“…Since we control for time-and district fixed effects, identification relies on the absence of time-varying omitted factors correlated with the timing of treatment. Regressing the treatment indicator on observable characteristics, we find that, other than the expected positive correlation between CHX-NCP and CB-NCP, the program on which CHX-NCP "piggy-backed" (Hodgins et al, 2019), the treatment is only weakly correlated with observable characteristics. 12 Among the sample of births predicted to take place in an institution, treated babies are significantly less likely to be found in rural areas, to be their mother's third born and more likely to have a mother with an ethnicity from the residual "other" group.…”