Human capital has long been recognized as a crucial factor in economic growth, so it is essential to understand human capital, particularly in developing countries (e.g., Schultz 1961; Becker, Murphy, and Tamura 1990; Barro 1991, 2001). Indonesia is an interesting case because its real GDP has grown sharply along with its stock of human capital. When Indonesia won independence from the Netherlands in 1950, its GDP was only US$66.4 billion.1 Fifty years later, however, the figure soared to US$675.4 billion (van Leeuwen 2007, Appendix A. 12). During the same period, its stock of human capital jumped from US$116.4 billion to US$483.1 billion (van Leeuwen 2007, Appendix A. 2). In addition, the Indonesian government has stressed the importance of education in economic growth. A good example is a major school construction program initiated in 1973 in which the Indonesian government constructed more than 61,000 primary schools between 1973–74 and 1978–79 (Duflo 2001).
The issue of returns to education is one of the most discussed issues in the literature of human capital. Because so many studies have considered this issue, survey papers are updated continuously (Psacharopoulos and Patrinos 2004). Some studies have examined this issue by focusing specifically on Indonesia, as discussed in the next section, but many important aspects of the issue have yet to be elaborated. This paper contributes to the literature in relation to Indonesia in at least four ways. First, more recent data (recorded in 2007) are used to update the rate of return to education. Duflo (2001) offers the most recent estimates for Indonesia, but her data is from 1995.2 Indonesian GDP grew 45.6% in real terms between 1995 and 2007 (IMF 2012), so it is interesting to determine whether this astounding growth has impacted the rate. Second, a quantile regression is applied to assess whether the rate of return at the mean can be generalized to the entire conditional distribution of earnings. Some attempts have been made to apply this technique to other countries, but few studies have focused on Indonesia (e.g., Chamberlain 1994; Buchinsky 1994; Hartog, Pereira, and Vieira 2001; Martins and Pereira 2004; Patrinos, Ridao-Cano, and Sakellariou 2006). This exercise also provides empirical evidence of within-(education) group inequalities in earnings. Third, further light is shed on returns to education in self-employment. Previous studies have generally neglected this sector despite its numerical importance in the Indonesian labor market. Efforts are made not only to measure differences in returns to education between paid-employment and self-employment but also to discern whether these differences are attributed to person- or sector-specific reasons. Fourth, in addition to monetary returns to education, nonmonetary returns, specifically happiness returns, to education are explored. This exercise is timely because the economics of happiness has greatly expanded since the late 1990s and because it has been recognized that money and happiness do not go together. If happiness is the ultimate goal in life and if money is merely a means to it, then limiting attention to monetary returns to education misses a more important part of the story of returns to education.
The main findings are as follows. The rate of return to an additional year of schooling is 10.7%, which is in line with the estimates in the literature. In addition, with the private cost of education taken into account, the between-(education) group inequalities in earnings are not as severe as those in developed countries. The rate varies little across the conditional distribution of earnings, with a possible exception at high quantiles where the rate is lower than others. In turn, the latter point alleviates concerns about within-(education) group inequalities. The rate is lower in self-employment than in paid-employment, and it is argued that this difference results from both person- and sector-specific reasons. Finally, happiness returns to education are consistent with monetary returns to happiness in the main.
II. Literature Review
Relevant papers will be compared and contrasted, when results are discussed. This section focuses mainly on major studies on returns to education in Indonesia. Byron and Takahashi (1989) estimate a 15–17% rate of return per year of schooling from 1981 data for urban Java. Subsequently, McMahon, Jung, and Boediono (1992) compare the social rate of return to education for general and vocational schools in major regions for the year 1986. Their rate of return for men ranges widely from 5 to 22%, but when Central Java, the most populated region in Indonesia, is considered, the range narrows from 9 to 14%.
These studies, however, do not adjust for school repetition and dropouts, which are a major source of concern in developing countries. Behrman and Deolalikar (1991, 1991, 1993,1995) address this concern and additionally control for household fixed effects to eliminate unobserved household heterogeneity. Drawing on 1986 data, Behrman and Deolalikar (1993) bring down the rate of return to the range of 3.2–9.3% per year of schooling for males.
Although attempts to control for household heterogeneity have been made, strictly speaking, causality has not entered the literature for Indonesia. To tease out the causal effect of education on earnings, Duflo (2001) exploits the Sekolah Dasar INPRES program, a major school construction program launched by the Indonesian government in 1973. She compares male cohorts who benefited from the program with those who did not, but are otherwise similar to the former cohorts. These cohorts were born between 1950 and 1972, and their statistics were recorded for the year 1995. Her rates are considered the most rigorously estimated for Indonesia, lying in the range of 6.8 to 10.6%. More recently, Newhouse and Suryadarma (2011) use the same dataset that this paper draws on with regard to returns to education, but they do not focus on estimating returns to education in general but to comparing returns to education in vocational high schools with those in general high schools.
The main dataset for this paper is the Indonesia Family Life Survey, a longitudinal survey. This survey started with over 22,000 individuals in 7,224 households in 1993 (representative of 83% of the Indonesian population). Four follow-ups ensued in 1997 (IFLS2), 1998 (IFLS2+), 2000 (IFLS3), and 2007 (IFLS4). Because the recontact rate for the original families was 93.6% in IFLS4, attrition should be of little concern. This paper uses mainly IFLS4, but depending on the specifications, IFLS3 is also tapped. The year 2007 is of particular interest for this paper because it is characterized by the highest GDP growth rate of 6.3% since the Asian financial crisis in 1997/98 before 2011 (IMF 2012). The rationale is that when an economy is experiencing a boom, more people participate in the labor force. In such a case, even those workers vulnerable to economic cycles, usually workers with low education, record earnings.
In most cases, the analysis considers men aged between 15 and 65 whose primary activity during the past week was working, trying to work, or helping to earn income in IFLS4.3 Depending on the specifications, however, further restrictions are imposed. The focal gender is male because there are more complex factors to be taken into account for women such as their selection into the labor force and frequent interruptions in their career paths. In fact, only 50.3% of women participated in the labor force in 2007, whereas the figure was 83.7% for men (ADB 2012). Hourly earnings are approximated by salaries/wages during the last month for individuals with paid-employment (net profits during the last month for self-employed individuals) divided by four times the normal hours worked per week.4 Weekly and annual earnings are also provided in the dataset, but hourly earnings are adopted because theories of wage determination are relevant to hourly wages. Moreover, annual earnings can be contaminated by unemployment during the year.
Years of schooling are constructed from the highest level of education attended and the highest grade completed at that school level.5 Such construction is attractive because it reduces measurement errors from school repetition and dropouts, which are a major source of concern in developing countries (Behrman and Deolalikar 1991). Another measure of education is a series of dummies for the education level categorized by no schooling, elementary school, junior high school, senior high school, and college or above. When necessary, junior and senior high schools are further divided into general and vocational education. Each category represents the education level that the respondent attended or graduated from. In addition, following previous research, potential work experience is calculated by age – years of schooling – 6.
Some caveats are in order with respect to a possible downward bias in the coefficient on education. This coefficient can be biased downward if hourly earnings are constructed as above and hours worked are positively correlated with education. The latter possibility is indirectly checked by regressing hours worked on ethnicity, potential work experience, its squared term, and years of schooling. The coefficient on years of schooling is statistically insignificant at conventional levels of significance, so the downward bias arising from this source is minimal. Another source of downward bias is attributed to potential work experience. Mincer (1974) demonstrates that this measure, instead of actual work experience, biases the estimate of years of schooling downward. Hence, the source of this bias needs to be kept in mind along with the upward bias coming from ability. A possible upward bias coming from ability is discussed in the next section.
The main specification is the conventional one proposed by Mincer (1974). The variable for earnings is regressed on potential work experience, its squared term, and years of schooling; ethnicity is additionally controlled for. Previous studies have controlled for more variables: demographic variables such as marital status; labor market outcomes such as union status, occupations, and industries; and even geographic variables such as urbanity and states. All these variables are potentially endogenous. More important, they are likely to have causal relationships with years of schooling, with the causality running from years of schooling to the extended variables. In such a case, controlling for the extended variables underestimates the coefficient on years of schooling. Depending on the purpose, it is necessary to control for other variables. For example, adding union status to the specification is necessary to estimate union premiums. It is also essential to hold occupation dummies constant to understand within-occupation returns to education. However, this paper investigates total returns to education, so it allows for education-related mobility. Mobility can be made across anything, be it marital status, union status, occupations, industries, urbanity, and state. For this reason, the simple Mincerian specification is employed throughout the analysis.
One of the most frequently mentioned concerns in the Mincerian specification is ability bias. According to this concern, if ability is positively correlated with schooling, but left out of the specification, then the coefficient on years of schooling is overestimated. Since the initiation of the empirical literature on returns to education, a large number of studies have measured the size of the bias by typically controlling for proxies for ability. In his early review, Griliches (1977, p. 18) takes into account not just an upward bias but also a downward bias to conclude that “the implied net bias is either nil or negative.” Later, a wide range of instrument variables have been exploited to estimate the causal effects of education on earnings. Card (2001, p. 1158) similarly concludes in his review that “a causal effect of education that is as big or bigger than the OLS-estimated return at least for people whose schooling choices are affected by the supply-side innovations that have been studied so far.” This paper does not assert that the issue of ability bias does not exist in Indonesia. Based on the surveys, however, the ability bias inherent in this paper's OLS is likely to be minimal, and the estimate can even be considered a conservative one. To further check possible biases, this paper compares its estimates with those in the literature.
The coefficient on years of schooling is typically interpreted as the rate of return to schooling. This paper follows this interpretation for the purposes of comparison with other studies and for its simplicity. At the outset, however, it needs to be emphasized that this interpretation is correct only under certain assumptions. Aside from the conventional assumptions related to the Mincerian specification (e.g., returns to one year of schooling are the same regardless of grade), Becker and Chiswick (1966) expound that this interpretation is correct when the cost at time t is equal to earnings at time t-1. Costs include direct costs for items such as textbooks and tuition and indirect costs such as opportunity costs. If the assumption is not satisfied, then the interpretation is incorrect. For example, if the cost at time t exceeds earnings at time t-1, then the coefficient overestimates the rate of return to education. Hence, if education is publicly subsidized, then private returns and social returns diverge, with social returns being lower than private ones. In fact, social returns are lower than private returns on average (Psacharopoulos 1994).
A Descriptive Statistics
One needs to understand returns to education in the context of developmental stages. Psacharopoulos and Patrinos (2004) review the literature to find that a rate of return to education tends to be lower in developed countries. Indonesia's per capita GDP was US$1,897 in 2007 (IMF 2012), making Indonesia a middle-income country according to the classification of Psacharopoulos and Patrinos (2004).6 In their review, the mean of years of schooling was 8.2 years for middle-income countries. As shown in Table 1, the Indonesian mean, 8.6 years, is very close to this figure. One may anticipate from this similarity that the rate of return to education in Indonesia would also be similar to that in other middle-income countries. As discussed below, this is the case.
Table 1. Descriptive Statistics
|Hourly earnings in 2000 (rupiah)||3,134||5,468|
|Hourly earnings in 2007 (rupiah)||6,044||8,045|
|Work experience (years)||22.2||13.7|
|Years of schooling (years)||8.6||4.3|
|Junior high school||16.8|
|Senior high school||30.9|
|College or above||12.0|
|Vocational junior high school||1.1|
|General junior high school||15.7|
|Vocational senior high school||14.5|
|General senior high school||16.4|
|College or above||12.0|
|Paid-employment in 2000||60.7|
|Self-employment in 2000||39.3|
|Paid-employment in 2007||60.4|
|Self-employment in 2007||39.6|
|Relative income ladder 1||5.2|
|Relative income ladder 2||24.6|
|Relative income ladder 3||54.3|
|Relative income ladder 4||14.8|
|Relative income ladder 5||0.9|
|Relative income ladder 6||0.2|
The mean and standard deviation of years of schooling are useful for summarizing education standards, but they are not useful for understanding the distribution of education levels. When education levels are listed, two modes stand out. One is attending or graduating from elementary school (37.0%), and the other is attending or graduating from senior high school (30.9%). The former group tends to include older individuals, and the latter group, younger ones. Another way to appreciate the distribution of education levels is to further divide education levels according to the type of education. This type of classification is particularly relevant for Indonesia because vocational education has been emphasized in Indonesia. Vocational education was available in junior high schools in Indonesia, but the proportion of the category is only 1.1% because the government phased out and transferred this type of education to senior high schools in the early 1990s. Thus, vocational education is popular in high schools. Of the sample, 16.4% attended or graduated from general senior high schools, whereas 14.5% did so for vocational high schools. The proportion of individuals who attended or graduated from college or above is not small, considering Indonesia's status as a middle-income country.7
Table 1 also reveals one important aspect of the Indonesian labor market. The proportion of self-employed workers is large, nearly 40% in 2000 and 2007. Previous studies of returns to education in developed countries have typically ignored this sector. One reason is its relatively small size. Given the large size of the sector, however, ignoring this sector would not shed proper light on returns to education in Indonesia. What follows explores whether a rate of return to education varies according to self-employment status and if so, why.
One of the main dependent variables is hourly earnings. The mean hourly earnings in 2007 were almost twice those in 2000. Although rising labor productivity might have contributed to this increase, the high inflation rate was the main force behind the rise. The Indonesia GDP deflator rose from 100 in 2000 to 201 in 2007 (IMF 2012). In addition, the mean potential work experience is 21.4 years, which indicates that most of the respondents were strongly attached to the labor market.
As mentioned above, happiness is also considered, along with hourly earnings to estimate returns to education. Most respondents answered that they were happy, but some expressed that they were very happy (6.1%) or very unhappy/unhappy (8.2%).8 The general tendency toward happiness is not unique to Indonesia. Although there tend to be more unhappy people in extremely poor countries such as some in Africa or in politically unstable countries such as those in Central Asia and Eastern Europe, people are mostly happy across the world (Graham 2009, chapters 2 and 3).
B Monetary Returns to Education at the Mean
Table 2 reports the rate of return to education. As shown across the columns (except column 2), the variable for potential work experience and its squared term display the usual pattern of concavity. These robust results suggest that Indonesian workers are similar to those in other countries in terms of the pattern of returns to work experience. When the simple Mincerian specification is applied as in column 1, the rate of return to education is estimated to be 10.7%.9 This rate is very close to the range between 6.8 and 10.6% that Duflo (2001) rigorously estimates. The slight difference may be attributed to different cohorts: whereas the ages of her sample range from 23 to 45, ours lie between 15 and 65. Another reason may be different years: her data concern the year 1995, which is 12 years before the year considered in this paper. Alternatively, the representativeness of the sample may account for the difference: her sample is nationally representative, whereas our sample is representative of 83% of the Indonesian population. In addition, the rate in column 1 is identical to the rate for the middle-income group in the review of Psacharopoulos and Patrinos (2004). It seems that the Indonesian rate of return to education has remained stable over time and is comparable to that for its peer countries.
Table 2. Monetary Returns to Education at the Mean (Dependent Variable: Log of Hourly Earnings)
|Constant||6.850 (0.066)***||7.287 (0.503)***||6.840 (0.074)***||7.171 (0.104)***||7.209 (0.112)***||7.166 (0.103)***||7.204 (0.112)***|
|Javanese||−0.067 (0.030)**||−0.108 (0.283)||0.038 (0.037)||−0.051 (0.030)*||0.053 (0.036)||−0.051 (0.031)||0.053 (0.036)|
|Work experience||0.044 (0.003)***||0.045 (0.018)**||0.048 (0.004)***||0.048 (0.003)***||0.052 (0.004)***||0.048 (0.003)***||0.052 (0.004)***|
|Work experience2 (÷1,000)||−0.568 (0.059)***||−0.576 (0.361)||−0.635 (0.079)***||−0.716 (0.057)***||−0.772 (0.075)***||−0.717 (0.058)***||−0.772 (0.075)***|
|Years of schooling||0.107 (0.004)***||0.057 (0.049)||0.096 (0.004)***|| || || || |
|Elementary school|| || || ||0.203 (0.098)**||0.123 (0.102)|| || |
|Vocational junior high school|| || || ||0.600 (0.130)***||0.505 (0.133)***|| || |
|General junior high school|| || || ||0.462 (0.102)***||0.347 (0.103)***|| || |
|Vocational senior high school|| || || ||0.876 (0.108)***||0.675 (0.102)***|| || |
|General senior high school|| || || ||0.872 (0.107)***||0.657 (0.103)***|| || |
|College or above|| || || ||1.506 (0.100)***||1.320 (0.094)***|| || |
|Elementary school|| || || || || ||0.204 (0.097)**||0.124 (0.102)|
|Junior high school|| || || || || ||0.473 (0.100)***||0.360 (0.101)***|
|Senior high school|| || || || || ||0.877 (0.105)***||0.668 (0.101)***|
|College or above|| || || || || ||1.509 (0.099)***||1.323 (0.094)***|
|Household fixed effects||No||Yes||No||No||No||No||No|
|Community fixed effects||No||No||Yes||No||Yes||No||Yes|
Behrman and Deolalikar (1993) stress the importance of controlling for unobserved household and community heterogeneity. However, Griliches (1977) and Card (2001) argue that OLS estimates of returns to education are not likely to be biased, and Duflo (2001) argues that this is the case specifically for Indonesia. For the investigation of the size of the bias from unobserved household and community heterogeneity, household fixed effects are controlled for in column 2, and community fixed effects are entered in column 3. The coefficient on years of schooling is imprecisely estimated when household fixed effects are introduced. The main reason is that the estimation depends on the within-household variation, but only 18% of the observations provide the variation. Because the within-community variation is large enough, the rate with community fixed effects is more precisely estimated. Including community fixed effects is also attractive because it controls for costs of living by community. At the same time, however, one needs to be aware that community fixed effects control for many characteristics of communities, and costs of living are just one of them. The coefficient on years of schooling is 0.096, which is slightly smaller and less precisely estimated than that in Column 1 and yet, if education is correlated with migration across communities and if migration increases earnings (if not, one would not migrate), then the coefficient in column 3 should be biased downward. Even when the coefficient in column 3 is unbiased, both coefficients on years of schooling in columns 1 and 3 are not statistically different at conventional levels of significance. When variables for education are modified as below, similar results are gained (compare column 4 with column 5, and column 6 with column 7). That is, controlling for community fixed effects slightly reduces the coefficients for education and their precision, but the substance of the results does not change. Because of the concern of over-correction, estimates without community fixed effects are mainly reported throughout this paper. However, controlling for communitiy fixed effects does not alter the substance of results.
Thus far, it has been assumed that each year of schooling would yield the same rate of return regardless of the grade. This assumption is relaxed by replacing the variable for years of schooling with a series of dummies for the education level as shown in columns 4–7. Because explicit tracking is implemented in Indonesia and the government has attempted to increase the enrollment rate for vocational high schools, junior and senior high schools are further split into vocational and general education as shown in columns 4 and 5.10 For both junior and senior high schools, the rate of return is higher for vocational education, but they are not statistically different from each other at conventional levels of significance. McMahon, Jung, and Boediono (1992) estimate that the rates of return for men in Central Java are 11% and 9% for general and vocational junior high schools, respectively, and 12% and 14% for general and vocational senior high schools, respectively. Although they argue that one return is higher than the other, they do not check whether the difference is statistically significant. Given our results, however, it is unlikely that they are statistically different.
Because there is no difference between general and vocational education in terms of the rate of return, general education and vocational education are to be combined. Their estimated coefficients of interest are displayed in columns 6 and 7. The results suggest that relaxing the linearity assumption is of some importance. Because one category consists of more than one year of schooling, it would be arbitrary to select one year over another. Nevertheless, for expositional purposes, some examples would be helpful to appreciate the importance of relaxing the linearity assumption. According to column 6, when the category of college or above is considered as 14 years of schooling, the rate is 10.8%. Similarly, when the coefficient on senior high school is divided by 11, the rate is 8.0%. This result is inconsistent with the reverse order observed by Psacharopoulos and Patrinos (2004). Unconventional as it may be, this result is consistent with the findings of Belzil and Hansen's (2002) US study. Although more solid evidence is needed, the results suggest that an increasing rate of return has important implications for earnings inequalities in general and between-(education) group earnings inequalities in particular.11 However, as far as private returns are concerned, caution is required for taking the rate at its face value. Recall the assumption made in interpreting the coefficient on education in the previous section. If private costs account for a large portion of total costs at higher levels of education, which is the case in Indonesia, the rate needs to be adjusted downward. Hence, between-group inequalities should be lower than those implied by the raw figures, although the extent of the adjustment is left for future research.
C Monetary Returns to Education at Quantiles
In the previous subsection, the linearity assumption for years of schooling is relaxed by replacing the variable for years of schooling with dummies for schooling. In this subsection, another assumption is relaxed. Estimating returns to education at the mean provides a simple, convenient way to understand the rate of return. However, it assumes that the rate is constant across the conditional distribution of earnings. And yet, the rate can vary according to the level of earnings, which has implications for within-group earnings inequalities and possibly earnings mobility. If the rate is higher at higher levels of earnings, then within-group earnings inequalities may be severe and earnings mobility may not be facilitated through education. For example, if the rate of return is higher for high school graduates in the high-income group than those in the low-income group, then relative to the case when the rates are the same, the inequality between the two groups would be larger, and the same level of education would not be of much help to the latter joining the former. Because inequalities consist of between- and within-group inequalities, it is important to investigate the rate across the conditional distribution of earnings.
This investigation is timely and compelling because within-group inequalities have become an important issue in developed countries (e.g., Juhn, Murphy, and Pierce 1993; Card and DiNardo 2002; Lemieux 2006). In particular, within-group inequalities grew rapidly in the United States in the first half of the 1980s and then resumed in the early 2000s (Lemieux 2006). It is notable that income inequalities are usually underestimated in recessions, and the early 2000s are characterized by the burst of the dot-com bubble (Solon, Barsky, and Parker 1994). Reflecting on this, one may suspect that the upturn seen in the early 2000s is itself understated. Specifically, through a quantile regression, Chamberlain (1994) and Buchinsky (1994) find that the rate is higher at higher quantiles for the 1963–87 US sample. Like Buchinsky (1994), Hartog, Pereira, and Vieira (2001) find not only that the rate of return is higher at higher quantiles but also that the gap between low and high quantiles grew in Portugal in the 1980s and early 1990s. Martins and Pereira (2004) confirm this pattern in 16 industrialized countries except for Greece.
Table 3 shows the results estimated by the quantile regression (Koenker and Bassett 1978). When years of schooling are taken into account as in Panel A, the rate of return varies according to the quantile, and the null hypothesis of the same coefficients on years of schooling for all quantiles is rejected at conventional levels of significance.12 However, the sizes of differences are small. Even the largest difference between the rate at quantile 0.5 (or quantile 0.1) and that at quantile 0.9 is 2.5 percentage points. This result indicates that within-group earnings inequalities in Indonesia are not severe, or more conservatively, they are not as severe as those in the United States. When the linearity assumption is relaxed as shown in Panel B, similar results are obtained. The null hypothesis of the same coefficients on the dummies for schooling for all quantiles is rejected only for elementary schools at conventional levels of significance.13
Table 3. Monetary Returns to Education at Quantiles (Dependent Variable: Log of Hourly Earnings)
|Panel A:|| || || || || |
|Years of schooling||0.121 (0.007)***||0.115 (0.004)***||0.121 (0.003)***||0.114 (0.003)***||0.096 (0.005)***|
|Panel B:|| || || || || |
|Elementary school||0.283 (0.224)||0.149 (0.069)**||0.249 (0.084)***||0.404 (0.105)***||0.109 (0.161)|
|Junior high school||0.546 (0.230)**||0.389 (0.073)***||0.519 (0.089)***||0.651 (0.111)***||0.380 (0.162)**|
|Senior high school||1.058 (0.229)***||0.872 (0.073)***||0.965 (0.088)***||1.035 (0.112)***||0.731 (0.162)***|
|College or above||1.571 (0.233)***||1.556 (0.085)***||1.647 (0.088)***||1.671 (0.109)***||1.261 (0.165)***|
The difference between our results and previous findings on the industrialized countries may be attributable to different levels of development. This possibility is consistent with the findings of Mwabu and Schultz (1996), who consider South Africa and estimate that the rate of return does not vary across deciles for Africans at higher education levels but that the rate rises with deciles for Whites at higher education in 1993. Apartheid segregated Africans and Whites not only occupationally but also geographically. If Africans are considered as living in a developing region and the opposite is the case for Whites, which is not far from reality, then the difference between our results and previous findings on the industrialized countries can be reconciled. Overall, the results in Table 3 suggest that at least the issue of within-group inequalities can be put aside in the promotion of education in Indonesia in the short term.
However, the following question remains unanswered: why is the rate of return similar across quantiles for developing countries or regions, whereas it is higher at higher quantiles for the industrialized countries? An answer to this question is beyond the scope of this paper, but we can speculate that education and income-generating unobserved factors such as ability and the quality of education are more complementary in developed countries than in developing ones. This question deserves further research, but the literature on skill-biased technology can be a promising starting point.
D. Monetary Returns to Education by Class of Worker
As shown in Table 1, Indonesia's self-employment sector is large, so ignoring this sector would reveal only part of the story of returns to education in Indonesia. In addition, Chiswick (1977) warns that this practice overestimates the rate of return for developing countries. Psacharopoulos (1994) reviews that the rate of return in self-employment is lower than that in paid-employment: 10.8% vs. 12.2%.
Column 1 of Table 4 shows that the rate of return to years of schooling is 5.5 percentage points lower for the self-employed than for the paid-employed. The reason for this difference may be person-specific, sector-specific, or both. If the reason is person-specific, then one can speculate that the quality of education of the self-employed is not up to the one expected in paid-employment, so it yields lower returns for self-employment. If the reason is sector-specific, then it may be that education of even the same quality is utilized less effectively in self-employment than in paid-employment. Some studies have shed some light on various factors related to being self-employed. For example, Blau (1985) focuses on Malaysian heads of household, and Borjas (1986) considers male immigrants in the United States. Furthermore, Dabos and Psacharopoulos (1991) examine Brazilian males (for a review, see Parker 2004). And yet, few studies have investigated sorting for the rate of return, which is the issue addressed in this subsection.
Table 4. Monetary Returns to Education by Class of Worker
|Self-employed||0.393 (0.085)***|| || |
|Years of schooling||0.125 (0.004)***||0.061 (0.010)***||0.130 (0.005)***|
|Self-employed × years of schooling||−0.055 (0.009)***|| || |
|Self-employed in 2000 and paid-employed in 2007|| ||−0.206 (0.097)**|| |
|Self-employed in 2000 and paid-employed in 2007 × years of schooling|| ||0.026 (0.011)**|| |
|Paid-employed in 2000 and self-employed in 2007|| || ||0.163 (0.060)**|
|Paid-employed in 2000 and self-employed in 2007 × years of schooling|| || ||−0.019 (0.008)**|
An individual with a low rate of return tends to sort into self-employment. Alternatively, the rate is lower for self-employment for all workers. Although imperfect, both possibilities are tested with the sample. A small number of workers switched from self-employment in 2000 to paid-employment in 2007 or vice versa. If sorting takes place in the rate of return, then the rate should be higher for workers who switched from self-employment in 2000 to paid-employment in 2007 even before they did. Column 2 demonstrates that this is the case. When the dependent variable is the log of hourly earnings of self-employed workers in 2000, the result for the coefficient on the interaction term between years of schooling and the dummy for migration as such indicates that migrants enjoyed a return that was 2.6 percentage points higher. Conversely, if the logic is correct, then the rate should be lower for workers who switched from paid-employment in 2000 to self-employment in 2007 even before they did. The results in column 3 provide support for this conjecture. According to the coefficient on the interaction term between years of schooling and migration as such, the return for migrants is 1.9 percentage points lower. Note that the absolute values of the coefficients on both interaction terms are similar, implying that sorting took place symmetrically between the two sectors.
And yet, the coefficients on both interaction terms are less than half of the rate gap between both sectors presented in column 1. Hence, sorting cannot fully explain the lower rate for self-employment. Because the rate is lower for self-employment, one can hazard that the same individual gains a higher rate of return when switching from self-employment to paid-employment, and vice versa. The first two rows of Table 5 do not provide support for this conjecture, but the third and fourth rows do. When workers migrated from paid-employment in 2000 to self-employment in 2007, the rate dropped from 11.2 to 5.8%. Overall, Tables 4 and 5 suggest that the lower rate in self-employment was a result of both person- and sector-specific reasons. However, further research is necessary to definitively understand the gap between the two sectors. One reason is that the first two rows of Table 5 do not conform to the possibility of a sector-specific reason. In addition, Tables 4 and 5 do not consider the rate of return for the same person in both sectors at the same time, which is ideal in theory but not possible in practice. Migration took place over a time window of seven years, so it is possible that returns to education could change regardless of the education level for the two sectors.
Table 5. Monetary Returns to Education for Switchers
|From self-employment in 2000 to paid-employment in 2007:|| || |
|Log of hourly earnings in 2000||0.093 (0.012)***||435|
|Log of hourly earnings in 2007||0.084 (0.015)***||436|
|From paid-employment in 2000 to self-employment in 2007:|| || |
|Log of hourly earnings in 2000||0.112 (0.008)***||624|
|Log of hourly earnings in 2007||0.058 (0.017)***||601|
E. Nonmonetary Returns to Education
Happiness is usually considered the ultimate goal in life, whereas earnings are a means to it (Sen 1982). If income and happiness went together, then it would suffice to examine the monetary returns to education. However, since Easterlin's modern econometric approach to the relationship between happiness and income (Easterlin 1974), numerous studies have confirmed that happiness and income do not go together (for a review, see Kahneman, Diener, and Schwarz 1999; Frey and Stutzer 2002a, 2002b; Di Tella and MacCulloch 2006; Dolan, Peasgood, and White 2008; Graham 2009).14 Japan is, inter alia, a spectacular example in this regard. Despite a six-fold increase in its per capita income between 1958 and 1991, its level of life satisfaction remained the same (Frey and Stutzer 2002a, chap. 1). Therefore, it is useful to examine returns to education in terms of happiness.
Education is usually considered as a control variable in the literature on happiness, and furthermore, insufficient attention has been paid to happiness in the context of returns to education. For example, Blanchflower and Oswald (2004) argue that more education provides more happiness in Britain and the United States, but Hartog and Oosterbeek (1998) and Stutzer (2004) find from the Netherlands and Switzerland, respectively, that it is middle levels of education that show the strongest relationship with happiness. And yet, they control for many variables along with education that may be causally related to education, with causality running from education to those variables. Hence, it is likely that the effects of education in their studies are biased downward. More important, to the best of our knowledge, no published study considering Indonesia has examined the relationship between happiness and education in general and happiness returns to education in particular. Because Indonesia is the largest Islamic country, it should be interesting to compare our results with the findings of previous studies, which have focused mainly on Protestant countries.15
If the Mincerian approach is applied to happiness, then there should be little difference in interpretations in terms of optimal years of schooling. An individual stops schooling when marginal unhappiness equals marginal happiness. For estimation, an ordered probit model is used, as usually done in the literature.16 The dependent variable is an answer to the following question: “Taken all things together how would you say things are these days—would you say you were very happy, pretty happy, or not too happy?” This question is widely asked to measure an individual's happiness. For example, the identical question is asked in the US General Social Survey, and a nearly identical question is asked in the Euro-Barometer Survey Series. The respondent answers on a scale of one to four, with one indicating “very happy” and four indicating “very unhappy.” This order is reversed for ease of interpretation in this paper. In addition, because only 0.28% of the sample indicated being “very unhappy,” the categories of “very unhappy” and “unhappy” are combined.
As shown in column 1 of Table 6, education is positively related to happiness. As shown in column 2, we check whether the rate of happiness returns to education would be lower for self-employment than for paid-employment as in the case of monetary returns to education. The results indicate that the rate is lower for self-employment than for paid-employment. The statistical significance of the interaction term is weak, however. As shown in column 3, we relax the linearity assumption and obtain similar results. That is, an increase in education raises happiness. As shown in column 4, we check whether happiness returns to education would vary according to the sector. The results indicate no difference in happiness between the sectors. Hence, the results in columns 1 and 3 suggest that education does make a substantial difference in happiness, but those in columns 2 and 4 imply that it is uncertain whether education in self-employment is as effective as that in paid-employment at least in terms of happiness. These results clearly illustrate that considering only one aspect of returns to education can be misleading for the purpose of welfare.
Table 6. Nonmonetary Returns to Education (Dependent Variable: Happiness Index)
|Years of schooling||0.047 (0.004)*** [0.0048]||0.054 (0.005)***|| || ||0.036 (0.005)***||0.035 (0.004)***||0.027 (0.005)***|
|Self-employed|| ||0.142 (0.091)|| ||−0.187 (0.159)|| || || |
|Self-employed × years of schooling|| ||−0.018 (0.011)*|| || || || || |
|Elementary school|| || ||−0.095 (0.097) [−0.011]||0.035 (0.176)|| || || |
|Junior high school|| || ||0.088 (0.105) [0.012]||0.173 (0.172)|| || || |
|Senior high school|| || ||0.209 (0.101)** [0.022]||0.355 (0.177)**|| || || |
|College or above|| || ||0.486 (0.100)*** [0.086]||0.639 (0.169)***|| || || |
|Self-employed × elementary school|| || || ||−0.186 (0.166)|| || || |
|Self-employed × junior high school|| || || ||−0.074 (0.156)|| || || |
|Self-employed × senior high school|| || || ||−0.247 (0.195)|| || || |
|Self-employed × college or above|| || || ||−0.317 (0.198)|| || || |
|Logged earnings|| || || || ||0.122 (0.016)***|| ||0.102 (0.016)***|
|Relative income ladder 2|| || || || || ||0.454 (0.098)***||0.418 (0.105)***|
|Relative income ladder 3|| || || || || ||0.740 (0.080)***||0.667 (0.098)***|
|Relative income ladder 4|| || || || || ||0.818 (0.080)***||0.734 (0.092)***|
|Relative income ladder 5|| || || || || ||1.330 (0.291)***||1.016 (0.275)***|
|Relative income ladder 6|| || || || || ||1.158 (0.592)*||1.139 (0.722)|
|Cut 1||−1.044 (0.084)||−0.972 (0.093)||−1.345 (0.129)||−1.217 (0.208)||−0.192 (0.152)||−0.584 (0.119)||0.084 (0.191)|
|Cut 2||1.980 (0.094)||2.054 (0.108)||1.680 (0.137)||1.809 (0.207)||2.860 (0.157)||2.500 (0.137)||3.182 (0.207)|
Columns 5 to 7 are not intended to measure the total effects of schooling but have two goals. One is to check the robustness of the estimates for years of schooling.17 The other is to measure the possible downward bias from the inclusion of variables that are considered to be important for happiness and positively related to education, with causality probably running from education to the variables. Absolute levels of income are controlled for in column 5, and perceived relative levels of income are controlled for in column 6. The former are usually assumed to be a critical determinant of the utility level in mainstream economics, and the latter are usually claimed to be more important than the former in the economics of happiness. The two variables are held constant at the same time in column 7.
The results in columns 5 to 7 show that more income, whether absolute or relative, is related to more happiness. The highest level of relative income cannot be precisely estimated because the category contains only 0.2% of the sample. The inclusion of one or both variables reduces the size of the coefficient on years of schooling, indicating that education has an indirect positive effect on happiness through income, whether absolute, relative, or both.18 Education seems to enhance happiness and to have a positive effect above and beyond absolute and relative income levels.
The marginal effects of education are not negligible. For example, every one-year increase in schooling raises the probability of stating “very happy” by 0.48 percentage point for the reference group (see figures in brackets in column 1).19 Similarly, every one standard deviation increase in years of schooling, that is, 4.3 years, raises this probability by 2.1 percentage points. Given the predicted probability of 4.9% for the response category, the size is large. When the linearity assumption is relaxed as in column 3, the marginal effects are the largest for college or above at 8.6 percentage points per year of schooling.20
This paper elucidates some neglected aspects of returns to education in Indonesia. It starts by updating the rate of return to education in Indonesia with more recent data, finding that the updated rate is similar to existing ones. Another interesting result is that the rate increases with the level of education. In addition, this paper relaxes the assumption of a constant rate of return across the conditional distribution of earnings in the quantile regression. Overall, the assumption is consistent with the data, but there is suggestive evidence that the rate is lower at high quantiles. This evidence implies that within-(education) group inequalities are not as severe in Indonesia as they are in developed countries. In addition, the monetary rate of return is lower for self-employment than for paid-employment, which may be attributed to both person- and sector-specific reasons. Last but not least, education has positive, substantial, and robust effects on happiness above and beyond absolute and relative levels of income.
These results suggest that education is a profitable investment source in Indonesia. That is, unlike in developed countries, Indonesia can invest in education without being too concerned about within-group and possibly between-group inequalities. In addition, education is an important and robust measure for enhancing happiness, the ultimate goal in life. Of course, future research should thoroughly investigate the reasons for the lower monetary rate of return for self-employment. Nevertheless, most of the results in this paper provide compelling evidence that Indonesia should invest more in education. Education can be pursued through private measures, but given positive monetary and nonmonetary externalities of education and possible credit constraints, the Indonesian government should consider more generous public subsidies for basic education.21 Another source of funding is official development aid (ODA). Regardless of the funding method, recent studies in development economics have strongly recommended the implementation of measures based solidly on empirical evidence. Otherwise, well-intended measures may squander funds, and worse, may even lead to unintended adverse consequences (for an introduction, see Duflo and Kremer 2005; Banerjee and Duflo 2009).