Child Labor and Schooling Decisions in Ghana

In this paper we investigate the choices involved in the tradeoff between a child’s labor outside of household work and their schooling hours from a sample of data from Ghana in the late 1980’s. While households are interviewed only once during the survey, the data were collected over an 11 month period for both the Northern and Southern regions of Ghana. Northern and Southern Ghana have rather distinct long-run rainfall patterns over the year. These two facts allow us to derive an identiÞcation strategy with good power for the child labor decisions made by households. Essentially, we use the month by region variation in child labor intensities as a source of exogenous variation in child labor, while controlling for secular month and region variation in child labor intensities. This approach uses the monthly span of the data to construct an identiÞcation strategy that uses the synthetic panel approach of Deaton (1985) to average out the unobserved preference and short-run income ßuctuations across month by region cells. We also consider a reÞned version of this strategy, in using both the realized rainfall (including sets of month and region dummies), as well as realized rainfall deviated from its long-term month by region means, as direct rainfall shock measures. These differing ways of using the rainfall data extract the different behavioral responses to short run and long run rainfall, and 1Boozer is at the Department of Economics and the Economic Growth Center, Yale University. Email: [email protected]. Suri is at the Department of Economics, Yale University. Email: [email protected]. We are thankful to the Ghana Statistical Service for permission to use their data and also to David Lister at the University of East Anglia, UK for access to the rainfall data. We thank Markus Goldstein, Chris Udry and the Development Lunch participants at Yale for helpful comments. thus to a degree income patterns. We interpret the strategy that uses simply month by region variations as being indicative of an average of the long and short run effects and that using the rainfall shock as indicative of the pure short run effect. We Þnd that both strategies give statistically similar impacts: an hour of child labor reduces school attendance by approximately 0.38 hours. Surprisingly, this effect is actually larger than the OLS estimate of the effect, 0.21. Also, these results imply that income, or poverty, may not be as important in determining child labor. A direct analysis of this effect indeed provides no evidence of heterogeneity of this substitution effect by our observed income measure.