Provincial differences in skill premia : evidence for Indonesia

(1)

Provincial Differences in Skill Premia:

Evidence for Indonesia

Master Thesis

Supervised by Prof. Dr. Hessel Oosterbeek January 2018 Marco van Loenen Msc Economics (Development Economics) 10445358 Abstract

This paper examines provincial differences in returns to education and skill premia for Indonesia. The analysis aims to explain the existence of provincial differences in skill premia through differences in the supply and demand of labour. Finding a relation between provincial differences in skill premia and supply and demand factors confirms the supply and demand theory. For the analysis, five different time periods are used. The findings show evidence supporting the supply and demand theory, especially in the most recent periods. This research has both theoretical as practical implications. It contributes to the existing literature by showing the relationship between skill premia and supply and demand of labour on a provincial level. Understanding this relationship has practical implications for Indonesian policy makers who want to continue enhancing the quality of schooling. Following the obtained results more research is needed to explain the differences in results across periods and other factors that cause provincial differences.

(2)

Statement of originality

This document is written by student Marco van Loenen who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in

creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Introduction ... 4 2 Context ... 6 2.1 Indonesia on a National Level ... 6 2.1 Indonesia on a Provincial Level ... 7 3 Data ... 8 3.1 The Indonesian Family Life Survey ... 8 3.2 The Returns to Schooling ... 9 3.3 Supply & Demand of Labour ... 13 4 Empirical Method ... 14 4.1 The Estimation of the Returns to Schooling ... 14 4.2 Estimating Skill Premia Difference Caused by Differences in Supply and Demand of Labour ... 16 5 Results ... 19 5.1 Provincial Differences in the Returns to Schooling ... 19 5.2 Provincial Skill Premia Differences Resulting From Relative Differences in Supply & Demand ... 22 5.2.1 Results IFLS 1 ... 22 5.2.2 Results IFLS 2 ... 28 4.2.3 Results IFLS 3 ... 28 5.2.4 Results IFLS 4 ... 31 5.2.5 Results IFLS 5 ... 32 5.2.6 Overall Results ... 34 6 Conclusion ... 35 References ... 37 Appendix ... 39 Appendix A – Summary Statistics ... 39 Appendix B– Full Regression Returns to Schooling ... 44 Appendix C – Returns to Schooling, Sample Containing Outliers ... 49 Appendix D – Full Regressions, Marginal Returns to Schooling ... 52 Appendix E – Marginal Returns, Sample Containing Outliers ... 57

(4)

Introduction

Indonesia is the largest archipelago in the world counting over 17,000 islands. Due to its geographic nature, Indonesia is a diverse country with wide provincial differences. These are not only differences in race, religion and ecosystems but also in wealth, poverty and inequality. Due to its diversity and being a fast-growing economy, Indonesia has received a lot of attention in the field of economics over the past two decades. One of the topics concerning development economists is the return to schooling in Indonesia. More specifically, how profitable is schooling as an investment decision in Indonesia? This paper seeks to explain whether returns to schooling in Indonesia have declined between 1993 and 2014 and most importantly, whether there are provincial differences in the average and marginal returns1

to schooling. In addition, this paper analyses if the differences in skill premia are attributable to provincial differences in the supply and demand of labour.

Indonesian poverty rates differ widely between provinces. In 2015 Papua, the province having the highest poverty rate nearly 30% of its population lived below the national poverty line. While in DKI Jakarta, the province having the lowest poverty rate, less than 4% of the population lived under the national poverty line (Poverty and Livelihood Map of Indonesia 2015). Besides poverty, Indonesia also suffers from severe income inequality. The Indonesian GINI coefficient equalled 0.407 in 2017, showing high-income inequality (Gini Ratio Provinsi 2002-2017). Besides the ethical concerns regarding poverty and inequality both factors also impede economic growth (Ravallion, 2016). It is therefore important for poverty and inequality, these to be alleviated in Indonesia. Differences in skill premia can enhance or reduce inequality both within and between provinces. Thus, understanding how the differences in skill premia arise can be used as a tool by Indonesian policymakers to reduce both poverty and inequality.

A number of papers attempt to estimate the Indonesian returns to schooling by using different methodological approaches. The most basic approach is called the Mincer equation. This Ordinary Least Squares (OLS) regression estimates the average return to schooling. The equation does not allow for differing returns across schooling levels. Purnastuti et al. (2012) find that in 2014 the return in Indonesia equalled 5.5%. Another paper using a similar approach for 2004 finds returns of 10.3% (Comola & Mello, 2010), almost double than what found by Purnastuti et al. This is an indication of declining returns to schooling in Indonesia. In another paper written by Purnastuti

et al. (2013) they claim to confirm this declining trend. This research uses a slightly

(5)

years of schooling variable by a dummy variable indicating the schooling level. By doing so the estimated coefficients provide information on the skill premia of different levels of schooling. More precisely, the estimated coefficients provide information on the differences in the returns between schooling levels2_{, not what the}

returns actually are. Although Purnastuti et al. (2013) claim to find evidence for declining returns, what they find in reality are declining skill premia. The use of the Mincerian equation has raised concerns regarding endogeneity and ability bias in the estimated returns the schooling. This concern has motivated Duflo (2001) to estimate the returns to schooling by using an Instrumental Variables (IV) approach. She exploited an exogenous variation in a large school building program in Indonesia. Her results find little difference between the OLS and IV estimations of the returns to schooling in Indonesia. Finally, the returns to schooling in Indonesia have been estimated by using quantile regression. The aim of this methodological approach is to find whether the rates vary in the conditional distribution of earnings. Sohn (2013) finds that these vary only little across the conditional distribution of earnings.

To explain changes in the relative wages between skill groups during 1963 to 1987 in the U.S. Katz and Murphy (1992) developed a supply and demand of labour model. They find evidence of supply and demand factors having an influence on relative wages confirming the supply and demand theory. Leuven et al. (2004) confirm these results when using their application of the supply and demand model to estimate international differences in male wage differentials. For the same research Blau and Kahn (2001) find contradicting results. Leuven et al. (2004) argue that the reason behind differing results is that Blau and Kahn do not use an internationally comparable measure of skill. In a more simplistic analysis Rojas3

(2006) decomposes between and within industry effects of the supply and demand model to explain changes in the skill premia in Mexico between 1991 and 1993. His results find evidence confirming the supply and demand theory.

This paper contributes to the existing literature in the following ways. First, it analyses five different periods over time. By doing so this paper aims to identify a trend and generalize the findings to future periods. Only Purnastuti et al. (2013) have analysed multiple Indonesian periods, though only two. Evidence from two different points in time is insufficient to speak of a trend. Second, there is extensive research on the returns to schooling on a national level and international comparisons. However, no research has investigated provincial differences thus far. Not for Indonesia or any other country. Third, this research aims to explain provincial differences in the supply

2_{Note: the differences in returns indicate how much more or less a given schooling level earns relative} to a chosen baseline schooling level.

(6)

and demand of label as a determinant of provincial differences skill premia. Again, the supply and demand model has been used to make national and international but not provincial comparisons.

The results show a declining trend in both the returns to schooling and skill premia in Indonesia as well as large provincial differences in the average returns to schooling and skill premia. In the first analysed period no evidence is found explaining provincial differences in skill premia by differences in the supply and demand of labour. In the subsequent three analysed periods there is evidence explaining provincial differences in skill premia through differences in the supply and demand of labour, however, only explaining the differences of one out of two analysed skill premia (the skill premium that is explained by the analysis is differing between the periods). In addition, the relation between the provincial differences in the skill premium and differences in supply and demand of labour in the second and third period is weak. In the fourth period, the found relationship is stronger. It is only for the fifth analysed period that a relationship is found between the provincial differences in skill premia and supply and demand of labour for both analysed skill premia. Similar as the result in the fourth period this relationship is increasing in strength compared to the previous periods.

The outline of this thesis is as follows. The next section will provide background characteristics of Indonesia on a national and provincial level. Section 3 presents and describes the data that is used in this research. Section 4 introduces the methodologies used to estimate the returns to schooling and the skill premia and for the supply and demand of labour analysis respectively. In section 5 the results of the research will be presented and discussed. Finally, section 6 concludes.

2 Context

In this section background information is presented regarding Indonesia on a national and provincial level respectively.

2.1 Indonesia on a National Level

Since 1990 the Indonesian Gross Domestic Product (GDP) increased from approximately $113 billion dollars to approximately $932 billion dollars in 2016. During the same period the GDP per capita has increased from $623 dollars to $3,570 dollars (GDP per capita (current US$), 2018). This has moved Indonesia from a developing towards emerging economy. This growth might sound like a fairy-tale,

(7)

however, this growth was not without setbacks and difficulties. The Indonesian economy collapsed in 1998 due to the Asian financial crisis. Resulting in inflation rates reaching a maximum of 72% in 1998 before slowing down to 2% in 1999. It took the economy until 2004 to fully recover from this collapse. Agriculture is the most import sector in Indonesia.

Since the 20th_{century the Indonesian government has put emphasis on}

increasing the years of schooling of its population. In 1973 the Indonesian government launched a major school construction program called the Sekolah Dasar INPRES program to increase enrolment rates. Between 1973 and 1979 more than 61,000 primary schools were constructed. By 1984 the government introduced a legislation making 6 years of primary school compulsory for children. This has led to substantially increased enrolment rates in primary schooling. In 1994 the government introduced another compulsory schooling legislation. The so-called Nine-Year Basic Education Program extended compulsory education to those aged between 13 and 15. More recently efforts from the Indonesian government move from increasing the quantity and access to education towards enhancing the quality. These efforts resulted in the launch of Bantuan Operational Sekolah program in 2005. The program aims to ensure that schools have sufficient funds to operate, reduce the education costs faced by households and improve school-based management. Nonetheless, despite all these efforts, Indonesian students had the second lowest test results out of 65 countries in 20124

. Since then the Indonesian government has made additional efforts to improve the quality of its education by mandating 20% of its yearly national budget (Clark, 2014).

2.1 Indonesia on a Provincial Level

Indonesia has 34 provinces counting a total of more than 17,000 islands (though mostly inhabited). Each province is different in terms of geographic location/sizes, distribution of religions and economic structures. The Indonesian population is mainly Muslim5_{(90%). Although most of the provinces are in majority Muslim, there are a}

lot of different forms of Islam practised between provinces. There are, however, pockets of Christians scattered throughout the country. Particularly in Flores, Timor, northern Celebes, the interior of Kalimantan and the Moluccas. Most of these Christians are protestant. A small part of the Indonesian population is Hindu. Bali is the only Indonesian province in which the primary religion is not Muslim but Hindu. More than half of Indonesia’s population lives on the island of Java. The second largest part of the population lives on the island of Sumatra. Each island widely varies in their population sizes, the smallest being South Kalimantan. Indonesian provinces

4_{The test was the International Student Assement (PISA) developed by the Organisation for Economic} Cooperation and Development’s (OECD).

5_{Forms of Indonesian Islam are different from the Middle-Eastern Islam and have Buddhism and} Hinduism influences.

(8)

have varying poverty rates ranging from nearly 4% to nearly 30% (Poverty and Livelihood Map of Indonesia 2015).

Economic sectors have different importance across differing provinces. Hariyanti and Utha (2016) have analysed which sector contributes the most per province. They find that agriculture is most important in the majority of the provinces, followed by the mining and manufacturing sector in most of the remaining provinces. Nonetheless, some provinces have another sector that contributes most. In DKI Jakarta finance and services is the most important sector, while on the island of Bali tourism is the most important source of income. While it is not the aim to specifically explain which sector is most important in which province, this comes to show that the economic dependency of sectors widely varies between provinces.

3 Data

This section presents and provides information on the data used for the analyses. First the source of the data is explained. Followed by an explanation of the data used to estimate the returns to schooling. Finally, the data used to perform the supply and demand analysis is provided and presented.

3.1 The Indonesian Family Life Survey

The Indonesian Family Life Survey (IFLS) is an on-going longitudinal household and community survey that collects information from households including their

Figure 1. Indonesian Provinces Covered by IFLS

consumption behaviour, income and assets. It also collects information on individuals including their education and labour market information. The sample is representative

(9)

of 83% of the Indonesian population and covers 13 of the 37 Indonesian provinces. These provinces are on the islands of Java, Sumatra, Bali, West Nusa Tenggara, Kalimantan and Sulawesi. Figure 1 shows a map of Indonesia indicating the provinces included in the IFLS. The survey began in 1993, since then four follow up surveys have been conducted6

. Each survey wave is conducted by RAND in collaboration with differing institutions7_{. The IFLS surveys and their procedures are properly}

reviewed and approved by the Institutional Review Boards in the United States and their collaborators. The IFLS data can be used as panel data. Nevertheless, this research uses the data as cross-sectional.

3.2 The Returns to Schooling

To analyse the returns to schooling the sample is restricted to workers aged 15-65, who were not full-time students and who reported net labour income. This is especially important for IFLS 1 and IFLS 2 where a portion of the survey respondents reported gross income. Having two different measures of income negatively influences comparability across respondents. In addition, respondents that did not report full schooling information were dropped from the sample. The dependent variable in this analysis is the natural logarithm of the hourly earnings8_{. The unit of}

measurement is the Indonesian Rupiah, which is Indonesia’s national currency. Although the construction of the hourly wage can create measurement error9

the more precise reflection of productivity through hourly earnings creates the preference to use hourly instead of monthly earnings. When constructing the hourly wages the data shows large outliers on both sides of the wage distribution. Further analysing the source of these outliers shows that some individuals have more hours worked per week reported than existent in a full week or have less than one hour worked per week10

. Due to these outliers, the mean wages of lower schooling levels are higher than those of more advanced schooling levels11

. This is highly improbable and so it is assumed that these values are either due to misreporting or data entry errors. Figure 2 presents the boxplots of the hourly wages from both IFLS 1 and IFLS 2 to show the magnitude of the outliers.

6_{IFLS 1 was conducted in 1993, IFLS 2 in 1997, IFLS 3 in 2000, IFLS 4 in 2007 and IFLS 5 in 2014} 7_{The first two periods are in collaboration with University of Indonesia. Starting from the third period} the collaboration is with the University of Gadjah Mada. In the fifth period Survey METRE

collaborates as well.

8_{The hourly wage is constructed by dividing annual (net earnings) by the normal hours worked per} week times the weeks worked during the year.

9_{Respondents were asked what their last monthly income was, how many hours they normally work} per week and how many weeks they worked last year.

10_{Note that this variable denotes the amount of hours worked in a normal week.}

11_{This paper distinguishes three different schooling levels, 1) primary schooling, 2) secondary} schooling and 3) tertiary schooling. These are discussed further in the paper.

(10)

Individuals having an hourly wage in the outlier range are omitted from the sample. To determine the outlier cut-offs the Tukey method is used12

. I am aware of the fact that it is arbitrary and subject to discussion to remove outliers. The estimated coefficients of the returns to schooling are hardly affected by outliers but these outliers can have influence on the supply and demand analysis. As explained further in this paper this analysis relies on the individual’s hours worked. Even though it is arbitrary and subject to discussion to remove outliers, it is therefore important to remove these obvious irregularities from the data. Figure 3 shows the boxplots of the hourly wages when the outliers are omitted from the sample.

Table 1 reports the summary statistics for the variables used to estimate the returns to schooling. For each variable the first row represents the Indonesian average followed by each of the provinces included in the IFLS. For each survey period the first column represents the mean. The second column represents the standard deviation. The mean of the hourly earnings increased by more than ten times during the full period. This increase is partly explained by the extreme inflation rates during the economic collapse but primarily by the high GDP growth per capita during the past two decades. Furthermore, the years of schooling have steadily increased. As well as the female participation rate which, slightly increased from 33% to 40% on average. The experience variable remained stable over time. Which is not surprising since it mostly depends on the respondent’s age. There are three main characteristics of the data that stand out. First the sample sizes differ widely across survey periods. The number of observations grows rapidly after the second survey period. Although the number of households in the survey increased over the years, the increase in sample size is mostly due to more complete information of respondents. Resulting in less individuals being omitted from the sample due to incomplete information. This brings the second point to attention. The means of the variables show a trend over the years. Except for the means reported in the second period. By looking at table 1, one can see that the sample from this period is different than that of the others. The average Indonesian years of schooling increased by 1.7 years in three years time, which is highly unlikely. Especially since this value drops again in the next survey period. Furthermore, the experience variable shows a large decline in mean value. Considering the fact that this variable is mostly dependent on the individuals’ age this is again highly unlikely13

. In this survey period a large part of the sample was dropped due to incomplete information. This has significantly changed the distribution of

12_{The Tukey method trims the lower and upper part of the distribution from the data. It removes the} values below the 25th_{percentile minus 1.5 times the interquartile range (IQR) and the values above the} 75th_{percentile plus 1.5 times the IQR.}

13_{This mean value would implicate that the average age of workers would drop in three years after} which it would increase again. This drop would be due to much lower retirement ages or largely decreased life expectancies. Both of which are unlikely.

(11)

Table 1. Summary Statistics – Data For the Estimation of the Returns to Schooling

IFLS 1 IFLS 2 IFLS 3 IFLS 4 IFLS 5

Mean SD Mean SD Mean SD Mean SD Mean SD

Dependent Variable Hourly wage Indonesia 751 770 1,014 933 1,807 1,788 4,215 4,307 8,855 8,777 North Sumatra 733 624 1,058 924 1,753 1,520 4,424 4,220 9,180 8,945 West Sumatra 875 854 963 902 1,830 1,752 4,909 4,466 10,096 9,854 South Sumatra 812 832 959 989 1,749 1,852 4,569 4,572 8,172 8,366 Lampung 361 390 834 736 1,520 1,388 3,442 3,493 7,653 7,625 D.K.I Jakarta 1186 959 1,466 1,013 2,363 2,112 5,894 5,255 11,731 9,217 West Java 799 761 1,207 1,022 2,042 1,969 4,526 4,445 9,840 9,576 Central Java 587 691 804 825 1,476 1,445 3,220 3,593 7,038 7,890 DI Jogyakarta 792 852 1,045 985 1,815 2,032 4,301 4,605 7,991 8,807 East Java 595 625 874 781 1,593 1,515 3,879 3,839 7,972 7,590 Bali 696 634 819 835 1,889 1,999 4,364 4,808 9,558 9,060 West Nusa Tenggara 618 676 766 797 1,604 1,614 3,452 3,788 8,736 8,752 South Kalimantan 803 840 983 791 1,878 1,530 4,861 4,084 9,762 9,088 South Sulawesi 671 749 688 831 1,744 2,013 3,721 4,390 9,201 8,751 Treatment Variable Years of schooling Indonesia 7.4 4.1 9.1 3.9 8.10 4.22 8.9 4.3 9.6 4.3 North Sumatra 7.6 3.8 9.2 3.6 8.2 3.9 9.2 3.9 10.1 3.9 West Sumatra 7.7 4.1 9.4 3.7 8.4 4.0 9.4 4.1 10.2 4.4 South Sumatra 7.7 4.0 8.7 4.1 8.0 4.1 8.3 4.5 8.8 4.5 Lampung 5.5 3.9 7.9 3.2 6.8 3.6 7.9 3.8 8.4 3.9 D.K.I. Jakarta 8.9 4.0 10.6 3.5 9.6 4.1 10.3 3.9 10.5 3.6 West Java 7.5 4.0 9.2 3.9 8.3 4.2 9.1 4.2 9.4 4.2 Central Java 6.7 3.8 8.4 4.0 7.3 4.1 8.2 4.3 8.9 4.3 DI Jogyakarta 8.6 4.4 10.4 3.9 9.3 4.3 10.4 4.3 10.9 4.1 East Java 6.8 4.0 8.6 3.8 7.6 4.1 8.6 4.3 9.2 4.3 Bali 7.2 4.3 8.7 3.9 8.3 4.3 8.9 4.5 9.8 4.5 West Nusa Tenggara 7.3 4.4 8.2 3.9 7.3 4.4 8.4 4.8 9.9 4.9 South Kalimantan 7.2 4.4 8.6 3.8 7.3 4.3 8.3 4.6 9.0 4.7 South Sulawesi 7.5 4.3 8.1 4.1 8.1 4.2 8.7 4.5 9.6 4.6 Control Variables Experience Indonesia 27.3 8.8 18.2 8.4 27.6 14.0 27.8 14.0 28 14 North Sumatra 26.6 8.9 19.3 9.6 30.0 14.0 27.9 14.2 27.2 13.9 West Sumatra 29.0 9.4 18.0 8.2 28.5 14.4 28.1 14.4 27.9 14.0 South Sumatra 28.0 8.7 18.8 8.6 28.4 13.6 27.8 13.8 27.7 13.1 Lampung 29.3 9.0 17.9 8.6 28.4 13.5 28.2 13.8 29.2 13.9 D.K.I. Jakarta 25.5 8.9 15.7 8.1 23.5 13.4 23.7 13.2 25.0 13.1

(12)

West Java 27.1 8.6 17.9 8.3 26.2 13.5 26.4 13.2 27.9 13.7 Central Java 28.5 8.2 18.2 8.0 29.1 14.5 29.9 14.6 30.0 14.4 DI Jogyakarta 26.8 8.8 19.5 8.5 29.4 14.9 28.6 14.5 29.6 14.6 East Java 27.7 8.1 19.6 8.4 28.5 13.8 28.6 13.8 29.5 14.5 Bali 27.0 9.7 18.3 8.0 26.5 13.7 27.2 14.1 27.2 14.0 West Nusa Tenggara 26.2 9.1 17.3 8.1 27.5 14.3 28.2 14.1 26.2 13.5 South Kalimantan 26.4 9.0 17.9 9.1 28.3 14.2 28.5 14.4 28.5 13.9 South Sulawesi 27.6 9.4 19.7 9.1 26.2 13.4 27.0 13.3 26.8 13.4 Female Indonesia 0.33 0.37 0.36 0.36 0.40 North Sumatra 0.31 0.34 0.39 0.38 0.38 West Sumatra 0.46 0.43 0.42 0.41 0.40 South Sumatra 0.32 0.28 0.32 0.27 0.31 Lampung 0.23 0.18 0.27 0.25 0.31 D.K.I. Jakarta 0.28 0.45 0.40 0.38 0.39 West Java 0.29 0.34 0.34 0.34 0.39 Central Java 0.36 0.42 0.39 0.40 0.44 DI Jogyakarta 0.44 0.42 0.43 0.41 0.42 East Java 0.34 0.31 0.35 0.36 0.41 Bali 0.34 0.41 0.36 0.40 0.41

West Nusa Tenggara 0.32 0.36 0.32 0.34 0.37

South Kalimantan 0.28 0.28 0.31 0.32 0.39 South Sulawesi 0.34 0.40 0.35 0.35 0.39 Number of observations Indonesia 3,383 2,881 11,066 12,855 15,978 North Sumatra 245 114 682 844 1,286 West Sumatra 188 140 496 662 839 South Sumatra 130 90 481 475 761 Lampung 87 76 401 563 712 D.K.I. Jakarta 378 332 1,122 1,053 1,108 West Java 460 585 1,928 1,993 2,257 Central Java 401 455 1,544 1,820 2,250 DI Jogyakarta 218 210 729 810 1,006 East Java 525 365 1569 1,983 2,211 Bali 247 140 571 704 956

West Nusa Tenggara 193 138 608 750 1,152

South Kalimantan 172 118 491 619 687

(13)

Figure 2. Boxplots of the Hourly Wages Containing Outliers

Figure 3. Boxplots of the Hourly Wages Without Outliers

characteristics of the sample. The third characteristic of the data that stands out is the provincial differences in mean values of the variables. For the hourly wage the provincial differences remain stable over time. Whereas the provincial differences in the years of schooling are declining over time. The provincial differences in female participation rates also decline over time. Sample sizes differ widely between provinces, which is not surprising regarding differences in population sizes.

3.3 Supply & Demand of Labour

To analyse the differences of supply and demand of labour between provinces the sample is the same as for the analysis of the returns to schooling. The analysis differs in the variables of interest. Appendix A presents the distribution of the highest

0 20000 40000 60000

Primary Secondary Tertiary Education Level

Hourly w

age

IFLS 1 − with outliers

0 30000 60000 90000

Primary Secondary Tertiary

Education Level

Hourly w

age

IFLS 2 − With Outliers

0 2000 4000 6000

Primary Secondary Tertiary Education Level

Hourly w

age

IFLS 1 − without outliers

0 2000 4000 6000

Primary Secondary Tertiary

Education Level

Hourly w

age

(14)

obtained schooling levels14

. Consistent with table 1 there are large provincial differences in the sample. In addition, Appendix A has the same disrupting pattern in IFLS 2 that does not follow the trend of the remaining periods. The portion of the sample having attained primary school as the highest level has decreased over time. Consequently, increased portions of the sample have attained secondary or tertiary schooling. In short, not only did the average years of schooling increase over time (resulting from table 1), the obtained level of schooling has also considerably increased.

Appendix A presents the employment shares of 6 different industry groups15

in Indonesia per province. The share of employment in agriculture has remained stable over time. This is consistent with the fact that agriculture is the most important sector in Indonesia.The fact that D.K.I. Jakarta has a virtually non-existent employment rate in the agricultural sector is not surprising considering that it is a fully urban province. The shares of employment in the Mining, Manufacturing and Construction industry have decreased on average, except for the provinces of Lampung and South Kalimantan. The finance, insurance, real estate and services industry group has seen a large rise in employment shares. Indicating that Indonesia is moving towards a service-oriented economy. Consistent with previously discussed tables, the IFLS 2 data shows a disrupting pattern. Furthermore, there are large provincial differences, indicating that the structure of provincial economies differs widely between regions.

4 Empirical Method

This section provides details about the empirical methods used. First, it provides the methodology to estimate the returns to schooling. Followed by the method used to estimate the skill premia and for the supply and demand analysis.

4.1 The Estimation of the Returns to Schooling

The empirical analysis to estimate the returns to schooling is based on the human capital earnings equation developed by Jacob Mincer16

(1973). This equation involves

14_{The data aggregates the Indonesian schooling grades in three different schooling levels: 1) primary} schooling, 2) secondary schooling and 3) tertiary schooling (college or higher)

15_{These are aggregated categories constructed from available employment data in the IFLS. These} groups are: 1) agriculture, 2) mining, manufacturing and construction, 3) transportation,

communication and public utilities, 4) trade, 5) finance, insurance, real estate and services, and 6) government.

16_{Although the human capital earnings function is often referred to as the Mincer equation other} authors have been important in the development. The human capital earnings function is a benchmark model from the human capital theory developed by Gary Becker (1964).

(15)

fitting an Ordinary Least Squares (OLS) regression by using the natural logarithm of earnings as the dependent variable, and years of schooling, potential years of labour market experience and its square as independent variables. In this research the model is augmented with a female dummy to capture gender differences in earnings. The equation17

is written as follows:

(1) ln 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠! = 𝛽!+ 𝛽!𝑠𝑐ℎ𝑜𝑜𝑙! + 𝛽!𝑒𝑥𝑝! + 𝛽!𝑒𝑥𝑝!!+ 𝛽!𝑓𝑒𝑚_! + 𝜀!

where 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠! is the constructed hourly wage variable of individual i, 𝑠𝑐ℎ𝑜𝑜𝑙!

denotes the number of completed years of schooling by individual i, 𝑒𝑥𝑝_! is an experience measure constructed by the individual i’s age minus the completed years of schooling, 𝑓𝑒𝑚! is a dummy variable indicating whether individual i is a female

and 𝜀! is the error term of individual i. In this equation the coefficient on the years of

schooling can be interpreted as the average private rate of return to one additional year of schooling. Regardless of the educational level to which this year of schooling refers. In other words, the Mincer equation doesn’t allow for different rates of returns to schooling for different schooling levels. Note that each province and each wave is estimated separately.

A potential limitation of the Mincer equation is the existence of endogeneity, creating a bias in the results18

. Harmon, Oosterbeek and Walker, Psacharopoulos and Purnastuti, Miller and Salim (2003; 1994; 2012) have investigated the existence of ability bias and find that it is virtually inexistent. A similar conclusion is found by Card (1999). Other research has tried to identify the magnitude of omitted variable bias by adding a wide range of control variables. Leigh and Ryan (2008) find considerably lower coefficients when estimating the schooling variable using control variables. Indicating an upward bias in the schooling coefficients in the Mincer equation. Meanwhile, other research suggest that adding control variables has a negligible effect on the estimation of the return to schooling (Harmon, Oosterbeek & Walker, 2003; Brauw & Rozelle 2008). Moreover, added control variables have the potential to be bad controls. Occupational status, marital status or regions of residence variables19

are likely to be affected by the level of schooling. Another approach to eliminate the possibility of biased estimates is used by Duflo (2001) in a well-known research. She exploited an exogenous variation in a large school building program in Indonesia. Her obtained Instrumental Variables (IV) estimations are very close to the OLS estimation.

17_{Note: Each period is analysed separately.}

18_{The equation suffers from omitted variable bias and ability bias. Ability bias stems from the fact that} people with greater ability tend to get more years of schooling. Thus the schooling coefficient partly captures unobserved ability resulting in an upward bias. The omitted variable bias stems from the fact that there are a lot of factors influencing income, not only the ones included in the Mincer equation. 19_{These are commonly used control variables in the Mincer equation.}

(16)

Evidence suggests that a bias in the estimate of the schooling coefficient is, if existent, small when using the Mincer equation. And so, the OLS estimates provide a good guide to the true returns to schooling. Moreover, this paper seeks to capture provincial differences in Indonesian returns to schooling. The goal is not to obtain the most precise measure of returns. In addition, the bias is assumed to be time-invariant and constant across regions. Resulting in the same proportion of bias in each of the estimated returns. Making these returns comparable between provinces and over time.

4.2 Estimating Skill Premia Difference Caused by Differences in Supply and Demand of Labour

To investigate whether provincial differences in skill premia are caused by differences in the supply and demand of labour this analysis uses the application of Leuven et al. (2004) on the model developed by Katz and Murphy (1992). This model constructs supply and demand indexes by skill group for each province relative to a base province. The three different skill groups are the three aggregated schooling levels. To estimate the skill premia of the schooling levels the Mincer equation is adjusted by changing the years of schooling variable to a dummy variable indicating the schooling level. The equation is written as follows:

(2) ln 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠! = 𝛽!+ !𝛽!!𝑙𝑒𝑣𝑒𝑙!"+ 𝛽!𝑒𝑥𝑝!+ 𝛽!𝑒𝑥𝑝!! + 𝛽!𝑓𝑒𝑚! + 𝜀!

where 𝑙𝑒𝑣𝑒𝑙!" is the schooling level dummy, indicating the schooling level of

individual i. The rest of the variables are the same as in equation (1). The estimated dummy coefficients will denote the skill premia of secondary and tertiary schooling20_.

Thus equation (2) allows me to compare two different sets of schooling levels, 1) primary vs. secondary schooling and 2) primary vs. tertiary schooling.

To estimate the relative demand of schooling level k in province j industry occupation cells are constructed21

. The demand index, ln 1 + 𝑑!,! , of each province

relative to the baseline province for each skill group is given by:

(3) 𝑑!,! = 𝑐!" !!_! !

!,!

!

o refers to the occupation-industry cell, 𝑐_!" is schooling level k’s share of employment in occupation-industry cell o in the base province, Δ𝐸! is the sum of

20_{Note: the skill premia of the secondary and tertiary schooling levels are the marginal returns of the} secondary and tertiary schooling relative to primary schooling, respectively.

21_{Occupations are aggregated in three different groups: 1) managers and professionals, 2) clerical and} sales workers and 3) craft workers, operatives labourers and service workers.

(17)

differences in shares of total labour input employed in cell o between province j and the base province, and 𝐸! is the share of total labour input accounted for by schooling

level k in the baseline province. The demand index measures the degree to which the occupation-industry structure favours schooling level k in province j relative to the base province.

The supply index for schooling level k in province j is given by:

(4) 𝑠!,! = ln

!!,! !!,!

where 𝐸!,! and 𝐸!,! are the shares of total labour input consisting of schooling level k

in province j and the baseline province b, respectively. The supply indexes express the relative share of each schooling level in a province’s labour force relative to the baseline province. The net supply index for schooling level k in province j is given by combining the demand and supply indexes. The equation is given by:

(5) 𝑛𝑠!,! = 𝑠!,!− 𝑑!,!.

The net supply of secondary or tertiary schooling groups k relative to the primary schooling group p in province j is given by22

:

(6) 𝑛𝑠!,!− 𝑛𝑠!,!.

The skill premium of schooling level k in province j is defined by 𝑠𝑝! !,!. The skill

premium of the schooling level k in province j relative to the skill premium of schooling level k in the baseline province is given by:

(7) 𝑠𝑝! !,!− 𝑠𝑝! !,!.

Equations (6) and (7) can now be compared. According to the supply and demand theory both equations should be negatively correlated. More precisely, when the net supply of labour in the secondary schooling group relative to the primary schooling group in province j is larger (smaller) then in the base province then the model predicts that the skill premium of the secondary schooling group is smaller (larger) than in the base province. When plotting equation (6) against (7) each combination is expected to lie in either the second or fourth quadrant. The number of possible combinations differs per IFLS2324

. The supply and demand theory implies that a

22_{Note: Until equation (6) the primary schooling group is also denoted by k. For clarity purposes this} schooling group is denoted by p starting from equation (6)

23_{When estimating the marginal returns to schooling some coefficients were imprecisely measured. It} is not possible to compare coefficients when there is no form of certainty what these are. Only

(18)

regression line through the combinations should have a negative slope. This hypothesis is tested by the following regression equation

(10) 𝑠𝑝_{𝑘 !,!}− 𝑠𝑝! !,! = 𝛽!+ 𝛽! 𝑛𝑠!,!− 𝑛𝑠!,! + 𝜀!.

The estimated 𝛽_! coefficient is expected to be negative and indicates the level of responsiveness (elasticity) of the skill premium to changes relative net supply of the schooling level. The higher the absolute value of 𝛽! the more responsive the skill

premium of the schooling level is to the relative net supply of schooling level.

It is important that the measure of skills (in this paper the schooling levels) is comparable across provinces. The research of Blau and Kahn (1996) suffers from this lack of comparability25

. As proven by Leuven et al. (2004) this has a negative influence on the outcome of the analysis26

. Considering the fact that the schooling system is equal in each province this is not considered an issue for this analysis. Even if there are provincial differences in the quality of schooling across provinces, these are assumed negligible. The research of both Blau and Kahn (1996) and Leuven et al. (2004) determine their cut-off points for the skill groups based on the baseline country27

. The skill groups in the baseline countries each have the same relative size of one-third of the total labour input. To clarify, the baseline country determines which score level serves as the cut-off between the different skill groups. Consequently, the choice of the baseline country (which is arbitrarily chosen) influences the outcome of the analysis. This analysis does not suffer from this issue since the cut-off value for each schooling group is determined by the years of schooling and not by a value obtained in the baseline province. Nonetheless, the choice of the baseline province determines the benchmark for the occupation-industry structure, which can influence the results. In addition, due to the few possible combinations when plotting and/or regressing equation (6) to (7) the choice of the baseline province can alter the outcome of this analysis as well. Therefore this paper follows Leuven et al. (2004) and repeats the analysis 13 times, using each province as a baseline province to increase the number of combinations.

As explained by Leuven et al. using this model does not implicate that supply and demand factors are the sole determinants of provincial differences in skill premia.

coefficients with a significance level of 5% or lower are compared (most have a significance level of 1% or lower). The marginal returns (skill premia) are presented later.

24_{There are 13 provinces, one of which is the base province, making 12 possible combinations per} comparison group, there are two comparison groups resulting in a maximum of 24 combinations per full (both comparisons groups pooled) IFLS.

25_{Blau and Kahn (1996) have developed a regression equation to estimate a skill value for a given} person. This value is based on the years of schooling and potential experience. By doing so they wrongfully assume that years of schooling and experience have a similar influence on skill across different countries.

26_{To solve this issue faced by Blau and Kahn (1996) Leuven et al. (2004) have used IALS scores for} making internationally comparable skill groups.

(19)

It is also not possible to decompose the explanation of provincial differences in an institution and supply and demand effect as done by Blau and Kahn (2001) and Devroye and Freeman (2001). Though provincial institutional differences will be less strong than international differences in institutions. In addition, it is not possible to further decompose the explanation of provincial differences through supply and demand factors in within and between industry effects as done by Katz and Murphy (1992). This analysis only seeks to show whether there is a relation between the provincial differences in skill premia and relative differences of net supply of labour.

5 Results

This section presents and discusses the results found. The first part generally covers the estimated returns to schooling over all the periods simultaneously. The second part starts by briefly discussing the estimated skill premia of all the periods simultaneously after which it discusses the results of the supply and demand analysis separately for each period.

5.1 Provincial Differences in the Returns to Schooling

Table 2 presents the average returns to schooling for an additional year of schooling per province for each IFLS. For clarity purposes solely the estimated coefficient of the returns to schooling28

variable (including the standard error) is presented in the table. The full regression results for each IFLS are presented in Appendix B. Table 2 includes the overall Indonesian returns to schooling to compare the returns of the investigated provinces with the national average29

. All the estimated coefficients are significant at the 1% level. Except for Lampung in IFLS 1 and East Java in IFLS 5, which are significant at the 5% level. Overall the estimated coefficients found in this analysis for Indonesia as a whole are decreasing and higher than found in previous papers (Duflo, 2001; Purnastuti, Miller, & Salim, Ecnomic Returns to Schooling in a Less Developed Country: Evidence for Indonesia, 2012)30_{. This suggests the existence}

of an upward bias in the OLS results presented in Table 2. Anyhow, as previously discussed it is beyond the scope of this paper to precisely measure the returns to schooling. Figure 4 visually shows the differences per province and between periods. Table 2 shows large provincial differences in the returns to schooling. The highest and lowest returns in IFLS 1 differ as much as 11.6 percentage points.

28_{This is the “years of schooling” variable from equation (1).}

29_{The return to schooling for Indonesia is a weighted national average.} 30_{Patrinos et al. (2006) find similar results in their OLS estimation.}

(20)

Notes: Standard errors in parentheses. *significant at 10% level, ** significant at 5% level, *** significant at 1% level.

Table 2. Estimated Returns to Schooling

Indonesia 0.150 (0.004)*** 0.135 (0.004)*** 0.117 (0.002)*** 0.118 (0.002)*** 0.097 (0.002)*** North Sumatra 0.117 (0.017)*** 0.108 (0.022)*** 0.083 (0.010)*** 0.099 (0.011)*** 0.093 (0.009)*** West Sumatra 0.149 (0.019)*** 0.109 (0.021)*** 0.105 (0.012)*** 0.121 (0.011)*** 0.102 (0.010)*** South Sumatra 0.216 (0.028)*** 0.142 (0.028)*** 0.118 (0.012)*** 0.105 (0.011)*** 0.083 (0.010)*** Lampung 0.100 (0.041)** 0.114 (0.042)*** 0.101 (0.015)*** 0.107 (0.013)*** 0.073 (0.012)*** D.K.I. Jakarta 0.141 (0.011)*** 0.139 (0.011)*** 0.134 (0.007)*** 0.123 (0.008)*** 0.108 (0.008)*** West Java 0.163 (0.011)*** 0.133 (0.008)*** 0.128 (0.005)*** 0.129 (0.006)*** 0.128 (0.006)*** Central Java 0.170 (0.014)*** 0.143 (0.010)*** 0.101 (0.007)*** 0.113 (0.006)*** 0.102 (0.006)*** DI Jogyakarta 0.149 (0.017)*** 0.122 (0.016)*** 0.139 (0.010)*** 0.132 (0.010)*** 0.105 (0.009)*** East Java 0.125 (0.011)*** 0.132 (0.011)*** 0.101 (0.006)*** 0.120 (0.006)*** 0.091 (0.007)** Bali 0.116 (0.016)*** 0.121 (0.018)*** 0.141 (0.011)*** 0.128 (0.009)*** 0.113 (0.008)***

West Nusa Tenggara 0.157 (0.019)*** 0.122 (0.020)*** 0.130 (0.009)*** 0.116 (0.008)*** 0.079 (0.007)***

South Kalimantan 0.113 (0.018)*** 0.097 (0.022)*** 0.080 (0.010)*** 0.074 (0.009)*** 0.078 (0.009)***

(21)

Although the sign of the change in returns from one IFLS to another is varying for each province, the returns follow a decreasing trend over the years. Except for Bali, which has maintained a relatively stable return to schooling. From having one of the lowest in IFLS 1 Bali moved to having one of the highest in IFLS 5. Along with the overall decrease in returns the provincial percentage point differences have decreased as well.

The full regression results presented in Appendix B show no surprising facts in the estimated coefficients. Consistent with previous research on returns to schooling the female variable has a negative sign (Purnastuti, Miller, & Salim, 2012; Johnson & Chow, 1997). This indicates that female workers earn less than their male counterparts. This result must not be confused with female returns to schooling. Furthermore experience has a positive impact on the hourly wage. This impact is declining in strength over the years, which is indicated by the negative sign of the quadratic experience variable. Appendix C shows the estimated returns to schooling for the sample containing the outliers. The estimated coefficients differ slightly from the sample without outliers. Providing evidence that removing the outliers doesn’t have an important impact on the estimation of the returns to schooling.

Figure 4. Returns to Schooling Per Province and Per IFLS

0 0.05 0.1 0.15 0.2 0.25 IFLS 1 IFLS 2 IFLS 3 IFLS 4 IFLS 5

(22)

5.2 Provincial Skill Premia Differences Resulting From Relative Differences in Supply & Demand

Table 3 shows the skill premia of the secondary and tertiary schooling groups31

. These results are consistent with the results presented in table 2 (section 4.1). There are large differences in skill premia across provinces and these premia tend to decline over time, as well as the differences in the premia between provinces. The estimated coefficients are mostly significant at the 1% level32

. Appendix D shows the full results of the regression of equation (2). Appendix E presents the marginal returns for the sample containing outliers. Again, the differences in the estimated coefficients between the sample containing outliers and the sample of interest are small. The rest of this section analyses whether the provincial differences in skill premia are due to relative differences in supply and demand of labour for each IFLS separately.

5.2.1 Results IFLS 1

Figure 5 shows plots of the skill premium of schooling level k in province j relative to the skill premium of schooling level k in the baseline province against the relative net supply of schooling level k relative to the primary schooling level in province j33_{. The}

skill premia of the secondary and tertiary schooling levels are analysed. Resulting in two different groups, 1) secondary schooling versus primary schooling and 2) tertiary schooling versus primary schooling. According to the supply and demand theory the observations are expected to lie in the second or fourth quadrant of the plots. In addition, a regression line is included in the plots, which is expected to have a downward slope. The first plot shows the relation between the primary versus secondary schooling group. The second plot shows the relation between the primary versus tertiary schooling group. When a quadrant is randomly chosen the probability of a good prediction is 1/234

. This probability stays the same for each new combination. Resulting in the fact that the probability of k correct predictions out of n comparisons follows a binomial distribution. Table 4 presents the estimated coefficient when regressing the skill premium of schooling level k in province j relative the skill premium of schooling level k in the baseline province against the net

31_{Estimated by adjusting the Minder equation with schooling level dummies, equation (2)}

32_{Except for the marginal returns of secondary schooling for Lampung in IFLS 1 and IFLS 2 and for} the marginal returns of tertiary schooling for Lampung in IFLS 1 and DI Jogyakarta IFLS 5 which are insignificant. Furthermore, the marginal returns to schooling for DI Jogyakarta if IFLS 2 is only significant at the 10% level.

33_{Thus, plotting equation (6) against equation (7)}

34_{There are 4 quadrants. Each combination is expected to lie in either the second or the fourth} quadrant. Making combinations in 2 out 4 (equals ½) quadrants the right one (according to supply and demand theory).

(23)

Table 3. Provincial Skill Premia per IFLS

Indonesia Secondary schooling 0.864 (0.037)*** 0.624 (0.035)*** 0.557 (0.020)*** 0.492 (0.021)*** 0.383 (0.021)*** Tertiary schooling 1.791 (0.065)*** 1.597 (0.050)*** 1.520 (0.031)*** 1.487 (0.029)*** 1.175 (0.027)*** North Sumatra Secondary schooling 0.627 (0.127)*** 0.486 (0.170)*** 0.371 (0.079)*** 0.250 (0.088)*** 0.209 (0.080)*** Tertiary schooling 1.411 (0.284)*** 1.538 (0.297)*** 1.177 (0.152)*** 1.317 (0.131)*** 1.123 (0.105)*** West Sumatra Secondary schooling 0.674 (0.141)*** 0.499 (0.178)*** 0.468 (0.101)*** 0.374 (0.098)*** 0.244 (0.095)*** Tertiary schooling 1.867 (0.260)*** 1.644 (0.255)*** 1.499 (0.171)*** 1.564 (0.134)*** 1.241 (0.120)*** South Sumatra Secondary schooling 1.247 (0.212)*** 0.941 (0.234)*** 0.475 (0.102)*** 0.507 (0.101)*** 0.246 (0.089)*** Tertiary schooling 2.537 (0.390)*** 1.909 (0.343)*** 1.556 (0.164)*** 1.507 (0.146)*** 1.092 (0.130)*** Lampung Secondary schooling 0.438 (0.293) 0.432 (0.268) 0.530 (0.105)*** 0.497 (0.099)*** 0.184 (0.087)** Tertiary schooling 1.390 (0.669)** 1.645 (0.632) 1.536 (0.261)*** 1.494 (0.182)*** 1.255 (0.150)*** D.K.I. Jakarta Secondary schooling 0.785 (0.089)*** 0.778 (0.094)*** 0.678 (0.059)*** 0.417 (0.072)*** 0.404 (0.074)*** Tertiary schooling 1.609 (0.134)*** 1.518 (0.121)*** 1.583 (0.080)*** 1.417 (0.093)*** 1.166 (0.096)*** West Java Secondary schooling 1.012 (0.092)*** 0.637 (0.069)*** 0.697 (0.047)*** 0.613 (0.049)*** 0.580 (0.054)*** Tertiary schooling 1.819 (0.157)*** 1.500 (0.094)*** 1.560 (0.067)*** 1.504 (0.068)*** 1.426 (0.073)***

(24)

Notes: Standard errors in parentheses. *significant at 10% level, ** significant at 5% level, *** significant at 1% level. Central Java Secondary schooling 1.005 (0.113)*** 0.598 (0.088)*** 0.518 (0.056)*** 0.443 (0.053)*** 0.329 (0.056)*** Tertiary schooling 1.845 (0.265)*** 1.569 (0.127)*** 1.349 (0.094)*** 1.467 (0.075)*** 1.262 (0.077)*** DI Jogyakarta Secondary schooling 0.721 (0.148)*** 0.279 (0.147)* 0.525 (0.087)*** 0.369 (0.098)*** 0.038 (0.097) Tertiary schooling 1.902 (0.213)*** 1.357 (0.173)*** 1.611 (0.112)*** 1.418 (0.117)*** 0.924 (0.113)*** East Java Secondary schooling 0.714 (0.096)*** 0.449 (0.088)*** 0.407 (0.052)*** 0.470 (0.051)*** 0.474 (0.058)*** Tertiary schooling 1.684 (0.181)*** 1.668 (0.133)*** 1.424 (0.086)*** 1.505 (0.074)*** 1.078 (0.081)*** Bali Secondary schooling 0.719 (0.137)*** 0.742 (0.144)*** 0.778 (0.096)*** 0.706 (0.088)*** 0.501 (0.078)*** Tertiary schooling 1.148 (0.242)*** 1.623 (0.236)*** 1.789 (0.136)*** 1.686 (0.114)*** 1.367 (0.099)*** West Nusa Tenggara

Secondary schooling 1.093 (0.180)*** 0.348 (0.158)*** 0.618 (0.086)*** 0.446 (0.079)*** 0.441 (0.075)*** Tertiary schooling 2.046 (0.302)*** 1.667 (0.236)*** 1.632 (0.130)*** 1.411 (0.103)*** 1.033 (0.089)*** South Kalimantan Secondary schooling 0.697 (0.159)*** 0.315 (0.121)*** 0.343 (0.084)*** 0.288 (0.084)*** 0.364 (0.085)*** Tertiary schooling 1.622 (0.253)*** 1.185 (0.278)*** 1.145 (0.119)*** 1.069 (0.119)*** 1.148 (0.115)*** South Sulawesi Secondary schooling 0.879 (0.179)*** 0.688 (0.197)*** 0.675 (0.104)*** 0.562 (0.103)*** 0.451 (0.095)*** Tertiary schooling 1.797 (0.294)*** 1.337 (0.311)*** 1.746 (0.162)*** 1.761 (0.144)*** 1.246 (0.124)***

(25)

Figure 5. Skill Premium Plotted Against Relative Net Supply - IFLS 1 −1 0 1 −2 −1 0 1 2 Net Supply Relativ e Skill Premium

Primary vs. Secondary. 77 correct out of 132, p=0.034

−1 0 1 −2 −1 0 1 2 Net Supply Relativ e Skill Premium

Primary vs. Tertiary. 78 correct out of 156, p=0.53

(26)

Figure 6. Skill Premium Plotted Against Relative Net Supply - IFLS 2. −1 0 1 −2 −1 0 1 2 Net Supply Relativ e Skill Premium

Primary vs. Tertiary. 96 correct out of 156, p=0

(27)

supply of schooling level k relative to the primary schooling level in province j35

. The first plot in figure 5 shows that there are 77 good predictions out of 132 possible combinations. This probability of correct predictions when done randomly equals 0.034. This rejects the hypothesis that the provincial difference in skill premium of secondary schooling is unrelated to the relative net supply. Table 4 shows opposite evidence. Although the estimated coefficient has the expected negative sign it is insignificant. Following the regression result the hypothesis that the provincial difference in skill premium of secondary schooling is unrelated to the relative net supply is not rejected. According to the binomial distribution the probability that the hypothesis is rejected is .966. Nonetheless, the regression result shows that this relation is unrelated. The second plot in figure 5 shows a similar result. Presenting 78 correct predictions out of 156 possible combinations. This does not reject the hypothesis that the provincial difference in skill premium of tertiary schooling is unrelated to the relative net supply. Table 4 confirms this result. The estimated elasticity is even slightly positive, presenting an opposite view on the theory of supply and demand, though insignificant. The last plot pools all the possible combinations in IFLS 1 and finds a slight negative elasticity. Again, this elasticity is insignificant. Overall the results in IFLS 1 show no evidence supporting the supply and demand theory.

Table 4. Skill Premium Regressed Against the Relative Net Supply – IFLS 1

𝑛𝑠!,!− 𝑛𝑠!,! R-Squared

Secondary schooling -0.091 (0.104) 0.011

Tertiary schooling 0.011 (0.111) 0

Total -0.032 (0.081) 0.001

Notes: Robust standard that take clustering at provincial level into account are reported. * significant at 10% level,

** significant at 5% level, *** significant at 1% level.

35_{Table 5 presents the estimated 𝛽}

(28)

Table 5. Skill Premium Regressed Against the Relative Net Supply – IFLS2

𝑛𝑠_!,!− 𝑛𝑠_!,! R-Squared

Secondary schooling -0.012 (0.036) 0.001

Tertiary schooling -0.103 (0.029)*** 0.055

Total -0.072 (0.023) 0.024

Figure 5 shows the plots of the provincial difference in skill premia against the relative net supply when using the IFLS 2 sample. Table 5 shows the results of regression of the provincial difference in skill premia against the relative net supply for the IFLS 2 sample. The first plot in figure shows 52 good predictions out of 110 combinations. This probability of correct predictions when done randomly is 0.68. This does not reject the hypothesis that the provincial difference in the skill premium of secondary schooling is unrelated to the relative net supply. The estimated coefficients presented in table 5 confirm this. Although the coefficient has the expected negative sign it is not significant. The second plot in figure 3 shows a different point of view. There are 96 correct predictions out of 156. This rejects the hypothesis that the provincial difference in the skill premium of tertiary schooling is unrelated to the relative net supply. Table 5 confirms this result. The estimated elasticity is significant and has the expected negative sign. The last plot in figure 5 pools all the possible combinations from IFLS 2 and finds a slight, significant, negative elasticity. This result is confirmed in table 5. The results from IFLS 2 show that there is evidence supporting the supply and demand analysis. Though this is only supported through the secondary schooling skill premium.

Figure 7 shows the plots of the provincial differences in skill premia against the relative net supply when using the IFLS 3 sample. Table 6 shows the results of regressing the provincial differences in skill premia against the relative net supply for the IFLS 3 sample. The first plot in figure shows 91 good predictions out of 156 combinations. This probability of correct predictions when done randomly is 0.022. This rejects the hypothesis that the provincial difference in the skill premium of secondary schooling is unrelated to the relative net supply. This result is confirmed by the estimated coefficient shown in table 6. The second plot shows a different view by

(29)

−1 0 1

Relativ

e Skill Premium

(30)

Primary vs. Secondary. 100 correct out of 156, p=0

−1 0 1

Relativ

e Skill Premium

(31)

Secondary schooling -0.093 (0.035)*** 0.045

Tertiary schooling -0.017 (0.052) 0

Total -0.060 (0.036)* 0.011

showing that there are 82 good predictions out of 156 combinations. This does not reject the hypothesis that the provincial difference in skill premium of tertiary schooling is unrelated to the relative net supply. This evidence is supported by the elasticity reported in table 6. Overall there are 173 correct predications out of 312 combinations in IFLS 3. This rejects the hypothesis that the provincial difference in skill premia are unrelated to the relative net supply. Again this result is confirmed by the estimated coefficient in table 6, although this estimated elasticity is small and only significant at the 10% level.

Figure 8 shows the plots of the provincial differences in skill premia against the relative net supply when using the IFLS 4 sample. Table 7 shows the results of regressing the provincial differences in skill premia against the relative net supply for the IFLS 4 sample. The first plot in figure 5 shows 100 good predictions out of 156 combinations. This rejects the hypothesis that the provincial differences in the skill premium of secondary schooling are unrelated to the relative net supply. This result is confirmed by the estimated coefficient shown in table 7. In addition the variation in the relative skill premium is increasingly explained by the variation in net supply compared to the elasticity’s36

in previous sections. The second plot shows a different story by showing that there are 75 good predictions out of 156 combinations. This does not support the supply and demand theory. Table 7 confirms this result. The estimated coefficient is slightly negative but insignificant. Overall there are 175 correct predications out of 312 combinations in IFLS 4. This rejects the hypothesis that the provincial differences in skill premia are unrelated to the relative net supply. Again this result is confirmed by the estimated coefficient in table 7.

(32)

Table 7.Skill Premium Regressed Against the Relative Net Supply – IFLS 4.

Tertiary schooling -0.026 (0.043) 0.002

Total -0.100 (0.037)*** 0.048

Figure 9 shows the plots of the provincial differences in skill premia against the relative net supply when using the IFLS 5 sample. Table 9 shows the results of regressing differences in skill premia against the relative net supply for the IFLS 5 sample. The first plot in figure shows 100 good predictions out of 156 combinations. This supports the supply and demand theory. This result is confirmed by the estimated coefficient shown in table 8. Table 8 shows an elasticity of -0.123, which is considerably stronger than found in previous periods. The second plot also rejects the hypothesis that the provincial differences in the skill premium of tertiary schooling are unrelated to the relative net supply. The plot shows 105 correct predictions out 156 combinations. Table 8 confirms this by showing a negative and significant coefficient for the elasticity. In this case 29.3% of the variation in the skill premium is explained by the variation in the relative net supply. When plotting all the possible IFLS 5 combinations there are 186 correct predictions. Conforming the supply and demand theory. Again this is confirmed by the estimated elasticity in table 8.

𝑛𝑠!,!− 𝑛𝑠!,! R-Squared

Tertiary schooling -0.171 (0.029)*** 0.293

Total -0.159 (0.03)*** 0.193

(33)

Primary vs. Tertiary. 105 correct out of 156, p=0

−1 0 1

Relativ

e Skill Premium

(34)

5.2.6 Overall Results

Following the previously discussed results the evidence is mixed to whether the supply and demand theory holds for the Indonesian provinces. In IFLS 1 there is no evidence confirming the theory. The estimated elasticity’s are all insignificant. When moving to the results found for IFLS 2 the picture is different. These results show evidence in favour of the supply and demand theory for the tertiary schooling level37

. IFLS 3 finds evidence in favour of the supply and demand theory as well. However, here the evidence is based on the secondary instead of tertiary schooling level. When looking at the full samples, both IFLS 2 and IFLS 338

show a significant negative elasticity. Nevertheless, when looking at the R-squared results, very little of the variation in skill premia is explained by the variation in relative net supply. In addition, the estimated elasticity’s are small. When turning to IFLS 4, again, the supply and demand theory holds for the secondary but not for the tertiary schooling level, as is the case for IFLS 2. Again, for the full sample, the elasticity is significant and confirms the theory. For IFLS 5 all the estimated elasticity's have the expected sign and are significant. Favouring the theory. It is noticeable that both the elasticity's and the R-squared results increase in both IFLS 4 and IFLS 5. This implicates that over the years in Indonesia the provincial differences in skill premia across schooling levels have become more responsive to relative differences in net supply and the variation is skill premia is increasingly explained by the variations in the relative net supply. One explanation for this striking result is that schooling is not a comparable measure of skill between provinces in the earlier periods but has increasingly become so. In other words, the quality of education was significantly different across Indonesian provinces, but over time schooling has become a comparable measure of skill. Implying that the quality of schooling has become similar across Indonesian provinces over time. If this hypothesis is true then my assumption that schooling is a comparable measure proves falls in earlier analysed periods. This hypothesis would support the results found by Leuven et

al. (2004), that differences in skill premia can be explained through a supply and

demand analysis when a comparable measure of skill is used. Furthermore, this would coincide with Indonesian governmental efforts to increase the quality of schooling. With the first program aimed to increase the quality being introduced just before the fielding of IFLS 4. Due to the focus on the quantity and access to schooling the quality can have developed differently across provinces. But as soon as the government has introduced programs to enhance the quality all the provinces received equal efforts. Possibly converging the quality of education towards the national mean and making schooling a comparable measure.

37_{Not for the secondary schooling group.}

Provincial differences in skill premia : evidence for Indonesia