Using a bivariate expected utility framework, we develop a two-period model where households determine, in the presence of risks, the parents' migration duration when children are left behind. Our model suggests that the optimal migration duration may respond differently to an increase in a given risk. We provide conditions under which it is optimal for households to decrease the parents' migration duration despite an income risk in the place of origin, and to increase it even though the income in the place of destination is risky. The idea of preference for "harm disaggregation" is used to explain the results. In the absence of uncertainty, we also show the role of the interaction between child human capital and wealth in the household's utility function in determining the optimal migration duration of parents. Empirical implications of this analysis are presented in the last part of the paper.