The role of birth order in child labour and schooling

Abstract

Does when a child was born relative to his or her siblings affect whether the child attends school or participates in child labour? We investigate this question by estimating the causal effect of birth order on the probabilities of school attendance and child labour participation. To address the potential endogeneity of family size, we use instrumental variable approach where the proportion of boys in the family is used to instrument family size. Using a longitudinal household survey data from Ethiopia, we estimate unobserved effects bivariate probit instrumental variable model of school attendance and child labour choices. The results suggest that the probability of child labour participation decreases with birth order, but we find no evidence that suggests birth order affects the probability of school attendance. However, among children who are going to school, hours spent studying increases with birth order. Results from complementary time-use analysis reveal that there is no birth order effect on hours spent on household chore. However, hours spent on school increases with birth order, where the increase in hours spent on school seems to come from a decrease in hours spent on market work.

Keywords:

• child education
• child labour
• birth order
• unobserved effects instrumental variables model
• Ethiopia

JEL Classification:

• J13
• I25
• I24
• O10

Acknowledgements

We thank Rachana Bhatt and Barry Hirsch for their invaluable suggestions. We have also benefited from the comments of Roy Bahl, James Marton and Felix Rioja. In addition, we would like to thank the editor, two anonymous referees and seminar participants at Georgia State University, Clemson University and Southern Economic Association Annual Conference for their helpful comments. Finally, we thank Young Lives (www.younglives.org.uk) for providing us the data used in this article. Any errors are the responsibility of the authors.

Notes

¹ The report from International Labour Organisation reveals that there were 153 million child labourers in the world in 2008 (Diallo, 2010). In Ethiopia, the country which is also the focus of the present study, data from the 2011 Ethiopian Demographic and Health Survey have showed that about 27% of children between the age of 5 and 14 involved in child labour (EDHS, 2011).

² It is important to note that parents send their kids to work not because parents are selfish; it is because, for poor families, sending their kids to work is crucial for the households’ survival (Basu and Van, 1998).

³ Goux and Maurin (2005) also employ similar instrumental variable for family size when they assess the effect of overcrowded housing on children’s performance at school.

⁴ Given the history of war and less-developed police force, particularly in rural areas, Short and Kiros (2002) argue that bravery and physical strength are highly valued in Ethiopian families. Since men supposedly have these essential features, Ethiopian parents prefer boys to girls.

⁵ Though the legal school starting age is seven in Ethiopia, it is not uncommon for most children in developing countries like Ethiopia to delay primary school enrolment by few years beyond the legal school starting age (Barro and Lee, 2000).

⁶ Since we are interested in exploring birth order effect, it is crucial to observe at least two kids in a given household. In countries like Ethiopia where parents have an average of five kids, this type of sample restriction does not create a serious selection issue.

⁷ The 14 hours per week cut-off is chosen to be in line with ILO’s definition of ‘light work’ which is working for 14 hours per week or less.

⁸ Some argue (e.g. Williamson, 1976) that the relationship between the proportion of boys and family size holds if parents have a taste for small or moderate family size since in large families a mix of both genders is more likely to happen due to mere biological probability. This argument is valid if parents care only about having at least one child of each gender. However, if parents prefer a specific proportion of boys – say, more boys than girls – then preference for sons affect fertility even if parents have a taste for larger family.

⁹ By construction, family size appears on both sides of Equation 3: as a dependent variable and a denominator of the excluded variable, proportion of boys in the household. Generally, this could lead to a well-known bias in labour economics called Borjas’ division bias (Borjas, 1980) if there is measurement error in family size. As in most household survey data, measurement error in family size is not a serious problem in our data to make Borjas’ division bias a serious concern.

¹⁰ Ethiopia is characterized by high fertility rate, with, for example, more than five kids per woman in our sample. Given the high fertility rate and the presence of son preference, the magnitude of the coefficient estimate of the ‘proportion of boys in the family’ variable (i.e. having 2.5 fewer children) is not surprising.

¹¹ Data from the 2011 Ethiopian Demographic and Health Survey show that divorce rate in Ethiopia is about 2.5% for men and 7.4% for women.

¹² The coefficient estimate of $\rho$ in the bivariate probit model is –0.111 and is statistically significant at 1% level, implying bivariate probit model is a better fit than univariate independent probit models.

¹³ The compete regression results from the other models presented in Table 7 are available upon request or at the web link at http://www2.gsu.edu/~ecosgg/research/pdf/Seid&Gurmu_AE2015.pdf

¹⁴ Mean and SDs of study hours are 1.75 and 1.01, respectively.

¹⁵ The time-use models are also estimated by including those children that are not going to school. The results are comparable to those reported here. These results are available upon request or at the web link given in footnote 13.

¹⁶ Mean (and SDs) of hours students in our sample spend on school, market work and household chore are 4.98 (0.19), 3.94 (2.02) and 1.31 (1.74), respectively.

¹⁷ In the fixed-effect 3SLS IV model, Equation 3, as before, denotes the first-stage equation where family size is the dependent variable and proportion of boys is the exclusion restriction. Results reported in the fixed-effect 3SLS IV model are obtained by running 3SLS IV regression on time-demeaned data.

¹⁸ The compete regression results from the three time-use models reported in Table 9 are available upon request or at the web link given in footnote 13.

¹⁹ The regression results from the various ordered probit models are available upon request or at the web link given in footnote 13.

²⁰ Regression results from unobserved effect bivariate probit IV model where the first-stage equation is modelled as count data model using exponential mean are available upon request or at the web link given in footnote 13.