In the Same Boat, but not Equals: The Heterogeneous Effects of Parental Income on Child Labour

Abstract

This paper examines the impact of parental income on child labour. The empirical literature has found conflicting results regarding whether poverty leads parents to send their children to work. Most of this literature, however, treats child labourers as a single homogeneous group, ignoring differences among working children in work intensity, hazard exposure, and type of employer. This paper argues that accounting for the heterogeneity in child’s working conditions may explain the conflicting results in the literature. Specifically, the existence of this heterogeneity may reflect heterogeneity in parents’ perceptions about the returns to child’s work, and hence in parental reasons to send their children to work. To test this hypothesis, I estimate the effects of parental income on child labour for various working conditions, using data from the 2010 Egypt National Child Labour Survey. This dataset provides rich information on the working conditions of child labourers. The findings show that the effect of parental income on child labour is minimal among children who work in non-hazardous jobs, jobs that are not highly physical, or in family businesses. In contrast, higher parental income does decrease the likelihood of child labour in market work, jobs that are physical and hazardous jobs.

Acknowledgements

I thank Shiferaw Gurmu, Barry Hirsch, May Gadallah, and Mahmoud Elsayed for valuable comments. This paper has also benefitted from helpful comments by Rachana Bhatt, Madeline Zavodny, and seminar participants at the 2015 Southern Economic Association Conference. I appreciate funding received from the Dan E. Sweat Dissertation Fellowship through the Georgia State University Andrew Young School of Policy Studies. The dataset analysed in this study was obtained from the Egyptian Central Agency for Public Mobilization and Statistics (CAPMAS) through a restricted-use data application.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. The total sample is composed of 66,922 children between the ages of five and 17. I restrict the sample to children whose relationship to their heads of households are either sons or daughters, and where at least one parent is present in the household. This limits the sample to 64,387 children. Since this paper focuses on child labour and schooling decisions, I exclude children who were too young to attend school (based on households’ answers to the question of why a child does not go to school) because attending school, in this case, is not a choice that families can make. This restricts the sample to 58,884 children of age 6–17. I have also dropped observations where household monthly income is missing or below 100 EGP. This restricts the sample to 54,671 children. Finally, I dropped two observations where child’s school status was missing. Thus, the total analytical sample is 54,669 children.

2. The definition of child labour of the ILO is guided by the principles enshrined in the ILO’s Minimum Age Convention No. 138 and the Worst Forms of Child Labour Convention No. 182.

3. I measure household income throughout this paper as the total income brought by all household members excluding income brought by children under the age of 18.

4. It is worth mentioning that there are some overlaps between these four dimensions. Specifically, there is an overlap between work intensity, hazard exposure, and age at first job. Specifically, the majority of child labourers (using the ILO definition) are engaged in hazardous work and had started work when they were kids under 12. In contrast, none of the children who are engaged in light economic activity face hazardous conditions during work and the majority of them had started to work when they were adolescents age 12–17. This implies that child labourers and children engaged in light economic activity are not only different in work intensity, but they also differ in exposure to hazards and their ages at first job. Therefore, the difference in the effect of income between these two groups will capture the effect of three dimensions without being able to isolate the effect of each of them. This is not, however, expected to be problematic as the differences between the two groups are consistent in a way that differentiates between favourable versus unfavourable working conditions. Therefore, the estimated combined effect will be quite informative.

5. I also tried another specification where I used income shocks over the past 12 months to instrument for household income. I focused on two sources of income shocks that are least likely to be correlated with the households’ behaviours. Namely, I focused on illness/serious accident of a working member and the death of a working member. Because poor parents are expected to have higher probabilities of sickness and death due to lack of healthcare access compared to the rich, I created a wealth index using households’ ownerships of assets and durable goods and added the quintiles of this index to the set of control variables. I implement the instrumental variable analysis using the two-stage residual inclusion approach introduced by Terza, Basu, and Rathouz (2008). The results of this analysis (available upon request) are qualitatively similar to the main findings in Section 6. Specifically, household income has little effect on types of work with relatively favourable working conditions including working for their families, working with light workloads, or working in non-hazardous jobs. On the other hand, household income does deter types of work that are most likely to be harmful to children such as hazardous jobs and excessive workloads.

6. To further explore whether the correlation between the subpopulation regressions is driving the main findings of this paper, as an alternative approach, I have also run my regressions on a sample of children with no siblings under 18 (n = 8747). In this case, children in the subpopulation regressions belong to different households, and hence the subpopulations regressions are independent of each other. The results of this analysis are shown in Table A6 in the Supplementary Materials and are consistent with the main findings in Section 6.

7. For example, for the work intensity criteria, I run a bivariate probit regression for child labour-ILO and a bivariate probit regression for light economic activity. Again, I use nonworking children as a comparison group in both regressions. I control in both regressions for the number of siblings who participate in the child labour-ILO and the number of siblings who participate in light economic activity.