Abstract/Summary

We examine the effect of domestic violence on child mortality using Demographic and Health Surveys from thirty-two developing countries. We first examine conditional associations between violence faced by the mother and child mortality after controlling for observable confounders. Children of (ever) physically victimized mothers are 0.4, 0.7, and 1.0 pp more likely to die within thirty days, a year, and five years of being born. We find similar associations when examining violence experienced in the last twelve months, although these are no longer statistically significant. The association is statistically significant, and larger, if the mother experiences violence in the last twelve months often, rather than sometimes. Violence is significantly associated with pregnancy loss, suggesting the true effect on mortality is larger than estimates based on live-births would suggest. We investigate robustness of associations to omitted variable bias, assessing the role of selection on unobservables to estimate lower bounds on the true effect. These continue to indicate economically meaningful positive effects, suggesting selection on unobservables would need to be 2.4-3 times that of selection on observables to nullify the estimated effect. We provide evidence that maternal smoking and breastfeeding practices are mediators in the relationship between domestic violence and child mortality.