Measuring Child Labor: The Who's, the Where's, the When's, and the Why's.

Measuring child labor accurately is a major challenge: parents' and children's reports tend to differ dramatically, and there is typically no way to verify whose reports are truthful (if any). To overcome this challenge, this paper uses novel data from a cocoa certifier in Côte d'Ivoire that draws on satellite imagery to minimize under-reporting. Concretely, aerial photos allow them to select remote and hard-to-reach communities-where parents typically have not been sensitized by government or NGOs, averting social desirability biases-and to visit these communities while cocoa is being harvested-precisely when children in employment are very visible, making it easier for enumerators to impute it if parents still fail to report it. We compare their figures with those obtained from business-as-usual surveys with parents and children in these regions, and find that (1) reporting inconsistencies between parents and their children in fact decrease with household remoteness; (2) adults dramatically under-report child labor relative to the certifier data, by a factor of at least 60%; and (3) in turn, children self-reports are statistically identical to the latter. Taking advantage of an experiment that randomly assigned a text-message campaign to discourage child labor, we further show that parents' reports not only underestimate its prevalence, but can even lead to the wrong conclusions about the effects of policy interventions.