Determinants of Indonesian rural secondary school enrolment: gender, neighbourhood and school characteristics

Abstract

In recent years the school enrolment rates of children aged 13–15 and 16–18 years have increased sharply in Indonesia, not only in urban but also in rural areas. Using various data sets spanning the years from 1993 to 2007, this paper investigates changes in factors associated with the enrolment of secondary school aged children in rural areas. It sheds light on the roles of gender and of neighbourhood and school characteristics, which have rarely been examined in the Indonesian context. The study finds that the disappearance of a gender gap in secondary enrolments between 1993 and 2007 contributed significantly to the rise in the overall enrolment rate. The findings also show that children living in wealthier communities and communities with a high proportion of enrolled children are more likely to attend school. Finally, various school characteristics are shown not to be strongly or consistently correlated with school enrolment.

Keywords:

• schooling investment
• gender
• rural development

Acknowledgements

This paper is a product of a research project undertaken at the Institute of Developing Economies – Japan External Trade Organization (IDE–JETRO). I am grateful to the editor and two anonymous referees for useful comments and suggestions. I also wish to acknowledge the financial support provided by IDE–JETRO. The views expressed in this paper are the sole responsibility of the author and do not necessarily reflect the views of IDE–JETRO.

Notes

¹In addition, there are smaller numbers of other religious schools, such as Christian and Buddhist institutions.

²The highest unit of Indonesian local administration is the province (propinsi), followed by the district and municipality (kabupaten and kota, respectively, referred to collectively as ‘local governments’), then the sub-district (kecamatan) and finally the community or village (desa).

³The figures presented in table 1 are from the Podes data set, and differ from the number of schools reported by the Ministry of Education and Culture. Podes does not include data for this study's target year of 2007, so I use 2008 Podes data to reflect school numbers in 2007.

⁴In 1993 there were 27 provinces and 303 local governments.

⁵Kevane and Levine (2003) find, however, that Indonesian parents treat sons and daughters roughly equally in schooling investment and that the investment level is not affected by whether daughters move away from the original household.

⁶Quisumbing, Estudillo and Otsuka (2004) argue that recent increased investment in female schooling in the rural Philippines reflects judgments about the comparative advantage of males and females: males are relatively well suited to farming, and thus receive more farm land from their parents, while females are relatively better suited to non-farm activities, and so receive more investment in schooling.

⁷If there is no spouse, the years of spouse education are set at zero.

⁸Because there are households with no farm land, I convert this variable to log (value of farmland + 1).

⁹Additional controls such as the factory and bank dummies are selected because they are thought likely to affect expected returns to child education and costs of schooling, and thus influence current school enrolment. The factory dummy, for example, will capture the opportunity cost of schooling: having a factory in a community may increase the opportunity cost, and thus reduce current enrolment. On the other hand, having a bank in the community may increase the availability of credit, and thus have a positive impact on enrolment.

¹⁰Another possible solution is to construct the average enrolment rate excluding the individual observation in question.

¹¹As explained, the dependent variable takes the value of 1 if a child was enrolled in school at the time of survey. Because grade repetition and delayed enrolment were not negligible in rural Indonesia, children aged 13–15 were not necessarily in junior secondary school and children aged 16–18 were not necessarily in senior secondary school. I confirmed that the key findings remained the same, however, if the sample children were restricted to those who graduated from primary school (in panel A) and from junior secondary school (in panel B).

¹²Ideally, household fixed effects would be included in the estimation model to control for time-invariant unobserved household characteristics. To do this, however, one needs to have multiple observations of the same age categories (13–15 and 16–18 years) from the same household over the three survey rounds. This significantly reduces the number of observations and makes estimation less efficient. Therefore, I estimated the functions separately by year, treating the data as repeated cross-section observations. I also tested whether estimated coefficients differ significantly across years, by pooling all three time periods and interacting year variables of the survey with each of the covariates. Some of these results are explained in the main text.

¹³The full estimation results are available from the author upon request.

¹⁴I estimated the effect of the education of the head and the spouse interacted with the male dummy, and found that the interaction terms were statistically insignificant.

¹⁵These neighbourhood characteristics are potentially endogenous, because unobserved parental attitudes toward child education can be correlated with the choice of neighbours: parents who want their children to have a good education may move to communities where rich neighbours live and enrolment rates are high. To eliminate this potential endogeneity bias, one might use the instrumental variable (IV) method. However, the IV approach was not feasible in this study, owing to the lack of appropriate instrumental variables, so the reduced-form estimation was maintained. As a robustness check, I also attempted to estimate the schooling demand functions without these two potentially endogenous variables, and found that all other coefficients yielded qualitatively similar results to those presented in table 5.

¹⁶It would have been better to include in the estimation a variable representing proximity to the nearest school, to indicate the opportunity cost of travel. However, this was not possible because the IFLS does not capture such data.

¹⁷A possible reason for the weak significance of teacher salaries is that this variable does not capture ‘incentive’ effects well, in that salary levels are not well tied to performance. Even so, the findings show that raising salaries alone, as was attempted by the 2005 law, may not be effective in increasing enrolment rates.