Estimating the causal impact of child work on the contemporaneous health of a child has proven quite challenging given non-random selection of children (i.e., selection based on unobserved attributes) into the labor market and the inability to find credible instruments. Recognizing this difficulty, we instead assess the robustness of the impact of child work estimated under the assumption of random selection conditional on covariates to varying degrees of non-random selection using a methodology proposed in Altonji et al. (JPE, 2005). The nature of the selection process is also identified using parametric assumptions. The results indicate that the relationship between child work and health is extremely sensitive to non-random selection. There is evidence of positive selection into work for children with BMI-for-age below the fifth percentile and above the median. There is, however, negative selection into work for the children with BMI-for-age between fifth and 50th percentile. Moreover, there is also evidence suggesting a heterogeneous impact of work on the health of the children 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. Important differences by gender are also uncovered.

Is a joint decision by parents regarding the children's health expenditure more beneficial for the children's health, instead of a sole decision made by a mother (or father)? We use data from the Indonesian Family and Life Survey (2000) where the reports from the parents about the decider regarding health expenditure is sometimes conflicting. Thus, we have measurement error in our binary regressor, which implies that our measurement error is nonclassical (i.e., the measurement error is correlated with the regressor). The regressor is also assumed to be correlated with the random error term due to unobserved factors. Since we allow for endogeneity arising due to nonclassical measurement error and unobserved heterogeneity, the Instrumental Variables (IV) and Ordinary Least Squares (OLS) estimates are biased. We prove that under certain conditions the IV and OLS estimates provide upper and lower bounds to the true value of the regressor. These results generalize the results obtained in Black et al. (2000) for non-classical measurement error and is of independent interest. Using our theoretical results, we can conclude from the empirical analysis that there is an indication of positive impact of the joint decision by parents, while there is an indication of a negative impact of the mother's or father's sole decision on children's health.

Using data from 1997 and 2000 wave of Indonesian Family Life Survey, we investigate the effect of mother's marital bargaining power on the health of her children. We construct a new measure of bargaining power using the information on bride price paid by the father's family to the mother's family at the time of the parents' marriage. The stochastic frontier method proposed by Polachek and Yoon (1987, 1996) is used to construct the bargaining power for each mother-father pair. The reports on bride price from both the parents may not agree resulting in measurement error in the relative bargaining power. Hence there is endogeneity due to measurement error and we use the instrumental variables technique to obtain consistent estimates. Since there is a lack of valid instruments in our dataset, we construct instruments from simple functions of the model data exploiting third moments of the data following Lewbel (Econometrica, 1997). The results indicate a positive effect of the mother's marital bargaining power on the children's health. This is similar to the effect obtained using traditional bargaining power variables of mother. Similar results are also obtained by analyzing the sample by gender.