Estimating the causal impact of child work on the contemporaneous health of a child has proven quite challenging given non-random selection into the labor market and the inability to find strong and valid instruments. Our data, the Indonesian Family Life Survey is no different. Recognizing the lack of a credible instrument, we instead pursue a different strategy based on the methodology of Altonji et al. (JPE, 2005). This method assesses the robustness of the impact of child work estimated under the assumption of random selection (i.e., selection into child work on observable attributes only) to varying degrees of non-random selection (i.e., selection into child labor on unobservable attributes). If the estimated effect is found to be extremely sensitive to selection on unobservables, then one should be wary about inferring an adverse causal effect of child work. In addition, the nature of the selection process is identified using parametric assumptions. The results are striking, suggesting positive selection of children into work when we consider underweight and high weight status as dependent variables. This indicates that there is both healthy worker selection effect as well as unhealthy worker selection effect. There is however negative selection into work for the children belonging to the intermediate weight range. This heterogeneity in the selection process across the distribution has not been previously identified in the literature. Moreover, we also find evidence suggesting a heterogeneous impact of child work on health once we allow for a modest amount of selection on unobservables. Specifically, we find evidence of a negative causal effect of work on healthier children, but evidence of beneficial impact of work on the least healthy children.
