Early Marriage and Child Cognition: Empirical Evidence from Indonesia

Abstract

As a country with the fourth largest population in the world, the high number of child marriages remains a major concern in Indonesia. The country is positioned in the top 10 highest child marriage rates in the world and the second highest in Southeast Asia. The high rate of early marriage in Indonesia aligns with the low educational achievement among children. Therefore, this study aims to analyze the effect of early marriage on children’s educational achievement as measured by their cognitive abilities. Using the fifth round of the Indonesia Family Life Survey dataset in 2014, this study employs ordinary least squares with clustered standard errors at the household level in multiple regression analysis. The dependent variable is child cognition as measured by the percentage of correct answers achieved by a sample of 4542 children aged 7 to 14 years on the Raven test, while the main independent variables are the parents’ age of marriage. The result suggests that the age of marriage of mothers had a positive and significant effect on their children’s educational achievement. This finding implies that children born to mothers who married earlier tend to have lower cognitive abilities. However, this result is statistically detectable for daughters only.

Keywords:

• Early marriage
• education
• cognition
• Indonesia

JEL Classification:

• J12
• J13
• I25

Acknowledgments

We would like to thank James Henshall and Novia Iskandar for proofreading our article.

Notes

1 Notes

https://www.unicef.org/protection/child-marriage (Accessed on 2 September 2022)

2 Centre on Child Protection and Wellbeing at Universitas Indonesia, 2020.

3 Statistics Indonesia, 2015.

4 Statistics Indonesia, 2015.

5 Statistics Indonesia, 2015.

6 Ministry of Education and Culture, 2018.

7 The most common issue in cross-section data is heteroskedasticity or spatial correlation (Gujarati, 2004), and this can be corrected with clustered standard errors. However, it is not appropriate to perform a heteroskedasticity test after estimation using clustered standard error, nor the Durbin–Watson test for autocorrelation since residuals cannot be serially correlated without a time component.

8 We thank the anonymous reviewer who provides this suggestion.