~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)
Number of Observation' 73,715 86,133 254,784
*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
1996Household Head Education
Higher than
Primarv Primary All
Panel A: Household Characteristics
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)
Age of Household head 42.34 39.98 45.06(12.57) (11.32) (13.82)
Education share expenses** 0.02 0.03 0.02(0.04) (0.06) (0.04)
Food Share expenses** 0.57 0.50 0.56(0.12) (0.14) (0.13)
Total Expenditures 253,417.00 462,127.60 286,849.00(194,656.80) (472,035.90) (309,105.90)
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)
Number of Observation' 17,688 19,034 60,584
Panel B: Individual Characteristics
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 Primary** 0.38 0.07 0.18(0.48) (0.26) (0.39)
Only Completed Junior level** 0.11 0.19 0.12(0.32) (0.39) (0.32)
Completed higher than Junior level** 0.24 0.48 0.36(0.43) (0.50) (0.48)
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)
Number of Observation' 79,493 85,144 264,345
*AII Monetary values are in Rupiah/month (1.00 USD = 2,500 Rupiah, 1996 exchange rate).
Standard Deviations are in parentheses. ** in percentage terms.
28
Table 1.3 (Continued) Mean Value by Education of Household Head
1999Household Head Education
Higher than
PrimaN Primary All
Panel A: Household Characteristics
Number of Children 2.01 1.88 1.86(1.48) (1.44) (1.51 )
Household head year of Education
Age of Household head 43.38 39.81 45.52(12.82) (11.67) (14.11)
Education share expenses** 0.02 0.03 0.02(0.03) (0.05) (0.03)
Food Share expenses** 0.64 0.58 0.63(0.11 ) (0.13) (0.12)
Total Expenditures 498,305.10 758,096.00 548,413.80(304,078.10) (542,950.00) (414,820.40)
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
Panel B: Individual Characteristics
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 Primary** 0.61 0.23 0.47(0.49) (0.42) (0.50)
Only Completed Junior level** 0.13 0.21 0.14(0.34) (0.41 ) (0.35)
Completed higher than Junior level** 0.18 0.31 0.27(0.39) (0.46) (0.45)
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
*AII Monetary values are in Rupiah/month (1.00 USD = 2,500 Rupiah, 1996 exchange rate).
Standard Deviations are in parentheses. ** in percentage terms.
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*)
Private Public Private Public Private Public
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*)
Private Public Private Public Private Public
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.
2. Literature Review
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
Number of Observations 57117 56087 51928 51627
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)
Number of Hours Worked* 0.89 0.69 0.84 0.44(4.25) (3.86) (4.15) (3.36)
Panel C Working Children
Number of Observations 14533 12825 12620 13120
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)
Number of Hours Worked* 39.27 39.32 39.51 40.79(14.74) (14.35) (14.52) (13.82)
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)
Number of Hours Worked* 40.65 39.48 40.30 36.48(13.53) (13.52) (13.82) (13.74)
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)
Number of Hours Worked* 31.05 31.76 32.27 15.74(18.26) (17.64) (11.95) (20.65)
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
Number ofObservation 119,477 84,071 57,626 82,474 101,130R-Square 0.070 0.179 0.283 0.279 0.532
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
Number ofObservation 119,477 84,071 57,626 82,474 101,130R-Square 0.07 0.17 0.25 0.24 0.28
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
Number ofObservation 119,477 84,071 57,626 82,474 101,130R-Square 0.06 0.17 0.27 0.26 0.25
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
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 ** 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
before after DID before after DIDControl Group' 0.409 0.450 0.883 0.864Treatment Group 0.491 0.545 0.913 0.913
0.082 0.095 0.013 0.030 0.049 0.019
Number of WorkingHours
before after DID before after DIDControl Group' 13.736 15.872 38.656 37.171Treatment Group 17.072 19.118 40.248 39.529
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.
2. Literature Review
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
Figure 3.5 Estimates of the Effects of Education Policy on Non-Education
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
Figure 3.6 Estimates of the Effects of Education Policy on Non-Education
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