Does Interjurisdiction Relationship Affect Education
Spending Decision?
A Spatial Economics Approach in Indonesia’s State Level
Husnirokhim Nurdin Alim
(S2504502)
h.n.alim@student.rug.nl
Supervisor:
Abstract
his study investigates 491 Indonesia’s state level behavior in education spending formulation. Four groups of variables, namely demographic, economic, geographic and politic, are introduced. Spatial interaction effect is embedded into the model by using Maximum Likelihood (ML) and Generalized Spatial Two Stage Least Squares (GS2SLS) estimation. The result reveals that Spatial Durbin Model (SDM) is better in representing state’s behavior. State’s education spending decision is not only determined by its own exogenous variables but also its neighbors’ decision through an endogenous and an exogenous interaction effect. The result also confirms the existence of spillover hypothesis in Indonesia’s state level.
Keywords : Government behavior, education spending, spatial dependency, state level JEL Codes : C21, H75, I22
Table of Contents Abstract ... 1 Table of Contents ... 2 List of Figure ... 3 List of Appendix ... 3 Chapter 1 Introduction ... 4 Problem Statement ... 6 Research Structure ... 6
Chapter 2 Theoretical Frameworks ... 7
The Importance of Education ... 7
The Role Government in Education ... 8
How Government Spending Spill to Other Neighbors? ... 9
Formal Form of Interjurisdiction Spillovers ... 12
Defining Weight Matrix ... 15
Formulating Government Spending on Education ... 16
Chapter 3 Indonesia as a Field of Study ... 19
Education System in Indonesia ... 19
Framework Decentralization in Indonesia ... 20
The Complexity of Government Role in Education ... 23
Chapter 4 Data and Methodology ... 26
Data and Variables ... 26
Defining the Weight Matrix ... 30
Methodology ... 31
Chapter 5 Results and Analyses ... 33
The Ordinary Least Square (OLS) Estimation ... 33
The Spatial Model ... 36
Modifying the Weight Matrix: A Robustness Check ... 38
Chapter 6 Conclusion and Discussion ... 39
Conclusion ... 39
Discussion... 40
References ... 41
List of Figure
Figure 2.1 Illustration of Government Spending Spillover ... 12 Figure 2.2 The Relationships between Different Spatial Dependence Models for Cross-section 14 Figure 2.3 Contiguity Matrix ... 17 Figure 3.1 School Enrollment Rate (%) ... 21 Figure 4.1 Cross-State Per Capita Deviation on Education Spending from 20 percent budget 28 Figure 4.2 Proportion of State’s Revenue in 2011 ... 29 Figure 4.3 Average Percentage State’s Spending by Sector in 2011 ... 30
List of Appendix
Appendix 1 Education System in Indonesia ... 47 Appendix 2 Summary Statistics ... 47 Appendix 3 Determinants of Education Spending Deviation Per Capita: OLS Results... 49 Appendix 4a Determinants of Education Spending Deviation Per Capita: OLS and ML Estimation
Result ... 50 Appendix 4b LR and Wald Test for ML Estimation ... 51 Appendix 5a Determinants of Education Spending Deviation Per Capita: OLS and GS2SLS
Estimation ... 51 Appendix 5b Wald Test for GS2SLS Estimation ... 53 Appendix 6a Determinants of Education Spending Deviation Per Capita Excluding
Chapter 1
Introduction
In according to socio-economy literature, education is commonly recognized as an essential part of human life because of its benefits. Investment in education mostly guarantees positive private rate of return on the job market (Granado et al., 2007; Hanushek and Höβmann, 2007) and decreases probability of being below the low-income cut-off unemployed (Oreopoulos, 2006). Education also have indirect effects on people’s life such as increasing life expectancy (Gathmmann et al., 2012; Cutler and Muney, 2006), reducing probability to be in poor health status, and improving welfare (Oreopoulos, 2003). In macroeconomic standpoint, increasing education quality would be a prominent input for national production factor, thus let national output to grow (Michaelowa, 2000). Public sector intervention on education is an everlasting discussion for economists, researchers and policy makers. Although education does not satisfy non-rivalry and non-exclusivity assumptions on pure public goods, government existence in this sector is still necessary because of some reasons (Gruber, 2011). First, unlike purchasing other goods, purchasing education faces difficulties in term of accessing loan in private market due to collaterality problem. Hence, education sector would not efficient in pure private market. Second, parental altruism problem makes parents to fail in maximizing family utility. Children will directly receive benefits from education investment. Nevertheless, they are not responsible for deciding until which level of schooling they should have. This responsibility falls to their parents. If parents put a low value on improvements in their children’s future earning potential, then they may underinvest in children’s education that results low level family utility. Additionally, without public intervention, education inequality may be passed on to each consecutive generation. It is much related to the universal education concept which means that education should be equally enjoyed by all citizens (Bedi and Garg, 2000). All in all, this intervention is believed to guarantee the abundant potential pecuniary and non-pecuniary impacts from education attainment.
Since 1998, most of the government of Indonesia’s affairs are devolved from central to local level; provinces, regencies, and municipalities (Legislation on Local Government No. 22/ 1999 further renewed with No. 32/ 2004). The ultimate goal of this policy is to bring government closer to its citizens hence government can better support their needs. This mechanism provides more space for local governments in managing their fiscal. However, it does not mean central government loses all of its authorities. Central still has power in making nation-wide policy and financing in particular aspects. One of the decentralized affairs is education. At least 51 percent of total national education spending in 2006 was allocated through state level (regency and municipality). Central government allocated another 44 percent and the last 5 percent by province level. This confirms state level plays important role in education (The World Bank, 2009).
Additional fiscal room in state level could change its behavior in fiscal decision making. According to Caldeira (2012) and Schaltegger and Küttel (2002), decentralization exhibits intergovernment competition that can increase efficiency of public spending. This theory allows the existence of information spillovers among citizens of neighboring jurisdictions that can influence their own policy maker’s decisions. Strategic interaction between local governments has become popular after Case et al. (1993) workhorse study and is better in defining government behavior. Stansel (2006) points out that the effect of interjurisdictional interaction should be most obvious in lower level of government.
Studies on government spending formulation that takes into account interjurisdiction spillover issues has been done by Case et al. (1993) and Stansel (2006) for total spending and Elhorst and Fréret (2009) for social expenditure, among others. They empirically show the spillover exists and affects local government’s behavior in budget setting policy. Those studies do not mainly focus their attention on education affair. Case et al. (1993) is one of studies that analyze the education sector. They provide an evidence that spatial linkage appears in education sector spending. Nonetheless, they do not intensively discuss this matter.
Problem Statement
This study investigates the factors that affect state’s behavior in determining education spending per capita. To provide a wider perspective, this study also tests the occurrence of spatial interdependency among states in education spending setting by incorporating spatial interaction effects. In the end, the spillover presence possibility will be elaborated.
Research Structure
This thesis comprises of six chapters. The first chapter introduces basic background of the research and followed by problem statement and research structure. The second chapter discusses conceptual framework of the topic that elaborates previous studies as the theoretical benchmark. It begins with the general public theory on the importance of government in education sector. Then, it is followed by spatial interdependency concept in public sector, and ended up with government determinants in providing education. To give some background of the place of study, chapter 3 exhibits Indonesia snapshot on political structure of decentralization regime, education structure and complexity of government role in education to provide basic understanding about education in Indonesia.
Chapter 2
Theoretical Frameworks
This chapter provides some main theoretical backgrounds to give prior understanding of the topic. In the beginning, the importance of education will be presented. The second part addresses the role of government in education, and is followed by the existence of public goods spillover across regions. The last part discusses how government formulates their spending on education.
The Importance of Education
Education is universally believed as important part of human life. According to Michaelowa (2000), without basic skills on literacy and numeracy, individual will face difficulties on their daily life situation. It affects societies in various ways, from personal contribution in economic activities to overall economic development.
In a microeconomic perspective, a large number of literatures confirm that education directly boosts individual earnings. Study conducted by Brunello et al. (2009), to illustrate, shows the estimated returns to education are ranging from 5 to 7 percent for male and 7.5 to 9.5 percent for female. In the Netherlands case, the return to schooling for man is even higher to 15 percent (Kalwij, 2000). Furthermore, education also brings indirect outcome on people’s life such as better health status and lower crime rate. Gruber (2011) argues higher education level may potentially increase productivity, and then society reaps benefits from education in terms of higher standards of living. Higher productivity is reflected in higher salary paid, and then the government collects more tax revenues. Hence, education can be viewed as economic generator.
In a macro level, a well-educated society leads to a sustainable economic growth as proposed by the endogenous growth theory. Higher education is viewed as improvement in human capital quality that can incline national productivity level, and in the end leads to higher economic growth (Nelson and Pack, 1999). They use this theory to explain Asian economic growth miracle in 1990s decade. Using US data, Irenzo and Peri (2006) empirically confirm that higher education level has a significantly positive consequence toward production externalities through increasing Total Factor of Productivity (TFP).
The Role Government in Education
Public sector intervention on education becomes point of interest for economists, researchers and policy makers. Although does not satisfy non-rivalry and non-excludable assumption on pure public goods, education sector needs government intervention because of the credit market failure in education and parents failure in maximizing their family utility (Gruber, 2011). In the first place, if parents somehow do not have sufficient resources to send their children to school, a private credit market should be able to lend them money. Nonetheless, this kind of credit is rarely available due to collaterality problem. Unlike borrowing for obtaining tangible goods, credit for education purposes do not have collateral document such as deed, certificate, or physical collateral that could make lenders feel secure in lending money. Hence, a purely private system cannot function efficiently without perfect capital markets, and capital market imperfections are likely to be severe in the developing countries (Bedi and Garg, 2000). The second problem is what so-called a family utility maximization problem. The direct beneficiaries from education are children but they cannot make their own schooling decision. This responsibility should be taken over by their parents. If parents place a low value on improvements in their children’s future earning potential, then they may underinvest in their children education that makes low level family utility (Gruber, 2011).
Not less important, government intervention comes due to the universal education concept. This means that every individual in a society should receive education. Government intervention has an equalizing effect on income distribution and compensate for families’ financial differences, particularly for an unlucky one (Sylwester, 2002). Without public intervention, inequality may be passed on to each consecutive generation. This concept is essentially based on the assumption that education is a normal good; higher-income parents will purchase the more education for their children than the poor one.
and Chude (2013), to illustrate, conclude every one percent increase in total government spending on education will leads GDP to grow about 0.2 – 0.3 percent per annum. However, to where government should interfere is also an important issue. If the intervention is only to raise school attainment, the economic impact will not so promised. Glewwe and Kremer (2005) argue that a public policy should also concern on the education system as a whole package covering quantity and quality so the prominent economic impact can be reaped. The education system in the developing countries is weak in almost all aspects, including weak incentives for teachers and inappropriate curriculums. The education quality approach would be more effective way to guarantee better economic condition (Hanushek and Höβmann, 2007; Hanushek et al., 2008). Using 43 developing countries, Devarajan et al. (1996) find a surprising result. Their analyses exhibit neither positive nor significant impact of education on economic growth. This striking result is because of misallocation on public expenditure; a productive public capital could become an unproductive spending if it is too much.
The political regime tends to delegate education spending decision into the lowest government tier. According to UNESCO (2005), the educational decentralization is essentially based on some ideas. This enables local government to develop policies based on a particular contexts and requirements of educational beneficiaries. Desire to improve the efficiency and effectiveness of education management supports decentralization as well. The decentralization in education that commonly followed by fiscal decentralization can be viewed as supplementation of inadequate national resources. Local governments can find additional own revenue to finance their education needs. UNESCO (2005) differentiates decentralization as de-concentration of task and outright devolution. The De-concentration of task is formed by creating a regional office that represents national ministry. The policy is taken by central afterward the implementation is delegated to a particular local. On the other hand, the outright devolution gives full responsibility for decision-making process in lower lever. In some cases, the educational decentralization involves all educational management and administration. However it does not mean the central loss all of its powers. The central can manage teacher deployment and finance some national priorities.
How Government Spending Spill to Other Neighbors?
political situation, demographic condition, and other own circumstances. The local voters will be the only beneficiaries of their state’s spending. In other word, each state is independently from other states’ spending decision (Case et al., 1993). In fact, locally public facilities do not only benefit the representative residents but also their neighboring residents, and vice versa. For that reason, the representative citizens could obtain utility from their local and neighboring states (Solé-Ollé, 2006). Formally, LeSage and Pace (2009) state spatial dependence reflects a situation where observed values at certain location i depends on the values of neighboring observations at nearby locations. The interaction among governments generates a reaction function that enables decision variables for a given jurisdiction is directly depends on other jurisdictions variables (Brueckner, 2003).
This issue becomes widely popular after Anselin (1988) workhorse study. He uses the spatial econometrics method as a subfield of econometrics in order to incorporate the spatial effects in econometrics. The spatial effects may be resulted from spatial dependence, a special case of cross-sectional dependence, or from spatial heterogeneity, a special case of cross-cross-sectional heterogeneity. When spatial interdependency exists, the least squares estimation could lead to inconsistent estimation. In contrast, maximum likelihood estimation gives consistent estimation (LeSage and Pace, 2009).
Brueckner (2003) provides three explanations why the strategic horizontal interaction among government or called as mimicking behavior exist. In the first place, this phenomenon arrives for the reason of spillover effects, in which spending on local public services may have beneficial or detrimental effect to its surrounding neighbors such that it is spatially correlated among jurisdictions. The police expenditure could be a good example of this situation. As proposed by Elhorst and Fréret (2009), additional spending on one particular state’s police department will push-out criminals from its jurisdiction to other jurisdictions. Seeing this situation, neighboring states for sure do not want criminals move toward their local. Hence, surrounding states allocate additional spending on police department as well.
welfare, they need more tax income that can be generated by attracting more people to come. In the end, the governments are engaged in fiscal competition.
The last explanation of government mimicking behaviors is the political yardstick competition. This model employs asymmetric information between voters and their government. According to Revelli and Tovmo (2007), the informational spillovers of fiscal policies from surrounding jurisdictions affect another region’s belief of imperfectly informed voters with respect to competency and honesty of their own government. The local voters learn the quality and efficiency of their own government by using the other governments’ performances as a yardstick. The governments are assumed as a rational agent and perfectly-informed on how voters make calculation. Hence, they could modify their fiscal decisions to influence voters’ inference (Bordignon et al., 2004; Allers and Elhorst, 2005). This paper will mainly focus on the first mimicking behavior; spillover model. To illustrate, consider five points on a circle with the same distance. Each point represents state 1 up to 5 such that each state has two shared-border neighbors as shown on Figure 2.1. State’s 1 neighbors are state 2 and 5, and the neighbors of state 2 are state 3 and of course state 1. This rule also applied for other states. Imagine the policy maker in state 1 provides new public facilities. State 1’s local inhabitant will certainly enjoy this policy. In addition, the surrounding inhabitants in state 5 and 2 receive benefits from it as well. This situation is called as spillovers. On the one hand, state 1’s spending spills out to those two states. On the other hand, state 2 and 5 get spill in from state 1.
Figure 2.1
Illustration of Government Spending Spillover
Source: author’s illustration
Solé-Ollé (2006) differentiates two spillover effects i.e. benefit spillovers and crowding spillovers. The benefit spillovers occur when a fraction of the locally-produced public goods in one particular jurisdiction is enjoyed by its representative and surrounding neighbor citizens, and is a perfect
s1
s2
s3
s4
substitute for their own provision of public goods. He uses radio and TV broadcasting as an example. On the opposite, the crowding spillovers occur when publicly provided facilities in a jurisdiction are crowded by residents in neighboring jurisdictions. The crowding of museums and parks by commuters and visitors is the example.
Empirically, literatures have shown the occurrence of spillover model, especially in the lowest tier of government. Case et al. (1993) evaluate the spillover effect of state’s spending on its neighbors. They use the terminology “neighbor” to represent similarly situated states that include not only the geographic proximity but also demography and economic similarities. Given those three different neighbor definitions, they find total state government’s spending is significantly affected by their neighbors’ decisions. Furthermore, they also propose that state’s expenditure should be analyzed in sectoral-based rather than in term of total expenditure to see the real effect per sector. Baicker (2001) confirms Case et al. (1993) study: every dollar increase in one state’s spending rises other states’ spending by 37 to 88 percents. The other studies in Spain (Solé-Ollé, 2006), France (Elhorst and Fréret, 2009), and the Philippines (Capuno, et al., 2013) also conclude the same result; interregional dependency affect local government public service spending decision.
Formal Form of Interjurisdiction Spillovers
The spatial interaction effect can adhere in endogenous variable, exogenous variables and disturbances as proposed by Manski (1993) and Elhorst (2010). The endogenous interaction effect exists when the decision making behavior of a particular spatial unit in some way depends on the other spatial units’ decision. The model has an exogenous interaction effect if the decision behavior made by a spatial unit in some way depends on independent explanatory variables of the decision taken by others. The last one is correlated effect. This effect exists when the units in the same group have similar unobserved environmental such that they behave similarly. Those three interaction effects can be bundled into a single model as (Elhorst (2010) calls this model as Manski Model):
(2.1a)
(2.1b)
~ 0, (2.1c)
⋮ , ⋯ ⋮ ⋱ ⋮ … " , # $ $ $
% &0 & ⋯ & , ' &
0 ⋯ &, ' &
⋮ ⋮ ⋱ ⋮ ⋮
& ' , & ' , ⋯ 0 & ' ,
& & ⋯ & , ' ( 0 )
* * * + , and 0 ⋮ 0 . (2.2)
The denotes the dependent variable, the is a nonstochastic spatial weight, the is set of independent variables, the represents spatially autocorrelated error term, and is independently and identically distributed error term for all observations with zero mean and variance . The interaction term represents the endogenous interaction effect among dependent variables, is the exogenous interaction effects among the independent variables, and is the interaction effects among the disturbance of the different spatial units. The coefficient , , and are the magnitude of endogenous interaction effect, exogenous interaction effect, and error interaction effect, respectively.1 Parameter has usual meaning as coefficient of the independent variables. The matrix quantifies the connections between neighbors. contains zero and non-negative elements if observations are neighbors and zero otherwise. The has dimension 1 2 1 and is usually row-normalized, such that the 1 2 1 spatial lag vector contains values constructed from an average of neighboring observations (LeSage and Pace, 2010). In another publication, LeSage and Pace (2009) give rule of thumb where in one set of spatial location consists of n locations, the maximum possibility relations that could arise are n2-n. The subtraction n from the potential n2 dependence relations is to rule out the dependency of an observation on itself.
1 Stata statistical software uses term
Figure 2.2
The Relationships between Different Spatial Dependence Models for Cross-section
Source: Elhorst (2010) (author’s modification)
0 0 0 0 0 0 0 General (Manski) Model
SAC Model SDM Spatial Durbin Error Model
All those three interaction effects not always come together into a single model. Hence we can make models with some combination effects as purposed by Elhorst (2010) (Figure 2.2). A model with combination of a spatially lagged dependent variable (WY) and a spatially autocorrelated error term (Wu) is called the SAC model (LeSage and Pace, 2009) or the Kelejian-Prucha model (Elhorst, 2010) or the spatial-autoregressive model with (spatial) autoregressive residuals (SARAR) (Drukker et al., 2013). A model consists of a spatially lagged dependent variable (WY) and a spatially lagged independent variables (WX) is called the spatial Durbin model (SDM). A rarely used model with a spatially lagged independent variable (WX) and a spatially autocorrelated error term (Wu) is called the spatial Durbin error model.
Getting more specific model with only one spatial interaction effect, this can be classified as a spatial lag model/ spatial-autoregressive model (SAR)2, a spatial lag X variable (SLX), and a spatial error model (SEM). The spatial lag model is defined as a single equation model where the dependent variable is affected by spatially lagged value on other spatial units and set of their own control variables (Elhorst and Fréret, 2009).
4 , (2.3)
The spatial lag X variable (SLX) is defined as a model with a spatial dependence in the independent variables.
5 , (2.4)
The spatial error model (SEM) is a single equation model with a spatial dependence in the error term (LeSage and Pace, 2009). This model formally can be written as:
(2.5a)
6 (2.5b)
The last model that does not incorporate any spatial interaction can be reduced into the Ordinary Least Squared (OLS) model. Choosing the most appropriate model can be through either the general-to-specific or the specific-to-general approach (Elhorst, 2010).
Defining Weight Matrix
LeSage and Pace (2010), Vidyattama (2012), and Arbia and Fingleton (2008) argue that neighborhood definition expressed in the weight matrix is an essential part on spatial econometrics. The estimation and inferences from spatial regression models are sensitive to particular
2
specifications of spatial weight structure. The weight matrix formulation in a spatial model, however, is tricky and open to multiple definitions within a single study. According to Arbia and Fingleton (2008), critics of spatial econometrics are almost always on the weight matrix definition, degree of accuracy, and the meaning.
There are a lot of ways to define neighborhood. Tobler’s first law of geography states that “everything is related to everything else, but near things are more related than distant things” (Tobler, 1970). Every local could have interrelation with others; however the nearest neighbor will affect most. For simplicity reason, most of researchers use the contiguity-based spatial weights, where the definition of neighbor is based on sharing a common boundary. Following Elobaid et al. (2009), bishop continuity is when two regions meet at a “point” diagonally. This is the spatial analog of two elements of a graph meeting at a vertex. The rook contiguity is condition when two regions are neighbors if they share (part of) a common border vertically and horizontally. The queen’s or king’s is the combination of rook and bishop contiguity. Two regions are neighbors in this sense if they share any part of a common border, no matter how short. Those criterions then can be combined with first order neighbor (regions are neighborhood if their borders directly touch each other) or second order neighbors (regions that are neighbors to the first order neighbors).
Figure 2.3 Contiguity Matrix
a. bischop contiguity b. rook contiguity c. queen contiguity
Source: Elobaid et al. (2009)
Formulating Government Spending on Education
The importance of government function in education becomes interesting topic to be discussed. One of these is government’s consideration in education spending formulation. Most of researchers cluster variables into four big groups i.e. demographic, economic, politic, and geographic situations. Economic variables are commonly approached by GDP or income per capita, and are usually identic in different observations. The other three group variables, on the other hand, really depend on its specific characteristics of an observation unit.
the distribution of the students’ population into ethnic groups (Jews and non-Jews), and the distribution of the Jewish student population by religious affiliation (secular, orthodox, and ultra-orthodox). Per capita GDP, the relative price of education, the distribution of income across the population, and the return on education represent the economic circumstances. The political variable is measured by public deficit/GDP ratio as overall budgetary pressure approach, and sometimes by the defense expenditure/GDP ratio. Their main findings are population size and per capita GDP are the most important components in defining education spending. Surprisingly, the proportion population in secondary school age is negatively correlated with education. This finding indicates expenditures in primary and higher education crowd out expenditure on secondary education.
In decentralized administration, the education spending decision is mostly delegated to the lower tier government, and no longer becomes central government’s authority. Numerous studies have dealt with this issue and ended up with mixture outcomes due to state’s heterogenity characteristics. Study in India’s states level by Chakrabarti and Joglekar (2006) tries to combine three aspects; economic, demographic and politic. The economic variables, income per capita and central government transfer have a positive and significant impact toward education spending. The rural population variable represents demographic aspect depicts a positive and significant coefficient. The premise is people living in a village depend more on government spending to access education institutions. Intuitively, rural population on average are poor therefore they need other resources to send their children to school. They also divide population age 5-9, 10-14, 15-19, and 20-24 to cover the schooling age, however the results are neither positive nor statistically significant. Unfortunately, they do not give sufficient explanations why populations in certain ages are not significant. The caste3 variable is also employed in the model to cover country unique demography characteristic. The reform variable, covering policy-changing, displays a negative and significant coefficient. Generally, education spending declines in the post-reform era.
Verbina and Chowdhury (2002) study the education spending determinants in Russia’s state level. They implicitly classified independent variables into economic, demographic, and geographic. The per capita revenue as a proxy for income affects education spending decision positively. This also means education is a normal good in Russia. Numerically, one percent change in per capita budget changes per capita education spending by 0.56 per cent, ceteris paribus. The student-population ratio, as a demographic variable, positively affects educational expenditures. The population density has a negative impact on total education expenditures. The justification of this negative correlation
3
is a denser-populated region leads its government to shrink its educational cost because of the developed infrastructure and/ or economies of scale. To cover the geographical factors, they use eleven dummy variables, although only three of them are statistically significant. It shows that geographical differences are matters in determining education spending.
Chapter 3
Indonesia as a Field of Study
Recently, Indonesia is in the middle of transition from middle-income to high-income country. Improving in education level means national productivity will be in premium level which benefits for economic development. Managing education itself, however, is not an easy task for Indonesia covering more than 237 million population4, 34 provinces, more than 500 states, and across over 17,000 islands5. This chapter highlights Indonesian facts in education system, decentralization framework, and complexity of government’s financing system in education.
Education System in Indonesia
Under Indonesian constitution, the national education system can be classified into formal, non-formal and innon-formal education which can complement and enrich each other. Formal education is structured educational pathways and tiered begins with two years kindergarten (TK/ RA) followed by six years primary education (SD/ MI), three years junior high school (SMP/ MTs), three years senior high school, and the higher education. Primary education and junior high school belong to 9 years compulsory education system. Senior high school student can choose either general (SMA/ MA) or vocational school (SMK/ MAK) with different curriculum. Usually general high school’s alumnae are prepared to pursue their study in university. Vocational school graduates are mainly focused to be ready in workplace; however they also can pursue study to higher education. The higher education system also provides both general and technical options. Polytechnics and professional schools offer diplomas with range of study from 1 year up to 4 years (D1, D2, D3, and D4), depending upon the program. For higher education, university and institute provide four-years study to obtain undergraduate education (S1), two years for Master (S2), and four years for Doctorate (S3).
The non-formal education can be practiced in anywhere and anytime. It enables lifelong learning to substitute and/or complement formal education; such as early childhood education, life skill training, on the job training, and community-based learning centers. In addition, citizens can substitute their primary, junior high and senior high education by following package A, B, and C equivalent programs, respectively. Those programs do not have age constraint and rigid schedule as formal education. The last scheme is informal education. This can be carried out by family and environment education in the form of independent learning activities. The whole picture of Indonesian education system is depicted in Appendix 1.
4 Number in 2010 based on Population Census by Statistics Indonesia 5
Most of education indicators in Indonesia show positive trend. One of these measurements is schooling enrollment rate6 as is illustrated in Figure 3.1. In period 1994 to 2012, this indicator increased gradually in all formal education levels. School enrollment rate at age 7-12, represent enrolment in primary education, slightly rose from 94.06 to 97.88 percentage point. Children enrolled at junior high and senior high school levels exhibited the most notable improvement; increased 17.13 percent and 15.56 percent, respectively. These trends indicate that citizens recently concern in educational improvement more than just primary school. Furthermore, government policy on nine years compulsory education could be one of the stimulants of this high enrollment rate in junior high school. Senior high school attainment is expected to grow in the following years due to government plan to extend compulsory education from nine to twelve years. Higher education level enrollment (19 – 24 years old) was steadily increased between 12 – 15 percent during 1994 and 2012.
Figure 3.1
School Enrollment Rate (%)
Source: Statistics Indonesia (author’s calculation)
Framework Decentralization in Indonesia
Indonesian government administration system is generally divided into central (pusat) and local (daerah) governments. The local government represents province (provinsi), regency (kabupaten) and municipality (kota)7 level. Regency and municipality essentially belong toward the same level
6
School enrollment rate is defined as proportion of all enrolled children who are in school at a certain age group of the population of the corresponding age group (www.bps.go.id).
7 The lowest tier of autonomous government level is at regency and municipality level (except for the Jakarta Special Capital Region Province). The state is called as regency when it is characterized by rural condition with mostly homogeneous population and has greater area. Administratively it is led by regent (bupati). On the
94.06 97.88 72.39 89.52 45.31 60.87 12.8 15.73 0.00 20.00 40.00 60.00 80.00 100.00 120.00 1 9 9 4 1 9 9 5 1 9 9 6 1 9 9 7 1 9 9 8 1 9 9 9 2 0 0 0 2 0 0 1 2 0 0 2 2 0 0 3 2 0 0 4 2 0 0 5 2 0 0 6 2 0 0 7 2 0 0 8 2 0 0 9 2 0 1 0 2 0 1 1 2 0 1 2
School Enrollment Rate (7-12 yo)
School Enrollment Rate (13-15 yo)
School Enrollment Rate (16-18 yo)
under province’s level coordination. Hereafter, this paper will use terminology “states” to indicate regency and municipality and “local” to indicate province, regency, and municipality.
Decentralization in Indonesia has taken place more than a century since 1903, before its independent. The prominent decentralization milestone, however, started in 1998 (effectively started in 2001) when Indonesia faced rapid political and administration reform due to the demand for democracy mechanism8. Legislation on Local Government No. 22/ 1999 and Central and Local Fiscal Balance No. 25/ 1999 become the key features of decentralization in Indonesia; latter was readdressed by Legislation No. 32/ 2004 and 33/ 2004, respectively. Formally, decentralization means delegating the authority from the central government to local autonomous government in term of right, power, and obligation to organize and manage their own government and public interests. The aim of decentralization is to bring governments and communities closer. This concept is essentially in accordance with Oates (1999) in which public services will be efficiently provided when it is held at the closest government level to the community. The reasons are local government has better understanding on community’s needs, local government is efficient in managing public funds, and competition between regions will promote innovation.
Hofman and Kaiser (2002) analogize Indonesia’s 2001 decentralization as a “big bang” since at the same occasion most of government authority was shifted to the lower government level, regional share in national government spending jumped steeply, and introduced a completely new intergovernmental fiscal mechanism. Ter-Minassian and Fedelino (2007) define this devolution of expenditure functions and revenue raising powers to sub national governments as fiscal decentralization. Decentralization provides wide division of authorities where each local government is given rights to control and manage some particular mandatory sectors by providing basic services to their citizens and elective sectors by developing local leading sector that is based on national minimum service requirements. At least 31 sectors are decentralized; inter alia education, health, public works, human settlement, investment, and labor management. Foreign politic, national defense, national security, law, monetary and national fiscal, and religion that belong to national strategic sectors are not devolved.
other hand, the state is called as municipality when it is characterized by urban condition with heterogeneous population and usually smaller jurisdiction area. Municipal is administratively led by city major (walikota). Others things for regency and municipality are the same including the autonomy policy and election mechanism.
8
In the new intergovernmental system, central government is mandated to share almost one-third of its national revenue for local governments based on money follow function mechanism. This policy is aimed to finance local authority, lessen fiscal inequality between central and local level government (vertical fiscal imbalance), and fiscal inequality among local governments (horizontal fiscal imbalance). The transfer consists of five elements; namely Revenue Sharing (Dana Bagi Hasil, abbreviated as DBH), General Transfer Fund (Dana Alokasi Umum, abbreviated as DAU), Specific Transfer Fund (Dana Alokasi Khusus, abbreviated as DAK), Special Autonomy Fund (Dana Otonomi Khusus) and Adjustment Fund (Dana Penyesuaian).
Get into the detail of the central transfer system, the lion’s share of it goes to DAU or non-earmarked transfer. It provides full authority for locals to spend and manage the grant. In 2011, at least 76 percent of state’s revenue and 65 percent of province’s revenue rely on this block grant. DAU is essentially calculated under formula that accommodate basic allocation and fiscal gap. Basic allocation is largely based on salary bill for local civil servant, and fiscal gap is calculated by taking the difference between estimated fiscal needs and fiscal capacity of local government. Hence, not surprisingly, about 45 percent of DAU goes for civil servant salaries (The World Bank, 2013). Moreover, the strong dependency on this grant somehow creates disincentive to collect local own revenue (Pendapatan Asli Daerah, abbreviated as PAD) through intensify local tax collection (Brodjonegoro and Martinez-Vazquez, 2002).
The second scheme is DAK. DAK is an earmarked grant designed for specific purposes funding that are aligned with national development priorities and implemented by local governments. This is mainly allocated for investment purposes. DAK in education sector, to illustrate, is mainly purposed to build and renovate classes and libraries. In 2011, the proportion of earmarked transfer is very small only accounted for 9 percent on state and 3 percent on province level. The third is a non-earmarked and a vertical equalization grant what so-called as DBH. This grant consists of revenue sharing from natural resources and taxes. In fiscal year 2011, DBH is allocated approximately 15 percent and 30 percent of state’s and province’s budget.
additional allowances for teachers, a school operational assistance program (Bantuan Operasional Sekolah, abbreviated BOS), and local incentive grants (Dana Insentif Daerah, abbreviated DID)9. While all the central transfer are managed and allocated by the Ministry of Finance (MoF), almost every ministry has other mechanisms to support local government by providing de-concentration funds (Dekonsentrasi, abbreviated as DK) and co-administered tasks (Tugas Pembantuan, abbreviated as TP). Both are delegation of authority from the central to lower tier of government to conduct certain central priorities. De-concentration is a delegation of authority from the central to the governor as a representative of the central government and/ or the vertical institutions in particular regions. Co-administered tasks is a delegation of authority from central to the local government and/ or directly to the village, from the provincial to the state level and/ or village, as well as from state to the village to carry out certain duties and obligations (Government Regulation No. 7/ 2008).
Another decentralization issue is regional expansion. In 1999, Indonesia only had 319 local governments, which consist of 26 provinces, 234 regencies and 59 municipalities. Those numbers rise steeply into 542 local governments in 2013; consist of 34 provinces, 410 regencies, and 98 municipalities10. The idea is fundamentally to enhance citizens’ welfare in its jurisdiction. Hopefully this could reduce the span of control between governments and citizens; consequently government can serve their citizens’ demand directly. This expansion is designed to recline development inequality since resources can flow to regions which previously was underdeveloped not only to regions that closed to the center of seat powers. Nevertheless, study by Bappenas and UNDP (2007) reveal that expansion effect is still far from the ultimate goals: public service delivery does not improve significantly after expansion. Lack of institution infrastructure ability and capability in newly autonomy locals probably become the reasons behind this. Thereafter, central government applies regional expansion moratorium program from 2010 to 2013 and prepares the new expansion guidelines.
The Complexity of Government Role in Education
Education sector as part of social infrastructures becomes Indonesia’s strategic priorities as well as on physical infrastructure and health sector. This objective is mandated by the constitution that government should ensure sufficient education for all citizens. Under fourth amendment of Indonesian constitution (UUD 1945) and legislation on National Education Standard No. 20/ 2002,
9 The focus of DID in 2013 is for educational capacity improvement for civil servants to improve government financial report. (Regulation of the Minister of Finance 202/PMK.07/2013)
10
government (central, province, and state) are ordered to allocate minimum 20 percent of their budget for education purposes. Furthermore, almost in every planning document such as Masterplan for the Acceleration and Expansion of Indonesia Economic (MP3EI)11, five years Medium Term National Development Plan (RPJMN), and 25 years Long Term National Development Plan (RPJPN), education is always concerned as a key development sector. Education becomes an important element to accomplish Millennium Development Goals12 in which by 2015 all children could obtain nine years (primary and junior high school) compulsory education.
The education institution in Indonesia is very large with nearly 60 million students, 3.5 million teachers, and 567,752 schools in all education levels13. Given highly decentralized system, education management becomes more complicated. In 2006, for example, more than 51 percent of total national spending in education was conducted by state level. Central and province level allocated another 44 and 5 percent, respectively (The World Bank, 2009). States are responsible for managing and budgeting schools and teachers at primary and junior high school level, excluding schools that belong to Ministry of Religious Affairs (MoRA)14. Provincial governments have limited authority, mostly for coordination purpose with states at the compulsory education level including staff development and the provision for education facilities. Central government sets national policy, issues guidelines and standard and still directly controls the higher education. The Ministry of Education and Culture (MoEC) and MoRA are responsible for setting policies and managing the system. Both ministries directly support education through providing civil service teachers (in all levels) and school grants (in compulsory education). Households’ contributions are fairly high in senior high school and incredibly high in higher education since government funding limitation (The World Bank, 2013). Beside these two ministries, the Ministry of State Personnel and Bureaucracy Reform (Menpan), the Ministry of Finance (MoF), and National Civil Service Board (BKN) also play important role in education such as hiring and determining the quotas of civil service teachers whereas the selection, deployment and management are handled by state level.
The education financing mechanism is already complicated as it is due to multiple sources and transfers across tiers of government. Spending for education comes from central government funds,
11 The Masterplan for Acceleration and Expansion of Indonesia's Economic Development (abbreviated MP3EI) is part of planning document with ambitious objectives to accelerate the realization of becoming a developed country of which the fruits and prosperity will be enjoyed equally among the people.
12
Millennium Development Goals (MDGs) is the global development paradigm agreed by 189 member states of the United Nations (UN). This declaration recommends concrete action plan for the world to overcome poverty, hunger and disease with the time limit of 2015. Indonesia commits to improve community welfare. (Cahyandito, 2011)
13 Data in 2010/2011 taken from Statistical Yearbook of Indonesia 2012 by Statistics Indonesia 14
Chapter 4
Data and Methodology
This section elaborates the data, variables and spatial weight matrix. The last part describes analytical method for intergovernment mimicking behavior in determining spending on education.
Data and Variables
This study tries to figure out the evidence of education spending spillover in Indonesian regencies (kabupaten) and municipalities (kota). For simplicity reason, from now on the terminology “state” will be applied to represent both regency and municipality. This study explores 491 Indonesian state governments level across 32 provinces in 2011; excluding 6 states belong to Jakarta Special Capital Region Province15. The fiscal year 2011 is chosen due to the rich state data set based on population census in 2010 and the sociopolitical stability. Furthermore, during 2010 the number of state is steady due to regional-expansion moratorium.
State’s spending on education should follow the fourth amendment of the constitution and legislation on National Education System no. 20/ 2003. Those set of laws require all government levels to allocate minimum 20 percent of their budget for education purpose. To reduce the biasness of education spending behavior, rather than use the overall education expenditure, this paper uses the amount of deviation between education spending and 20 percent of state’s total budget. This method allows to obtain positive and negative deviation values which mean state government allocates their education spending greater or below than 20 percent. In the model, per capita term of this variable (EDDEVCAP) is set as dependent variable, and it represents the education spending behavior.
78879:;<= >?@A>BC'(.EFE@A>BAFAC C , (4.1)
where 78G<7 =, HIHG<7 =, and <I<= represent education spending, total budget spending, and population at state i, respectively.
In the fiscal year 2011, most of the states allocate education spending above the national requirement. The highest positive deviation is mainly dominated by states located in Java Island.
However, out of 491 states, there are 61 states still spend below the requirement level. Most of them are located in east Indonesia; namely Maluku, Maluku Utara, Papua and Papua Barat Province (Figure 4.1). States in east Indonesia are characterized as poor regions with limited resources availability and quite big area to be served. Given the limitations, they have to build underdeveloped basic physical infrastructure, poor health care facilities, and limited education facilities in the same time. Consequently, education do not become the main consideration. On the other hand, states in Java Island mostly are better developed, such that they can allocate their budget for human capital quality improvement through increasing education spending.
Figure 4.1
Cross-State Per Capita Deviation on Education Spending from 20 Percent Budget (in thousand Rp)
Source: Directorate General for Fiscal Balance, the Ministry of Finance (author’s calculation)
The independent variables are divided into four broad categories i.e. demographic, economic, politic, and geographic situations. The demographic variables are taken from Population Census 2010 conducted by Statistics Indonesia (Badan Pusat Statistik). The first variable is population density (POPDEN). The denser the state, the lower education per capita demand will be due to economies of scale (Verbina and Chodhury, 2002). The second variable is urban population (URPOP) is employed under the hypothesis that the more population in urban area; the less demand on government spending on education will be (Chakrabarti and Joglekar, 2006). They usually have enough own resources to send their children to non-government subsidized school. The third is student density per age group. It is expected to give positive affect of education spending. Under Indonesian law, compulsory education is 9 years; 6 years for elementary school and other 3 years for
junior-high school. Two group variables schooling age 7 – 12 (DEN7_12) and 13 – 15 (DEN13_15) are included. In addition, schooling age 5 – 6 (DEN5_6) years and 16 – 18 (DEN16_18) years are incorporated as well to reflect demand of education for early childhood education and senior high school. The last one is average years of schooling (AVSCHOOL), and it is expected to give positive impact to education spending. All the demographic data are taken from 2010 data (lag t-1) due to the budgeting process – budgeting for year 2011 is calculated at 2010 and based on 2010 data16.
Figure 4.2
Proportion of State’s Revenue in 2011
Source: Directorate General for Fiscal Balance, the Ministry of Finance (author’s calculation)
The second category is economic circumstances that consists of Gross Domestic Regional Product (GDRP) and government revenues. Gross Domestic Regional Product (GDRP) per capita is incorporated to capture the economic well-being of the state. On average, state’s GDRP per capita in 2010 is Rp18.565 million. Assuming education as normal goods, demand will rise as per capita GDRP rises. State’s revenue is embodied in our model to measure resource availability. Figure 4.2 depicts the proportion of five-source per capita revenues in 2011. General transfer fund (DAU) and revenue sharing fund (DBH) play dominant role in local government spending in almost all locals. Both transfers on average are accounted 57 and 14 percentage point of state budget income, respectively. State own revenue (PAD) that represents ability to generate revenues from its jurisdiction is only contribute around 9 percent. Other revenue – mostly come from province-state
16
The Minister of Home Affairs Law No. 37/2010 Guidelines for Formulation Local Government’s Budget 2011 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1 1 7 3 3 4 9 6 5 8 1 9 7 1 1 3 1 2 9 1 4 5 1 6 1 1 7 7 1 9 3 2 0 9 2 2 5 2 4 1 2 5 7 2 7 3 2 8 9 3 0 5 3 2 1 3 3 7 3 5 3 3 6 9 3 8 5 4 0 1 4 1 7 4 3 3 4 4 9 4 6 5 4 8 1
revenue sharing and adjustment fund – supports 14 percent for state’s budget. Due to earmark characteristic, other revenues in this model is subtracted by adjustment fund. Special transfer fund (DAK) contributes another 7 percent to state’s income. DAK is excluded as well from our model due to its earmarked characteristic for specific purposes based on central government priority which does not represent the state behavior in budget allocation. Those four resources (excluding DAK) are expected to have positive impact on education spending. Co-administration (TP) and de-concentration (DK) funds are not involved in this model because those two resources are not considered into state level budget.
Figure 4.3
Average Percentage State’s Spending by Sector in 2011
Source: Directorate General for Fiscal Balance, the Ministry of Finance (author’s calculation)
To capture geographical characteristics, seven variables are introduced. The first is dummy municipality to cover different demand of education spending between municipality and regency. If the observed state is a municipality, it is given value one, and zero if it is regency. The premise suggests that when a state classified as a city, the demand of government spending on education will be lower (Chakrabarti and Joglekar, 2006). The second variable is area in km2. The more area they have, the more demand on education spending provided by government. The sub-nation characteristic is included by clustering nation into six groups based on big islands; namely Sumatera, Java, Kalimantan, Sulawesi, Papua and Maluku, and Nusa Tenggara. Five dummy variables are introduced: dummy Sumatera (DSUM), dummy Java (DJAVA), dummy Kalimantan (DKAL), dummy Sulawesi (DSUL), and dummy Papua and Maluku (DPM). Nusa Tenggara (DNUS) is set as the control variable to get rid of from perfect multicollinearity problem. If the observed state is located in Java Island, for example, the DJAVA value for this state is set as one, and zero otherwise. (Detail of summary statistics is in Appendix 2).
Defining the Weight Matrix
Spatial weight matrix reflects the conditions of each neighboring states. For Indonesian case, the spatial weight matrix is slightly unique due to its archipelago-country characteristic with 17,504 islands17, in which 27 states do not have in-land neighbors. Several spatial econometrics studies have been implemented in Indonesia and use different methods to define the specification of the spatial weighting matrix. Sugiharti (2014) tests spatial econometrics on regional income convergence in East Java. Since her location of analyses only in East Java, which is located in the same island, she does not face neighborless states problem. Hence, first-order rook contiguity matrix can be applied directly. Her finding is spatial interrelation is important part on regional convergence analyses. In nation-wide scope, Wandani and Yoshida (2013) evaluate the existence of spatial interdependency on national road traffic. They define the spatial weight matrix by using queen contiguity approach. For cities whose centers are not more than 100 km apart, a pair-of-cities is categorized as neighbor and not otherwise. They do not use first-order contiguity matrix because there are many small states that do not share the same border but close to each other. It makes sense in term of land transportation since automobile and motorcycle able to transport cross border in Indonesian main islands as long as in the threshold range. This approach, however, could not appropriate to evaluate an island state, which is why they do not taken into account the neighborless island states. Their findings are automobile trips exhibit spatial correlation among neighbors but motorcycle trips do not.
17
Vidyattama (2012), focusing in regional disparity, uses first-order rook contiguity criterion in which two states are considered as neighborhood even if there is only one connecting point as their shared boundary, such as corner to corner. For 23 neighborless states, located in independent island, are ignored from the model and assumed the impact does not significant. However he notes that other proper contiguity criteria should be formulated to capture spatial impact from those neighborless states. He concludes the distribution of the Human Development Index (HDI) confirms the spatial development gap in Indonesia. Widiastuti (2013) tests the relationship between economic development and entrepreneurship in Indonesia state level uses the modified rook contiguity matrix to cover neighborless island states. She reformulate Vidyattama’s weight matrix concept for neighborless states by manually evaluating sea transportation interconnection between island state and the nearest states. They are categorized as a neighborhood if there is a ship serving those states and not otherwise.
Under the previous studies, this paper applies modified first-order queen contiguity matrix W, with row normalized. For more precise analyses, this paper employes Indonesian map in 2010 published by Statistics Indonesia with the help of arcGIS software. For a state that has neighbor in the same island, the usual weight matrix definition is employed; 1 if they share the same border, and zero otherwise. For 27 island-states that do not share the in-land border, the weight matrix is adjusted manually by looking at sea transportation interconnection to the nearest neighbor (Widiastuti, 2013). If they are interconnected by sea transportation, it is defined as neighbor and not otherwise.
Methodology
This paper employs spatial econometrics procedures with the help of Stata statistical package software. In the first place, we run the OLS estimation in order to get the most fitted model from twenty-one variables. R squared, Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) are implemented in model selection. Heteroscedasticity and multicollinearity test are performed. The chosen OLS model is further used as the benchmark for spatial model.
To start spatial analyses, we define neighborhood relationship by weighting between locations of the weight matrix (W). The size of the matrix W is 491 x 491 that represents interaction among 491 states. The queen binary-contiguity-method is applied and is row normalized. Each element matrix J&=KL is weighed as 1 if they share the common border and 0 otherwise. Maximum Likelihood (ML)
Analyzing the spatial econometrics, we employ general-to-specific approach as proposed by Elhorst (2010). Departing from general Manski model, we test the possibility of the existence of spatial interaction effect. Stata allows us to get endogenous and error () interaction effect. For exogenous interaction effect (θ), the manual method is employed by multiplying the weight matrix with all exogenous variables (WX). Therefore, we can analyze the full Manski Model (see Table 2.2). In model selection, there are three specification tests can be applied; Wald tests, Likelihood ratio (LR) test, and Lagrange multiplier (LM) test (Verbeek, 2012). Those tests basically evaluate models that nested in other model. One model is considered nested in another if the first model can be generated by imposing restrictions on the parameters of the second. The null hypothesis is restricted and unrestricted model are the same, such that restricted model is better. Rejecting Ho means general model is more appropriate. Stata provides Wald test and LR test for MLE, and Wald test for GS2SLS.
Chapter 5
Results and Analyses
This chapter provides results of the analysis of state level education spending determinants. It begins with OLS regression results and is followed by spatial econometrics results. Some simulation is conducted to get the most parsimonious model. Brief conclusions will follow in the end of section.
The Ordinary Least Square (OLS) Estimation
Using OLS regression, we look for the most fitted model to explain state’s behavior in education spending, represented in per capita education deviation from 20 percent of total state’s budget (EDDEVCAP). The independent variables are clustered into four groups; namely demographic, economic, geographic, and politic. This paper simulates five models from general to specific to select the most parsimonious model. In the first place, Model 1 is consisted of broad range of variables as expected to influence state’s behavior. The Model 5 ends up with the fewest number of variables. (Appendix 3)
Age-specific density variable, then, is replaced by population density per km2 (POPDEN) variable. This reveals negative and significant coefficient to education deviation per capita although the magnitude is relatively small. Holding other things constant, additional one inhabitant per km2 reduces education deviation per capita by Rp18.5 – 22. The result confirms Verbina and Chodhury (2002) study that more populated state tends to reduce per capita education spending. The explanation is because of better developed infrastructure and/or economies of scale – as population density increases, state’s average cost for education purpose is likely to decline. Average schooling year (AVSCHOOL) gives significantly positive impact to education spending decision. Increasing one year average schooling leads to higher education deviation spending per person by Rp43,000 – 45,000, ceteris paribus. This means that the more year children attending school, the more government spending on education should be allocated. The last demographic variable is an urban population (URPOP). One thousand additional people in a particular urban area reduce education deviation spending by Rp104-148, ceteris paribus. Sufficient privately-owned schools in urban areas lessen state’s intervention in terms of provision of educational facilities (Chakrabarti and Joglekar, 2006).
To cover geographical condition, six broad areas are clustered: Sumatera, Java, Kalimantan, Sulawesi, Papua and Maluku, and Nusa Tenggara. Hence, five dummy variables are employed; i.e. dummy Sumatera (DSUM), dummy Java (DJAVA), dummy Kalimantan (DKAL), dummy Sulawesi (DSUL), and dummy Papua and Maluku (DPM). Nusa Tenggara area is set as the control to get rid of perfect multicollinearity problem. DSUM, DJAVA, and DPM are statistically significant and negatively correlated. This informs that the deviations from education spending in those three areas are lower than others. Papua and Maluku areas are among the lowest deviation. Underdeveloped circumstances could be the reasons why the spending in education in these area very small, even some of states have not complied 20 percent national rules. Dummy municipality (DMUNI) is introduced to see municipality and regency difference. However in our analyses this variable is not significant. Area size in km2 (AREA) reveals negatively and statistically significant toward education deviation. Holding other things constant, the larger area by 1 km2, the lower deviation on education spending per capita approximately by Rp7.5 due to economic of scale reason.
Rp110,000 to 117,000, ceteris paribus. For DBH, the coefficient is slightly lower than other revenue around Rp55,000. The magnitude of DAU coefficient is 0.036; one million rupiah additional revenue from DAU increases education deviation spending by Rp36,000. The positive signs of those three variables confirm the hypothesis that education is normal goods at the government’s perspective. Public works, administration, and health per capita spending are categorized as political variables due to the allocation for those spending should pass the state’s council. Although not significant, health spending per capita (HEALCAP) gives positive impact toward education spending per capita. It indicates that health spending has complementary effect with education. The other two variables, public works (PWCAP) and administration (ADMCAP) spending per capita are strongly significant at one percent confidence level for all models. The negative signs tell that those two sectors spending have substitution effect to education. The substitution degree is higher in administration spending rather than public works. The more spending on administration and/ or public works sector, the lower spending on education will be. Numerically, reducing one million rupiah spending per capita on administration affair will be translated to incline education spending per capita by two thousand rupiah. Reducing one million rupiah per capita spending on public works leads education spending to increase by Rp1,610.
Three model selection criterion are employed i.e. R2, Bayesian Information Criterion (BIC), and Akaike Information Criterion (AIC). Out of five models, the model 5 is relatively better than others under those criterions. Firstly, this model can explain the variation of education spending deviation per capita 87.47percent with only ten variables. Adding more variables only slightly increases R2. Secondly, this model has the lowest value of BIC among other models, although the AIC is not at the lowest level. Afterward, we will use the Model 5 as the benchmark for further spatial analyses. 78879:;<= 0.162O2.20P'Q<I<87 = 0.044;9G:SIIT=O 1.48P'VWX<I<=O
0.1138<Z=O 7.41P'\;X7;= 0.0368;W:;<= 0.0558^S:;<=
0.117IH:;<=O 0.161<:;<=O 0.205;8Z:;<= . (5.1)
