Gender differences in numeracy in Indonesia: evidence from a longitudinal dataset

Abstract

This paper uses a rich longitudinal dataset to measure the evolution of the gender differences in numeracy among school-age children in Indonesia. Girls outperformed boys by 0.08 standard deviations when the sample was around 11 years old. Seven years later, the gap has widened to 0.19 standard deviations, equivalent to around 18 months of schooling. I find no evidence that households invest more resources in girls relative to boys. However, there is suggestive evidence that schools play a role in fostering the gender gap in numeracy. Given the importance of numeracy in later life, there may be some scope for public policies to address the widening numeracy gap between genders.

Keywords:

• numeracy
• gender gap
• Indonesia

JEL Classification:

• I21
• J16
• O15

Acknowledgement

I have benefited from comments and suggestions provided by Evgenia Dechter, Paul Frijters, Arya Gaduh, Andrew Leigh, Peter Orazem, Robert Sparrow, Asep Suryahadi, Yuji Tamura, Julia Tobias, and seminar participants at Australian National University, the Labour Econometrics Workshop at Deakin University, SMERU Research Institute, and University of Indonesia.

Notes

1. The four countries are Brazil, Tunisia, Turkey, and Uruguay. Information on PISA is available at http://www.oecd.org/pisa/.

2. Separating all three effects requires applying some restrictions on the econometric model (see, for example, Hall, Mairesse, and Turner 2007).

3. This information is taken from https://www.cia.gov/library/publications/the-world-factbook/geos/id.html.

4. The IFLS is publicly available at http://www.rand.org/labor/FLS/IFLS/.

5. The EK modules and the rest of IFLS questionnaires are available at http://www.rand.org/labor/FLS/IFLS.html.

6. In contrast to the large and significant gender gap on numeracy, I find no gender gap in skills measured through the second part of the EK module, the Raven's Test.

7. Information on TIMSS can be accessed from http://timss.bc.edu/.

8. All correlations between the independent variables are below 0.5. Therefore, multicollinearity is not an issue.

9. I also estimate an individual fixed effects estimation, which removes all time-invariant sources of bias. As a comparison to the fixed effects results, I also estimate a pooled OLS estimation (results available upon request). The results using all three methods (OLS, panel fixed effects, and pooled OLS) are qualitatively similar.

10. There may be a worry that the expenditure variables are endogenous to the outcome variable. I test for the robustness of the results by excluding these variables. The resulting gender gap estimates (not shown) remain robust to excluding these variables.

11. This calculation is done by estimating an OLS regression with standardized EK2 2000 score as the dependent variable and years of schooling as the independent variable on a sample of out-of-school adults, controlling for age.

12. This hypothesis carries an implicit assumption that household fertility decision does not depend on the sex of the children, as endogenizing fertility decisions is beyond the scope of this paper.