Experimental Design Details
Our setting is a subset of villages covered by Evidence Action’s Dispensers for Safe Water (DSW) program. Evidence Action, starting with a pilot of 20 dispensers in western Kenya in 2007, has since grown to operating 27,500 dispensers serving 4 million people across Kenya, Uganda, and Malawi today. We focus on their operations in western Kenya, specifically Kakamega and Bungoma county, where we are able to exploit random variation in dispenser provision from the Kenya WASH Benefits study (Null et al., 2018), a RCT which examined the effects of water chlorination on early child development. In particular, we study children born to women after January 2008, living in 132 villages (grouped into 82 “clusters” of 1-3 villages each). Of these villages, 65 had been assigned to a “water” (W) treatment and 67 to a “passive control” (C) condition in the WASH study. In the W villages, the study randomly allocated the provision of dilute chlorine solution in two ways: through point-of-collection chlorine dispensers, and provision of dilute chlorine solution (“WaterGuard”) to households with children for one year. Specifically, chlorine dispensers were installed at a subset of public water sources in each village, on average 5 per cluster. The dispensers provided a dosed amount of chlorine solution, enough to safely disinfect the typical amount of water collected. In addition, all family compounds in dispenser villages which had children under 5 years of age at the time of the original WASH study were given 1-liter bottles of dilute chlorine solution for point-of-use water disinfection every 6 months for the first year of the intervention, where each 1-liter bottle expires within 6 months. Our analysis uses survey data (Haushofer et al., 2018) that allow us to examine mortality for a sample of children over twice as large as that of Null et al. (2018) and for a longer period of time (between 1 to 3 additional years, depending on the area).
This study estimates the intention-to-treat effect of the community-wide provision of dilute chlorine solution through dispensers and household delivery on all-cause child mortality. Following McKenzie (2012), we condition on the pre-intervention under-5 (under-2) mortality rate of the cluster to improve statistical power (ANCOVA framework). The pre-intervention mortality rate is computed with all children born between January 1, 2008 and before the roll-out of the intervention. We also estimate a difference-in-difference model which compares changes in mortality rates (before and after the intervention) across treatment and control areas for children born since January 1, 2008. For each model, we first estimate the effect of the community-wide provision of dilute chlorine solution through dispensers and WaterGuard on under-2 and under-5 (years) mortality. , We will look at the impact of both bottled chlorination and dispensers together, as well as just dispensers by focusing on children born at least one year after the rollout of the intervention, who therefore would not have benefitted from the provision of WaterGuard. Because it is possible that the provision of WaterGuard during the first year of the WASH-B study also influenced these women’s use of WaterGuard later on, we also conduct a robustness check in which we exclude these women from the sample entirely (including all of their children, rather than just those born shortly after the beginning of the intervention). Second, we split up the previous results further in order to examine mortality effects at different times after birth among children who survived until a certain age. In particular, we examine effects on neonatal (< 1 month), post-neonatal (1≤ months <12) and child (1 ≤ years <5) mortality. We chose these periods following the analysis of the global burden of disease due to diarrhea (and other infectious diseases) by Gakidou et al. (2017). This strategy will identify the post-natal periods where observed mortality is reduced and shed insight on survival dynamics.