Internal migration, children’s schooling and gender gaps in education

ABSTRACT

In many sub-Saharan African countries, a large number of people migrate internally due to economic, demographic and political factors. We ask whether household mobility affects child educational outcomes and in what direction. Because girls and boys face different challenges in achieving education, and families often underinvest in girls’ human capital, we focus on gender differentials. We study the case of Uganda, a country where universal basic education and gender equality are still complicated goals. We use the National Panel Survey of 2005, 2009, 2010 and 2011 to develop fixed effects models of the relationship between mobility and children’s educational outcomes: school delay and attendance. We find that moving with the family during primary or lower secondary school decreases delay in schooling and increases school attendance for boys, while we do not find any significant effect for girls. We argue that our results confirm pro-male preferences of families, where girls do not benefit equally from the opportunities offered by internal migration. Expanding girls’ opportunities, as well as providing adequate services, are crucial steps in achieving gender equality in education.

KEYWORDS:

• Internal migration
• gender disparity
• school delay
• Uganda
• child education

JEL CLASSIFICATION:

• I25
• J13
• O15

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Chea and Wongboonsin (2020) investigates the effects of internal rural-urban migration in Cambodia on the education of children left behind compared to the effects on children migrating with their parents. The study finds that children migrating with their parents are disadvantaged, but this result is not differentiated by gender since the survey does not ask about the sex of the child.

² Source: UNESCO http://data.uis.unesco.org. The gross enrolment rate is the total number of students attending primary school, regardless of age, expressed as a percentage of the official primary school-age population.

³ Gross national income per capita is lower than the average of other SSA countries, and even though poverty levels have been falling in recent years, Uganda still has a low Human Development Index (HDI), standing at 159th out of 189 countries. Human Development Report 2020, 2019 ranking. http://hdr.undp.org/en/data.

⁴ These statistics are available on request.

⁵ See Section 4 of the questionnaire. Questions about education are asked to all persons aged 5 years and above.

⁶ Source: UNESCO http://data.uis.unesco.org.

⁷ We discarded use of the latest waves of the UNPS, since the attrition rate with the baseline year was too high.

⁸ For the same reasons, given that our age range of interest involves very young children at baseline, it makes little sense to investigate school dropout, since in this context, children up to age 15 might go back to school any time.

⁹ See Section 3 of the questionnaire, question h3q15: “How many years has [NAME] lived in this place/village?” We have chosen five years, since this is the time span between the first and second wave. After 2009, the time span between the following waves is one year, so that under the same conditions, we capture the movements of children who did not move before 2009, but moved afterwards.

¹⁰ We use a dummy variable, since the very few multiple movements observed in the panel time span due to its structure (five-year time span between the first and the second wave, and then a one-year time span between each successive wave) did not allow us to build a continuous variable.

¹¹ Question h3q18.

¹² On a more practical side, using the within estimator, we smooth some measurement errors due to inconsistencies in the reported age and completed school grade, and therefore in the theoretical grade for the age variable used to build Delay.

¹³ Ch. 5, appendix 5.4, p.246-7: “…with a continuous outcome and a binary treatment one mistakenly estimate a first-differenced equation in an effort to kill the fixed effect”.

¹⁴ This information is drawn from the Community/Facility questionnaire.

¹⁵ IFPRI SPEED Database https://www.ifpri.org/project/speed (Accessed: July 2020).

¹⁶ World Bank Private Participation in Infrastructure (PPI) Project Database https://ppi.worldbank.org/en/ppidata (accessed May 10, 2021).

¹⁷ A tentative explanation might be that literate mothers have more chances to work, so that children have to replace them in the household chores, thus reducing attendance. We cannot control for parents’ hours worked because in years 2010 and 2011 it is not possible anymore to link children with parents. For education, we use information derived in 2005 assuming that the education of parents is concluded.

¹⁸ We have treated Urban as a fixed variable interacted with the time dummies because we could not detect enough variability—namely, changes of residence from urban to rural areas or vice versa in the panel time span—to estimate a time varying coefficient. The interpretation of its coefficients is as if the child had always resided in the same area as the baseline area. As mentioned in the data section, urban areas are underrepresented in the data because of the survey design and this might represent a problem for the estimated coefficient of this variable.

¹⁹ Full results are available on request.

²⁰ SFull results are available on requestSo that the coefficient of Delay is not constrained to be one as in first differences. Of course, the lagged value of Moved is zero.