Safe Drinking Water at Home: Evidence from a randomised trial in Uganda

The intervention consists of the random distribution of three different water treatment methods over the selected treatment population and training in its correct use. The treatments are a rocket stove to boil water, ceramic filtration and membrane filtration. Each drinking water filter is accompanied by a leaflet with instructions on the use and maintenance needs. The drinking water filters and cook stoves are distributed to the dwellings of the selected households by members of HEWASA and Kiima foods, with the help of local guides of the Village Health Teams (VHTs). Members of HEWASA provide small group training in each village for the responsible household members in use and maintenance of the assigned water treatment method.

2023-04-16

2023-04-30

1. Drinking water quality
2. Health
3. User acceptance
4. Affordability
5. Labor burden

1. The participating households will be asked to identify the main drinking water storage container in the dwelling and a water sample will be taken from the designated water container. To assess the microbiological improvements in drinking water quality, household water samples will be analyzed in the lab on the presence of total thermotolerant coliforms (TTC), and in particular the bacterium E.coli. The assessment will be performed using the membrane filtration technique on membrane lauryl sulphate medium. The unit of measurement is Colony Forming Units (CFU). A second aspect of water analysis is turbidity, i.e. a measure for the presence of particles in the drinking water. The unit of analysis is Nephelometric Turbidity Units (NTU). The values of TTC, E.coli and turbidity will both be reported in levels (CFU/NTU). All microbiological data will be log transformed.

2. Health will be measured by surveying the female household head on medical care seeking and medical costs in the last month. Furthermore, for each household member the incidence of illness in the last seven days will be recorded.
Diarrhea prevalence in young children will be closely monitored by self-recording cases daily among children below five years of age in a health diary. A diarrhea case is defined according to the WHO-definition as “three or more loose stools passed within 24 hours, or more frequently than is normal for the individual''. The diaries also include the symptoms of vomiting and fever.

3. Apart from water quality and health benefits, the products are evaluated from a user-perspective. At each household visit the enumerators observes the presence and functionality of the product, surveys on frequency of usage and untreated water consumption and observes whether the filter contains treated drinking water and the cook stove is used. Furthermore, the user is asked to rate the product on easiness, safety, robustness, effectiveness and speed.

4. To address potential energy savings due to water filtration or energy efficient boiling, all respondents will be surveyed on the household expenses for charcoal, firewood or other fuels used in the previous seven days. Furthermore, a willingness to pay questionnaire is conducted to elicit users' product value under real market circumstances.

5. The labor burden is estimated by surveying the main operating household member on the time spend for treating drinking water. We measure, the volume of drinking water fetched and treated, as well as the time spend on and the frequency of each task involved in the process. This allows to compute a total time per volume of water. Furthermore, the respondents are questioned on the main operators of the product to detect gender biases in labor burden.

/

/

Experimental evidence to answer the four research questions will be obtained by a four-arm randomized controlled trial over an 18-month period. The trial will include three treatment groups that each receive either a cook stove for energy efficient boiling, the ceramic filter or a membrane filter, and one control group that continues business as usual. The RCT design allows to identify and compare the long run causal impacts of the three treatments on the outcomes listed above. Data will be obtained by a baseline survey and four follow-up surveys, accompanied by drinking water quality analysis at point of source and point of use. The treatment will be phased-in i.e., all households in the control group will receive a water filter at the end of the experiment.

Not available

The randomization is done per strata, which is in this study the two districts of interest. All eligible households in the two districts are randomly assigned to the four trial arms (boiling, ceramic filter, membrane filter and business as usual) with an assignment rate of 25% in each arm. The study arms are non-overlapping arms. The randomization is done by means of the random number generator in Stata 17.

Randomization is done at household level. This is individual randomization for the outcomes drinking water quality, uptake and household savings. This is group level randomization for the outcome diarrheal disease, since a household often counts multiple children below five years.

No

The households can be considered as clusters for the outcome diarrheal disease (for reasons outlined above).
600 households are involved in this study.

600 households and approximately 900 children under five years.

The four treatment arms all contain 150 households and approximately 300 children.

We estimate the minimal detectable effect sizes (MDEs) at each follow-up for microbiological water quality and health outcomes in children under five years. We design an experiment with 80% power and 5% significance level and rely on the standard deviations of the outcome variables in the baseline data. The MDEs are conservative since they account for 10% attrition and 5% non-compliance. In the first scenario, we find MDEs of 51% and 49% for E.coli and TTC, which are very likely given the 99% efficacy of the household water treatment methods. We are able to detect a reduction of 44% of E.coli and 43% in TTC in the pooled scenario. There is a 40% reduction in turbidity detectable in the first scenario and 39% In the second scenario. Considering child health in the first scenario, we are able to detect reductions of 69%, 39% and 89% in respectively diarrhea, fever and vomiting in the last seven days. The number of days ill should decrease by 43% in order to be detected. Given that household water treatment is believed to reduce diarrhea by a half, these MDEs are on the high end. We are more likely to detect an effect when pooling the treatment arms. The MDEs as percentage of the mean are then estimated at 46% for diarrhea incidence, 26% for fever, 59% for vomiting and 29% for number of days ill. Remark that we can as well pool the observations over time given the panel structure of the final dataset. Pooling the observation over time increases the sample size and reduces the MDEs for the given significance level and power. aseline parameters One treatment arm vs control Pooled treatment arms vs control Variables Mean SD MDE MDE std. MDE % MDE MDE std. MDE % Ecoli (10logCF U/100mL) .85866147 .86257356 .3094416 .35874227 50.959103 .25265801 .29291183 44.108986 TTC (10logCF U/100mL) 1.4163516 .8232103 .29532033 .35874227 49.33831 .24112804 .29291183 42.605278 Turbidity (10logN T U/100mL) .60001236 .61352879 .2200987 .35874227 39.757736 .17970984 .29291183 33.886497 900 children under five evenly divided over four treatment arms account for a minimal detectable effect size of 0.13 in diarrheal disease incidence in the past seven days. The standardized minimal detectable effect size is 0.32. Baseline parameters One treatment arm vs control Pooled treatment arms vs control Variables Mean SD ICC MDE MDE std. MDE % MDE MDE std. MDE % Diarrhea .23825504 .42625439 .32675844 .16493471 .38693962 .69226116 .10995647 .25795975 .46150744 Fever .49328858 .50023478 .25861463 .1907052 .38123137 .38659966 .1271368 .25415424 .25773311 Vomiting .16442953 .37087238 .40639234 .1459403 .39350545 .8875553 .09729353 .26233697 .59170353 Days ill 2.2940516 2.5927639 .28176934 .99349666 .3831805 .4330751 .6623311 .25545368 .28871673 The estimated effect sizes are rather conservative as they already account for 5% non-compliance in the treatment group and 10% attrition. Moreover, these calculations do not apply to the pooled estimator over the four follow up rounds. The pooled estimator will have higher power stemming from the multiple rounds of post-treatment data collection, as suggested by McKenzie (2012).