Motivation and Background
While the outcomes of most public investments schemes are not completely foreseeable, the benefits of investing in education to both children and broader society could not be more clear: education is strongly correlated with improvements in health and nutrition, it is one of the best protections against poverty, and it fosters civic participation and democratization.
Although the importance of education to individuals and society is apparent, access to education differs substantially between and within countries. In particular in developing countries the access to education is extremely restricted. Looking at within country disparities, these barriers to education often disproportionately affect girls. In the 1970s the Indonesia’s central government launched the Sekolah Dasar INPRES program (henceforth SD INPRES), one of the largest primary school construction programs in history in order to counteract this trend of stagnating primary school enrollment rates. This large-scale construction program led many economists to study the numerous impacts on the Indonesian population by taking advantage of its form as a natural experiment.
Previous literature on the impact of SD INPRES suggests that until recently only little attention has been paid to the effects of the primary school construction program on girls. This is not very surprising since Indonesia constantly performed poorly among international gender equality measurements such as the Gender Inequality Index by the United Nations or the Global Gender Gap Report by the World Economic Forum. Work by Hertz and Jayasundera (2007), Pettersson (2012), and Ashraf et al. (2014), take differing approaches in studying the effects of the SD INPRES program in women. The novelty in our approach lies in the fact that with our data set, containing more cohorts than previous analyses, we are able to check for persistency of the resulting effects of the school construction program on later cohorts by extending the time range used in Duflo’s pioneering analysis. We also investigate whether the effect of the program on women might not be reflected on the intensive margin, i.e. the actual duration of education, but rather on the extensive margin, in other words the likelihood of completing primary school education.
A key element of our identification strategy to identify causality of the school construction program on educational outcomes relies on variations of an individual’s exposure to the program based on date and region of birth. For the purpose of our analysis, we only treat the combination of these two variations as exogenous. By its nature, the SD INPRES program was designed to allocate more schools to regions where primary enrollment was particularly low, inducing heterogeneity in the number of average schools built per district. A second source of variation is reflected in the age of the students. A child of 12 years of age or older in 1974 when the SD INPRES schools started to operate, could not benefit from the program. On the other hand, a child aged 6 or younger in 1974 was young enough to fully benefit from the newly constructed schools. Similarly, children between 6 and 12 years of age in 1974 only enjoyed partial exposure to the program. If the program had an effect on an individual’s years of education, we would expect the program to have the largest effects for fully exposed cohorts, a somewhat mediocre effect on the partially exposed individuals, and no effect on children that were too old in 1974 to benefit from the program. The large, exogenous shock of the SD INPRES program enables us to differentiate between treatment and comparison groups, in the manner of a quasi-experiment using a difference-in-differences approach.
In line with previous literature, we find that the SD INPRES program had significant effects on schooling outcomes of Indonesia’s children. However, children did not benefit from the primary school construction program equally. In fact, we find substantial heterogeneity of the program’s effect between genders, mostly favoring boys over girls. On average, one more school per 1,000 children in the district of birth increased schooling duration by 0.12 to 0.21 years for boys. However, this effect is less clear-cut for women: depending on the sample specification and type of analysis, we were able to obtain significant results for women albeit substantially smaller in magnitude. In particular when breaking down the analysis on each birth year cohort, our estimates suggest that only the youngest cohorts hitherto benefited from the program, with effects ranging from 0.1 to 0.17 additional years of schooling for an additional school built in the district of birth. Such suggestive late onset of the program’s effectiveness on women motivated us to perform a persistency analysis on later female cohorts. Including eight additional birth year cohorts, we find increasing and significant effects of on average 0.2 additional years of schooling. These findings suggest that a potential effect on women’s schooling might have set in later. Our mixed evidence thus far gives rise to doubts whether the SD INPRES program worked on the extensive margin of schooling attainment. Due to possible selection disadvantages at the transition between primary and secondary school, the effect of the program for girls might not be reflected in the in years schooling but merely in their likelihood of primary school completion. We find significant but negligible effects on women’s likelihood to graduate from primary school.
Ideally, a policy trying to increase education should be targeted at disadvantaged groups of society in order to decrease inequality and break the vicious cycle of poverty. Our findings suggest that women in Indonesia are particularly disadvantaged when analyzing educational outcomes. Further research should hence try to identify other possible heterogeneities when stratifying the sample interest. During our analysis, we made first attempts by looking at different specifications such as the difference in effects between individuals living in rural versus urban areas, or looking at potential heterogeneity varying by ethnicity. Such deviations of our analysis might unmask further heterogeneity, which will be essential to identify when assessing and improving Indonesia’s education policy. It will therefore be necessary to devote further research and policy attention towards the long-term impacts of the program by analyzing the educational and labor market outcomes of later cohorts, captured in the SUPAS 2005 and 2015 polls. On the basis of these results it will be necessary to identify and study potential policy remedies to overcome the tremendous challenges Indonesia’s education system faces.