2D1 ESSAYS ON EDUCATION AND INTERGENERATIONAL TRANSFERS IN INDONESIA A DISSERTATION SUBMITTED TO THE GRADUATE DIVISION OF THE UNIVERSITY 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, Chairperson Timothy J. Halliday Sang-Hyop Lee Gerard Russo Robert D. Retherford
213
Embed
ESSAYS ON EDUCATION AND INTERGENERATIONAL TRANSFERS …
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
~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
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
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
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
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
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
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
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.
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
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)
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
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
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
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
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
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
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).
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
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
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.