ABSTRACT
This paper examines the relationship between armed conflict intensity and child labor using household level data from Iraq and taking advantage of a quasi-experimental setup. Armed conflict intensity is measured by the number of deaths related to conflict, and child labor is separated by type of work: economic and household. After controlling for individual and household characteristics that determine child labor, we find that armed conflict intensity is associated with a higher likelihood of entry into economic work sufficient to qualify as child labor, but is not associated with entry into household child labor. However, conflict intensity is associated with marginal increases in hours worked for both types of activity. We also explore gender differences. These results provide further evidence of the long-term costs of war on households.
KEYWORDS:
Acknowledgement
The authors would like to thank Maia Sieverding, Semih Tumen, participants at the Youth Vulnerability 2017 Workshop in the MENA Region in Cairo, and seminar participants at Texas Christian University and at the American Economic Association meetings for comments on earlier drafts.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 Author’s calculation based on Harbom and Wallensteen’s (Citation2007) report of 232 armed conflicts since World War II. The average MENA share of the world’s population is around 5.3% for the 1992–2015 period.
2 A 1996 wave is not publicly available, and UNICEF is currently working on the 2017 wave.
3 See UNICEF’s definition of child labor at https://www.unicef.org/infobycountry/stats_popup9.html (accessed on June 11, 2018).
4 The school attendance rate of the pooled sample is 91%, so we do not lose a large share of the full sample with this limitation. An earlier version of this paper includes the full sample. The results are similar to those presented here.
5 We also allowed the 75th percentile threshold to vary by year, and the result holds.
6 Given the existing data we cannot check the parallel trends assumption. To test the orthogonality assumption of the treatment and control variables, we present 15 estimates of the difference-in-differences coefficient in Appendix Table A1, with each control variable separately as a dependent variable. Out of 15 coefficients, 8 are statistically significant (7 statistically significant at the 5% level). Ideally, all 15 coefficients would not be statistically significant. That said, all coefficients are small (with the exception of the number of household members).
7 For more details on this debate see Ai and Norton (Citation2003) which first suggested a way to calculate marginal effects for interaction terms in nonlinear models, and Greene (Citation2010) which cautions against their interpretation of partial effects in this case.