Determinants of Child Labour and School Attendance: The Role of Household Unobservables

We develop a random effects multinomial logit model to distinguish between unobserved and observed household characteristics as determinants of child labor and school attendance. Using a semi-parametric approach, the random effect is drawn from a discrete distribution of latent classes of households. The results show that household-level unobserved heterogeneity is substantial. Household-level unobserved heterogeneity swamps observed income and wealth heterogeneity. Households that belong to the class with a high latent propensity to send their children to work are not influenced by marginal changes in the explanatory variables. Households most sensitive to changes in explanatory variables are those with a high propensity to have their children neither in school nor working. Policy interventions and changes in external conditions are likely to produce large changes in the behavior of this group of families.