.

1992 1995

~ M ~ ~ m ~ M ~ ~ m ~ M ~~ ~ ~ ~ ~ N N N N N M M M

Age

~ M ~ ~ m ~ M ~ ~ m ~ M ~~ ~ ~ ~ ~ N N N N N M M M

o.g

0.8

0.7

0.3 -others

0.2

0.1

a 0.6

~ 0.5 - parents8" ...•... other relatives

Ii: 0.4 --self

- ..... ". -.".".

-parents•. "•... other relatives

--Self

-others

, ' -..Age

0.9

0.8

0.7

= 0.6

i 0.5a.£ 0.4

0.3

0.2

0.1

~~~~~~~~

Figure 1.4 Private Education Transfers Resources

30

5. Methodology

The estimation of private educational transfers includes educational transfer

inflows, educational transfer outflows and net educational transfers. Educational

transfer inflows per month, denoted by qr+, is the private transfer received or

education expenditure spent by school age groups of age i. A positive superscript

indicates positive fund flows. Educational transfer outflow, denoted by qr- ,is the

total cost of all education services received by all enrolled members in the

household. A negative superscript indicates outflows.

Estimation of public education transfers consists of the estimation of educational

transfer inflows and outflows. All students of a particular school level are

assumed to have the same average educational cost. Included in the estimation

are four levels of formal education, from elementary to higher education.

Vocational schools and general education schools are considered identical. Out

of-school programs, training programs, and schools that are not registered at the

Ministry of National Education are assumed insignificant. Most education

financing was centralized before fiscal year 2000. That is, the source from the

central government dominated education financing during that time. The district

government covers for about 5 - 10% out of the total education budget. It should

be noted that, due to difficulties in collecting data of education financing from

district governments, education financing that comes from district governments

are ignored.

31

5.1 Estimation of Private Education Transfers

5.1.1 Private Education Inflow

Private education inflow, Eir+, is defined to be transfers received by household

member i for educational expenditure purposes from a principal agent. "Principal

agent" is defined as an agent in the household that bears all the education

expenditures for the members. It is important to distinguish the agent who pays

the education transfers. Therefore, in addition to the age profile of education

transfer inflow, the age profile of the transfer outflow can also be estimated.

Private education inflow is an explicit individual educational cost. For each

household j and household member i, the individual education expenditure is

estimated by regressing, at the household level, total household educational

costs on the number of enrolled household members in each age group. The

relationship is as follows:

1.1

Assuming that the production function is homogenous of degree one, qj denotes

educational expenses for household j, N; is the number of members of age

group f enrolled from household j. The regression includes age groups 5 to 25

and older. Children are expected to start primary education at the age of 7, but a

significant number of 5 and 6 year-olds are already enrolled in kindergarten. No

distinction is made between boys and girls in the regression.

32

Coefficient fJt obtained from regression equation (1.1) is interpreted as average

cost of education expenses for each household member. This coefficient is

employed to calculate the share of education expenditures of each enrolled

member. I allocate education expenditure to each enrolled member i of the

household j as follows:

1.2

D~ dummy variable for household member i in age group fthat is enrolled, zero

otherwise. I drop subscript j for household for simplicity. Ej?+ is treated as an

estimate of education transfers received by member i in age group f in household

j. Superscript e+ indicates transfer inflows or transfers received.

5.1.2 Private Education Outflow

Gross educational transfer outflow is the total of educational funds transferred by

the principal agent or household members to other household members. This

can be estimated by assuming that household agents can be principal earners in

the household or household heads, but are not necessarily both. In Indonesia,

most principal earners are also household heads. Based on Susenas 1992 and

1995 (Figure 1.4), most children receive education transfers from their parents.Vi

Gross educational outflow is calculated as follows:

33

1.3

Et!- is private education outflow of household t ii from principal agent i. This is

the net of all enrolled member education expenses, Eir+, in the household j. The

negative sign indicates the reverse flow of education transfers.

5.1.3 Net Private Education Transfers

The net education flow of age group f is estimated by summing up education

transfers inflows of age group f, q7+ and education outflow of the same age

group f, q7- as follows:

1.4

This is a net education transfer borne by age group f.

5.1.4 Estimating Private Non-education Transfers

To estimate non-education transfers, estimating the consumption allocation to

household members is important. There are several existing methodologies that

are applied to estimate consumption allocation for household members. Two

more established methodologies are the Engel Method and Rothbarth Method.

The estimation of non-education transfers departs from these existing

methodologies, which are applied to further develop our specific needs.

34

Small household

Large household

5.1.4.1 Engel Method

The Engel method suggests that the food share can be a means by which to

approximate an adult's welfare. By assuming that food is a necessary good, the

wealthier a family is the lesser the proportion of their budget is devoted to food.

The Engel method estimates equivalence scales by equalizing the food

expenditures share of a couple with a child relative to the share of food

expenditure for a childless couple. An additional child will in fact increase food

share expenses. Hence, by some increase in their budget, a couple with a child

can reach a food share equal to the childless couple's by spending more. Figure

1.5 illustrates the compensation required to equate the welfare level of the adults

with a child (large household) to that of childless couple.

Outlay

Figure 1.5 Illustration of The Engel Method

35

Given Figure 1.5, the compensation required is the difference between X1 and Xo

and the equivalence scale (es ) is defined as;

1.5

The equivalence scale (es) is the ratio between the total expenditures of two

adults with one child to the total expenditures of reference adults. Both levels of

consumption depend on their demographic composition (a or ao) given the same

price and utility level. This also implies that the equivalence scale depends on the

level of consumption.

This paper closely follows the extended Working's model to estimate individual

allocated consumption. The model includes demographic variables as followsviii;

[xoJ () F-1 [n"oJwj =a+jJln _1 +7]ln nj + L r, -ry +rz+Cj'nj '=1 nj

1.6

Food share (Wj) of household j is linear with logarithmic per capita expenditure

(x!nj) , logarithmic household size (nj) , and proportion of demographic variable

(nfj) to the household size, and Z is socio-economic characteristics of the

household. As for demographic variables, Deaton and Muellbauer (1986) uses

two age groups of children variables, under 5 years old for younger children and

over 5 years old for older children, and one adult group variable. This paper uses

demographic variables of age groups 0-4, 5-9,10-14,15-19, and 65+.

36

By collecting terms of equation (1.6) we obtain;

1.7

If the Engel method has implications as previously mentioned, the food share will

be a negative function of per capita expenditure given constant household size,

and the coefficient ,8will be negative. The second implication, that food share is a

positive function of household size, is reflected by a positive sign for (17 - ,8). The

coefficient 17 is elasticity of food share with respect to household size.

The Engel method estimates consumption allocation based on the food share

allocated to each household in order to maintain their utility level, when an

additional member is introduced. The consumption allocation of the k-th member

is the total compensation required for the family with a k-th member to maintain

the same utility level as the family without k-th member. The food share (w1) for

family with k-th member is illustrated with the following equation:

1.8

x1 is total expenditure of family with k-th member. n is household size and z is a

vector of household characteristics.

37

The food share without k-th family member is as indicated below;

o kk-1 I ( k-1) ( ) (k) r, rwj = fJo n xj + fJ1 - fJo In nj -1 + I k + k + Z •

, (n j -1) (n. -1)1.9

Equivalence scale of k-th member is obtained by equating both equations (1.8)

and (1.9) as follows:

Finally the share of k-th member can be expressed as:

The share will be one for nk, otherwise as follows;

[

(/1 a) (nO -1J Ir? k Js =1- exp - 1 - fJo In _j_ _ , + r .k /30 nj fJonj(nj -1) fJo(nj -1)

1.10

1.11

1.12

The share of consumption of the particular age group in the family uses the

predicted demographic variable's coefficient obtained from regression (1.7).

5.1.4.2 Rothbarth Method

The Rothbarth method, similar to the Engel method, replaces the food

expenditures in the Engel method with the adult goods share in the construction

of equivalence scales. This method implies that the marginal rate of substitution

38

between adult goods and child goods is equal. Deaton and Muellbauer argue

that the Rothbarth method's assumption is more plausible in estimating child

cost. Adult goods refer to adult clothes, alcohol products, or tobacco. In addition

to these classifications, Bradbury (1994) also uses saving as an alternative for

adult goods expenditures. However, unlike the Engel method, the Rothbarth

method cannot estimate the cost of children over 15 years of age as they tend to

consume adult goods.

As shown in Figure 1.6, if compensated by the amount of X1 - Xo such that they

are able to consume the same amount of adult goods, the adults with a child will

have the same preferences as the adults without a child. The presence of

children only has an income effect on adult goods consumption. This is a strong

assumption since the presence of children may not only have an income effect.

For example, adults who smoke when they do not have children have to re

consider their behavior when children are present, because of health effects of

smoking for children. Adults who used to go to the cinema have to take into

account the additional cost of baby-sitting when children are present.

39

Large household

Small household

Total Outlay

Figure 1.6 Illustration of The Rothbarth Method

Similar to equation (1.7) in the Engel method, the Rothbarth method follows the

relationship

1.13

Adult share as a proportion of adult expenditures to total expenditures is denoted

by 1Ij. The remaining independent variables are the same as in the Engel method.

However, the implications are slightly different.

40

5.1.4.3 Other Method: Ray's Method

Ray's method assumes that, in addition to the food share as a measure of

welfare, other expenditure shares such as housing, goods and services, and

durable goods, also can be incorporated as welfare indicators. Ray's uses the

following relationships:

1.14

Where B = I e,n, is the equivalence scale; e, is the equivalence scale of specifici

age group f; xi is real income; number of children is denoted by N; Wj

represents food share, housing, goods and services, durable goods. Generalized

from the Engel method, equation (1.14) is a non-linear simultaneous equation

that incorporates commodities share (Wj), rather than only food share.

5.1.4.4 Other Method: Split Method

Both the Engel method and Rothbarth method of estimations depend heavily on

the assumption of the form of household welfare preferences. This assumption

causes the estimation to suffer from bias. Engel's method uses food-share as the

welfare indicator, causing an upward bias in the estimation of child cost. It is

widely accepted that the equivalence scale estimated by the Engel method is

41

likely to be higher than the true equivalence scale. On the other hand, the

Rothbarth method, which is based on the adults' good share, concludes that

children are relatively costless in Indonesia. Deaton argues that the Rothbarth

method has more plausible assumptions than the Engel method. However, the

Rothbarth method cannot be applied in estimating cost of children older than 14

years old since they may consume the adult goods used for the analysis.

The split method of allocation estimation is based on an assumption that some

goods are directly assignable to certain groups in the household. If the household

consists of adults and children, there should be goods that can be directly

assigned to each of the groups. Education can be assigned to children who are

enrolled in school. Adult clothes are designated to adults, while children's clothes

are assigned to children. Other non-assignable goods, such as food, furniture,

and others, are allocated with certain accepted rules. Some socio-economic

sUNeys have detailed and constructive data to accommodate this method.

Estimation of education transfers as done above is one way to estimate

assignable goods. The other assignable goods /, either for adult or child, can be

estimated by the following;

1.15

eft is expenditure of obseNed household j on assignable good /. The number of

adults or children of age group f in the household j is denoted by Nfj' Nfj is the

42

number of individuals of age group f in the household. By estimating equation

(1.15), we can estimate Pj , which is the age group specific average cost of

assignable good expenditure. The expenditure share of age group f of good I is

estimated using Pj similar to estimation of individual education expenditure

(Equation 1.2). That isCj~ = Cft (jJfi/~jJfi ). The individual assignable goods'

L

consumption is calculated as ct =L Ci~ . L is the number of assignable goods to';1

individual i. The non-assignable goods are allocated by using a-priori

assignments to children and adults.

Thus, allocation of non-education consumption to individual i is total consumption

of assignable goods and portion of non-assignable goods to individual i. That is

Cj =Cr + yjCr;a , where C;a is consumption of non-assignable good of household

j and Yi is the individual fraction of consumption of these goods. Finally, non~

education transfers to children are defined as T =Cj - Yf ' where yk' is child labor

income.

5.2 Estimation of Public Education Transfers

5.2.1 Estimation of Public Education Transfers Inflow

Public educational transfer inflow per age group f, qg:, is calculated by assuming

that all age groups at the same school level face the same average cost of

43

education. The educational transfer inflow is estimated in several steps. First, I

calculate the total budget per school level by summing up all the budgets of the

responsible ministries for each school level. Next, I calculate the average cost of

education per student per school level k, qgk' That is, I divide the total budget

per school level by its number of students. Next, I calculate the enrollment rate

per age group f per school level k based on Susenas data. The number of

students per age group f per school level k, Pkt, is then estimated using the

calculated enrollment rate weighted by the total number of students per level.

The usual age range for elementary school students is 7-12, junior high school is

13-15, senior high school is 16-18, and higher education is 19 and over. The

early entrants, late entrants, and repeaters mean that some of the age groups

are counted in several education levels, which affects the school enrollment

distribution. Total Cost per school level is obtained by multiplying the average

cost of education per student in each school level by the number of students per

age group. I then find total education cost per age group by summing up the total

cost for all school levels per age group. Finally, I calculate the average public

transfers inflow per student in each age group by dividing the calculated total

education cost per age group by the total number of students in age group f, Pt.

The average per capita of public transfers per age group f, qgt is expressed as

follows:

44

1.16

Where, q~t is per student public education transfer to age group f. The average

of public transfer of education level k is Cfgk' k is the education level from

elementary school (k =1) to higher education (k =4). The enrolled population in

school per cohort f at education level k is denoted by Pkf, while P f expresses the

population of age group f. To estimate Pkf, I use the enrollment rate profile by

cohort f, Ekf, for education level k from Susenas multiplied by Pf, or

The elementary school level age distribution starts from 5 and finishes at around

18 years of age (Susenas 1993, 1996). For primary school students, early

entrants tend to increase during the last three years of observations. On the

other hand, repeaters tend to decrease at the same time frame. Those who are

older than 12 years of age and are still in elementary schools in 1993 make up

more than 9% of the sample. The peak elementary school age is around 9.5

years, and the age-profile is normally distributed from 5 years of age to around

16 years of age. From the proportion predicted, I calculate the number of

students at each age based on the total population in the particular age groups

(Pi).

45

5.2.2 Estimation of Public Education Transfers Outflow

The government re-allocates some proportion of taxes paid by the productive

age groups to the school aged groups to finance public education. Public

education transfers outflow is estimated as follows:

1.17

Where, ~e is proportion of their tax re-allocated to public education, r is the tax

rate and ~e is earnings of individual i.

6. Results and Discussion

6.1 Estimation of Private Education Transfers Results

6.1.1 Test on Estimation of Private Education Transfers

I use module-Susenas data from 1992 and 1995 to test the estimation that uses

equation (1.1). Module-Susenas 1992 and 1995 contain detailed individual

education data, including individual level education expenditures by the

household. I apply equation (1.1) to re-estimate the individual education

expenditure and compare it with the surveyed individual education expenditure

data. Finally, I can test whether equation (1.1) results in a close estimation of

individual education expenditures.

Figure 1.7 shows coefficients of regression results of equation (1.1) using

module-Susenas 1992 and 1995 data with variation of one-standard deviations.

46

The figure exhibits the age profile of the regression results. Coefficients for all

age groups are significantly different from zero at the 1% significance level.

These coefficients are used to estimate the education expenditure share of

individual household members. Older age groups are associated with higher

coefficients, implying that their education share is relatively higher, as compared

to the younger age groups' share. When comparing 1992 and 1995, it is noted

that coefficients of the younger age group do not differ by much and begin to

diverge from age 15.

70000

~ 60000::::lVIQ)

0::: 50000c:

.!2VI

40000VIQ)...0'1Q)

0::: 30000J!lc:Q) 20000'13

!EQ)0 10000 __ 1992u

~1995

05 10 15 20 25

Age

Figure 1.7 Regression Results for Education Expenditures on Enrolled Age

Groups: Estimated Coefficients jJ by Age Group

Individual education expenditure profiles for both estimated and actual data are

presented in Figure 1.8. Although the estimated values vary slightly from the

actual data values, there is no indication that the estimated profile is consistently

47

biased upward or downward. The graph indicates a downward bias for young

age groups and upward bias for older age groups, when inspecting the estimated

data relative to the survey data. The biases reach up to 7% difference from the

real data. The estimated profile appears to fluctuate around the real individual

data, with noise occurring particularly among the teenage age groups and older

age groups. More school choices may be the source of this increased fluctuation.

High school level education can be either general or vocational, and there exist

choices between private and public schools. However, I assume all senior

education level schooling to have the same unit cost regardless of their type. The

gap between tuition fees for public and private schools are greater for education

levels above elementary school

16000

14000>.:i:- 12000e0:iE

.c 10000III'Q.::l

~ 8000~Ql....IIIe 6000~l-e0 4000:;;IIIl.)::l

"C 2000 data 92 • estirrated 92w....... data 95 '" estirrated 95 Age

o~--,-;~~~~:;::::::;::~~~~~~---,-----,~~~~~

~ ~ ~ ~ ~ ~ 0 ~ ~ 0/ ~ V ~ ~ 0/ ~

Figure 1.8 Comparison Between Actual Data and Predicted Individual Education

Expenditure

48

I regress the fJp as parametric coefficients obtained from regression on equation

(1.1) over iJnp obtained from direct calculation from the dataix.

- -f3np =ao + alf3p + v

The null hypothesis is that the slope of the regression, ai' should not significantly

be different from one. The intercept, ao, should not be significantly different from

zero. Table 1.4 presents the regression results. The coefficients a 1 is higher than

0.96 for both years. A good fit is indicated by a high R-Square value (higher than

0.9) and results of an F-test indicate that the coefficients are not significantly

different than one. The constant also has value that is small and approaching

zero. Therefore, it is determined that estimated f3p is not biased and is a good fit.

Table 1.4 Goodness-of-fit Average Cost (Coefficients of Regression) Over Non

Parametric Average Cost

Dependent Variable: 1992 1995

Education Share

Dependent Variable:Beta Est. 0.971* 0.964*

(0.0004) (0.0006)Constant 0.008 0.012

(0.0001 ) (0.0002)

N Observation 357,334 163,244R-Square 0.94 0.92

F-Test of Beta* 5667 3459

Note: * significance with 1% confidence level. ** standard deviation in parentheses

49

As a more convincing test of the estimation methodology, actual individual

education expenditure data was regressed on the allocated individual education

expenditures, q; as follows:

The null-hypothesis is similar to that of the above regression. That is, the

coefficient r 1 is not significantly different than one. Regression results are

presented in Table 1.5. The coefficients for both years exceed 0.87, with a high

R-squared value.

Table 1.5 Regression Results of Estimated Data on Real Data

Dependent Variables: Estimated Data

1992 1995

Independent Variable:Actual Education 0.87 0.97

Expenditures (0.02) (0.04)Constant 382.23 117.95

(52.24) (140.89)

N Observation 144,958 148,794R Square 0.85 0.84

Note: standard deviations are in parentheses

The confidence interval for both coefficients is unbiased, where the coefficients

are within the desired range. Confidence intervalsx on slope coefficients indicate

an unbiased estimation. Estimation using equation (1.1) produced to produce

50

estimates in close proximity to the real data. Therefore, equation (1.1) can be

applied to estimate individual education transfers from household transfers for

years where individual level data is not available.

6.1.2 Estimation of Private Education Transfers Inflow Results

Susenas 1993, 1996, 1999 and 2002 surveys covered detailed household

education expenditures. By using equation (1.1), I can estimate individual

education expenditures (iit+). The regression results are shown in Figure 1.9.

The regressions are based on the same age profiles as previous regressions. All

age groups have significant coefficients at a 1% confidence level. The

coefficients are increasing over the four years Susenas. The coefficients are

higher for more recent years, where higher per unit education costs are

observed.

The older age groups correspond to higher coefficients. I treat these coefficients

as the average cost of education for a particular age group. I calculate the share

of individual education expenditures using this average cost and finally estimate

the individual education expenditures based on these shares. The regression

results shown in Figure 1.9 indicate a higher unit cost for the college age group

(older than 20 years old). This implies that, once enrolled in higher education, an

individual's share in the household education expenditures is relatively high. In

general, however, average education inflow transfers at the college level are low,

primarily reflecting low enrollment rates at the college level.

51

140000.00

40000.00

60000.00

20000.00

• 1993

)I( 1996

- - .. - -1999

I 2002

80000.00

100000.00

~ 120000.00:;,UI

~s:::o'iii

~Cl

~UI1:G)'(3

:e~(J

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Age Group

Note: Education transfers estimation regression results on average monthly household educationexpenditure. All coefficients are significant at a 99% confidence level.

Figure 1.9 Regression Results of Education Expenditures on Enrolled Age

Groups: Susenas 1993, 1996, 1999, and 2002

Age profiles of education transfers, which are the average per capita of education

transfers received per cohort (q,e+), are presented in Figure 1.10. These profiles

are similar to those in Figure 1.8. Each profile is a concave curve with a peak,

varying from age 15 to 17 years old. Fifteen years old marks a transition to the

level of senior high school. The peak could reflect expenditures necessary for

entering this new level of school. In addition to entry fees, parents also have to

spend money on new clothes and books.

52

There is a jump before the peak age that may represent the high burden to

families by sending their children to senior levels, which is beyond compulsory

education requirements. Beyond nine-year compulsory education, the

government allocates funds primarily to public schools and small amounts to

private schools. Therefore, those who are enrolled in private schools have to

spend more of their own budget to finance their children's education. For children

exceeding 17 years of age, education transfer inflow is declining. This is due to

the low enrollment of children in higher education, partially due to the higher

average cost of education at this level.

25000

~I

.clI:l 20000.i5.:::::s ...... -1993r:t:- --1996"CQl --1999> 15000'Qj(.) -2002Qlr:t:~

oS! 10000CIIc~I-QlCl 5000 _.. ~ ..lI:l... --Ql

~...

0 .

5 10 15 Age 20 25 30

Figure 1.10 Private Education Transfers Profiles 1993, 1996, and 1999

53

6.1.3 Estimation of Private Education Transfers Outflows

Figure 1.11 presents the gross education outflow using the household head as

the principal agent. The profiles display a peak at around the early fifties. The

profiles tend to display high variation. Household heads older than forty years

provided higher education transfers. This may be due to a larger number of

children attending either elementary or junior high schools. They may also have

members who are starting to enroll in levels higher than senior high school.

Beyond age groups in the fifties, education transfer outflows start to decline.

In the most recent year of transfer profiles, a small peak occurs at the younger

age groups, 25 to 28 years of age. This may occur due to the initiation of

transfers to pre-school aged children. The sample covers school aged children

from 5 years. They tend to attend school at the pre-school level, where the cost

is relatively more expensive than at primary school. Therefore, they spend more

on these pre-school aged children. For households with heads beyond 29 years

of age, the peak disappears and the estimates follow the path of the previous

year to reach a peak at approximately the same average age.

54

Age

....... 1993

--1996

~1999

-2002

-30000

Figure 1.11 Monthly Education Transfers Outflow by Household Head as

Principal Earners

6.1.4 Estimation of Private Education Transfers Net Flow Results

Combining education outflows and inflows using Equation (1.5) provides a net

education transfer profile per cohort as shown in Figure 1.12. The patterns for all

years are similar. Positive net education transfers peak at age around 18, and

the minimum point is at an age between 45 and 50. This leads to a point of zero

net education transfers being reached between 25 and 30 years of age;

indicating that, at this age, education transfers breakeven. At these ages,

individuals graduate from higher education and start to be productive. In addition

to these higher educated individuals, those who graduated from senior high

school at this age already have stable earnings. They are also taxed to cover the

education expenditures.

55

The intersection points and extreme points do not change significantly from year

to year. However, a rightward tendency appears to exist. This indicates that, over

time, individuals obtain a longer education and become a net receiver at slightly

older age than previously. This indicates that the age structure and family

structure do not change significantly during the period of analysis. To clearly

examine this age structure, I calculate the weighted average age of both

transfers' recipients and providers.

30000

----.cco 20000 "... """ Head 1993'0..:::J --Head 19960:::.......-(/) 10000 --Head 1999I-Q) -+-- Head 2002- Age(/)cco 0l-

I- I{)

...... OJ OJQ)

Z -10000Q)OJcoI-Q)

-20000~

-30000

Figure 1.12 Net Education Transfers Flow with Household Head as Principal

Agents

Table 1.6 presents the average age of recipients. Average age is weighted by the

amount of average education transfers of the particular age group. The average

56

age of recipients is relatively stable. That is, around 16 to 17 years of age. There

is no tendency for the average age to change over the years. There is nearly 39

years of age difference between recipients and the household head. It takes 39

years in average for the individuals to payoff their education expenditures by

retransferring them to the next generation.

Table 1.6 Average Age of Transfers' Recipients and Providers

1993 1996 1999 2002

Average of age recipients 16.80 17.05 17.16 16.73

Average of age transfers giversHousehold Head 55.61 56.87 54.21 53.33

A clear depiction of the transfer flow is shown in Figure 1.13. Arrows are

constructed to indicate the direction and magnitude of flow of education transfers.

The average of age household heads who act as principal agents and who

perform transfers is located in the base. The average of age of those who enroll

and receive education transfers is located at the arrow's head. The widths of the

arrows indicate the averages of education inflow or outflow. The transfer profile

of private education of the United States is provided for comparison.

The arrow widths are increasing over time, which means that education transfers

are increasing nominally every year. Indonesian base ages are similar to those of

the United States with the net recipients of the United States being older than

those of Indonesia. There are several explanations for these differences. The

57

profile of the United States accounts only for education transfers at the higher

education level. Therefore, the age of recipients tend to be older. Both

estimations assume household head as the principal agent. Because the sample

of the United States only accounts for higher education level transfers, the

household head that transfers to higher education students tends to be younger.

Another explanation is the different of age structure between two countries. The

United States has a relatively older age structure than that of Indonesia. The

school enrolled populations are also relatively older. In addition to this, the

profiles suggest that years of education in the United States are relatively higher

than years of education in Indonesia.

L...--+---h Rp. 21,992,00

1999

Rp.1 0,401,00

1996

Rp. 6,383,001993

USA 1987 (Lee et al. 1994)

6 45 52 Age

Note: Hollow arrow assumes that household head as principal earners;Case for USA refers to private higher education transfers

Figure 1.13 Private Education Transfers Flow

58

6.2 Estimation of Non-education Transfers Results

6.2.1 Equivalence Scale Comparison

Equivalence scale results are exhibited in Table 1.7, taking the age group 30 -

34 as reference. Estimation using the Engle method produces child costs that

exceed adult costs. That is, children in the 0-4 age group are estimated to cost

about 114% of reference adults. The older the children are, the more expensive

they become. On the other hand, the Rothbarth method results in less expensive

children. The children of the 10- 14 age groups are estimated to cost about 64%

of the reference adult groups when adult clothing is used as the dependent

variable. However, the younger age groups cost much less and children in the

youngest age group are estimated to be free. Ray's method concludes that the

cost of children ranges from 88% to 94% with the youngest age group being the

most expensive.

Table 1.7 Equivalence Scale Comparison

Method Age Group Notes

0-4 5-9 10-14

Engels 114% 144% 152%

Rothbarth <0 22% 64% Adult clothing

Rothbarth <0 <0 38% Adult food

Ray's 94% 96% 88% Food-share,housing, good

and services anddurable goods

* Reference adult 30 -34

59

6.2.2 Consumption Allocation Results Using Split-Method

Employing the Engel method to allocate household expenditures to individuals

produces over estimates of consumption allocation to children, while the

Rothbarth method has the opposite results. The split method offers an

intermediary between Engel and Rothbarth methods, without being restricted by

the assumption on preferences.

Figure 1.14 exhibits the allocated consumption results using the split method.

Two assignable goods are estimated: education and adult health expenditures.

The non-assignable goods are allocated by using a-priori shares for both children

and adults. First, the allocation is made assuming that children are as expensive

as adults. That is, the child equivalence scale is assumed to be 1. Second, the

allocation is done assuming that the cost of children has positive linear

relationship with age. The older the children the more expensive they are. The

scale used ranges from 0.2 to 0.8 for children aged between 0 and 14 with the

average equivalence scale of 0.5. This is consistent with Deaton and Muellbauer

investigation on Indonesia's children cost.

The allocated consumption exhibits a peak at about 30 years of age for allocation

of non-assignable goods using a linear scale (Figure 1.14). Using per capita

consumption, or a scale of 1, the consumption peak is at an earlier age. There is

a sudden increase from age 11 to 15 years, while profiles for the ages from 0 to

10 are relatively flat.

60

- - - - linear 0.2 - 0.8

, .. - ... - ..

...

Age

- ....... .. .. - .. - .... ...

Consumption Allocation (Mean) 1996Split Method

I-+--constant =1

1--constant 1 withouteducation and health

-- linear 0.2 - 0.8 withouteducation and health

, ...... ..

100

90

1: 80.~& 70

"EIII

60

~ 50 •

i,,

40 ,::J

~

8 30

f 20

10

0

Figure 1.14 Consumption Allocation Using Split Method

Figure 1.15 illustrates the consumption allocation using the split method

combined with linear proportion for children younger than 15 for three years of

survey data. I estimated the education and health expenditure of individuals in

the household combined with other expenditures allocated by a-priori proportion.

Panel A shows their nominal value per month, while Panel B presents the relative

value to maximum for each year. Panel A exhibits a growing monthly nominal

value consumption allocation over three years of analysis. Profiles shown in

Panel B, on the other hand, enable one to distinguish the age variation of

consumption allocation. Panel A shows that the peak of allocation is at the late

20's. It occurs for all the years. Panel B shows that age profiles until early 40's

are quite similar, while the profile at later ages displays slight variation over the

three years of analysis.

61

Age

Panel A:. Consumption Allocation Profile using Split fv1ethod200

:2 180.!!!g- 160

a::-g 140coUl

5 120Ec: 100,gc.E 80::IUl

6 60()

~;; 406

:::2: 20 I I--1999 -1996 --1993O.).,.,..,"TTT"'r"T'TTTTTTT"";;:;:;::;::;:;::;:;:;:;:;:;:;:;;::;:;:;:;:::;:;:;:;:;::;::;:;::;::;:;:;::;::;:;:;::;:;:;::;:;:;:;:;:;:;:;:;:;:;:;:;:;:;;""r-Trrn-rT1rrTnrrrT1rrrrr......-r-rn

o It) 0~

It) 0 It) 0It) <0 <0 ,...

Panel B: Consumption Allocation Profile using Split fv1ethodRelative to tv1aximum

0.9

x 0.8co:::2:.9 0.7CD.~ 0.612CD

0.5~c::8 0.4c.E 0.3::IUlc:0 0.2() 1-1999 -1993[-1996 Age

0.1

O-hnerrrrTTTTrrTTTTTr"nTn.".,..,,..,.,,TrnT'TTTTTTr"I'TT"T1r'TTT'1'1TTTTrnT'TTTTTT1'"1TrTrrrrrrrnOTT1,...,..,TrTTTTTTn

o ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ @ ~

Note: linear proportion uses 0.2 to 0.8 weights for children whose ages are between 0 - 14 yearsold, while the adults whose ages are older than 15 years old are weighted by one.

Figure 1.15 Consumption Allocation Profile Using the Split Method and Linear

Proportion Allocation (0.2 - 0.8) for Children

62

6.3 Estimation of Public Education Transfers

Table 1.8 presents a summary of public education expenditure allocations. More

detailed allocation information is attached in the appendix. Included in the table

are data from fiscal years 1993/1994, 1995/1996, 1998/1999 and 1999/2000.

The table is summarized on the basis of responsible ministries and type of

programs. The five levels of education shown in the table receive major

government support. Additional types of schools are not shown, including training

in the government department and the education system outside the Ministry of

National Education system. The government subsidizes these training schools,

but the allocated budget and number of students is considered minor and

negligible.

The recurrent budget administered by Ministry of National Education (MONE)

and Ministry of Religion Affair (MORA) is always higher than its development

budget (Table 1.8). This reflects the fact that most of the education budget

remains focused on teachers' salaries, school administration, and other routine

allocations. MONE managed the development budget, which is almost half of the

recurrent budget, with the exception of fiscal year 1999/2000. There is a

tremendous rise in the development budget due to the financial crisis in 1997. As

previously mentioned, MONE is primarily an executive ministry for junior, senior,

and higher education levels. MONE also administers out-of-school education

63

support, which accounts for a smaller proportion than that of the formal schools.

The junior high school level receives a higher recurrent budget than senior high

school levels; and higher education levels receive even smaller recurrent

budgets, but a higher development budget. This is because MONE distributes

the development budget not only to public universities, but also to private

universities. Even though each private university receives a lesser development

budget compared to each public university, there are more private universities

than public universities.

The Ministry of Home Affairs manages teachers' primary salaries, one of the

largest budget allocations to the primary level, and collaborates with Bappenas to

manage the SD-INPRES program. The number of teachers at the elementary

level is considerably greater than at the junior high level or even senior high

schools. Therefore, the allocation is also a major part of the total primary level

budget allocation. In addition, the Ministry of Finance (MOF) directly allocates

subsidies for the primary school level as part of their recurrent budget. Finally,

the Social Safety Net Program contributes to primary school level financing,

beginning in the 1998/1999 fiscal year. The main purpose of the Social Safety

Net Program is to safeguard students from dropout due to financial hardships

resulting from the financial crisis during 1997. It was mainly focused at the basic

education level. The program is a collaborative effort among MONE, MORA,

MOF, and Bappenas.

64

Table 1.8 Education Financing by Ministries and School Level (in Billion Rupiah)

Ministry ofTotal

Fiscal NationalSocial Safety

Primary School Ministry of Ministry of Budget PerPercapita Student

YearSchool Level Enrollment Rate

EducationNet (JPS or

SubsidyHome Affairs Religion School

Transfers (Rupiah)(MONE)

PKM-BBM) (MOHA) Affairs (MORA) Level(Billions)

1993/1994 Kindegarten 34.96%Primary Level 109.92% 56.43 87.02 4,747.90 - 4,891.35 164,694.31Junior High School 53.86% 1,118.58 - - 1,118.58 156,827.44Senior High School 33.87% 998.52 - . 998.52 238,187.36High Education 14.23% 721.11 - - 721.11 352,898.78

1995/1996 Kindegarten 39.15% - -Primary Level 111.88% 42.49 110.00 5,108.90 4.00 5,265.39 178,803.10Junior High School 62.32% 1,776.94 - 128.00 1,904.94 226,692.99Senior High School 35.97% 1,182.23 - 228.00 1,410.23 302,155.30High Education 16.96% 1,132.53 - 94.00 1,226.53 462,853.44

-1998/1999 Kindegarten 37.63% . -

Primary Level 114.52% 45.11 207.28 204.87 5,296.29 6.16 5,759.71 196,611.02Junior High School 70.43% 2,408.75 72.51 - 274.39 2,755.65 295,019.82Senior High School 38.31% 1,408.95 104.41 - 376.31 1,889.67 366,428.73High Education 18.09% 1,653.32 - 190.35 1,843.67 602,573.24

-1999/2000 Kindegarten 38.32% - -

Primary Level 111.99% 452.78 1,653.77 234.98 5,740.00 8.19 8,089.72 283,760.61Junior High School 73.27% 3,292.99 73.09 - 366.60 3,732.68 396,535.62Senior High School 39.48% 2,609.27 102.92 - 501.24 3,213.43 605,705.25High Education 19.43% 3,047.55 - 254.17 3,301.72 949,108.82

65

Per student public transfers for each school level is obtained by dividing the total

allocated bUdget by number of students per school level. The number of

kindergarten students is still small and the public transfers given to this pre

school level are also small. Therefore, this cell is left blank. The gross enrollment

rate at each school level is provided for easy comparison. The gross enrollment

rate for the elementary level reaches exceeds 100% in fiscal year 1993/1994. On

the other hand, the enrollment rate at the junior level is relatively low, at slightly

higher than 50% in the same fiscal year. However, as previously mentioned, the

number of students in this level is steadily increasing, as is the enrollment rate.

By the fiscal year 1999/2000 the enrollment rate exceeded 70%.

Primary school has slightly higher average per student public transfers compared

to junior high school during the fiscal year 1993/1994. However, it still remains

lower than the average per student transfers of senior high school and higher

education. On average, the elementary level received a lower and relatively less

stable per student transfer than that of other levels. The junior high school level,

in contrast, obtains a higher average per student transfer following fiscal year

1993/1994 with the gap between junior and elementary level constantly

increasing. The senior high level and higher education display a similar trend,

with their average public transfers significantly higher than those of elementary

schools, and those at the junior high school level.

66

Higher average per student public education transfers for the higher school levels

is due to slow growth in enrollment rates at the senior high school and higher

education levels, and increases in the government's total budget for those levels.

Although the government also increased the total budget for elementary and

junior high school levels, total budget growth is proportionally smaller than the

growth in the number of students. Teacher salaries are the main component of

the elementary school budget. There is a slow attempt by the government to

increase the salary of elementary level teachers. In general, average per student

public education transfers have experienced slower growth in elementary schools

than higher school levels.

Figure 1.16 summarizes the trend of annual public and private education

transfers per student. The average of private education transfers gradually

increases over the years, and is consistently lower than the public education

expenditures even at the lowest school level. Over time, averages indicate an

increasing trend for all school levels. There is a large jump of in public education

transfers, particularly for senior high school and higher education levels, starting

from fiscal year 1998/1999 to 1999/2000. The primary level and junior high level

expenditures are similar during the earlier fiscal periods and start to converge.

Junior high level expenditures lead the expenditures of the primary level. The

difference in per student public expenditures between junior and senior levels is

relatively stable, but differences with the higher education levels are relatively

increasing.

67

1,000,000.00

900,000.00

g. 800,000.00

~'" 700,000.00.....l 600,000.00os........ 500,000.00bIlos..... 400,000.00;.-<-; 300,000.00===-< 200,000.00

100,000.00

--primary

-junior

...•... senior

- - - ....- - - higher education .

---trprivate .' ."' "' '" .'...'

.'" .. ... ....... ~..

A-----A--~.•..........•..........•..........•..........•..... -

~ .~------. . . . .

1993/1994 1994/1995 1995/1996 1996/1997 1997/1998 1998/1999 1999/2000

Fiscal Year

Figure 1.16 Average Public and Private Expenditure Per Capita Per SchoolLevel

400.00

(400.00)

(500.00)

(200.00)

A e

o(100.00)

:2 300.00.S!!a.Ii 200.00'0

§ 100.00t/)::Jo.ct:.'-

.S1~Cll

t=Cll

""~ (300.00).!..~"1ii::Jc:

~(600.00) __outflow 1999

__ Inflow 1999

---outflow 1996

----inflow 1996

)( outflow 1993

--- inflow 1993

Figure 1.17 Per Capita Public Education Transfers Outflow and Inflow

68

Public education transfer outflow are estimated by assuming a flat rate tax using

equation (1 .17), L qgi =L qgt =L?e r Y/ =Qnational' If the tax rate (r ) is flat

the profile of public education transfer outflows will follow the earnings' profile,

regardless their earnings. Thus, as shown in Figure 1.17 and Table 1.9, the

population weighted average age of transfer providers is in the mid thirties. If the

peak of transfer receivers is in their fifties, the difference between transfer

provider and receiver is about 20 years. That is the time required to pay back the

government transfers by taxing their earnings and redistributing them to the

school age generations is about 20 years. The time required is relatively shorter

than in the private education transfers where more than 35 years are required to

pay back the education funds. This is due to the assumption made in the case of

private education transfers, that household heads are the main providers for

private education transfers. On the other hand, the public education transfers are

assumed to evenly tax earnings of all productive populations.

Table 1.9 Average Age of Public Education Transfers

1993 1996 1999

Average of age recipients 15.99 14.99 15.95

Average of age transfers givers 35.82 35.45 36.25

69

The average of public education expenditures was taken, qgj' and equation

(1.17) was used to calculate per capita public transfers per cohort t, q+ . I alsogf

obtain the level of enrolled population at school level k per age or cohort t, Nfk'

from the enrollment profile provided by Susenas. Finally, I calculated q+ whosegf

profile is shown in Figure 1.18.

Public education transfers significantly dominate private education transfers

(Figure 1.18). Though both transfers experience increasing patterns, public

education transfers disproportionately increase in fiscal year 1998/1999. The

high increase in fiscal year 1998/1999 is a result of the financial crisis in 1997.

The government expanded the education budget with the implementation of new

programs and social safety nets for all levels of education; with emphasis on the

elementary and junior high school levels, groups vulnerable to dropping-out. The

private sector, however, did not decrease their expenditure on education and

eventually increased it at rates consistent with previous years.

70

--AJblic93

....... AJblic 96

-AJblic99

__A-ivate 93

~A-ivate96

--A-ivate 99

350

:2 300.ma.'"a:::"0 250c:coIII

'"0E 200l!!~IIIc: 150e!I-,lga. 100coell)a-lii 50'"c:~

05 6 7 8 9 10 11 12 13 14 15 16 17 18 19 2021 22 2324 25 2627 28 2930 31 3233 34 35

Age

Figure 1.18 Per Capita Public and Private Education Transfers by Age of

Recipient 1993, 1996, 1999

Public education transfers reach a peak at around 7 to 12 years of age in fiscal

year 1993/1994, and expand to reach a peak at age 15 in fiscal year 1995/1996.

This was to accomplish the nine-year compulsory education policy. The

government transfers were spread evenly to older ages in fiscal year 1998/1999.

Private education transfers experience a peak at the later age to compensate for

the lack of government subsidies at the senior and higher education level. The

government subsidizes the nine-year basic education levels, and provides a

relatively smaller amount of subsidy for senior high school. The government also

subsidizes public universities and a very small amount of private universities.

Higher education enrollment is dominated by the private universities. Households

have to increase education expenditures if their children go to college,

71

particularly private universities. This explains why, on average, private education

transfers reach a peak at a later age than that of public education transfers.

Panel A of Table 1.10 summarizes average cost of private and public education

transfers. The same table also presents total annual public and private transfers

by school level (Panel B) and age groups (Panel C) in 1993, 1996, and 1999.

The total budget is calculated by multiplying the per student cost for each school

level by the number of students at the respective school level. I estimate both

private and public transfers, and also divide the profile by age groups.

Inconsistencies between school levels and age groups are due to the fact the

presence of late entries and repeaters. Late entries or repeaters expand the

enrollment numbers in one school level. Those who are enrolled in the primary

school level include members of age group 13 to 15 due to this reason. The

same situation occurs for those in the junior high school level, where ages range

from 16 to 19. There is about 15% to 18% of the total national education budget

that cannot be allocated, as the nature of application of the budget tends to be

more complicated. The unallocated budget is categorized as non-formal

education and training for civil servants.

Total public transfers are significantly higher than total private transfers at the

primary level. Average of private education transfers for primary and junior high

school is considered smaller than the average of public education transfers for

the same school level. There is a large increase in public education transfers at

72

the junior school level in 1996, compared to those in 1993. This may be due to

the initiation of the nine-year compulsory education program in 1994. The

government increased subsidies to the junior high level to accommodate the new

policy. The government constructed more junior high school buildings during

1994 to 1996. While private education transfers are slightly lower than public

education transfers for the senior school level for all years, the accumulated

private education transfers is relatively higher than the public education transfers

for the higher education level in 1996 and 1999 (Panel B).

Comparing among the school levels, higher public education transfers reduce the

private transfers because of high subsidies in the primary level. Subsidies occur

at junior, senior and even higher education levels. However, because the amount

of the subsidy at higher school levels is not as high as at primary level, the

reduction of private education transfers at these level is not as significant as the

reduction at the primary level. This is due to the government priority placed on

basic education levels. Junior and senior high schools receive subsidies as well

but the subsidy per capita is not as high as that for primary schools. In addition,

higher education costs are greater. Higher enrollment rates at the higher

education level may not be followed by proportional increment increase in public

education transfers. Last, but not least, is the greater role of private universities,

enabling replacement of the public universities.

Higher education schools experienced rapid development in Indonesia over the

last few decades. Private colleges are growing more rapidly than public colleges.

73

Private universities are usually more expensive than public colleges. While the

government increases its higher education budget, most of the funding is going

to public universities, where a small proportion of college students enroll.

Therefore, when parents choose to send their children to a university, the

majority bear the higher cost of private universities. Comparing the three years of

the analysis, private education transfers increase with public education transfers

for every school level. The private education transfers at the university level

increases at higher percentages exceeding the public transfers.

Accumulated transfers by age group provide similar profiles (Panel C). Private

contributions are increasing over time. In the fiscal year 1998/1999 private

contributions are about 40% of total transfers, while over 55% of transfers to the

age cohort exceeding 19 years are from private resources. Accumulated private

education transfers are significantly lower for 5 - 12 age groups. Public

education transfers are the highest at these age groups. A large increase in

public education transfers occurs for the age group 13 - 15. Public education

transfers play an important role among the younger age groups, while private

education transfers significantly affect the older age groups. Private contributions

are around 13% to education investment for those whose age are between 5 and

12 in fiscal year 1993/1994 and are increasing over the years. Private

contributions are also higher for older age groups, especially those who are older

than 16 years. This is also true of the profiles by school levels. The higher the

school level is the higher private contributions.

74

Table 1.10 Average and Accumulated Private and Public Education Transfers

1993** 1996** 1999**School Level or

Age GroupPanel A: Average Transfers by School Level (in Rupiah)

Private Public Private Public Private Public

Primary 24,924 164,694 38,999 178,803 60,829 196,611

Junior 65,311 156,827 95,066 226,693 265,173 295,020

Senior 132,442 238,187 200,569 302,155 290,767 366,429

Higher Education 317,792 352,899 624,314 462,853 763,489 602,573

Panel B: Total Transfers Given (in Billion Rupiah*)


Primary 740.24 4,891.35 1,140.30 5,228.06 1,734.17 5,605.18

Junior 465.74 1,118.35 882.09 2,103.42 2,496.39 2,777.37

Senior 554.33 996.92 987.62 1,487.84 1,542.29 1,943.61

Higher Education 644.17 715.34 1,684.17 1,248.61 2,059.61 1,625.52

Panel C: Total Transfers Given (in Billion Rupiah*)


5 -12 673.17 4,448.20 1,106.79 4,929.64 1,599.77 5,170.78

13 - 15 455.64 1,345.12 875.29 2,217.92 2,197.45 2,736.63

16 - 19 648.09 1,165.37 1,262.22 1,802.30 2,098.07 2,369.82

> 19 728.57 883.38 1,445.03 1,106.50 1,772.43 1,425.59

Note: * 1.00 USD =Rp. 2,500,- (1996 exchange rate). ** Public transfers use fiscal year

1993/1994,1995/1996, and 1998/1999.

75

Comparing the profiles by education levels and age groups reveal that repeaters

or late entrants in primary schools may increase the burden of public transfers by

as much as 1 billion Rupiah per year. The accumulated public transfers at the

primary level is about 4,891,00 billion Rupiah in fiscal year 1993/1994, while

public transfers in the same year for age group 5 - 12 at about 3,734,00 billion

Rupiah. If the net age of enrollees in the primary level is about 5 - 12, there is an

excess burden on public transfers to the primary level due to students who are

older than 12. They are primarily repeaters or late entrants. Other years have the

same profile and a similar amount of excess burden. This excess burden is

inefficient. Both excess private and public transfers could be reallocated to higher

school levels if the repeaters or late entrants did not exist.

7. Conclusions

Private education transfers flow from older ages to younger. In Indonesia, while

the average education transfers provider is about 55 years of age, the average

education transfers receiver is around 12 years of age. This is relatively younger

than the average age of education transfers receivers in the United States,

around 20 years old. Private education transfers in the United States are

primarily for higher education purposes. The average age of transfer recipients is

relatively higher for developed countries compared to those of developing

countries. This reflects differences in the age structure as well as the education

level distributions.

76

Education investment in Indonesia, as in any other country, comes from private

and public resources. Public education transfers dominate the private transfers at

the primary school level, while private transfers started to playa larger role in

covering education expenditures at the higher school level. Private education

transfers account for around 13% of education expenditures at the primary level

in fiscal year 1993/1994, and tends to increase over the years.

Most citizens earn a basic education and a small percentage pursues a higher

degree. In junior, senior and higher education, private contributions are

respectively 25%, 36%, and 45% in the fiscal year 1993/1994. In general, private

contributions tend to grow over time, particularly at the higher education level.

Indonesia continues to focus on basic education. In the fiscal year 1995/1996

and 1998/1999, private contributions for higher education are relatively larger

than government transfers. Household experiences indicate that higher

education is relatively more expensive, especially at private institutions. The

government at this time remains focused on basic education, from the

elementary to junior levels. While the senior high school level receives attention,

the government has just started to raise support for higher education. Yet, private

institutions are still relatively lacking in government support.

The estimation methodology utilized here has several limitations. In estimation of

private education transfers, the variation of junior and senior high school levels

are not taken into account. Even though in the aggregate this will not result in a

77

significant difference, the individual estimation may be biased at that particular

school age. The estimation of public education transfers covers formal education

excluding pre-school level. The increasing popularity of the pre-school level will

require urgency in estimation of the transfers given to this level by the

government. The public transfers to non-formal education also need to be

examined more carefully. The estimation of this school type will change the

profile of public transfers recipients since most non-formal education students are

from older generations. Last but not least, public transfers outflow require more

detailed age profiles of tax payers to accommodate the tax rate regulation by the

government. In addition to this, estimation of public education transfers in the

decentralized context is also a challenge for future research.

78

ESSAY 2: EDUCATION POLICY, CHILDREN'S SCHOOLING, AND LABORDECISIONS

79

1. Background and Objective

The objective of this paper is to empirically investigate the influence of education

policies on the trade-off between enrolling children in school and having them

work. Economists have examined the determinants of child labor, the effect of

government policies on banning child labor (Basu 1999), and the link between

international trade policies and child labor. Literature on the trade-off between

schooling and child labor in specific countries is also quite extensive . A model

developed by Basu and Van has been utilized in explaining how households

decide whether to put children to work, and the welfare implications of such

decisions.

Despite this extensive literature, theoretical and formal empirical investigations

on the effects of education policies on child labor are lacking. This gap in the

literature is addressed here, with the examination of the effects of education

policy on school enrollment rates and child labor supplies in Indonesia, using

Indonesian Socio-Economic Survey Data (Susenas). In particular, policies

directing school construction by the Indonesian government, in accommodation

of nine-year compulsory education laws, are considered. Duflo investigates the

effects of the construction of large primary school buildings in Indonesia between

1974 to 1984 on years of schooling and wages. This analysis specifically utilizes

distance to school as an approximation of building construction between 1993

and 1995.

80

Indonesia has passed several education policies in an attempt to enhance the

quality of its human capital. The Indonesian government imposed six-year

compulsory education laws in 1984, after constructing a large number of primary

school buildings beginning in 1974. Ten years later, an extended nine-year

compulsory basic education program was passed, following the construction of a

considerably high number of junior high schools, in 1994. The government then

instituted scholarships and block grant programs in response to the financial

crisis of 1998. Finally, the education system was completely decentralized,

effective from 2002.

In response to the implemented policies, the primary school gross enrollment

rate increased to 99.6% at the start of the 1989/1990 school year, and to higher

than 100%xi at the end of the 1990/1991 school year. The junior high school

enrollment rate also increased to 73.8% by 2001/2002 from 53.6% in 1993/1994.

When school enrollment increases, child labor supply usually decreases.

However, statistics for Indonesia indicate this does not necessarily occur on a

one-to-one basis. In 1993, more than 13.7% of junior high school children

(between 10 and 14 years old) were on the job market, 5.86% were working at

home, and 13.6% were not employed (Susenas 1993). In 1998 and 1999, these

child laborii rates were approximately 8.24% and 7.09%, respectively (according

to Sakernas 1998, 1999); or 10.96% and 10.04% respectively, when using 100

Village Survey data.

81

Emphasized here is the nine-year compulsory education policy initiated in 1994.

To facilitate implementation of this policy, the Indonesian government

constructed approximately 1,000 junior high school buildings. This paper

employs the difference-in-differences method in examining how the school

construction affected household decisions. This methodology follows Duflo

(2001) and is done to eliminate the problem of non-randomness and to

accommodate the intensity of the program. It is found that school construction in

Indonesia increased enrollment by 3%, two years after the program introduction,

and by about 5% after eight years. Results indicate that labor decisions are

influenced less than schooling decisions, that rural households are affected more

than urban ones, and boys more than girls. Household heads compensate for

declining child labor income by increasing their own labor supply, but the labor

supply of their spouses is ambiguously affected.

This paper contributes to the understanding of how the family determines

individual labor supply and how mandatory education policies change these

decisions. The analysis demonstrates that education policies implemented in

Indonesia in the 1990s increased parental labor supplies, to compensate for the

decline in child labor and the consequent loss of income. This paper provides an

essential background on the effects of education policies on family decisions

regarding child school enrollment and labor supply. An empirical analysis follows,

leading to the critical conclusion that the education policies implemented in

Indonesia increased parents' labor supply to compensate for the decline in child

82

labor, which eventually decreased child labor income. The program's effects on

the level of intra-household transfers from parents to children for both

educational and non-educational purposes are investigated in the next chapter of

this dissertation.

The following section outlines existing literature on education policies, schooling,

and child labor. Section 3 formulates a theoretical model for examining the effect

of the education subsidy on child and parental labor supply. Section 4 and 5

present an empirical investigation of the effects of compulsory education on

schooling enrollment, child labor, and parental labor supply. Section 6 provides a

summary of results and concludes this analysis.


Basu and Van (1998) propose two hypotheses on the causes of child labor.

First, child labor helps maintain family expenditures at the subsistence level.

Second, child labor is a substitute for adult labor. Households end up with

reduced welfare when child labor is banned. Duryea, Lam, and Levison argue

that banning child labor causes adverse effects as households can no longer

maintain a subsistence level of consumption. Ranjan suggests a ban on

products produced by factories employing children causes them to work in more

hazardous and unpleasant environments. Providing a credit market, on the other

hand, helps solve the child labor problem and increases school enrollment.

83

Policies such as subsidies or compulsory education promise to eradicate child

labor. While child labor laws strongly affect child labor force participation , Basu

(1999) suggests an integrative approach to reducing child labor, such as the

installation of education policies combined with economic development to

increase adult wages.

Baland and Robinson analyze child labor by comparing one-sided and two-sided

altruism models. They conclude that child labor is inefficient in an imperfect

capital market if altruistic parents fail to provide bequests, assuming there is a

trade-off between child labor and human capital accumulation. That is, poor

parents are too poor to save money and provide children with bequests.

Introducing child labor disutility in the model by Bommier and Dubois generalizes

Baland and Robinson's conclusions that child labor is not efficient. This result

occurs, with a perfect market, altruistic household members and no corner

solutions for transfers. Bommier and Dubois (2004) also argue that child labor

will not be efficient when net transfers flow from children to parents. In other

words, child labor is inefficient under negative net transfers.

Most literature on child labor discusses its efficiency and how child labor laws

affect the welfare of the household. Recently, economists have attempted to

explain household decisions regarding schooling and child labor. Determinants of

schooling have strong relationships with determinants of child labor with

opposing implications. Beegle, Dejia, and Gatti show the trade-off between child

84

labor and school enrollment. Child labor reduces school enrollment. A strong

relationship between household income, parental levels of education, and

children's schooling has been established . Poverty is the major reason for child

labor. The persistence of poverty is reflected in a less sensitive child labor

reaction to education policies among impoverished households. Evidence

across countries confirm this assertion. Priyambada, Suryahadi, and Sumarto

(2002) also show that the financial crisis in Indonesia caused stagnation in the

reduction of child labor.

Manacorda (2004) investigates the effects of the presence of other household

members on the household decision regarding child labor participation. He uses

data on child labor laws and household labor supply in the United States during

the 1920's, and confirms that the increase in the number of household members

who were working eventually increased the school enrollment of other household

members. Large families cope with having more children by choosing to send

more of them to school. Diversification between schooling and labor, however,

biases the substitutability between quantity and quality of children as older

children may have to work more to support younger siblings.

The effects of education policies on enrollment rates has been widely examined .

It has been found that the compulsory education program in the United States

raises the enrollment rate by 5%. Duflo (2001) investigates the effect of the

construction of a large number of primary school buildings in Indonesia on years

85

of education, as well as on the labor supply. Using the difference-in-differences

method, she finds that the large amount of construction improves the years of

education of the affected cohort. She also confirms that the policy eventually

increases their wages. Spohr demonstrates that Taiwan's education reforms

improved enrollment rates and years of schooling. In the long run, the program

also increased female labor force participation. Other researchers have

investigated the impact of schooling on child's health (Harold, et al. 2001).

Fitzsimons investigates the effects of risk on education and child labor in

Indonesia, using the 1993 Indonesian Family Life Survey (IFLS). She finds that

children in rural areas play an important role in smoothing family consumption

when their parents lack insurance or access to credit markets. Uncertainty hits

the 10 to 14 age group harder, such that they are more likely to accumulate

fewer years of education. Fitzsimons demonstrates that some parents in

Indonesia resist sending their children to school and do not consider education a

priority. Trade-offs between schooling and child labor has been well documented

across countries. Grootaert and Kanbur find that a basic education program

strategically reduces the incidence of child labor. Priyambada, Suryahadi, and

Sumarto (2002) conclude that children often have to work part-time to finance

their schooling. Banning child labor eventually results in children dropping out of

school.

86

Child labor and intergenerational transfers are strongly related. If intra

household allocation is defined as a child's consumption less his labor income,

child labor affects the intra-household allocation profile at early ages. Child labor

also reduces education transfers resulting from shortened human capital

accumulation. Emerson relates child labor to intergenerational redistribution.

He argues that a benevolent government that promotes social security programs

eventually decreases child labor and increases human capital accumulation.

3. Conceptual Framework

Basu and Van (1998) argue that parents allow their children to work in order to

maintain the household's utility level. Parents who treat child labor as a

safeguard for household income have to consider the effects of a decline in

family resources, should they choose to enroll their children in school. They must

rely on new resources to replace child labor earnings and to finance their

children's education. Several resources may be available to parents: inter

household transfers, savings, bank loans, or increased income through additional

work. In developing countries, where capital markets are not always available,

parents often have to substitute for the lost labor of their children. Some

households sell their assets to finance their children's education. The following

model attempts to explain the effects of education subsidies through school

construction on school demand, child and parental labor supply.

87

Assume a household has one child. Educational expenditure is the focus of this

analysis, with households only consuming public goods (xg) at price pg. Leisure

time of parents is denoted by Lh and a parent works (1- Lh) . Children may go to

school or work. If parents decide to send their child to school, they have to

spend q*p9 for his education where pe is the education price and q is the

education level. A child works (1 - LC) where LC is the amount of time spent at

school. A child who attends school, therefore, can also work.

Parents' utility depends on their leisure time Lh, consumption of public goods ~,

and their child's utility. Altruistic parents care for their child by incorporating the

child's utility into household utility. The child's utility is a function of school time

Lc, which is assumed to also be leisure time. Therefore, child labor or working

time (1-LC) will cause disutility for the child. As with parents' utility, public goods'

consumption enters into the child's utility. Consumption of education is a positive

factor in the child's utility. Thus, education is perceived here as both consumption

and investment. A child does not have decision power and parents maximize

utility by making decisions for the child; regarding consumption, investment in

education, and employment as follows:

2.1

Household utility depends on consumption of public goods, school time, and

educational expenditures. The household budget constraint is as follows:

88

2.2

Parental wages are 'I and household non-labor income is,t. Income is pooled

and re-distributed to purchase public goods and pay for education. The child's

earnings yC follow the relationship:

2.3

Where s is a child's initial endowment, dependent upon parental education.

Substituting the earnings function into the budget constraint, then maximizing the

utility function subject to the constraint, forms the Lagrange function as follows:

L=V(X) =u(xg ,Lh,UC(xg,q,LC))-

At (pg xg + peq_(1- Lh)yh - (1- LC)yC - yn) - ~ (yC - sql-a (Lr r).

Parents' utility is assumed to be separable. In general, parents maximize their

weighted utility and their child's utility with weights (1) and 1- (1) respectively.

Parents maximize V(X) = {f)UP (xg ,Lh,) +(1- (1))U C (xg ,q,Lc ). However, for

simplicity, assume that parents place equal weight on their child's and on their

own utility. Utility is also quasi-concave for every argument and optimization

produces the following first order conditions:

q: ug-Atpe+~s(l-a)q-a(Lrr=0

x g: U + UC

- ~pg = 01 xg xg

Lh: Uh

h - ~yh =0~

89

( )

a-lL~: U~ - Aryf + Azsaql-a Lf =0

At: pg xg + peq- (1- Lh) yh _ (1_ LC) yC _ yn = 0

Az: yC - sql-a (Lfr = o.

The first order conditions imply:

A = u~ + Azs(l- a )q-a (Lc)a = U:c + Azeql-a (Lct-1

= U;h = U:g + U:g

1 pe yC yh pg 2.4

The first component of the above equation is the ratio of the sum of marginal

utility of education investment and its return on education investment. This is due

to the household's perception of education as both consumption and investment.

The level of education is chosen at the level where its marginal utility and its

return equal the ratio of parents' marginal utility of leisure to earnings. Parents

trade off higher education for their children (including higher education

investment (q) and longer years of schooling) with an increase in their own labor.

Solved for optimum child labor, parental supply and education demand, the

Marshallian demand functions are:

LC* = LC(pe pg yh yC.z)1 1 ' , , ,

q* =q(pe,pg ,yh,yC;Z)

Lh* =Lh(pe,pg,yh,yC;Z)

2.5

Government policy on education may take several forms: construction of

buildings, provision of educational supplies, teacher training, scholarships, or

vouchers. This paper analyzes how school construction affects demand for

90

education and child labor. The Indonesian government accommodated the nine

year compulsory education policy by constructing additional junior high school

buildings.

Construction of additional schools increases the school supply (Figure 2.1).

School construction decreases transportation costs and reduces the opportunity

cost of education. If the market price of schooling is pm and the government

subsidy is PO, cost of schooling (pe) will bepe =pm - pg. That is, the education

subsidy, PO, directly reduces the education price, pe. This price effect allows

parents to invest in the same level of education at a lower cost. Therefore, the

policy increases incentives for parents to send their children to school. At the

same time, parents can invest in a higher level of education at the same previous

price. The demand for schooling increases. Thus, the more time children spend

in school, the less time they have to work.

The dotted vertical lines in Figure 2.1 indicate compulsory education as amended

by the Indonesian government. The government set a minimum education level,

in this case a minimum of nine years of education. If the government efficiently

enforces the policy, a case indicated by line 2A, school demand would be

censored at the minimum required level. However, inefficient or weak

enforcement, as indicated by line 28, will provide more opportunities for

households to avoid adhering to the law. Thus, in this case school demand will

depend on household choices.

91

2B. in case of inefficient

compulsory education

/ 81

1. before

compulsory education

2A. efficient compulsory

education

/3. construction

more buildings

4. Lower

opportunity cost

D1

Figure 2.1 The Effect of Education Subsidy and Compulsory Education onSchool Demand

Under the assumption of perfectly separable utility, inspection of comparative

static conditions confirm that dq/dPe is negative under the condition that

'e = (1- a)yC /q? a/G. If the return on education is higher than a/G ' decreases

in the price of education due to the government subsidy will raise education

demand. Otherwise, the sign is ambiguous. The implication for child labor, (1-LC),

on the other hand is the opposite, where d (1- LC) / dPe is positive.

92

The own price elasticity of education demand (sPe,q) is the percentage change in

education demand due to a unit percentage change in education price. That

is the marginal return to schooling time. The comparative static,xiii produces

The cross price elasticity of child labor is the change in child labor resulting from

a one unit change in education price. That is, sec =, efL (Pe,Pg,yh,yC,a,s),P ,r. r.

where 'e =s(l-a)q-a(L~r is the marginal return of education demand

previously defined and fL(Pe,Pg,yh,yC,a,s) = f(Pe,Pg,yh,yC ,a,s)pe/l-Lc . Thus

2.6

The ratio of the price elasticity of education demand to the cross-price elasticity

of child labor is a function of a and labor time.

The effect of changes in education price on parents' labor supply is ambiguous in

any case. A reduction in education price that leads to a decrease of child labor

will result in price and income effects which simultaneously impact the

households' decisions on their labor supply. A reduced cost of education leads

to an increase of education demand. This eventually results in a decline in child

labor supply. Schooling requires almost full-time attention, and thereby reduces

93

full-time working. However, being a full-time student is not meant to completely

eliminate the family's intention of sending their children to the labor market, even

for a more limited time. With the same expenditure, parents can invest in a

higher level of education for their children. Increased schooling time will reduce

children's working time. Therefore, there is a reduction of family income that

should be compensated for, in order for the family to maintain its welfare level.

School demand is higher after the program's implementation indicating that

parents wish to send their children to school more after the program. However,

the child labor supply depends on factors other than the education price. For

parents who chose not to send their children to school before the program, but

send their children to labor market instead, the decision to send their children to

school means higher education expenditures. To maintain the same level of

welfare, families must find other resources, which will be difficult if the credit

market is under-developed. Thus, parents have to work more or continue to

depend on child labor. Therefore, although enrolled in school, the children still

have to work and contribute to the household income, so that they can maintain

their welfare level. In this situation, the response of child labor is less elastic. A

possible change is in the distribution of working hours. The child works fewer

hours and spends more time at schools. Jobs available for these children are

those that indirectly assist them with remaining enrolled at school. However,

some children may quit their jobs, and compel their mother to work or the

household head to increase his/her labor supply.

94

Another situation occurs with parents whose children attending school and

working prior to the program's implementation. Cheaper transportation costs due

to distance changes may leave at least three options for these households. First,

if parents choose the same level of education, they are able to reduce child labor

and maintain the same level of welfare. Second, parents can send their children

to school with lower education expenditures, maintain their child labor levels, and

acquire a higher utility level. Third, parents may choose to increase their

education demand and acquire higher levels of education for their children,

maintain their child labor levels, and receive at least the same level of utility. The

bottom line is that the lower education price leads to a higher school demand and

allows households to maintain the same welfare level, if child labor options are

open. Some options are corner solutions. Households are only able to invest at

the minimal school level required by the government.

Assume that there is no tax re-distribution effect; then the education subsidy

produces a positive price effect as shown in Figure 2.2. The education subsidy

increases education demand and reduces the child labor supply. The income

effect, on the other hand, increases both the child labor supply and school

demand. Overall, the education subsidy increases school demand and reduces

child labor non-proportionally. In addition, although child labor and school

demand are not perfect substitutes, the education subsidy brings the household

to a higher utility level with lower child labor levels.

95

......

~ ::::::::'::::::::'::::::::~ .~ y .~"

........................

·..····.. ····......school Demand..................

Figure 2.2 The Welfare Analysis of the Effects of Education Subsidy on School

and Child Labor Demand

The following empirical section places an emphasis on how education policies

affect both child schooling and child labor. Even though the unitary model above

does not differentiate between fathers and mothers, the empirical analysis

estimates the effects of the education policy on father and mother's labor force

participation. Education policy is approximated by the changes in distance

between households and the nearest junior high school. This measure is also

assumed to reflect changes in education prices.

96

4. Data and Empirical Strategy

4.1 Data Description

This analysis uses Indonesia socio-economic survey data (Susenas) for 1993,

1996, 1999, and 2002. These data sets capture the period of implementation of

the nine-year compulsory education program, which began in 1994. The

Susenas-core collects the main characteristics of households, including

expenditures and income, and information on individuals in the households,

including gender, relation to household head, school enrollment status, the

highest level of school attended or completed, and household head

characteristics. A more detailed survey of socio-economic status, including

education, expenditure, and income information, was collected, and is known as

the Susenas-module. This special module covers approximately 65,000

households and 225,000 individuals, fewer than that of the Susenas-core sample

size.

Table 2.1 presents the average of variables used in this analysis. Displayed in

the table are children's activities, divided into school and working, for four years

of data. Housework and other activities are not provided. Working refers to full

time or part-time employment. The second category of working presented in the

table includes both those who work full-time and those who work part-time.

Children can attend school as a main activity and continue to work part-time. The

number of working hours is also presented Table 2.1. Panel A includes data on

all children aged between 8 and 25 years. Panel B presents information for

97

children who are attending school; and Panel C displays data for children who

are working full-time.

More than half of children are enrolled in school; with about 15% working full-time

as shown in Panel A. In addition to working full-time, children can also work part

time. Average working hours for children who work part-time is about six hours

per week. Between 1993 and 2002, enrollment and full-time work figures have

changed slightly. However, when including part-time workers, the proportion of

working children has declined considerably. Their working hours, however, does

not change significantly during the years of analysis.

School children also work on part-time bases (Panel B). The proportion of part

time and full-time child workers fluctuates during the years of analysis. In 1999,

this value increased, nearly reaching the initial figure in 1993 due to the financial

crisis in Indonesia. By 2002, the percentage of school children that worked had

declined considerably to 2.3%.

Panel C shows that the percentage of full-time working children enrolled in

school fluctuates slightly over the years of analysis. The percentage of working

children who were also enrolled in school declined in 1996 compared to 1993.

This percentage fluctuated between 1999 and 2002. Average hours worked was

at about 39 to 40 hours per week. Research indicates that a compulsory

education policy effectively increases the enrollment rate. If the inverse

98

relationship between school enrollment and child labor applies, this education

policy should negatively affect child labor. However, the number of working

children remains high even though education policies successfully increase

school enrollment. The statistics shown here confirm that some children both

work and attend school. Thus, education policies may alter school enrollment

without making part-time child labor vanish.

Table 2.2 presents information on variable means for employment data of

household heads and their spouses. Panel A provides working statistics for

household heads, while Panel B displays the same statistics for spouses. Similar

to Table 2.1, Table 2.2 also presents working conditions, as a percentage of

household heads or spouses who are working full-time or part-time. In addition,

the number of hours worked are displayed for both groups.

99

Table 2.1 Variable Means on School Enrollment and Employment

1993 1996 1999 2002

Panel A All Children

Number of Observations 95474 91325 87613 85332

Age 13.03 13.20 13.58 13.59(5.32) (5.38) (5.46) (5.54)

School 59.8% 61.4% 61.3% 60.5%(0.49) (0.49) (0.49) (0.49)

Full-Time Working 15.2% 14.0% 14.4% 15.4%(0.36) (0.35) (0.35) (0.36)

Part-Time or Full-Time Working 20.0% 18.4% 19.3% 17.9%(0.40) (0.39) (0.39) (0.38)

Number of Hours Worked* 6.78 6.21 6.51 6.75(15.47) (14.89) (15.20) (15.87)

Panel B School Children


Age 11.76 11.87 12.03 11.89(3.65) (3.71) (3.79) (3.82)

Full-Time Working 0.4% 0.3% 0.5% 0.4%(0.06) (0.06) (0.07) (0.07)

Part-Time or Full-Time Working 5.5% 4.4% 5.1% 2.3%(0.23) (0.21 ) (0.22) (0.15)


Panel C Working Children


Age 19.23 19.66 19.94 20.13(3.40) (3.39) (3.24) (3.15)

School 1.6% 1.4% 2.1% 1.7%(0.12) (0.12) (0.14) (0.13)


Note: *Includes part-time and full-time workers. Standard deviations are in parentheses.

100

The majority of household heads were full-time workers. There are slight

fluctuations in the trends of the percentages of full-time workers or all types of

workers including part-timers; as both percentages declined in 1999. The

percentage including part-timers was about 86.8% in 1993 and declined slightly

in 1999. The financial crisis in 1998 may have caused the high unemployment

rate afterwards. The average hours worked for the household head was about 40

hours, with slight fluctuations experienced between the years.

Similar situation to household head occurred to spouses. Spouses' employment

rate declined in 1999. Spouses' employment participation was at about 35% in

1993, and declined to 31.5% in 2002. A larger percentage of spouses worked on

a part-time basis. When part-time worker is included, the working participation

rate was about 15% higher for almost all years of analysis. On average, spouses

work fewer hours than household heads.

101

Table 2.2 Variable Mean on Employment of Household Heads and Spouses

1993 1996 1999 2002

Panel A Household Head

Age 44.11 44.51 45.06 44.79(13.53) (13.52) (13.82) (13.74)

Full-time Working 84.5% 82.8% 81.8% 83.3%(0.36) (0.35) (0.39) (0.38)

Part-time or Full-time Working86.8% 86.2% 85.7% 86.0%(0.34) (0.34) (0.35) (0.35)


Panel B Spouses

Age 37.63 31.76 38.92 38.79(11.84) (17.64) (11.95) (11.96)

Full-time Working 35.0% 33.4% 33.1% 31.5%(0.48) (0.47) (0.47) (0.46)

Part-time or Full-time Working49.0% 48.5% 50.9% 44.8%(0.50) (0.50) (0.49) (0.49)


Note: * Includes part-time and full-time workers. Standard deviations are in parentheses.

If school enrollment (Ei) or child labor (Wi) are a function of the price of education

and other characteristics of the individuals, household or region, a simple

regression enables us to reveal which factors have the greatest influence on

demand. Consider a simple linear relationship:

2.7

102

The child labor variable, Wi, has the same specification. That is the independent

variables are similar. Following equation (2.5) child school enrollment or child

employment status is a function of the price of education (Pe). X represents

individual characteristics, household characteristics, and sub-district

characteristics. Education price, Pe, is unobserved and approximated by

distance and travel time to the nearest school. This distance approximates

transportation costs and reflects the opportunity cost of going to school. If the

relationship is linear, education price can be approximated as

Pe; = rpt + rp2d; + rp3t;+ K;. Where distance to the nearest school is reflected by d;

and time to reach the school is t. Thus, the reduced form of equation (2.7) is:

2.8

The assumption that the distance or travel time to school is uncorrelated with X

may hold before the compulsory education program is enforced. After the policy

is imposed, the government decides where to place schools, depending on sub

district characteristics, such as primary school enrollment of the previous year.

Therefore, non-randomness and endogeneity problems may occur if the

regression is applied to cross-sectional data the year after the program is

imposed. The control variables X may be correlated with school distance, d;.

The government systematically builds schools to accommodate the compulsory

education policy parameter. Estimates of equation (2.8) will be biased if applied

after the program was begun (i.e., using 1996 data). The distance variable is

endogenous and depends on the school enrollment of the previous year. If junior

103

high school distance is assumed to change only after imposition of the policy in

1994. Equation (2.8) can be used to examine the determinants of child labor and

enrollment using 1993 data. The simple regression in Equation (2.8), restricted

to the 1993 sample is thus done prior to analyzing the program's effect on

enrollment and child labor.

Table 2.3 presents the bivariate probit regression results using the sample of

children between 10 and 20 years old from 1993 survey data. The regressions

are applied to school enrollment and child employment decisions. Results

represent the probability of being enrolled or involved in the labor market in

percentage terms.

The results show that residence in urban areas is positively associated with both

school enrollment and child labor. In urban areas, a child has a higher probability

of enrolling at school than going to work. It is generally accepted that households

with females as the household head are vulnerable to poverty. However, the

regression results indicate that female household heads are associated with

increased school enrollment, rather than higher child labor supply. The

probability of a child attending school is positively high when the household head

is female. Parental education is an important determinant of school enrollment

and child labor. The years of education of both heads and spouses have a

positive effect on school enrollment; as their years of education increase, the

probability of child labor decreases.

104

Children's characteristics are important in determining their activities. Male

children tend to attend school more than go to work, implying that households

prefer to send their female children to work. Child labor primarily occurs in the

agriculture industry. Even though industrialization is underway, Indonesia

continues to have a large agricultural industry. Therefore, agriculture households

prefer for their children to work in the paddy field. These children tend to have a

lower educational level because of this preference.

In addition to the agriculture industry, households who have their own business

or a family business tend to have their children participate as family workers. For

example, a child may have to watch over the family shop after returning from

school. As the child ages, he/she may be required to work full time and

discontinue his/her education. Households prefer for the eldest child to work and

contribute to family's income. As previously mentioned, female children are more

vulnerable to child labor. Those who are the oldest child and female are the most

vulnerable to child labor. These children usually have lower educational level, as

they have to work from an earlier age.

105

Table 2.3 Regression Results on Determinants of School Enrollment and

Employment Decisions, Susenas 1993

School WorkingEnrollment Decisions

Honsehold Characteristics

Urban 2.39 1.19(0.02) (0.01)

Household Head Female 2.03 0.57(0.06) (0.03)

Household Head Single -0.08 -0.74

1-(0.06) - (0.02)

Years ofEducation Father 1.24 -0.37(0.01) (0.01)

Years ofEducation Mother 0.61 -0.37(0.01

1-(0.01

Concrete Wall -5.09 1.46(0.13) (0.06)

ChUdren Characteristics

Male 4.02 0.23

-(0.02) _. (0.01)

Agriculture FieldDummy -20.05 22.39(0.05) (0.03)

Family Worker Dummy 3.50 4.37(0.04) - (0.02)

Oldest Child -17.48 2.32(0.03) (0.01)

Male· Oldest Child 0.76 2.49(0.03) (0.02)

Time took to school -0.02 0.03(0.01)

~(0.01)

Distance to schooJ 0.03 -0.02-0.01 (0.01)

R-Square 0.64 0.46Number ofObservations 13,967 13,967

Note: Coefficients and standard deviations are in percentage terms. Highlights indicatecoefficient is significant at the 99% confidence level; Standard deviations are in parentheses;Province dummies, other housing characteristics and other individual characteristics are not

displayed.

106

Distance to school does significantly affect decisions regarding school enrollment

and child labor. For every one kilometer reduction in distance between the

household and the nearest school, there is an increase of 3% in the probability

that children enroll at school. The presence of a nearby school reduces the

probability that households send their children to work. Regression results

indicate that the change in the probability that households send their children to

work is about 2% with a school distance change of one kilometer. There is a non

proportional effect of changes in distance on child labor compared to child school

enrollment.

4.2 Empirical Strategy

I employ the difference-in-differences method to measure the effects of an

exogenous policy on junior high school age groups, and to solve the bias in

estimation noted above. Randomized assignment of public policy on school

expansion is politically unpopular. In the case of the nine-year compulsory

education policy, the government constructed junior high school facilities based

on previous primary school enrollment levels, as well as junior high school

enrollment rates. Thus, the placements are not random. The difference-in

differences method makes it possible to evaluate the government program in

spite of this non-randomness . Duflo, Mullainathan, and Bertrand suggest to

avoid serial correlation bias in Difference-in-Differences method by collapsing

periods of survey into two, before and after the program. If age groups who are

107

affected by junior high school construction are used as treatment groups, while

other age groups who are not affected are designated as controls, the difference

in-differences method compares outcomes such as school enrollment or child

labor, before and after the program was started, between treatment and control

groups.

Changes in treatment groups' school enrollment during the period under

investigation is compared with those of control groups' school enrollment during

the same period. The same method is applied to their employment decisions and

the number of hours worked. In the case of the previous estimation in the

previous section, a sub-district variation or fixed effect might weaken the effect of

distance to school on school enrollment or working decisions. The expected

differences across control groups during the period are assumed to be negligible.

The difference-in-differences method eliminates the sub-district effects. Thus,

unbiased estimation using the difference-in-differences requires additively

separable period fixed effects and regional fixed effects.

Higher enrollment rates due to the lower opportunity costs of schooling follow the

construction of new buildings and classrooms. During the fiscal years 1994/1995

and 1995/1996, the Indonesian government built approximately 1,000 junior high

school buildings and 5,000 classrooms to facilitate extended compulsory

education. I use distance to the nearest junior high school as a proxy for this

108

policy as previously described. I identify school density at the sub-district level

and changes that occur between these two years.

There is a potential endogeneity problem, as families tend to move to areas

where schools are closer. If families can move to a region with more schools, the

estimation is bias upward. Duflo (2001) finds that "91.5 percent of the children in

the IFLS sample, were still living in the district they were born in at age 12 (p.8)".

She briefly discusses how families might move to benefit the program, and

suggests the region of school as instruments. However, the appropriate data are

not available. I assume the immigration rate (9.5%) is not large enough to lead to

biased in estimates.

I assume that in the absence of school construction, there would have been no

change in the enrollment rate. I modify equation (2.8) and estimate the

followingXiv:

2.9

Where Efi is school enrollment of individual i of treatment group f at sub-district I.

Subscript sub-district I is, again, dropped for simplicity. The difference-in

differences effect of building construction on the enrollment rate at the sub

district level is captured by the coefficientp2' This is the interaction between the

average distance to the nearest junior high school, d" and treatment group Afi.

Time, before and after the program, is indicated by dummy variable t.

109

Equation (2.9) is also used to estimate the program's effect on the decision to

work full- or part-time and the number of work hours. Provincial, sub-district

characteristics and household characteristics are included as controls. These

observable characteristics may affect the demand for school as well as working

decisions. Unobservable characteristics that may affect the choice of activities,

however, still lead to bias in the estimates.

The treatment group is represented a dummy variable, Afi" which equals 1 if

individual i is between 12 and 15 years of age and zero otherwise. Control

groups include those who are not constrained by the nine-year compulsory

program. There are two categories of control groups. Those who are currently

not affected by the program but may be affected in the future, make up the first

category. Included in this category are those in the 8 to 11 age group, and those

older than 15 years. The younger age group, 8 to 11, is not compelled by the

program during the survey but will be in the following one to two years. Those in

the age group 16 to 19 are still considered to be of school age, and may

experience higher enrollment rates triggered by the program. Thus, a spillover

effect may exist, which would bias the estimation results. The second control

group includes those who are not affected by the program for the entire period

under consideration. This category includes those older than 15 years of age. In

addition, late entries, drop-outs, and early entries may affect the division between

control and treatment groups.

110

The specification of equation (2.9) is interpreted in Table 2.4, which presents the

individual effects, being either treatment group or control group, of the change in

school distance before and after the program on school or working decisions.

The change can be calculated by a subtraction of the outcomes before and after

the program (horizontally) or between control and treatment groups (vertically).

The difference-in-differences effects are shown as Pz (d1- do)' the coefficient of

interaction. Generally, the final difference-in-differences is presented as

coefficient of interaction pz for every one kilometer distance changes.

Table 2.4 Interpretation of Difference-in-Differences of Equation (2.9)

Control Treatment Difference

Before Po + P1do Po + P1do + pzdo + P3 pzdo + P3

After Po + P1d1+ P4 Po + P1d1+ PZd1+ P3 + P4 PZd1+ P3

Difference P1d1- P1do + P4 (P1d1-P1do)+ pz (d1- do)

(pZd1- pzdo) + P4

Dividing observations into two groups- control and treatment- is difficult due to

repeaters, late entrants, and early entrants at the same school levels. Children

enroll in junior high school at a considerably wide range of ages from 11 to

around 20 (Susenas 1993, 1996). Dividing age groups into two may create an

estimation bias. To overcome this problem and to see the variation in enrollment

rates by age, I expand the treatment and control groups into age specific dummy

variables (Duflo 2001). I estimate the interaction coefficients between age

dummy variables, Afj, and distance to the nearest junior high school, dt:

111

f=25 f=25

Ej =Jil + Ji2dj + L: Ji3fdj *An + L: Ji4fA,j + Ji5t + Ji6 X + V •f=8 f=8

2.10

Household characteristics and sub-district characteristics in the previous

estimation are used in the estimation. Previously, the index f reflected whether an

individual was categorized as part of the control or treatment group. In this

specification, the index f reflects the individual's age group. Coefficients of

interactions, Afj reflect the program's effect on a specific age group, f. The

meaning of the coefficient of interaction term remains the same. That is, it

reflects the effect of a kilometer school distance change on the household

decision's regarding children's activities. For those categorized as part of the

treatment group, individuals in the 12 to 15 age groups, the coefficients of

interaction are expected to be significantly different from zero.

Distance changes should negatively affect school enrollment. The nearer the

school, the higher the probability that households send their children to school.

Alternatively, distance should be positively related to child labor in the treatment

group. The control group's coefficient of interaction is not expected to be

significantly different from zero.

4.3 Simple Differences

The non-parametric difference-in-differences results of demand for schooling and

work using 1993 and 1996 data are illustrated in Table 2.5. The treatment group

112

includes children between 12 to 15 years old. Two control groups are presented:

the first combines the young age group (8 to 11) with the old age group (16 to

25); the second only includes those older than 20. However, the program may

significantly affect the young, in which case including the young age group as

part of a control group leads to a biased estimation. Schooling increases for the

young age group, while it does not change for the older age group.

The program increased the enrollment rate of the treatment group. The treatment

group experienced a 69.1 % enrollment rate before the program, increasing to

76% after the program's implementation. Eliminating of the variation by use of

difference-in-differences between treatment group and control group provides the

effect of the program on the treatment group.

Effects of the nine-year compulsory program on the treatment group differ with

the use of the two control groups. Including the young age group as a control

reduces the effect of the program on the treatment group. This is because

enrollment rate of the young age group also significantly increased after the start

of the program. The program significantly affects the young age group during the

years of analysis. Excluding the young age group from the control increases the

effect of the program on the treatment group. The difference-in-differences

results for enrollment rates between treatment and control groups are about 4%

when compared with the first control group and 7% when compared with the

second control group. Child labor declines by only 4% when the older group is

113

the control. Including the young age group in the analysis may cause the true

effect of the program on the treatment group to be biased.

Table 2.5 Non-parametric Difference-in-Differences Tabulation on Child Labor

and Enrollment Between 1993 and 1996

Proportion EnrolledBefore After Difference DID

Control Group 8 -11 & 16 - 25 53.18% 55.67% 2.50% 4.43%

Control Group 20 - 25 4.13% 4.13% 0.00% 6.92%

Treatment Group 12 - 15 69.07% 75.99% 6.92%

Proportion WorkingBefore After Difference DID

Control Group 8 -11 & 16 - 25 27.59% 23.97% -3.62% 0.39%

Control Group 20 - 25 57.11% 57.11% 0.00% -4.01%

Treatment Group 12 -15 12.64% 8.62% -4.01%

Figure 2.3 shows the non-parametric difference-in-differences (DID) results by

age for three categories of activities: enrolled at school, working full-time, and

both full- and part-time. The figure compares the difference-in-differences

between 1993 and 1996 as well as between 1993 and 2002. Panel A of Figure

2.3 presents non-parametric DID results for the proportion who enroll at school

and who work full-time. Panel B exhibits non-parametric DID for the proportion

who enrolls at school and who work either part-time or full-time. For this purpose,

I use those aged 20 to 25 as a control group.

114

Panel A Nonparametric DIDSchool Enrollment vs Full-time working

Age

...•... Full-time Working 1993/1996

••• )1(••• Full-time Working 1993/2002

--Enroll at School 1993/1996

~ Enroll at school 1993/2002

1'2:::::.'1.'3.. 14 15 16 17 18 19

•• -::K •• '-:.' .-;.-••• '" ••••••••••••_. ' •• " ' ::: -:.-:-f.~.. ."~.

')1(... ,.''" ••::1( •••••••• )1(•••

0.2

0.15

0.1-c'0 0.050-CIltnIII- 0CCIlU...CIl0- -0.05

-0.1

-0.15

Panel B: Non-parametric DIDSchool Enrollment vs Including Part-time Working

17 1~-----16

--Including part-time working 1993 11996

--er-Including part-time working 1993/2002

12"--1''O"8---+14...~5------

--Enroll at School 1993/1996

--.- Enroll at school 1993 12002

0.2

0.15

0.1

-c 0.05'00-CIl 0~cCIl -0.05u...CIl0-

-0.1

-0.15

-0.2

Figure 2.3 Non-parametric Difference-in-Differences Results Using 1993/1996

and 1993/2002

115

The compulsory education program affects school enrollment more than it affects

child employment. For every age group, the program increased enrollment rates

by a higher percentage than the decrease in child labor. The effect of the

program gradually increased from age 12 to age 15 or 16, then gradually

declined for older age groups. The program decreased the percentage of part

time workers more significantly than the percentage of the full-time workers, in

the longer term. That is, examination of the profiles of full-time workers, and part

time workers in Panel A and Panel B, which includes 2002 survey data, indicates

that the program gradually influenced enrollment and part-time workers, but did

not similarly affect full-time workers over the long term.

5. Empirical Results

5.1 The Effect of School Distance on Children's Activities

Table 2.6 presents coefficient of interactions from estimation results for equation

(2.9). That is the difference-in-differences results when dividing individuals into

treatment and control groups. The first set of results compares 1993 with 1996.

The second set includes 1999, while the third set of results involves 2002 data.

The table consists of four panels and three columns. Panels A, B, C, and D

differentiate between results for children who are enrolled in school, working full

time, working full- or part-time, and the number of working hours, respectively.

Each column represents different control groups. The first column indicates

116

control groups of 8 to 11 and 16 to 25 years old. The second column eliminates

the younger age group from the control, and the third column uses only those

who are older than 20 years of age as a control. The same arguments regarding

the control group selection are applied. Only the third category is displayed for

analysis of 1999 and 2002 data. The coefficients indicate the effect of the

program on the probability on each of the child's activities per one kilometer

changes in distance.

Changes in distances to the nearest junior high school affect school enrollment

and child labor moderately. The distance changes do not affect the percentage of

child labor as much as the percentage of school enrollment. The coefficient of

interactions, when using school enrollment as the dependent variable, are

negative and significantly different from zero at the 1% level of confidence. The

effect of the program is lower when including younger control group. Enrollment

rates increased about 0.24% (percentage point) per 1 km distance change. The

reported change in distance varies from 6 km to 8 km during the survey period.

For an average 6 km distance change, the difference in enrollment between the

treatment group and control group, both before and after the program

implementation is about 1.44%. While using different control groups does not

significantly change the results of schooling decisions, choice of control groups

does affect the results for child labor. For the different categories of child labor,

the sign and significance of the coefficients differ according to the choice of

control groups.

117

Table 2.6 OLS Regression Results for Difference-in-Differences with Four

Different Dependent Variables: Coefficients of Interaction

Two-Year Analysis***Three-Year Four-YearAnalysis*** Analysis***

(1) (2) (3) (3) (3)

Panel A: Enroll at School

Coefficients-1.l9E-03 * -2.0lE-03 * -2.44E-Q3 * -3.00E-03 * -2.45E-03 *

Standard Deviation 3.39E-04 3.57E-04 3.73E-04 3.47E-04 3.44E-04

Number ofObservation 119,477 84,071 57,626 82,474 126,308R-Square 0.11 0.35 0.51 0.53 0.53

Panel B: Full-time Working

Coefficients-4.90E-04 * 1.24E-03 * 6.49E-04 ** 2.99E-03 *-5.16E-04

Standard Deviation 2.49E-04 3.18E-04 4.46E-04 4. 14E-04 3.4lE-04


Panel C: Part-time and Full-time Working

Coefficients 5.32E-04 ** 7.88E-04 * 1.55E-03 * 1.29E-03 * 6.95E-04 *Standard Deviation 2.94E-04 3.52E-04 4.66E-04 4.33E-04 4.05E-04


Panel D: Number ofHours Worked

Coefficients -5.53E-04 -7.41E-03 1.l3E-02 9.06E-04 1.4lE-03 *Standard Deviation 1.05E-02 1.39E-02 2.03E-02 1.91E-02 4.24E-04


Note: *Indicate significance at the 99% confidence level: ** denotes significance at the 95%confidence level (1) control group ages 8 - 11 and 16-25: (2) control group ages 16-25: (3)control group ages 20-25: *** two year analysis use 1993 and 1996, three year analysis use

1993, 1996, and 1999, four year analysis use 1993, 1996, 1999, and 2002

118

The junior high school distance changes may affect the younger control groups'

school enrollment. Inclusion of the younger control group results in a negative

and significant coefficient of interactions; excluding the younger group produces

a negative but insignificant coefficient. When including only the group of those 20

to 25 years old a positive coefficient results, which is as expected. This may be

due to the spillover effect of the program on younger children's labor supply, as

previously mentioned. Younger children work less than the treatment group

would, as they are constrained by both the six-year and nine-year compulsory

education programs. This result may also be due to the fact that younger children

are more likely to have older siblings who are working. The younger siblings are

able to attend school because there are older siblings that may have finished

school and are already participating in the labor market. Parents may also prefer

that the older siblings work rather than attend school, allowing the younger

siblings to enroll in school . More analysis on the effect of siblings on school

enrollment and labor participation are discussed in the later sections.

The program decreased child labor by a lower percentage than the increase in

school enrollment. The decline of child labor substitutes for about 50% of

schooling decisions. Including part-time workers in the analysis, the program

increases enrollment, but reduces part-time workers by only 60% of its effect on

school enrollment (Panel C). The program also does not affect hours worked

significantly. The results confirm that the education price elasticity of school

demand is higher than the cross-price elasticity of child labor. This may reflect

the fact that households continue to require children to work in order to maintain

119

their level of welfare. In summary, child employment as a main activity is

relatively inelastic, as compared to part-time employment when school facilities

are nearer. Small changes in the number of work hours may reflect the fact that

full-time workers dominate the sample.

Including data from the more recent years of 1999 and 2002 in the analysis

provides more robust results for school enrollment. The program causes a steady

improvement in school enrollment an average of about 0.2% per 1 km distance

change. However, including these recent surveys leads to differing results for

children employment decisions, for both full-timers and part-timers. The results

obtained when including 1999 show that shortening the distance to the nearest

school had a smaller effect on child employment, declining by about 0.07% for

full-time workers and 0.13% for part-timers; hours worked were not significantly

affected. The insignificant results were probably due to the financial crisis in

1998, which may not have affected enrollment, but did affect employment.

Children had to continue working to fulfill their families' basic needs. This

confirms results from Priyambada, Suharyadi, and Sumarto (2002), which

indicate that child labor was necessary during the crisis to finance their school

enrollment. By 2002 the program had a significant effect on full-time workers.

This may be a result of the economic recovery in Indonesia, which took place

before 2001, after the financial crisis. Even when students drop work as a main

activity and increase school attendance, they continue to work part-time.

Clustered regression results at the sub-district level, as a check for robustness,

120

confirm that clustering only affects the standard deviations (Table 2.7); the

coefficients are not affected. In general, enrollment increases by about 0.2% per

1 km change in distance and work declines by lesser amount.

Figure 2.4 presents the results for the predicted value of (Ej ) from the estimation

of equation (2.10). The effect of the program, allowing for individual variation, is

shown for all years. If f denotes the dummy representing the treatment group

(aged 11 to 16) and f denotes a dummy indicating whether the observation is

from before or after the program, the difference-in-differences is [F (f=1, f=1) - F

(f=1, f=O)J-[F (f=O, f=1)-F (f=O, f=O)]. Panel A presents the difference-in

differences of the predicted value and Panel B shows the interaction coefficients

of the regression results; that is, the effect of the program per 1 km change in

distance, holding everything else constant.

Child labor and schooling decision are nearly mirror images for non-parametric

difference~in-differences estimation (Panel A of Figure 2.4). Two categories of

employment are presented; full-time working, and full-time and part-time working.

Children either go to school or work. The employment levels vary by age. Two to

8% percent of the treatment group quit work; a higher percentage of the older

age groups also quit work. Effects begin to lessen upon reaching 17 to 18 years

of age. At the same time, school participation increases by almost the same

proportion for all age groups under 17 years.

121

The program can only explain about 30% of the total changes in both child

employment and school enrollment. Difference-in-differences' results of the

predicted value of the schooling decision, calculated from the OLS regression

results of equation (2.10), differs from those of the non-parametric estimations.

The program's effect is smaller. The school enrollment only increases 5% within

the same time frame as shown in Panel A. The child labor decreases by less

than 5%. Panel B graphs the coefficients of interaction. If distance is the only

factor allowed to vary and everything else is held constant, a 1 km change in

distance to the nearest school results in about 0.3% increase in enrollment rates

and a 0.2% decline in child labor for those aged between 14 and 17. These

estimation results confirm the program does not affect child labor and enrollment

equally. Schooling decisions are more elastic, while the employment decision,

either part- or full-time, is less elastic.

122

Table 2.7 OLS Regression Results with Employment, School, and Hours

Worked as the Dependent Variables, Clustered by Sub-district Level: Coefficients

of Interaction

Two-Year Analysis***Three-Year Four-YearAnalysis*** Analysis***

(1) (2) (3) (3) (3)

Panel A: Enroll at School

Coefficients -I.I9E-03 * -2.00E-03 * -2.44E-03 * -3.00E-03 * -2.45E-03 *Standard Deviation -4.64E-04 -4.58E-04 -5.06E-04 -4.7IE-04 -3.44E-04


Panel B: Full Time Working

Coefficients 4.90E-04 ** 5.16E-04 1.24E-03 * 6.49E-04 5.46E-04Standard Deviation -2.90E-04 -4.51E-04 -5.78E-04 -5.46E-04 -4. 12E-04

NumberofObsenation 119,477 84,071 57,626 82,474 126,308R-Square 0.Q7 0.18 0.28 0.28 0.28

Panel C: Full Time and Part Time Working

Coefficients 5.32E-04 ** 7.88E-04 l.55E-03 * l.29E-03 * 1.29E-02 *Standard Deviation -3.56E-04 -4.67E-04 -5.87E-04 -5.47E-04 -4.30E-04

Number ofObsenation 119,477 84,071 57,626 82,474 126,308R-Square 0.06 0.17 0.25 0.24 0.28

Panel D: Number or Bours Worked

Coefficients -5.55E-04 -7.4IE-03 I.I3E-02 9.06E-04 1.29E-02 *Standard Deviation -1.36E-02 -1.98E-02 -2.72E-02 -2.52E-02 -4.30E-04

Number ofObsenation 119,477 84,071 57,626 82,474 126,308R-Square 0.06 0.17 0.27 0.26 0.28

Note: * Indicates significance at the 99% confidence level: ** denotes significance at the 95%confidence level: (1) control group ages 8 - 11 and 16-25: (2) control group ages 16-25: (3)control group ages 20-25: *** two year analysis use 1993 and 1996, three year analysis use

1993, 1996, and 1999, four year analysis use 1993, 1996, 1999, and 2002

123

The response of number of hours worked to changes of school distance is

positively low (Panel A and Panel B of Figure 2.5). However, in the longer term,

children tended to work fewer hours. The hours worked declines by

approximately 0.4 to 1.2 hours per week for those aged 14 to 17 (Panel A), or 0.6

hours in average (Panel B). The treatment group decreased hours worked to

compensate for their enrollment in school. On the other hand, age groups below

14 years old are not significantly affected by the school distance changes. The

degree of variation amongst younger age groups is considerably low, and may

cause small effects of the program to these age groups as shown. The older

siblings in the household, as previously mentioned, may also cause decreases in

the effect on child labor, which may initially be low, of these younger age groups.

124

Panel A Difference-in-Differences of Predicted Value AllYears

0.2

0.15

- 0.1c'0ll. 0.05G)ClCI:l- 0cG)CJ...G)

-0.05ll.

••

.... ....................

-0.1

_a-Enroll at school

--..- Full-Time Working

..• lK' •• Non-parametric Working

'lK ••• ll:"• ". • lK' .' •• "lK' ......

--e-- Full-Time & Part-Time working

.. .•. .. Non-parametric Schooling

0.004

0.003

0.002

- 0.001c'0ll. 0G)ClJ!! -0.001cG)

~ -0.002G)

ll.-0.003

-0.004

-0.005

Panel B: Coefficients of Interaction All Years: per 1 kmDistance Change

. ".•... Emoll at school--e- Full-time Working- Full-Time & part-time working

Figure 2.4 Effect of the Junior High School Distance Changes on School

Enrollment and Employment Decisions by Age for All Years (1993, 1996, 1999,

2002)

125

- all year -e- 93, 96, & 99 -.- 93 & 96

Age

1817161514

Panel A: Difference-in-Difference of Predicted valueNumber of Working Hours

0.4

0.2

0

-0.2

~-0.4

::J0 -0.6I

-0.8

-1

-1.2

-1.4

0.12

0.1

0.08

~ 0.06::J0I

0.04

0.02

0

Panel B: Coefficient of Interactions: per 1 km DistanceChange

-all year

-e-- 93, 96, &99

-.-93 &96

11 12 13 14 15 16 17 18 19

Figure 2.5 Effect of the Junior High School Distance Changes on Number of

Hours Worked by Age

126

The program's effect on child employment decisions is less than its effects on

schooling decisions (Panel A and Panel B of Figure 2.6). Results also indicate

that the program influences the actions of boys more than those of girls. Boys

decrease their employment decision by about 3% to 6% (Panel A), depending on

their age group. Girls reduce their employment decision by only 2% to 3%. The

results confirm that most households still favor boys in enrollment decisions.

Parents continue to believe that it is unnecessary for girls to earn higher degrees.

Once girls graduate from elementary schools, it is thought that they are better off

working and contributing to family's business or resources. This gender disparity

does not only occur in the agriculture families, but also within families who have

their own business. Therefore, the results show a persistence of child labor

amongst girls since the majority of households either work in agriculture or open

their own businesses.

In addition to the gender disparity, the compulsory program may affect children in

rural and urban areas differently. Decreased distance to school decreases child

labor in rural areas more than that in the urban areas (Panel C and Panel D of

Figure 2.6). Decisions to work in rural areas decline by about 2% to 8%, and by

about 1% to 3% in urban areas. School demand in the rural areas is also higher

by about 2%, relative to urban areas. In general, the construction of junior high

schools affects school demand by as much as 2% to 6%, depending on the age

of child. The program benefits those in rural areas and boys, more than it

benefits those in urban areas and girls.

127

Children can benefit from the program if they have older siblings. Figure 2.7

presents how siblings affect school enrollment (Panel A and C) and child labor

(Panel B and D). The figure distinguishes children who are youngest or oldest in

the family or have younger or older siblings from those who do not have siblings.

The profiles show that there is no benefit in being the youngest child. Their

school enrollment after the program is similar to those who are the oldest child or

without siblings (Panel A). Similar results occur for the child labor decision. There

is no difference between the three categories of children. However, these figures

are limited to distinguishing those who are youngest or oldest, while other

members are not included.

Panel C and D show results of estimates using samples of all children who have

younger or older siblings. The oldest children are included as those who have

younger siblings and vice versa for youngest children. There is a benefit for

children in terms of their school enrollment or child labor if they have older

siblings. If they have younger siblings, their school enrollment is slightly lower

and their level of child labor is higher after the program.

128

Panel A: DID Predicted Value

"

Goo -, oo -E].,. .... _ -' - - -;j(, -, _, _ " _. ,)I(', .. , .[3, ':i!i. --

" .

0.08

0.06

0.04-c'0 0.02Q.Q)0) 0C'a-C~ -0.02Q)

Q.-0.04

-0.06

1i?- . ,.,-,118, 14 15

1::1'

16 17

Age

-0.UM--8-- Enroll at School tv1ale )I( Enroll at School Female

- - -E]- - - Part-time &FUll-time Working tv1ale - - -)I(- - - Part-time &Full-time Working Female

Age

19

.." .

1716151413

Panel B: DID Coefficients of Interaction:

per 1 km Distance Change

a.8.. .)1(.. ",." \

#)1(- . ..... .:'.... .. .. '..1....... .. .... ........ ' ..

,-:',0" "-;j(' '0'" ')1("

CI ,,':' ':JII::::::::~;

12

O+------,-------,-------,-------,--------,----r---cf---+-,--------,

0.004

0.001

0.002

0.003

-0.002

-0.003

-c'0Q.Q)

JcQ) -0.001~Q)Q.

-0 OOfu Enroll at School tv1ale

.. ·EJ- - - Part-time &Full-time Working tv1ale

)I( Enroll at School Female

- - -)I(- •• Part-time & Full-time Working Female

Figure 2.6 DID of School and Work Decisions Using 1993,1996,1999, and

2002 Survey Data: Effect of the Program by Age Comparing Boys vs. Girls and

Urban vs. Rural

129

Panel C: DID Predicted Value

--Enroll at school urban o Enroll at School rural

Age

17

' ..• )1('

')1(- - - - •••)1( •.

15 16" .....

, -'-'"12 13 14I::··:::t·· ........ ~-_.,

')1(.

0.08

0.06

0.04-c'0 0.02D.Q)en 0BcQ)

-0.02(JI-Q)

D.-0.04

-0.06

-0.08

........ Part-time &Full-time Working urban - - -)I(' •• Part-time &Full-time Working Rural

0.008

Panel 0: DID Coefficient of Interaction :

per 1 km. Distance Change

0.006

0.004

, -)I(.., • _ >0< •• - - •• )1(. - • • • •-.......... • - ')1( •

Age)1(' - , • - • ~ •

12 13

, .....,.-""-"- '"'-'''..... _--.-,' -..,_.,--

°l--==::r:=::::=---,---------,--------,----~--y~~/------,

0.002

-0.004

-0.006

-0.008

coD.Q)

Ec~ -0.002Q)

D.

--Enroll at school urban 8 Enroll at school rural

- - -.- - - Part-time & Full-time Working Urban - - -)1(- - - Part-time & Full-time Working Rural

Figure 2.6 (Continued) DID of School and Work Decisions Using 1993, 1996,

1999, and 2002 Survey Data: Effect of the Program by Age Comparing Boys vs.

Girls and Urban vs. Rural

130

Panel A: DID Predicted Value of Scholl Enrollment Siblings'Comparison0.1

0.08

.... 0.06c0

Q. 0.04CD

i....0.02c

CD~CD

0Q.

-0.02

-0.04 -without sibling ___ youngest ~oldest

Age

18

~oldest

171615

___ youngest

14

-without siblings

Panel B: DID Predicted Value of Working Activity Siblings'Comparison0.02

0.01

.... 0c'0

-0.01Q.CDC)

S -0.02cCDu...

-0.03CDQ.

-0.04

-0.05

Figure 2.7 DID of School and Work Decisions Using 1993, 1996, 1999, and

2002 Survey Data: Effect of the Program by Age Siblings' Effect on Household

Decisions

131

Panel C: DID Predicted Value of Scholl EnrollmentSiblings' Availability

0.08

0.07

0.06

t: 0.05

~ 0.04CI) 0.03Cl

~ 0.02

~ 0.01~ O+------r"~'----r--,------,------_r-____,_--,___-_,_-~\r_______,

-0.01

-0.02

-0.03

-without sibling --e- has younger siblings ~ has older siblings

0.03

0.02

0.01-l:0 0D.CI)Cl -0.01III-l:CI) -0.02u...CI)D.

-0.03

-0.04

-0.05

Panel D: DID Predicted Value of Working ActivitySiblings' Availability

Age

15

--without siblings --e- has younger siblings~ has older siblings

Figure 2.7 (continued) DID of School and Work Decisions Using 1993, 1996,

1999, and 2002 Survey Data: Effect of the Program by Age Siblings' Effect on

Household Decisions

132

5.2 The Effect of Education Policies on Parental Labor Decisions

Changes in distance to school significantly affect the labor supply of children.

Enrolling children in school also increases the educational transfers from parents

to children. Parents then must find a way by which to compensate for the decline

in household income, resulting from the loss of child labor and an increase in

household expenditures from costs of schooling. This section examines how

parents' labor supply is affected by the program, using results from the non

parametric difference-in-differences, and then estimation using Equation (2.10).

Households are divided into treatment and control groups, where the groups are

differentiated based on whether they are affected by the education policy. The

treatment group encompasses households who have junior high school age

children, while the control group includes those without children in that age

group. Table 2.8 exhibits non-parametric difference-in-differences on the effect

of the change in distance on employment decisions, hours worked, and labor

income for both household heads and their spouses. In general, household

heads experience a significantly larger change than their spouses do. Labor

supply increases by 2% among household heads; their spouses increase their

decisions to work by a tenth of that, only by 0.2%. Both parents' labor income

increases, and the household heads' labor income increases by more than six

times the amount of their spouse's. These overall trends imply that households

compensate for higher education expenditures and foregone income with

133

increased employment by both parents; although spouses work more on a part

time basis than household heads.

Household heads increase work levels slightly, but the labor supply of their

spouses remains relatively stable. A possible explanation for the inelasticity of

mother's labor supply is the lack of employment opportunities for women. It is

easier for a household head to work more, than for a mother to find a new job.

Therefore, mother's labor force participation is less responsive to the policy. It is

also possible that women's or mother's labor force participation is not a substitute

for child labor. Thus, once the child quits a job, it is the head or adult male who

are able to substitute for the lost labor and income. In addition, mothers'

education levels are relatively lower than fathers', making it relatively more

difficult for mothers to find jobs.

Distance to school significantly affects the parents' labor supply. The distance

particularly affects household heads; their employment decisions, working hours,

and labor income changes are associated with changes in school distance.

However, spouses' labor supply is not significantly influenced. Table 2.9

presents the regression results for Equation (2.10) using the treatment and

control groups previously described. Four panels are shown: work statuses,

working at least 1 hour, work hours, and labor income. The above findings are

consistent with the results shown in Table 2.8. The household head's labor

supply decisions are relatively more elastic than their spouses' labor supply

decisions.

134

Table 2.8 Non-parametric DID: Effect of a Change in Distance on Parental Labor

Supply, 1993 to 1996

Household head wife Household head

Full-time Working

before after DID before after DIDControl Group' 0.291 0.299 0.867 0.829Treatment Group 0.356 0.366 0.898 0.880

0.064 0.067 0.002 0.031 0.052 0.020

Part-time and Full-time Working


0.082 0.095 0.013 0.030 0.049 0.019

Number of WorkingHours


3.337 3.247 -0.090 1.592 2.358 0.766

Labor Income

before after DID before after DIDControl Group' 43,883.32 52,325.43 178,079.20 447,227.60Treatment Group 70,121.37 87,002.36 230,904.40 548,116.80

26,238.05 34,676.93 8,438.88 52,825.20 100,889.20 48,064.00

Note: * Indicates control group =household without members at Junior high school age; treatmentgroup includes households with members of junior high school age: monetary value of laborincome is in Rupiah (1USD =2,500,00,· Rupiah)

135

Table 2.9 Coefficients of Interaction Difference-in-Differences: the Effect of

Education Policy on Parental Labor Supply

Distance***

Head Wife

Number of Observation121,164 106,651

Panel A: Full-time Working

Coefficient -7.58£-04 * 3.02£-05Standard Deviations 3.11£-04 4.96£-03

R-Square 0.02 0.01

Panel B: Number of Hours Worked

Coefficient -0.04 * 0.02Standard Deviations 0.02 0.02

R-Square 0.04 0.01

Panel C: Labor Income ****

Coefficient -1510.48 * -452.45 **Standard Deviations (294.60) (177.18)

R-Square 0.275 0.024

Notes: * Indicates significance at the 99% confidence level: ** denotes significance at the 95%confidence level: *** the coefficients of distance is per 1 km change of distance: **** Labor

income is per month in Indonesian Rupiah

136

A 1 km decrease in the distance to the nearest school increases the employment

status of household heads by about 0.08%. The change in employment status for

an average 6 km change is 0.5% for household heads, which is quite low. The

effect is almost negligible for spouses. The hours worked of household heads

change by approximately 0.85 hours per week, while spouses' work hours

increase by 0.5 hours per week, though the change is not significantly different

from zero. The program positively affects labor income. The closer the distance

to the nearest junior high school, the higher the labor income of both parents is.

Household heads' labor income is affected significantly at a 99% confidence

level. Labor income for household heads increase by approximately USD 3 per

month for 6 kilometer changes in school distance.

6. Conclusions

The household head's educational level is the most important factor in

determining children's school enrollment. Child labor and school enrollment rates

share similar determinants with opposite implications. The education of spouses

also plays a significant role, but to a lesser degree. If the household head is a

woman, children in the household have a higher probability of going to school

rather than participating in the labor market. Most child labor occurs in agriculture

and family businesses. The distance to school is an essential factor in household

decisions to send children to school.

137

School enrollment is more elastic than child labor, in regards to changes in the

school distance. School enrollment increases by 2% to 4% for every 1 km school

distance changes, while child labor is reduced by less. In the long run, the

change in distance explains an increase in the enrollment rate of 5% to 6%, but

has little effect on child labor supply. By using the school distance as a proxy for

education subsidies, the difference-in-differences method explains 30% of the

total increase in enrollment rates and declines in child labor.

Educational subsidies, estimated by the distance to school, affects boys

enrollment more positively than it does girls. The gender bias of the program can

be explained in several ways. First, household operations remain gender biased.

They depend heavily on girls, and the belief that it is not necessary for girls to

obtain higher education persists. Thus, boys are sent to school, while girls

continue to contribute to the household's resources. In particular, there is strong

and vast evidence that child labor mainly occurs in the agriculture industry and

family businesses. There exists extensive literature in which this possible

explanation is discussed. Second, school construction is not based on gender

variation. They do not build schools based on where enrolled girls are more

concentrated or in the agricultural areas.

Parents have to work more to compensate for the cost of sending their child to

school and the simultaneous reduction in child labor income. The education

138

subsidy increases labor participation of household heads, but does not increase

that of their spouses. Spouses may find it difficult to find jobs because of their

education levels, lack of employment opportunities, and constraints imposed by

the necessity of performing other housework. Spouses are constrained by

household activities, making them unable to perform additional market oriented

labor. In addition to this, women have fewer opportunities in the job market due

to their skills and education.

The results show that child labor is relatively insensitive to the education subsidy.

This may be associated with the high opportunity cost of child labor for some

families. In addition, the education cost borne by the families remains too

burdensome. The household head has to compensate for both costs by

increasing his labor supply. However, if there is no opportunity in the labor

market, or the reduction of child labor cannot be substituted for with increases in

the adults' labor supply, the household may be unable to compensate. Therefore,

removing the child from the labor market may reduce the household welfare.

Child labor income may be required to maintain the child's enrollment status.

Hence, the education subsidy increases school enrollment, and reduces child

labor supply, but by a smaller percentage.

139

ESSAY 3: THE EFFECT OF EDUCATION POLICY ONINTERGENERATIONAL TRANSFERS

140

1. Motivations and Objectives

The Indonesian government launched a six-year basic education program in

1984. The program focused on eradicating illiteracy and improving the quality of

human capital through the construction of many elementary schools, particularly

in rural areas. Making education compulsory led to a significantly higher

enrollment rate at the elementary level, which rose to 99% by the 1989/1990

school years. However, by 1993 more than 30% of elementary school graduates

did not continue their education: 13% of junior high school age children were on

the job market, 6% were working at home, and 13% were jobless (Susenas

1993). In light of the success of the six-year program and in order to fulfill the

demand for a more highly educated labor force, the government imposed a nine

year compulsory education program starting in 1994. The Indonesian

government constructed over 1,000 new junior high school buildings, resulting in

over 5,000 buildings and classrooms during fiscal years 1994/1995 and

1995/1996 in total, to extend its compulsory basic education program.

While the longer program may not affect parents with higher income or

education, it places financial constraints on parents with lower incomes or

education. Those in rural areas are often faced with liquidity constraints.

Sending their children to school instead of working can result in a significant

decrease in total family income. Parents who work in agriculture also need their

children to work in the fields. Furthermore, parents who themselves have had a

141

relatively low education may consider additional education for their children an

unnecessary investment. These parents believe children aged twelve to fifteen

are ready to assist in supporting their families by working. Children contribute to

smooth out the family consumption pattern. Extension of the schooling period

may lead to a shock in the consumption pattern.

This paper examines how the introduction of nine-year compulsory education has

influenced familial educational investment decisions and non-educational

transfers in Indonesia. Sending children to school involves opportunity costs that

are higher than the cost of the education itself. Compelling parents to send their

children to nine years of schooling may lead to a failure in allocating and

transferring sufficient funds into other expenditures such as health or nutrition.

Alternatively, parents may be forced to increase non-educational expenditures

that complement the longer period of education. For example, providing

sufficient food is an essential part of children's development. Household

resources tend to be highly allocated to food in most developing countries,

including Indonesia. In 1993 and 1996, the average share of income spent on

food in Indonesian households was higher than fifty percent. Therefore,

allocating the household budget to non-educational expenditures is as important

as education itself.

This paper contributes to the understanding of how educational and non

educational transfers at the individual level respond to government policy, a topic

142

few authors have previously addressed. There exists limited literature on the

effect of compulsory education on enrollment rates or earnings, with special

reference to the United States (Angrist and Krueger 1991; Acemoglu and Angrist

1999; Goldin 2003; L1eras-Muney 2002) and Taiwan (Spohr 2003). Papers

exploring the relationships between intergenerational transfer decisions and

policies such as compulsory education are unavailable. This paper therefore fills

a gap in the research on educational policy and family decision-making in

allocating resources to children, using constructed individual transfer data. This

paper also applies the pooled-cross-section difference-in-differences

methodology where intensity of the program is taken into account.

The following analysis relies on estimated individual education transfers dataXV

derived from the Indonesian Socio Economic Survey (Susenas) for the years

1993 and 1996. Susenas is an annual national representative survey that covers

approximately 65,000 households or 250,000 individuals. Susenas includes

detailed information on food and non-food expenditures and income sources.

This data facilitates estimation of intra-household allocation or transfers to

childrenxvi• The rapid increase in school building construction during the

expansion of the education program in 1994 and 1995 can also be estimated

based on Susenas data for 1992 and 1995. To obtain individual earnings data, I

impute Labor Force Survey (Sakernas) data for the same years onto Susenas

data.

143

This analysis uses the difference-in-differences method to identify changes in

non-educational and educational transfers among the affected groups due to the

installation of compulsory education by the government. The enrollment rate

increased by about 4% two years after the program was introduced. I find that

building construction during fiscal years 1993/1994 and 1995/1996 mainly

affected education transfers of the higher quartile families. By taking the

logarithm of education transfers, I find the transfers increase about 10% to 20%. I

also find that non-educational transfers are less sensitive. These transfers only

increase as much as 5% after the program's introduction by the government. The

increase of non-educational transfers is due to both a decline in income from

children's labor and an increase of non-education consumption.

The average educational level of household heads is slightly above 6 years. I

strongly suspect that the introduction of the nine-year compulsory education

program still binds most parents due to their low educational background.

Families in the low-income class as well as low educational backgrounds might

find that the policy suppresses family income significantly. In order to establish

an efficient educational policy, cash transfers or family subsidies may be

necessary to compensate for the income forgone by sending their children to

schoolxvii as well as higher children expenditures due to commodities'

complementary nature.

144

In the following section, I review the literature on compulsory education,

governmental policy-making, and family decisions regarding the allocation of

resources. Section III covers the conceptual framework that will form the basis of

my empirical analysis. Section IV describes the data. Finally, sections V and VI

discuss the empirical results and draws final conclusions.


Compulsory education raises several issues despite its role in enhancing

enrollment rates and a more even distribution of educational levels in a

population. First, if parents decide to send their children to school, for any

additional investment in education, they have to bear opportunity costs of having

their children away at school instead of working and the out-of-pocket expenses

of supporting the educational activities. This implies that families are forced to

transfer a higher proportion of their wealth to their children's education and

acquire other sources of income to compensate for foregone earnings. Parents

can increase their income by working more hours to replace their children's lost

hours. Parents can also reduce their savings and the amount of bequests to their

children. With a perfect credit market, parents can borrow money to invest in

education. Alternatively, parents can decrease their total expenditures, producing

a trade-off between educational and other expenditures including non

educational transfers to their children.

145

The second issue is child labor. Investing in education and relying on child labor

are two faces of the same coin, in that they depend on similar factors. The levels

of parental education and income are major determinants of both private

education investment and child labor. Haveman and Wolfe and Behrman and

Knowles find that household income and schooling are strongly related. Among

poor families, parents expect children to contribute to the family's total income

and smooth out family consumption.

Fitzsimons (2002) investigates the effect of risk on education and child labor in

Indonesia using the 1993 Indonesian Family Life Survey (IFLS). She finds that

children in rural areas play an important role in smoothing family consumption

when their parents lack insurance and/or access to credit markets. Uncertainty

hits the ten to fourteen year age groups harder and they are more likely to have

accumulated fewer years of education. Fitzsimons demonstrates that some

parents in Indonesia resist sending their children to school and do not consider

education a priority. However, the analysis did not consider the impact of policy

reform on the rigidity of response to schooling, as it took place a year before the

nine-year compulsory education program was promulgated.

Keane and Wolpin investigate the effect of borrowing constraints and parental

transfers on educational attainment, specifically the effect of the unavailability of

collateral assets on children's school enrollment. Using the 1979 youth cohort of

the National Longitudinal Surveys of Labor Market Experience (NLSY), they

146

construct an optimization model of young men's schooling, working, and savings

decisions. More highly educated parents make substantially higher transfers

while their children are attending college. On the other hand, lower educated

parents only transfer small amounts. If borrowing constraints are relaxed among

the youth, the enrollment decision is not affected regardless of parental

educational level. Carneiro and Heckman find a strong relationship between

family income and college enrollment using NLSY data. They distinguish

between short term and long term borrowing constraints and conclude that long

term borrowing constraints are a major determinant of family income and

enrollment.

The relationship between compulsory education enforcement and school

enrollment and earnings has been given considerable attention . Most

investigations consider compulsory education policy in the United States and its

ability to explain higher enrollment rates in secondary school. They argue that

compulsory education and child labor laws together explain higher enrollment

rates (Acemoglu and Agrist 1999: Angrist and Krueger 1990: L1eras-Muney

2001). Oreopoulos, Philip, and Stevens (2003) also show that compulsory

education enforced on an earlier generation eventually improves the educational

attainment of the next generation.

Spohr (2003) investigates the effect of compulsory junior high school to years of

education and workforce participation in Taiwan. He finds significantly longer

years of education for both males and females in Taiwan following educational

147

reform. He also shows a stronger work participation rate among females as a

result of tuition-free education at the junior high school level. Duflo investigates

the large construction of primary school buildings in Indonesia after the

Indonesian government promulgated six-year compulsory education. She finds

that the projects significantly increase years of education and earnings amongst

the affected groups.

Education has several important roles, as investment, as consumption (Schultz

1960), and as part of intergenerational transfers. The intergenerational effect of

education is significant. Governmental intervention is recognized as being

essential to shifting intergenerational mobility. The government's role in

redistributing wealth from more to less educated parents supports higher

educational achievement in the next generation.

The effect of governmentally mandated compulsory education has mixed results

when it comes to poor parents. Chevalier differentiates between parental income

and the unobserved characteristics of poor parents that prevent them from

spending on their children's education. Parents under pure liquidity constraints or

facing other unfavorable characteristics are more efficient targets for

governmental interventions such as the education maintenance allowance (EMA)

implemented in the United Kingdom. If unobserved characteristics are dominant,

they may hamper government policy so it produces no efficient results. Poor

parents therefore tend to keep educational investment at a subsistence level.

148

3. Conceptual Framework

This section covers the conceptual basis for my empirical analysis, beginning

with Becker and Tomes's theoretical modeling of parental decisions regarding

transfers to their children. This model spells out how parents allocate funds for

educational and non-educational expenses for children, to accommodate

changes in educational unit prices. The model can be used to examine the effect

of government subsidies and compulsory programs. Parents treat expenditures

on children as both consumption and investment. The unit of analysis is at the

household. Parents usually provide all educational expenses for their children ,

but only some of the non-educational expenditures.

Parents' utility is a function of their own consumption (cp), their leisure (Lp), their

children's non-education (Ck), and education consumption (q). The utility function

Uh = U(cP,Lp,ck,q) is quasi-concave for every argument, continuous, and twice

differentiable. The first derivative is positive for each argument. For simplicity, the

number of children per parent is assumed to be only one.

In addition to gaining satisfaction from a child's consumption, parents also look at

returns. Consumption is accommodated by child's labor income (Yk)

endogenously contributing to their parents' budget, following the earnings

function ~ =Ck-rqr &. That is, parents perceive children's education and non

education expenditures as both investment and consumption. Where 0 ?: r?: 1

149

and & is child initial skill endowment. The child's non-educational consumption

(Ck) is defined as consumption excluding educational expenses, with price per

unit denoted as Pk• Pk is therefore child's non-educational consumption unit cost.

Parents spend their income on part of children's non-educational consumption

through inter-vivos transfers. Non-educational transfers, T, are the total of a

child's non-educational consumption less his labor income (Yk ). Thus, we can

define non-educational transfers as T = Ck~ - ~ .

If parents work as much as (1-Lp) with wage wp , the budget constraint therefore

is as follows:

3.1

Educational unit cost is defined as Pe, which is the required unit cost to achieve a

certain level of education. Parents' own consumption unit price (Pp) can be

defined as adult unit cost. Government influences parents' decisions through two

channels. First, by subsidizing education, as:

3.2

Parents pay Pe for their children's education and the government pays Pg of the

education market price, Pm. The government can also intervene by regulating the

minimum level of education investment level (qb)' In the Indonesian case,

compulsory education requires parents to send their children to school for nine

years, to the junior high school level. I therefore assume that the government

150

subsidy (Pg) and minimum human capital level (qb) are exogenously given, as

follows:

3.3

Maximizing the household utility function Uh subject to the defined earning

function, budget constraint (3.1), and compulsory education constraint (3.3) form

the Langrage Function as follows:

L =Uh(cP,Lp,ck,q) -~ (cppp+ CkPk + qPe-(1-Lp)wp - Yk)

+~ (q - qb) - A:3 (Ck-rqr - Yk)

First order conditions are:

LC =Uc -~Pp =0p p

LLp = ULp -~wp = 0

LCk =UCk -~Pk-~(1-r)Ckrqh6'=O

Lq =Uq -~Pe +~ -~rcl-rqrl6' =0

LAl = -cpPp-CkPk -qhPe +(l-Lp)wp+ Yk = 0

LA2 =q-qb ~o

L - l-r r v - 0A3 - -ck q 6' + r k - .

From the first order condition, I can obtain:

3.4

3.5

,,1,2 is a Kuhn-Tucker multiplier. It follows the complementary slackness condition

that constrains it to being greater or equal to zero. If ,,1,2 is equal to zero,

compulsory education is not binding. Given the case that government regulations

151

are not binding, equation (3.5) means that parents will invest in education when

the net marginal utility and marginal returns of education investment are equal to

marginal utility gained by increasing their own consumption or their leisure.

The implicit Marshallian demand functions are as follows:

C~ =cp(pp,pk,Pe,wp)

c; =ck(pp,pk,Pe,wp)

q* =q (pp,pk,Pe,wp)

A; =~ (pp,pk,Pe,wp)

;.; =~ (pp,pk,Pe,wp)

A; =~ (pp,pk,Pe,wp)

By assuming the utility function Uh is perfectly separable in every component and

knowing that (dPe = -dPg ), the effect of government subsidy xviii Pg on education

transfers, an interior solution exists and parents invest optimally in their children's

education at a higher level than the regulated level (q > Clb ) or ..1,2 is equal to zero

for non-constrained parents. The education subsidy results in a positive changing

of education transfers (dqjdPg > 0). Non-education transfers are also positively

affected by the government subsidy (dTjdPg > 0). This means that non-

education transfers are complementary with education transfers.

When the government regulation is binding and ..1,2 is not equal to zero, parents

have to sacrifice more on the utility gained by increasing their leisure or own

consumption to meet the higher children education investment. Corner solutions

152

might exist. By regulating the minimum level of children education, the

government policy creates a shadow price Az that is larger than zero. Parents

invest in their children's education just as much as q =qb' That is, parents have

to invest more than they desire. dqjdPg and dTjdPg are equal to zero assuming

that parents perceive education and non-education transfers as investment. In

addition, if constrained parents perceive children's expenditure as consumption

alone (A3 =0), dqjdPg is equal to zero, while dTjdPg is positive. A complete

comparative static result is presented in the appendix.

Empirical results indicate whether parents are categorized as constrained or non

constrained towards education policy. If constrained, parents will not invest

optimally for their children's education. On the other hand, if non-constrained, the

family spends in an efficient way towards their children's education. If families are

bound by government policy, the government subsidy has zero effect on both

educational unit costs and parental income to children's expenditures, provided

parents perceive children's expenditures as both investment and consumption.

These parents are bound because they have insufficient resources to send their

children to school. If the family's resources depend partially on children

contributions, sending them to schools means losing some resources that may

be replaced by working more, inter-household transfers, government cash

transfers or any other subsidy that minimizes the distortion.

153

4. Data and Empirical Results

4.1 Data Sources

The Indonesian Socio-Economic Survey (Susenas) is a national representative

survey that includes almost 65,000 households or 225,000 individuals. Susenas

contains individual characteristics of household members, such as gender,

relation to household head, members' school enrollment status, the highest level

of school being attended or completed, and household head characteristics. Brief

household expenditures and income data are also included. This Susenas is

called the core-Susenas since it collects the main characteristics of households

and individuals in the households.

To complement the core-Susenas, Susenas conducts a comprehensive survey of

more specific topics such as health, education, expenditure, income, and tourism

every third year. The three-year cycle surveys are called modules. One of the

Susenas modules includes a detailed household expenditure and income survey.

For my analysis I use the module-Susenas 1993 and 1996. These surveys years

cover complete household food and non-food expenditures. The surveys also

included comprehensive data on familial education expenses. All expenditure

data are recorded at the household level. Combining module-Susenas and core

Susenas enables me to estimate individual consumption allocations, which is

essential for my analysisxix•

154

In addition to Susenas 1993 and 1996, I also use the module-Susenas 1992 and

1995. These module-Susenas consist of detailed education surveys, including

distance to the nearest schoolsxx, time required to reach the nearest schools, an

individual's education expenditures, and principal agents who support the

education expenses. I impute average distance to the nearest particular school

level at the sub-district level to the main database from module-Susenas 1993

and 1996.

Estimation of parents' inter-vivos transfers to their children requires labor income

data for individuals. module-Susenas 1993 and 1996 has comprehensive

household non-labor income data but poor individual earnings data. To obtain

individual earnings, I impute the average earnings of individual per hour at sub

district level from the Indonesian Labor Force Survey (Sakemas). Sakemas is an

annual national conducted survey that contains working age individual's earnings

data. I use the average of earnings at the sub-district level categorized by male

or female household head and non-head member. Labor income for children is

defined as earnings plus the individual non-labor incomexxi•

Table 3.1 presents the descriptive statistics of nominal annual household

education expenditures, individuals, household heads, and sub-district level

characteristics in 1993 and 1996. Panel A indicates itemized annual household

education expenditures using Rupiah currency. Tuition fees take up the biggest

portion of household education expenditures. Households also spend a

155

considerable amount on enrollment fees and books. Panel B summarizes

childrenxxii characteristics, including their labor income, consumption, education

expensesXXiii, and transfers with and without education. Education was only

approximately 6% - 7% of the total individual consumption. On the other hand,

the average of children's labor income was around 40% of their total

consumption in 1993 and 32% in 1996.

Panel C in the same table shows household head characteristics. Their average

age was 45 in 1996 and 44 in 1993. Average years of education were only 6

years and there was not an important change between the two years of

observation. Labor income and household expenditures increase slightly. The

sub-district characteristics are described in Panel D. Included are distances to

the nearest school for three categories of school, average household head labor

income, and enrollment rates at the primary school level. Distance to the nearest

school declines considerably for all school levels. The average household head

income also experiences a significant increase.

156

Table 3.1 Descriptive Statistics

(all monetary units are per year in Rupiah (1 USD =Rp.2,500 with 1996exchange rates) )

Panel A Annual Household Education Expenditures per Item (for those with expenditures)Number of Observation 32,457 (1993) 35,614 (1996)

1993 1996

Variable Mean Std. Deviation Mean Std. Deviation

Tuition 72,734.81 227,339.40 108,466.70 320,606.50

Course Fee 5,042.09 40,308.42 5,620.25 56,656.85

Enrollment Fee 13,831.22 101,666.90 27,010.82 168,157.50

Other Fee 7,219.70 37,086.45 13,702.71 69,571.76

Books 13,853.94 30,587.02 21,404.33 54,342.73

Stationary 9,609.01 17,520.80 10,693.00 31,712.47

Total Education 122,290.80 304,214.40 199,326.30 492,106.90

Panel B Consumption Allocation of the Children and Intrahousehold Transfers to the ChildrenNumber of Observation 128,902 (1993) 124,098 (1996)

Age 13.10 5.34 13.22 5.37

Male 52.96% 49.91% 52.76% 49.92%

School Enrollment 58.01% 49.36% 60.65% 48.85%

Labor Income 217,170.48 767,373.84 259,393.68 952,418.28Total Consumption 484,569.12 449,278.20 818,384.16 793,312.80Education Transfers Received 33,005.88 147,196.92 59,910.40 197,632.44Consumption without education 451,563.24 388,717.08 758,473.68 702,034.44Intrahousehold Transfers Received 267,398.64 891,247.92 558,990.36 1,243,597.20Intrahousehold transfers received without education 451,563.24 388,717.08 758,473.68 702,034.44

Panel C Head or Household CharacteristicsNumber of Observation 59,593 (1993) 60584 (1996)

Age 44.26 13.78 45.06 13.82

Male 87.77% 32.76% 87.72% 32.82%

Years of Education 6.00 4.26 6.05 4.20Total Expenditures 2,421,154.80 3,296,041.20 3,442,188.00 3,709,270.80Labor Income 2,031,472.80 2,992,058.40 2,713,821.60 4,068,073.20

Panel 0 Sub-district Level CharacteristicsNumber of Observation 1,774 (1993) 2,361 (1996)

Distance to the nearest primary school (km) 6.83 7.47 0.84 0.38Distance to the nearest junior high school (km) 8.70 9.43 2.55 1.41Distance to the nearest senior high school (km) 10.38 13.53 5.46 5.52Enrollment Rate at primary school level 34.04% 7.57% 33.50% 8.12%

157

4.2 Empirical Analysis

An empirical approach to examining the relationship between educational

subsidy and non-educational and educational transfers follows from the fact that

the two types of transfers are functions of education price Pe, non-education

price Pk, parental consumption unit price Pp , and parents' wage wp. The

government constructed many junior high school buildings and classrooms to

accommodate increased enrollment following its mandated nine-year compulsory

education program.

The increased number of schools in the neighborhood decreases the opportunity

costs of parents sending their children to school. I approximate education unit

cost by distance to the nearest junior high school. The distance is also a

proximate for government program intensity in the sub-district level and as

sources of variation. I assume that children and parents choose junior high

school distance as a major determinant. Instead of distance to the nearest junior

high school, the number of buildings constructed is actually a better measure of

sources of variation. However, these data are not available. Hence, I use the

distance variable as the best available measure and as the source of variation.

As described in Table 3.1 previously, the average distance changed considerably

from 1993 to 1996.

Families can move to a region with more schools and lead to bias in estimation.

Duflo (2001) finds that "91.5 percent of the children in the IFLS sample, were still

158

living in the district they were born in at age 12 (p.8)". She discusses briefly that

families might move to benefit the program and suggest the region of school as

instruments. However, there are no data available. I assume the immigration rate

(9.5%) is not large enough to lead to estimation bias.

I approximate the price of education with the following relationship:

3.6

Education unit cost, Pe,ji , is a function of the distance of a child i from his school j

(dji) and quality of the school j (kji) that the child i goes to. If the quality of school

(kj) is a function of observable criteria such as average household income at the

sub-district I, the quality of school can be expressed as kj = f.il + f.i2~/' ~I is

average income of household in the sub-district I. If education unit cost and

transfers are linear in price Peji and other prices, taking equation (3.6) and a

proxy for education quality provide a relationship between education and non

education transfers to the distance and quality of schooling at the individual level.

I drop sub-district and school subscripts to simplify notation. Further, I assume

that children's non-education expenditure price and parents' own consumption

price are constant.

T; = Yo + Y1dj + Y2~J + Y3 X + rp

q = Xo+ X1dj + X2YdJ + X3 X +~

159

3.7

Non-educational transfers for child i Ti is defined as in equation (3.2), which is the

difference between total food and non-food consumption excluding education

expenses, Ck,i, less labor income of the member, Yk,i. Oi is observed total

education expenditures as qi * Pe. Household characteristics used as control

variables are denoted by X.

I use the difference-in-differences method to estimate the effect of the education

subsidy through construction of school buildings. I start from equation (3.7) and

estimate the effect of a change in distance to a school on both educational and

non-educational transfers to children, with children divided into treatment and

control groups. The treatment group is considered affected by the compulsory

education policy, while the control group is not influenced by the policy. While the

focus group is between 12 and 15 years of age, age groups 11 and 12 could

overlap in that they may be enrolled either in junior high school or elementary

school. Children between 15 and 18 years old could be either in junior or senior

high school. I presume that the treatment age may be extended from 12 - 15 to

11 - 16 years of age. I use the comparison groups of 8 - 10 and 17 - 25 years of

age.

The difference-in-differences eliminate the variation of individuals, time, and

regions. The effects of the program on the non-education transfers and education

transfers to the old cohort are assumed to be negligible. I neglect the cohort

effects. Taking the difference between the treatment group and control group,

160

before and after, the program's effect can be captured by running the following

regressionsxxiv;

~t = 'I/o + 'l/1dit + 'l/2ditAit + 'l/3Ait + 'l/i + 'l/sX + Vi

0it = (Po + rP1dit + rP2ditAit + rP3Ait + rPi + rPs X + Si •

3.8

Tit and Oit denote the transfers of non-education or education transfers to child i,

sub-district I, and year t (1993 and 1996). The effect of building construction on

the education and non-education transfers is captured by the coefficient '1/2 and

rP2' which are difference-in-differences between treatment and control groups

before and after the program. These coefficients represent the interaction

between distance to the nearest junior high school, dit, and treatment group

dummy Ait. Treatment group Ait is age dummy variable and is one if individual i is

between 11 and 16 years of age, and zero otherwise. Table 3.2 describes how to

interpret the equation (3.8). The program affects the non-education transfers as

much as '1/2 per 1 km distance change. And, the program affects education

transfers as much as rh respectively per 1 km distance change.

Table 3.2 Interpretation of DID model

Control Treatment Difference

Before rPo + ¢J.do ¢o + ¢Jdo + rhdo + th rP2 do+ th

After rPo + ¢J.d1+ rPs ¢o + ¢Jd1 + rhd1 + th + ¢5 rhd1+th

Difference ¢J.d1-¢1do+rPs (¢Jd1 - ¢Jdo) + (rhd1 - rhdo) + ¢5 rh(d1-do)

161

Expanding the treatment age dummy, Ait' to age group dummy variables from 8

to 25 allows me to capture more variation and detailed effects of compulsory

education to specific age groups. The following regressions obtained by

expanding equation (3.8) enable us to investigate age variation in the compulsory

education effect due to constructing more school buildings (Ouflo 2001);

f=25 f=25

T;t = If/o + 1f/1/ + 1f/2dit + I 1f/3fditAfit + I 1f/4fAfit + 1f/5t + 1f/6 X + Vif=8 f=8

f=25 f=25

0it = rPo + rP11 + rP2dit + I rP3fdit Afit + I rP4fAfit + rP5t + rP6 X + 9if=8 f=8

3.9

I dropped the sub-district indicator for simplicity. Sub-district fixed effects are

indicated by 1f/1I and rPlI' The dependent variable is non-education transfers (Tit) or

transfers of education (Qit) to children. Tit and Qit denote the transfers of

respectively non-education and education transfers to child i, sub-district I, and

year t (1993 and 1996). dit is the distance to the nearest junior high school in a

sub-district I for child i. The dummy variable for age group f in year t for individual

i is described by Afit. I use dummy age group from 8 years of age to 25 years of

age. I use X to denote control variables.

Coefficients 1f/3j and rP3j of equation (3.9) capture the effects of distance changes

to the age group f on their non-education and education transfers. Age group f

experiences a decline or an increase in non-education transfers 1f/3j per 1 km.

distance change. Education transfers decrease or increase by rP3j for the same

distance change. Coefficients 1f/3j and rP3j should be significantly different from

162

zero for the treatment age group (11 - 16), while other ages' coefficients (8 - 10

and 17 - 25) should not be significantly different from zero. If the coefficients of

interaction are significant, junior high school construction affects the education

and non-education transfers of the treatment age groups.

The next section is divided into several sub-sections. First, I examine the change

in enrollment rates during the period of analysis. Second, I investigate the effect

of the policy on educational and non-educational transfers using a simple

approach by dividing the sample into the treatment group and the control group.

Third, I use more elaborate methods to capture age specific variations in the

effect of the compulsory education policy. Both approaches utilize regressions

restricted by rural/urban, per capita expenditure level, and household head years

of education to evaluate which groups benefit the most from the program.

4.2.1 Compulsory Education and Enrollment Rates

Table 3.3 illustrates a non-parametric difference-in-differences analysis of

demand for schooling and the employment decision. Schooling increases for all

groups. Treatment groups show a 76% enrollment rate before the program and

80% enrollment rate after the program. Eliminating variation between two groups,

treatment and control groups, indicates that after the program is introduced, the

enrollment rate increases by 4%.

163

Table 3.3 Non-parametric Difference-in-Differences Tabulation on Schooling

Demand and Employment Decisions

Schooling Decision AverageBefore Mter Difference DID*

Group 8 -10 & 17 - 25 0.52 0.53 0.01

Group 11-16 0.76 0.80 0.04 0.04

Working Decision AverageBefore Mter Difference DID*

Group 8 -10 & 17 -25 0.30 0.28 -0.02

Group 11-16 0.13 0.09 -0.D3 -0.01

Dividing the data into two groups, control and treatment groups, is difficult due to

repeaters in junior high school, late entries, and early entries to the same school

level. The age groups that still enroll in junior high school are very wide (Susenas

1993, 1996), starting from age 11 to around 20. Dividing age groups into two may

introduce a bias into the estimation. To overcome this problem and to see

variation in enrollment rates changing by age due to building construction, I

expand the treatment and control groups into age specific dummy variablexxv. I

use equation (3.9) and replace the dependent variable with individual schooling

decision (Eit) or employment decisions (Wit). I estimate the interaction coefficients

between age dummy variable, Afit' and nearest distance to junior high school, dit.

f=25 f=25

Eit = fl1 + fl2dit + L fl3fdit *Afit + L fl4fAfit + fl5 t + flaX + v .f =8 f =8

164

3.10

X, similar to the previous regression, is a vector of household characteristics and

sub-district characteristics. The employment decision (Wit) relationship is not

shown but it follows the same specification with equation (3.10).

Figure 3.1 presents the difference-in-differences of predicted values from

regression results of equation (3.10). That is, if f denotes the dummy of treatment

group and t denotes dummy of before and after the program, the difference-in

differences is [F(f=1, t=1) - F(f=1, t=O)J-[F(f=O, t=1)-F(f=0, t=O)]. There is an

increase of school enrollment and the profile depends on the age. The rise of

school enrollment starts from age of 12 to age of 17, at ages above than 17 the

magnitude of the increment change in school enrollment declines. There is a gap

between non-parametric profiles and the predicted value profiles. The DID model

explains 60% of the increase of school enrollment. The proportion of children

working declines by less than the proportion of enrolled increases. Thus, the

proportion working and enrolled apparently increases.

165

School Demand and Working Decision

Age

'-.19

-}

18

....... 'O'O .......

171615

". '. ............. ~ ....

14

...+ + ..+.....

" ..

.'

. . . •. . . non-parametric working

___ predicted working

...+... non-parametric schooling

__predicted schooling

.....

0.08

0.06

0.04

0.02

~ 0C

I

C

~-0.02

C-0.04

-0.06

-0.08

-0.1

Figure 3.1 School Demand and Working Decision

4.2.2 School Distance and Transfers: First Approach

Table 3.4 presents the coefficients of interactions of equation (3.8), with non-

education (Panel A) and education transfers (Panel B) as dependent variables. I

also use equation (3.8) to investigate the effect of distance to school on non-

education consumption (Panel C) and children's labor income (Panel D). To

examine how the effects of the program differ by levels of household per capita

expenditures and household head years of education, I estimate equation (3.8).

This is done by running separately restricted samples of two categories of

family's expenditure per capita and three categories of household head's years of

education. First, I run the regression including all families combined. Second,

results are reported separately for the two categories of family per capita

166

expenditures. The sample consisting of families with higher per capita

expenditure is categorized as the fourth quartile, while the sample of families of

lower expenditures is categorized as the first quartile. Third, results are reported

for three sub samples based on household head's years of education. The three

categories are families with household head years of education lower than 6

years, between 7 to 12 years of education, and higher than 12 years of

education.

Decreased distance to school reduces non-educational transfers to children aged

12 - 15, the treatment group, as shown in the first column of Panel A. After the

program, distance to school is considerably reduced. The closer the school is,

the higher the non-education transfers. Estimation results of unrestricted samples

indicate, although not significantly, that distance to school has the same effect on

educational transfers and non-education consumption, but at a smaller

magnitude. On the other hand, increased distance to school increases child

labor income (Panel D). The coefficient of interaction for the regression using

child labor as the dependent variable is significant and positive. Similar to both

non-education and education transfers, the shorter the distance is to the nearest

school, the lower the child labor income.

167

Table 3.4 Regression (OLS) Results for Difference-in-Differences Between Two

Types of Cohort (Treatment Group 12 - 15 and Control Group 20 - 25)

Children with families' Children with famille.'expenditures per expenditures per Children with head's Children with head's Children with head's

AU Children capUa Ie•• than Rp. capita higher than Rp. education is less than education is between education is higher20,000·" Jyear 60,000·" I year 6yrs 7 and 12yrs than 12 yr.(lowest quartile) (highest quartile)

Number of Observations 112,859 8,713 33,338 61,755 30,124 6,281

Penel A: D.~ndantV.....b.. : Non4duc::.tIon b'all8fer-.

Coefficient of interaction: distance to the nearest -467.47 • 47.79 -1438.19 • -278.75 •• -791.50 • -1115.96 •

junior high school" treatment dummy (96.86) (155.56) (252.15) (130.82) (205.26) (654.49)

Rsquare 0.167 0.1356 0.1052 0.14 0.1016 0.086

Panel B: Dependent Vertebkt : Education tr_r.r.

Coefficient of interaction: distance to the nearest -6.16 -5.43 1.90 -51.41 67.65 • 466.31 •

junior high school * treatment dummy (21.65) (2.85) (66.58) (38.31) (18.77) (115.93)

Rsquare 0.0758 0.097 0.089 0.0282 0.0759 0.0774

Pen. C: Dependent Van.IM: Non-Educ::etlon CoMumption

Coefficient of interaction: distance to the nearest-72.78 5.58 -301.44 • -96.37 • -17.79 -859.34 •

junior high school" treatment dummy(41.68) (7296.00) (130.19) (24.92) (95.71) (447.78)

Rsquare 0.3767 0.142 0.1545 0.4164 0.2723 0.233

P8nel 0: 08p4lnd8nt Varl8bh : Chldhln Labor Income

Coefficient of interaction: distance to the nearest 400.86 • -47.65 1138.65 • 130.97 • 841.35 • 722.92junior high school It treatment dummy

(85.81) (155.62) (209.54) (121.23) (178.18) (499.99)

Rsquare 0.133 0.1371 0.1066 0.154 0.1046 0.0793

Note: standard errors are in the parentheses; USD =Rp. 2,500 exchange rates in 1996* indicates significantly different from zero at a 1% confidence level; ** denotes significantlydifferent from zero at a 5% confidence level.

168

Changes in non-education transfers are mainly due to changes in child labor

income. While changes of school distance increase non-education consumption,

changes of school distance reduce labor income more significantly. Non

education transfers are non-education consumption less labor income. A change

in the transfers may come from one or both of the components. I employ non

education consumption and children's labor income as dependent variables

using the same regression method as presented in Panel C and D of Table 3.4.

Non-education consumption is significantly affected by changes of school

distance. Non-education consumption is higher when the school distance is

closer. However, the magnitude is small when compared to the non-education

transfers' coefficients. On the other hand, changes of school distance have a

positive effect on children's labor income. That is, the nearer the school the lower

their labor income. Reduction in the distance to school successfully increased

school enrollment, which was followed by a decline in children's hours worked.

Thus, distance to school apparently decreases labor income.

Changes in the distance to school moderately affect the non-education transfers

in general. Decreasing the distance to the nearest junior high school by 1 km

increases transfers as much as Rp. 467,00 per month or around USD 2.25 per

year, using 1996 exchange rates. On average there was a 6 km change in the

school distance during the period of analysis, as shown in Table 3.1. Parents

compensate for this difference by increasing non-education transfers by as much

as 5%xxvi of the average of non-education transfers between 1993 and 1996. On

the other hand, education transfers are insensitive to the distance change. The

169

coefficient of interaction, using education transfers as dependent variable, is not

significantly different from zero. These results confirm that non-educational

transfers complement the government program, while education transfers do not

change as a result of the government program. These results are consistent with

a binding type of family.

The response of families to change in the distance to school after the program

strongly correlates with per capita expenditure level and household head's years

of education. While the magnitude of the coefficients of interaction depends

heavily on parental categories, in general, signs of coefficient of interaction are

not sensitive to household head's educational level. The higher per capita

expenditure is, the greater the effect of the distance on the non-education and

education transfers. More highly educated household heads respond more

significantly to changes in school distance. Distance to school does not

significantly change non-education transfers of families with lower per capita

expenditure and lower head years of education. For these families, distance to

school does not affect education transfers. On the other hand, families with

higher per capita expenditure and higher household head education increase

their education and non-education transfers significantly due to changes in

school distance.

Parents do not change their education transfers as a result of changes in

distance to school before and after the program. Examining the estimation results

from restricted samples, the only significant responses are from higher educated

170

parents. The closer the distance (Le., the cheaper the unit cost of education), the

lower the educational transfers. This is indicated by coefficients of interaction

presented in Table 3.4 that are positive and significantly different than zero at the

1% confidence level. For these types of families, education transfers act as

substitutes for the government program rather than complement it. Responses by

other family types to change in the distance to school are negligible. Their

coefficient of interactions is not significantly different from zero.

Changes in non-education transfers are due to lower child labor income and

higher non-education consumption. Child labor declines after the program's

introduction. As a consequence, children's labor income also declines. If non

education transfers are non-education consumption less labor income, non

education transfers decrease as labor income declines. As households respond

to distance to school changes by increasing non-education consumption, non

education transfers' changes are greater.

Parents are constrained by the government policy. They provide an inefficient

level of education. These families treat children's education and non-education

expenditures as consumption. Preferences affect the consequences of changing

these expenditures. These constrained-type families increase non-educational

transfers when education prices are lower, but do not increase educational

transfers when education prices decrease. Parents compensate children who

experience a decline in their labor income by increasing non-education transfers.

On the other hand, educational transfers are insensitive to the policy changes.

171

The household head's educational level could be an important factor in

characterizing the constrained-type families. Lower educated household heads

may compose more of the constrained-type families than the non-constrained

type. The restricted regressions on children from higher educated household

heads show a different pattern: distance positively affects educational transfers

and negatively affects non-educational transfers. On the other hand, the unit cost

of education makes educational transfers decrease and non-educational

transfers increase. Higher educated parents may represent a more efficient type

of family, but are still inconsistent with the comparative static results discussed

earlier.

4.2.3 School Distance and Transfers: Second Approach

The first approach has several disadvantages: the method is sensitive to control

group selection and age variation is neglected. In this section, I use equation

(3.9) to closely examine the age variations. Initially, non-educational and

educational transfers per month are used as dependent variables. I then look at

the program's effects on labor income and non-educational consumption, and

which factors influence non-educational transfers. The coefficients should be

significantly different from zero for those who are influenced by the program,

while the coefficients of those who are not affected by the program should not be

significantly different from zero. Those that are obligated by the program may fall

within the 11 to 17 year old range due to repetition or late entries.

172

4.2.3.1 Estimates of the Effects of Education Policy on Non-education Transfers

Several variables are applied to control for variation, which may come from

individual, household, and district or sub-district. Table 3.5 presents estimation

results using different control variables with non-education transfers as the

dependent variable. Column 1 of Table 3.5 shows the regression results obtained

when controlling for the variation in the primary enrollment rate and average

labor income of the household head at the sub-district level. I control the primary

enrollment rate in 1993 at the sub-district level to capture the sub-district school

environment variation before the program was imposed. The average household

head's labor income at the sub-district level is used as an approximation for

school quality. In addition to controlling for school quality variation, this average

also represents economic conditions at the sub-district level.

Estimation results shown in column 1 of Table 3.5 only control sub-district

variation, which are the enrollment rates of the previous year and the average of

labor income at sub-district level. Estimation yields a low 0.08 R-square. At the

younger ages, these results confirm those previously asserted that the distance

to school reduces non-education transfers. The lesser the distance to school is,

the higher non-education transfers. However, only the coefficients of interaction

age dummy variables of 11, 13, and 15 and the distance are significantly different

from zero at a significance level of 1%.

173

Column 2 of Table 3.5 presents the results without controlling for any household

and neighborhood characteristics. Column 3 presents regression results obtained

by adding dummy variables for household characteristics. The first three dummy

categories are household head education at the sub-district level (i.e., less than 6

years, between 7 and 12 years, and more than 12 years of education). The

second three dummy categories are family expenditure levels (first quartile,

second quartile, or top quartile). These variables control for variation in family

characteristics.

Estimation results without controlling for variation in household head's years of

education are shown in column 4, while column 5 shows the estimation results

without using the expenditure level dummy variable for controlling household

variation. All results indicate that coefficient signs are robust and consistently

negative regardless of the addition of characteristics' control variables. Results

with the complete set of control variables (column 3) yield significant coefficients

of interaction from ages 8 to 16. The response to the change in school distance

varies by age.

174

Table 3.5 Estimates the Effects of Education Policy on Non-Education Transfers:

Coefficients of Interaction Between Age Variable Dummy at 1993 or 1996 and

Distance to the Nearest School at 1993 or 1996

Non-education TransfersAge in 1993 or 1996

(I) (2) (3) (4) (5)

8 -8.87 153.34 -243.Q\ • -134.18 -210.53 *(78.12) (81.67) (68.91 ) (69.41) (73.15)

9 31.11 209.66 • -241.29 • -100.02 -219.21 •(83.72) (87.34) (72.16) (75.36) (76.27)

10 -135.07 23.71 -472.24 • -317.15 • -422.91 •(71.31) (74.49) (64.18) (64.75) (67.57)

11 -295.79 • -139.14 -577.49 • -416.67 • -561.78 •(76.39) (78.89) (69.06) (69.41 ) (72.45)

12 -134.28 32.30 -501.20 • -356.55 • -428.06 •(75.51) (78.19) (68.79) (69.64) (71.67)

13 -266.05 • -119.93 -516.85 • -350.68 • -525.78 •(86.09) (89.19) (77.25) (78.63) (80.70)

14 -157.08 -17.30 -399.95 • -289.74 • -365.36 •(105.54) (108.93) (93.43) (97.93) (96.35)

15 -347.34 • -220.08 -494.42 • -396.11 • -502.02 •(104.80) (107.85) (95.12) (97.65) (98.07)

16 -259.91 -125.20 -489.44 • -334.58 • -498.72 •(134.42) (137.17) (127.95) (128.51) (130.29)

17 -192.11 -66.18 -203.25 -180.63 -218.05(171.53) (173.48) (163.30) (167.36) (164.58)

18 -306.96 -189.48 -321.14 -283.58 -362.25(228.42) (229.49) (223.36) (225.85) (224.04)

19 -111.82 51.04 -71.08 -74.20 -102.65

(197.96) (200.03) (183.88) (190.12) (187.01)

20 84.78 198.52 106.59 106.Q\ 83.09(187.97) (188.87) (179.44) (182.83) (181.61)

Primary Enrollment at 1993 yes no yes yes yes- - -

Average household head laboryes no yes yes yes

income at sulHlistriclleve1

- - .-

Household head years of educationno no yes no yes

dummy variables- ~

Expenditure level dummy variable no no yes yes no

N 139,705 139,705 139,705 139,705 139,705

R 0.079 0.069 0.160 0.125 0.143

Note: standard errors are in the parentheses, all regressions included age dummy variables, yeardummy variables, and distance to the nearest junior high school. Coefficients of age group 21 25 are not displayed. *Indicate coefficient is significantly different from zero at the 99%confidence level

175

Figure 3.2 presents the variation by age of the effects of changes in distance to

school on non-education transfers. Panel A in the figure plots the coefficients of

interactions with one standard deviation of variation for this regression. The

distance to school affects non-education transfers negatively among the young

cohort, but affects the older cohort positively. The pattern is relatively flat in the

age range of 11 to 16 but increases rapidly for the older cohorts. The results

confirm those of the first approach previously discussed. Panel B of Figure 3.2

also presents estimation results using logarithmic non-education transfers. The

estimation omits negative transfers, almost 10% of the total sample. Treatment

groups, from 11 to 16 years old, experience 0.2% to 0.3% higher non-education

transfers per 1 km. change of distance to the nearest junior high school. That is,

1.2% - 1.8% for 6 km. distance changes. This is considerably underestimated

when compared to results from the first approach. This may be due to the

omission of observations with negative transfers, which is necessary to apply the

logarithmic regression.

176

Age

":

...•.. +d

j.••••••~

......... -d

..•...

........JI..

.••••••.• l1li.

..~.

•

10 11 12 13 14 15 16 "· ... 1~: 1 20: '.

Unrestricted Regression the Compulsory Program Effectson Non-Education transfers

--coefficients

Panel A.200.00

100.00

0.00

8 9-100.00

J!} .......I: -200.00(I)

'uIE -300.00(I)0() -400.00

-500.00

-600.00

-700.00

Panel B. Unrestricted Regression the Compulsory Program Effectson Log(Non-Education Transfers)

.......8 11 12 13 14 15

Age

I

.......~

--coefficients ......... -d ....... +d

......•0.0040

0.0030

0.0020

0.0010

en 0.0000-I:(I)

'u -0.0010E(I) -0.00200()

-0.0030

-0.0040

-0.0050

-0.0060

Figure 3.2 Estimates of the Effects of Education Policy on Non-Education

Transfers: Coefficients of Interaction Between Age Variables Dummy and

Distance to the Nearest Junior High School

177

4.2.3.2 Estimates the Effects of Education Policy on Education Transfers

Regressions with education transfers as the dependent variable follow the same

methodology as the non-education transfers regressions. Column 1 of Table 3.6

presents the regression results when controlling for sub-district characteristics.

The coefficient signs are negative and significant at the 99% confidence level for

the age groups 13 to 19, but positive and significant for the age groups 8 to 10.

The effect of distance to school is increasing for older cohorts. This is consistent

with increasing of costs of education up to the college level. Among the junior

high school age groups with significant coefficients (13 - 15), education transfers

increase as the distance decreases. The signs of the coefficients of interaction

are robust for any absence or presence of control variables.

178

Table 3.6 Estimates of the Effects of Education Policy on Education transfers:

Coefficients of Interaction Between Age Variable Dummy at.1993 or 1996 and

Distance to the Nearest School at 1993 or 1996

Education Tun ferAge in 1993 or 1996

(I) (2) (3) (4) (5)

8 24.95 • 59.93 • -1.54 9.78 2.66(11.41) (11.63) (10.92) (11.05) (11.08)

9 25.01 • 63.95 • -5.34 9.86 -3.01(10.77) (11.06)

-(10.27) (10.3:'2 (10.47)

10 32.69 • 67.12 • -5.99 11.07 -0.07(9.99) (10.36) (9.54) (9.58) (9.73)

II 17.38 51.15 • -15.19 3.24 -13.50(11.16) (11.44) (10.99) (10.85) (11.1 6)

12 13.14 49.23 .. -28.94 • -12.63 -20.18 •(9.20) (9.54) (9.14) (8.96) (9.27)

13 -32.84 • -1.34 -60.99 .. 42.09 .. -62.73 ..(10.71) (10.95) (10.40) (10.22) (10.63)

14 -:47.71 .. -17.40 -75.55 .. -62.82 • -72.01 •(11.15) (11.45) (10.81) 10.'78) (1098)

15 -91.63 • -70.47 • -113.85 • -ld3.12 • .J 14.97 •(18-.58) (19.45) (16.92) 17.34) (17.46)

16 -88.73 • -59.67 .. -114.38 • -96.14 • -116.77 ..(23.49) 24.28 (21.73) 22.20) ('22.31 )

17 -121.13 • -93.91 • -121.81 • -119.07 • -124.18 •(22.26) (22.92) (20.72) (21.15) (21.21)

18 -123.62 • -99.10 • -123.17 • -119.55 • ·129.15 •(26.48) (27.14) (25.13) (25.64) (25.50)

19 -115.06 • -80.24 • -109.01 • -109.34 • -113.87 •(34.55) (34.99) (32.88) (33.60) (33.30)

20 -128.60 -104.78 -J24.55 -124.87 -128.39(110.45) (110.68) (109.95) (110.10) (110.11)

Primary Enrollment at 1993 yes no yes yes yes

Average household head laboryes no yes yes yesinoome at sub-district level

-- -I-Household head years of

no no yes no yeseducation dummy variables- ~~ 1- -

Bxpenditure level dummyno no yes yes novariable

-

N 139,705 139,705 139,705 139,705 139705R 0.059 0.039 0.105 0.085 0.095

Note :standard errors are In the parentheSes, all regressions Included age dummy variables, yeardummy variables, distance to the nearest junior high school. Coefficients of age group 21 - 25are not displayed. ·significantly different from zero at the 99% confidence level

179

Panel A. Unrestricted Regression the Compulsory Program Effects on EDUCATIONTransfers

50

-150

•.......°l~--::~~~~'----'----'---~----O------'----'---'-------'----=---'

~ .....~ .... 1tl.....~:"" . 13 14 15 16 17 18 19 !tJ. "'&. ". ,: Age

" ..•..."lI.•• ". ••..... '. ..,.....................

.................................&.

.l1c -50Q)

'0lEQ) -100o()

-200 I --Coefficient ...... +d ... & ... -d I

-250

Panel B. Unrestricted Regression the Compulsory Effect on Log(Education0.04 Transfers)

[ --Coefficient +d & ••• -d I

••

••••.•• &.

0.03en...-c: 0.0Q)

T5~ 0.01Q)00

0

-0.01

-0.0

-0.0

Figure 3.3 Estimates of the Effects of Education Policy on Education Transfers:

Coefficients of Interaction Between Age Variables Dummy and Distance to the

Nearest Junior High School

180

Education transfers to young cohorts up the age of 15 are more affected by the

school distance changes. Panels A and B of Figure 3.3 plots coefficients of

interactions along with their associated variation. The coefficients in Panel A

indicate a downward trend. Among the junior high school age groups, the

program positively affects the education transfers in a significant way. However,

the program does not affect the older cohorts. Panel B indicates results obtained

by using the logarithm of education as the dependent variable. Panel B indicates

that most treatment groups experience 1% to 2% higher education transfers per

1 km. change of distance. The transfers are 5% to 10% increases for 6 km.

changes. This is considerably higher than the non-education transfers'

responses. These results contradict those obtained previously.

4.2.3.3 Estimates of the Effects of Education Policy on Labor Income and Non

Education Consumption

The compulsory education program negatively affected the younger cohort's

labor income through the reduction of the distance to the nearest junior high

school, but did not affect the older age groups. Figure 3.4 shows coefficients of

interaction for Equation 3.9 with child labor income as the dependent variable. I

use the control variable for family characteristics and sub-district characteristics.

All signs are positive and significantly different from zero at the 1% level of

significance for age groups 8 to 16. The older age groups' coefficients decline but

are not significantly different from zero. Panel B of Figure 3.4 shows that ages 8

181

to 13 experience a decline in labor income by around 6% to 7% per 1 km.

change in the school distance. This means their labor income decreases by 36%

to 42 % for 6 km of distance changes.

Decreases in the distance to nearest junior high school negatively affect non

education transfers. Non-education transfers are non-education consumption

less child labor income. The regression results on child labor income confirm that

part of the effect of increasing non-education transfers is due to decreasing child

participation in labor market. Another part of the effect comes from changes in

child non-education consumption. Panel A and Panel B of Figure 3.5 present

estimates of regressions on non-education consumption. Non-education

consumption among the treatment groups change by 0.4% to 0.5% per 1 km

distance change or 2% to 2.5% for average of 5 km. changes in distance.

Coefficients of age groups older than 17 are larger in magnitude but are

insignificant.

182

Panel A: Unrestricted Regression the Compulsory Program Effects on LaborIncome

20

Age

""

9 10 11 12 13 14 15 16

I---coefficients ···EJ· ··-d . ··6···+d I

8

t!. ..•• ·6··'· '6'·,· '6···· .t;...... t;..",

[3 •••• ·EJ···· 'EJ'·,. 'EJ'·,. ,E)•••• 'E!.,

500.00

400.00

300.00

200.00

100.001!lc: 0.00Q)

'0tE -100.00Q)

8 -200.00

-300.00

-400.00

-500.00

-600.00

Panel B: Unrestricted Regression the Compulsory Program Effects onLog( Labor Income)

A,.ll .... .t!..

1--Coefficients ... E)--. -d ... t;.... +d I h, .EJ

"0'

8 9 10 11 12 13 14 15 16 17 18 19 20Age

t!. .•... t;.., ..• t:r .•.•.t!. •.... t;....•• 6 '.

[3 •••.• E). "', 0·· ...[3 .•.•. E)••••• f]

0.08

0.07

0.06Ul-c:Q)

"0 0.05~Q)0() 0.04

0.03

0.02

Figure 3.4 Estimates of the Effects of Education Policy on Labor income:

Coefficients of Interaction Between Age Variable Dummy and Distance to the

Nearest School

183

200.00

Panel A: Unrestricted Regression the Compulsory Program Effects on NonEducation Consumption

100.00Age

0.00

8 11 12 13 14 15 16 18 19 20.el -100.00cQ)

'13::E

-200.00Q)

8-300.00

0 ,. .1:1

-400.00 tiI --Coefficients . --s·---cJ --er-+d I

-500.00

Panel B: Unrestricted Regression the Compulsory Program Effects onLog( Non Education Consumption) Age

0

~'d"~ 10 11 12 13 14 15 16 17 18 19 20-0.001

--Coefficients -- ·fl--· -d --'A-'- +dI-0.002 A

.el ,A" ,A.' , ,-0.003 ts .....!:s.,·,·6, "

. "c 'tJ.' ,Q)

'13::E .Q) -0.0048 ,fiJ

-0.005 .,s., " ,0

G. .13 •• E1, .,. --[3'" " '0' , ,

-0.006 , , I:],'1:1'1:1

-0.007


Consumption: Coefficients of Interaction Between Age Variable Dummy and

Distance to the Nearest School

184

Taking logs of non-education transfers and education transfers omits non-zero

and negative transfers. Thus, the estimation results may suffer from selectivity

bias. If non-education consumption increases by 2% to 2.5% and children's labor

income declines by about 36% to 42%, non-education transfers should also

increase by more than 2% to 2.5%. The estimation results of logarithmic non

education transfers show an increase of only 1 to 1.5%. However, taking the

percentage change, based on average non-education transfers during both

years, from the results of the first approach indicates an increase by as much as

5% to 10%. The latter value seems more reasonable when looking at results of

both components of non-education transfers.

4.2.3.4 Estimates the Effects of Education Policy on Transfers Using Restricted

Sample

This section examines the program's effects on households with different per

capita expenditures or from different regions (urban vs rural). To examine which

groups are affected most by the program, regressions are conducted restricting

the sample by family per capita expenditure level and urban/rural categories.

Figure 3.6 to Figure 3.8 plot the restricted regression results for non-education,

education transfers, and labor income respectively.

As previously discussed with the first approach, the signs of estimation results

are not sensitive to the household's category of per capita expenditure. The

185

changes of school distance negatively affect non-education transfers for all per

capita expenditure categories (Figure 3.6). Households in all per capita

expenditures categories respond similarly to the change of school distance.

Unlike the previous approach, the magnitude of responses to changes in the

school distance does not indicate significant differences among the three

categories of household per capita expenditure.

Urban households increase their non-education transfers more than rural

households do in response to the change in school distance. Transfers of the two

areas begin to diverge starting from age 13 where urban is affected more than

rural. If urban and rural areas are correlated with household per capita

expenditure, such that urban households tend to have higher household per

capita expenditures than rural households do, the estimation results will be

consistent. Urban households reflect more efficient households that compensate

the changes of school distance by higher non-education transfers. Rural

households, on the other hand, tend to be more constrained by the program

imposed.

186

0.01

0.00

0.00

l/)

c 0.00Q)

'0~Q) 0.000()

0.00

-0.01

-0.01

Panel A: Log( Non-education Transfers) Categorized by Per-CapitaExpenditures

13 14

, ,)K

I--Iowest --e-- middle .. '::1:'" highest I

Age

0.003 Panel B: Log( Non-education Transfers) Categorized by Rural/Urban

--urban -......- rural I

0.002 '"

0.001

0

-0.001l/)-c: -0.002Q)

'0

~ -0.0030()

-0.004

-0.005

-0.006

-0.007

-0.008 I.... ···all samples


Transfers Using Restricted Sample: Coefficients of Interaction Between Age

Variable Dummy and Distance to the Nearest School

187

0.04

0.02

CJ)

C -0.02Q)

'13

~ -0.04()

-0.06

Panel A: Log (Education Transfers) Categorized by Per-CapitaExpenditures

•• .:1(. '. 'lK'

-0.08

.)K.)K.,',)K•• , ••)j( •••• )K".

.'

-0.1

0.05

0.04

0.03

0.02

CJ) 0.01CQ)

'13 0i:Q)0 -0.01()

-0.02

-0.03

-0.04

-0.05

I --lowest -s-- middle .. ')K- •• highest I

Panel B: Log(Education Transfers) Categorized by Rural/Urban

I all samples ---urban - rural I

Figure 3.7 Estimates of the Effects of Education Policy on Education Transfers

Using Restricted Sample: Coefficients of Interaction Between Age Variable

Dummy and Distance to the Nearest School

188

Figure 3.8 Estimates of the Effects of Education Policy on Labor Income Using

Restricted Sample: Coefficients of Interaction Between Age Variable Dummy and

Distance to the Nearest School

189

Similar to estimation results for non-education transfers responses to changes in

school distance are strongly correlated to household levels of per-capita

expenditure. Education transfers increase most for the highest per capita

expenditure category as shown in Panel A of Figure 3.7. The transfers increase

as much as 6% per 1 km. distance change or 30% per average 6 km, while the

lower per capita expenditure households are affected much less than that.

Similarly with non-education transfers, urban households' education transfers are

more affected than those of rural households. The responses tend to diverge

starting from around age 13. This may reflect the much higher cost of education

in urban areas relative to rural areas at older ages.

Labor income of children in households of highest per capita expenditure

category is also highly affected (Figure 3.8). Their labor income declines by

about 50% per average of 6 km of distance changes. Children from lower per

capita expenditure households are affected by as much as a 30% decline in labor

income with the same distance changes. The program tends to more efficiently

reduce child labor of higher quartile households, while lower quartile households

continue to struggle with the pressure to decrease child labor supplies. Lower

income households depend heavily on child contributions to family income. Thus,

they send their children to labor market. As a result, child labor income is

insensitive to the school distance changes for these households.

190

Child labor income of urban households is affected more than those of children in

rural households; a decline by about 8% per 1 km school distance change in

urban areas and 6% in rural areas. Older cohorts' labor incomes in both rural and

urban areas do not change significantly in response to school distance changes.

However, the effect of changes in the school distance on urban labor income

tends to decline as the children get older, while child labor income tends to

increase in the rural areas for children of the same age.

5. Conclusions

The compulsory education program moderately affects non-educational

transfers. Education transfers increase by as much as 5% to 10%, while non

education transfers increase only by about 1% to 2%. By decomposing the non

educational transfers into non-educational consumption and children's labor

income, it is shown that increasing non-educational transfers are due to both

factors. Parents have to bear higher child costs because of higher non

educational consumption and declining income contributions from children. In

general, the program significantly affects the treatment group Uunior high school

age), ranging from 11 to 16 years old, for both educational and non-educational

transfers. There is not much variation among these groups. The changes in

distance to nearest junior high school did not affect the older cohort. The effect of

the program on the age group 8 to 10 is also significant, particularly in its effect

on non-educational transfers.

191

Response of the households strongly correlates to their household per capita

expenditure levels. The program robustly affects non-education transfers and

education transfers of households in the higher quartile of per capita expenditure.

These households represent a more efficient type of household. They

compensate for the lower education price by increasing non-education transfers

and education transfers to their children. On the other hand, lower per capita

expenditure households tend to be more constrained by the program. They

increase moderately non-education transfers to their children, but they do not

change their education transfers as a result of lower education price. Lower per

capita expenditure households depend heavily on the government educational

subsidy.

The lack of school participation in low-income families is due to the

complementary nature of non-education transfers. A family burden increases if

they let their children go to school. Therefore, in order to successfully enhance

their school participation, the government directed transfers that relieve families'

burden and may assist those whose resources depend on their children's labor

income. The program promises to shift children's activities from working to

schooling. This analysis was based on survey data from 1996, only two years

after the compulsory education program was begun. Analysis over a longer

duration may require analyzing households' behavior, in particular, behaviors of

those who have lower incomes. As the longer duration is sufficient for the

192

households to adjust to the policy changes, results may differ. A comprehensive

analysis of government cash transfers, a new program implemented after the

financial crisis, may also be required to assess the efficiency of the government

schooling program. Households in the lower quartile income are eligible to

receive cash transfers. Analysis on how integrated program, through compulsory

school and cash transfers received by households, is a challenge left for future

research.

Urban households are more affected by the school distance changes compared

to rural households. Boys are more affected by the school distance changes

relative to girls. Insensitivity of rural households' responses to the program is

mostly due to their vulnerability in child labor. Most rural households work in the

agriculture industry. Parents need their children to supply labor for the farm.

Thus, child school participation is also low. As a result, their education transfers

are considerably insensitive. Investigation of gender related issues in the rural

and urban context are a future challenge. Households, particularly in the rural

areas, favor boys over girls. Girls are required to work rather than attend school,

while boys are allowed to go to school.

193

Appendix

Appendix Table 1 Education Policy Milestone in Indonesia

YEAR POLICY GOVERNMENT ACTION NOTE

Dutch Colonial Period

1684The Dutch East India Company managed Christian's faith school andregulated time of instruction and school fees

1800The Netherlands East Indies Goveroment replaced The Dutch company

School taught moral education, reading, and writingand started the native Indonesian education

1848 The government allocated special budget for native Indonesian education.

1867 Department ofEducation was established only responsible for Christian's schoolThe government spent more fWlds to build lOOre schools in Java but not in

1883 other islands. 'Sekolah Radja', secondary school for aristocrat, wasestablished and taught Dutch.

1899Sekolah Radja' changed to OSVIA (Training School for NativeGovernment Officials) taught in Dutch language.

1906Village schools for native Indonesian were established in several residentsin Java

After Independent 19451950 Compulsory education Government provided large scale teacher education (Kursus Pengajar Lack of fund to provide teachers

untuk Kursus Pengantar Kewajiban Belajar (KPKPKB) to fill about138.240 lack ofteachers

1959 Compulsory education Appointed 153 out of275 districts from 21 out of25 provinces First conference on Compulsory Education

1960 Curriculum 1960 Education Classified education as general education, special education,and education for disable

Begin of SoehartoPresidential

1968 Curriculum 1968 Politically changed the education philosophy. Due to lack of teachers,

government issue aMandatOlY Teaching Law; Banned foreign school

1973 Basic Education Contruction 6,000 new school building; that continued Wltil fiscal year INPRES No. I 1973Development Program 1993/1994 with total constructed building arouod 148.945. Provision of

new teachers totally arouod 1.001.604 until fiscal year 1993/1994

1975 Curriculum 1975 Ministry of Education and Culture issued Education Basic Memorandwn Centralize curiculumstated that education had to be developrneot oriented and progress oriented

1984 6 years compulsory Goveroment formally enforced tbe program thateducation actually already part ofBasic Education Development

Program started in 1973

1994 a 9 years compulsory Strengthening private jWlior high school to be more efficient; OpenjWlior a. Implemented through formal and infonnal school; b.education high school, infonnal basic education for certification (paket A for Inforce anyone age 16 - 44 to be literate, ahle to speak

elementmy level & Paket B forjunior high level) ; empowering religion Indonesian Language and basic knowledgebased school (MTs - Tsanawiyah) by graduallyrehahilitation their (Pernherantasan 3 buta)facilities under Ministry ofReligion Affair

b Curriculum 1994

1999 Decentralization

194

Appendix Table 2 Summary of Comparative Static

Note: Assuming the utility has a separable utility function In every argument for all three cases

of human capital, whether bound or unbound by compulsory education. Question mark

symbolizes ambiguous effect.

1. Not bound by compulsory education -1,2 = 0, 2. Bound by compulsory education -1,2 > 0

Comparative Static Children expenditure perceived as:

Consumption Investment Consumption and

Investment

1 2 1 2 1 2

dC~ ? - - 0 - 0dPe

d~ ord~ + + + 0 + 0dYp dwp

dq~ - 0 - 0 - 0dPe

d~ ord~ + 0 + 0 + 0dYp dwp

d% ? - - 0 - 0dPe

dj/d ordj/d + + + 0 + 0dYp dwp

.. ..

195

References

Acemoglu, Daron, and Joshua David Angrist. "How Large Are the Social Returnsto Education? Evidence from Compulsory Schooling Laws." NBERWorking Paper 7444 (1999).

Anderberg, Dan, and Alessandro Balestrino. "Self-Enforcing IntergenerationalTransfers and the Provision of Education." Economica 70, no. 277 (2003):[55]-71.

Angrist, Joshua David, and Alan B. Krueger. "Does Compulsory SchoolAttendance Affect Schooling and Earnings?" The Quarterly Journal ofEconomics (1991): 979 - 1014.

Asaad, Ragui, Deborah Levison, and Deborah Zibani. "The Effect of Child Workon Schooling in Egypt." Economic Research Forum Working Paper(2004).

Baland, Jean-Marie, and James A. Robinson. "Is Child Labor Inefficient?" TheJournal of Political Economy 108, no. 4 (2000): 663-79.

Basu, Kaushik. "Child Labor: Cause, Consequence, and Cure, with Remarks onInternational Standards." Journal of Economic Literature 37, no. 3 (1999):1083-119.

Basu, Kaushik, and Pham Hoang Van. "The Economics of Child Labor."American Economic Review 88, no. 3 (1998): 412 - 27.

Becker, Gary S. "Investment in Human Capital: A Theoritical Analysis." TheJournal of Political Economy 70, no. 5 (1962).

Becker, Gary S. , and Nigel Tomes. "Child Endowment and the Quantity ofChildren." The Journal of Political Economy 84, no. 4 (1976): S143-S62.

Becker, Gary S., and Kevin M. Murphy. "Family and the State." Journal of Lawand Economics 31 (1988): 1-18.

Beegle, Kathleen, Rajeev Dehejia, and Roberta Gatti. 'Why Should We CareAbout Child Labor? The Education Labor Market, and HealthConsequences of Child Labor." NBER Working Paper 10980 (2004).

Behrman, Jere R., and James C. Knowles. "How Strongly Is Child SchoolingAssociated with Household Income." PIER Working Paper 97-022 (1997).

Ben Porath, Yoram. "The Production of Human Capital and Life Cycle ofEarnings." The Journal of Political Economy 75 (1967): 352-65.

Benabou, Roland. "Tax and Education Policy in a Heterogeneous AgentEconomy: What Levels of Redistribition Maximize Growth and Efficiency?"Econometrica 70, no. 2 (2002): 481 - 517.

Bommier, Antoine, and Pierre Dubois. "Rotten Parents and Child Labor." Journalof Political Economy 112, no. 1 (2004): 240-48.

Bradbury, Bruce. "Measuring the Cost of Children." Working Paper ANSW(1994).

Bray, Mark, and Murray R. Thomas, eds. Financing of Education in Indonesia:Asian Development Bank, 1998.

196

Carneiro, Pedro, and James J. Heckman. "The Evidence on Credit Constraints inPost-Secondary Schooling." The Economic Journal 112 (2002): 705 - 34.

Chavalier, Arnaud, and Gauthier Lanot. "Financial Transfers and EducationalAchievement." Keele Economics Research Paper 2000/01 (2000).

Cremer, Helmuth, and Pierre Pestieau. "Intergenerational Transfer of HumanCapital and Optimal Education Policy." GEPR Discussion Paper No. 4201(2004).

De Fraja, Gianni. "The Design of Optimal Education Policies." Review ofEconomic Studies 69 (2002): 437 - 66.

Deaton, Angus. The Analysis of Household Surveys: A MicroeconometricApproach to Development Policy: John Hopkins University Press, 1997.

Deaton, Angus, and John Muellbauer. "On Measuring Child Costs: WithApplications to Poor Countries." Journal of Political Economy 94, no. 4(1986): 720-44.

Deolalikar, Anil B. "The Determinants of Primary School Enrollment andHousehold Schooling Expenditures in Kenya: Do They Vary by Income."(1997).

Duflo, Esther. "Schooling and Labor Market Consequences of SchoolConstruction in Indonesia: Evidence from an Unusual Policy Experiment."American Economic Review (2001): 795 - 813.

Duflo, Esther, Sendhil Mullainathan, and Marianne Bertrand. "How Much ShouldWe Trust Difference in Differences Estimates." Quarterly Journal ofEconomics 119, no. 1 (2004): 249-75.

Duryea, Suzanne, David Lam, and Deborah Levison. "Effects of EconomicShocks on Children's Employment and Schooling in Brazil." PSGResearch Report - University of Michigan 03-541 (2003).

Edmonds, Eric V. "Does Child Labor Decline with Improving Economic Status."NBER Working Paper 10134 (2003).

--. "Does Illiquidity Alter Child Labor and Schooling Decisions? Evidencefrom Household Responses to Anticipated Cash Transfers in SouthAfrica." NBER Working Paper 10265 (2004).

--. "Understanding Sibling Differences in Child Labor." Journal of PopulationEconomics forthcoming (2004).

Edmonds, Eric V, and Nina Pavcnik. "International Trade and Child Labor: CrossCountry Evidence." GEPR Discussion Paper No. 4309 (2004).

Emerson, Patrick, and Shawn D. Knabb. "Expectation Traps, IntergenerationalRedistribution and Child Labor." draft (2004).

Ersado, Lire. "Child Labor and School Decisions in Urban and Rural Areas:Cross Country Evidence." FGND Discussion Paper no. 145 (2002).

Fitzsimons, Emla. "The Effects of Risk on Education and Child Labor." TheInstitute for Fiscal Studies Working Paper WP02/07 (2002).

Gertler, Paul, David I. Levine, and Minnie Ames. "Schooling and Parental Death."(Forthcoming ) The Review of Economics and Statistics (2002).

Glomm, Gerhard, and B. Ravikumar. "Public Versus Private Investment inHuman Capital: Endogenous Growth and Income Inequality." The JournalofPolitical Economy 100, no. 4 (1992): 818 - 34.

197

Goldin, Claudia Dale, and Lawrence F. Katz. "Mass Secondary Schooling andthe State: The Role of State Compulsion in the High School Movement."NBER Working Paper 10075 (2003).

Grootaert, Christian, and Ravi Kanbur. "Child Labor: A Review." Policy ResearchWorking Paper 1454 (1995).

Hanushek, Eric, Charles Ka Yui Leung, and Kuzey Yilmaz. "Redistributionthrough Education and Other Transfer Mechanism." NBER Working Paper8588 (2001 ).

--. "Redistribution through Education and Other Transfer Mechanism."Journal of Monetary Economics 50, no. 8 (2003): 1719 - 50.

Haveman, Robert, and Barbara Wolfe. "The Determinants of Children'sAttainments: A Review of Methods and Findings." Journal of EconomicLiterature XXXIII (1995): 1829 - 78.

Heckman, James J., Lance Lochner, and Christopher Taber. "GeneralEquilibrium Treatment Effects: A Study of Tuition Policy." The AmericanEconomic Review 88, no. 2 (1998): 381-86.

Heckman, James J., and James R. Walker. "The Relationship between Wagesand Income and the Timing and Spacing of Births: Evidence from SwedishLongitudinal Data." Econometrica 58, no. 6 (1999): 1411-41.

Hutasoit. Compulsory Education in Indonesia: UNESCO, 1954.Keane, Michael P., and Kenneth I. Wolpin. "The Career Decisions of Young

Men." The Journal of Political Economy 105, no. 3 (1997).--. "The Effect of Parental Transfers and Borrowing Constraints on

Educational Attainment." International Economic Review 42, no. 4 (2001):1051 - 102.

King, Elizabeth, and Lee Lillard. "Determinants of Schooling Attainment andEnrollment Rates in the Philippines." RAND Note (1983).

Kruger, Diana. "The Effects of Coffee Production on Child Labor and Schooling inBraziL" Ph.D. Dissertation Manuscript, University of Maryland, 2002.

Lee, Ronald D. "The Formal Demography of Population Aging, Transfers, andthe Economic Life Cycle." In The Demography ofAging, edited by SamuelPreston, 8 - 49: National Academy Press, 1994.

Lee, Ronald D. , and Timothy Miller. "Interage Resource Flows in the Us: ADescriptive Account." Population Association ofAmerica, San Francisco(1995).

Lee, Ronald D., and R. Edwards. "The Fiscal Effects of Poplation Aging in theUs: Assessing the Uncertainties." In Tax Policy and Economy, edited byJames Poterba, 141-81: NBER MIT Press, 2002.

--. "The Fiscal Impact of Population Change." In Seismic Shifts: TheEconomic Impact of Demographic Change, edited by R. K. Triest: FederalREserve Bank of Boston Conference Series, 2001.

L1eras-Muney, Andriana. "Were Compulsory Attendance and Child Labor LawsEffective? An Analysis from 1915 to 1939." Journal of Law and EconomicsXLV (2002): 401 - 35.

198

Loth, Erik. Private Intergenerational Transfers and Population Aging: TheGerman Case, Contributions to Economics,. Heidelberg; New York:Physica-Verlag, 2001.

Manacorda, Marco. "Child Labor and the Labor Supply of Other HouseholdMember." CEP Discussion Paper 0590 (2003).

Mason, A , and T. Miller. "Dynasties, Intergenerational Transfers, and Life-CycleIncome: A Case StUdy of Taiwan." In Sharing the Wealth: DemographicChange and Economic Transfers between Generations., edited by AMason and G. Tapinos, 57 - 84. Oxford: New York: Oxford UniversityPress, 2000.

Mason, A, and N. Ogawa. "Aging, Intergenerational Transfers, and Saving inJapan." Paper presented at the Conference on Population AGing in theIndustrialized Countries: Challenges and Responses, Tokyo, Japan 2001.

Mincer, Jacob. "Investment in Human Capital and Personal Income Distribution."The Journal of Political Economy 66 (1958): 281-302.

Nicholson, J.L. "Appraisal of Different Methods of Estimating Equivalence Scalesand Their Results." Review of Income and WEalth 22 (1976): 1-11.

Oreopoulos, Philip, Marianne E. Page, and Ann Huff Stevens. "Does HumanCapital Transfer from Parent to Child? The Intergenerational Effects ofCompulsory Schooling." National Bureau of Economic Research. WorkingPaper Series NO.1 0164 (2003): 1-[46].

Peltzman, Sam. "The Effect of Government Subsidies-in-Kind on PrivateExpenditures: The Case of Higher Education." The Journal of PoliticalEconomy81, no. 1 (1973): 1-27.

Priyambada, Agus, Asep Suryahadi, and Sudarno Sumarto. 'What Happened toChild Labor in Indonesia During the Economic Crisis? The Trade-Offbetween School and Work." SMERU Working Paper (2002).

Ranjan, Priya. "An Economic Analysis of Child Labor." Economics Letters 64(1999): 99-105.

Rosenzweig, Mark Richard, and Kenneth I. Wolpin. "Evaluating the Effects ofOptimally Distributed Programs: Child Health and Family PlanningInterventions." American Economic Review 76, no. 3 (1988): 470 - 82.

--. "Natural 'Natural Experiments' in Economics." Journal of EconomicLiterature 38, no. 4 (2000): 827 - 74.

Schultz, T. Paul. "School Subsidies for the Poor: Evaluating the MexicanProgresa Poverty Program." Journal of Development Economics 74, no. 1,Special Issue (2004): 199-250.

Schultz, Theodore W. "Capital Formation by Education." The Journal of PoliticalEconomy 68, no. 8 (1960): 571-83.

Spohr, Chris A "Formal Schooling and Workforce Participation in a RapidlyDeveloping Economy: Evidence from Compulsory Junior High School inTaiwan." Journal of Development Economics 70 (2003): 291 - 327.

Sylwerster, Kevin. "Can Education Expenditures Reduce Income Inequality."Economics of Education Review 21 (2002): 43 - 52.

Tilaar, H.AR. Lima Puluh Tahun Pembangunan Pendidikan Nasiona11945-1995Suatu Analisis Kebijakan: Gramedia Widiasarana Indonesia, 1995.

199

Tzannatos, Zafiris. "Child Labor and School Enrollment in Thailand in 1990's."Economics of Education Review 22 (2003): 523-36.

Zilcha, Itzhak, and Jean-Marie Viaene. "Public and Private of Education andIncome Inequality." Discussion Paper No.7 (2003).

i National Transfers Account Project proposal is submitted to National Institute of Health byMason at the East West Center and Lee at UC Berkeley (2004).ii I use the average salary calculated by Clark (1999) and multiply it by number of teachers.Clark's results are considered a sufficient proxy and have been used in a previous study (Sutjiptoet al. 2001).

iii I use the same proportion of each school level budget available at Clark et al. (1999) tocalculate other year each school level.iv Receipts per level are 2 million Rupiah, 4 million Rupiah, and 10 million Rupiah for primaryschools, junior high schools, and secondary high schools respectively.

v Most statistics are population weighted. Estimation for the next section are not clustered since itis proven that clustering, either into household or sub-district level, does not change thesignificant level.vi Ninety percent of enrolled students whose age 5 - 20 years receive their education expenditurefrom parents (Susenas 1992 and 1995 - Education Module, Central Biro Statistics -BPSIr,ldonesia)VII I drop subscript j to reduce notationviii Consumption allocation by Engel's method is developed by the NTA Project Team

ix 7J is defined as enrolled household member education expenditure share. Direct calculation ofindividual education expenditures over total household education expenditures indicates

individual share_as the non-parametric fJ.x I calculate confidence interval on coefficient r0 as

Coefficient r0 should be fluctuated around

Nobs / Nobs ( )2Yo ±t(1-a/2;Nobs -2) MSE .L ql NObs.L qi -"iii and

1=1 1=1

YNObS( )2

Yl ±t(l-aI2;Nobs -2) MSE .L qi -qi1=1

zero, while coefficients r1 should be fluctuated around one.

confidence interval r1 as

xi Gross enrollment rate is number of all students, regardless their age, divided by the number ofpopulation of the respected school level agexii Child labor is defined as any paid working performed by children whose age is younger than 18.In addition, ILO defined child labor as working activities performed by children younger than 15that prevent them from schooling.

200

. The first

Yeh = I a,N, + 8 where Yeh is household'=5-9

entrepreneur income and N, is number of family member that belong to demographic group f, 1014, 15-19,20-24,25-29,30-34,35-39,40-44,45-49, 50-54, 55-59, 60+, that is working (s;=1) tofind the share of entrepreneur income a f and calculate the member income based on the

calculated share Yehi =Yeh afiS~ . Labor Income is defined as two-third of entrepreneur/ Lafis;

income plus earnings.xxii I define children as those whose status as children in the household between 5 and 25 yearsof age.xxiii Similar to estimation of individual non-labor income, individual education expenditures are

estimated by using a simple regression: I q~t = I fJ,N: + 8 Where qh;e+ is individuali '=5-9

education expenses for member i, N, is number of member i whose age belong to age groups fand enroll at school. The school age group is divided into 5-9,10-14,15-19,20-24,25-29 and 3034 age groups.xxiv See, for example, Wooldridge Introduction to Econometrics: A Modern Approach 2nd Ed.(2003) p. 433xxv See DufJo (2001), for examplexxvi Based on the average of non-education transfers shown in Table 3.1

( Uh + UC )Uh yC [_pe .!:.- + (1- a) yC ]xiii ( h C ) xgxg xgxg LhLh LC q

f Pe'pg,y ,y ,a,e = I

_eaql-a (Lf )a-I Al (yh yh (Uh , + UC ) + pgUh pg )xgxg xgxg LhLh

argument inside the bracket,(uh +Uc )Uhh hyc [-pe.!:.-+(l-a)lC..] determine the sign

xgxg xgxg L: L: Lf q

because the second and third argument are both certainly positive. The sign depends on thisC C

term _pe.!:.- +(1- a)L . The last term is return on education, re =(1- a)L. Thus, the sign is~ q q

positive when re ~ pe .!:.- . Otherwise the sign should be ambiguous.LC

Ixiv I use OLS regression for this purpose since I will make a linear difference. Difference-indifferences by using probit requires non-linear, more complicated, formula to interpret thecoefficient of interactions. However, both methods produce almost the same results.xv An elaborate estimation individual education transfers data refers to chapter 1xvi I use Engels' method to estimate the intra-household allocation or transfers to childrenxvii Investigation on government cash transfers in education transfers and intra-householdallocation is in progress.xviii Complete comparative static calculation is available upon requestxiXEstimated by using the Engel's Methodxx I use the average of nearest distance as a proxy of opportunity cost sending children to school.Junior high school students tend to go to the nearest distance school. Schultz (2004) supportsthis assumption.

xxi To obtain non-labor income, I regress

201

ESSAYS ON EDUCATION AND INTERGENERATIONAL TRANSFERS …

Documents