Improving household water collection and storage behaviour using a randomised video intervention in South Africa

The intervention is a locally produced water-safety video filmed in Siyanyanzela, where the study is based. The video addresses (i) safe water storage and handling practices, (ii) risks associated with makeshift household taps, and (iii) reassurance that municipal standpipe water is safe at source despite occasional discoloration. The video was conceptualised and produced using community members and with feedback from the community and includes messages from the local nurse and municipal ward councillor. The video is approximately 8 minutes long, delivered in isiXhosa and English with simple visual demonstrations.

In October 2025, trained fieldworkers will visit each household randomised to the treatment group. Using tablets, they will play the video in-home in the presence of household members. Immediately after viewing, fieldworkers will administer a recall checklist, record who was present, and attach a laminated reminder sticker summarising the 5 most important messages from the video to the water storage container of the treated household.

Since mobile data is expensive and network access is not universal in these communities, this mode of delivery ensures uniform exposure to the intervention. Control households will receive no video or sticker during the study period. Reminders about the video content will be sent via Whatsapp to a sub-sample of treated households, however this will be exploratory and not part of the main intervention.

2025-10-08

2025-11-01

Stored-water contamination (binary variable). This measures if the household’s main stored drinking-water sample tests positive for E. coli using the Colilert screening test at baseline and endline (1 if positive, 0 if not). The test is conducted in a laboratory on campus and the water sample is taken from the household’s stored water container at the same time of the interview and then transported in a cooler box to the laboratory within 4 hours.

Diarrhoea incidence (binary variable). Variable indicating whether any member of the household is reported to have experienced diarrhoea in the past 30 days at baseline and endline. The variable is calculated from the household roster with individual-level reports aggregated to the household level.

Safe storage and handling behaviour index (continuous variable). Standardised score (mean 0, SD 1) combining observed and self-reported behaviours: observed presence of lid on storage container; dedicated scoop/cup; container location; frequency of cleaning container; handwashing frequency; observed presence of soap.

Trust in municipal water supply (Likert scale). Respondent’s rating of trust in municipality’s ability to provide water that is safe to drink, measured on a 5-point scale (strongly distrust to strongly trust).

Source of drinking water (binary). Indicator = 1 if the household reports usually collecting drinking water from a municipal standpipe (vs. makeshift/illegally connected tap).

Individual-level diarrhoea incidence.

Household exposure to the video/sticker. To measure spillovers the endline survey will ask all households whether they have seen the video or the stickers.

This is a two-arm household-randomised controlled trial testing whether a short locally produced water-safety video (delivered in-home on tablets) plus a visual reminder improves (i) microbiological quality of stored drinking water and (ii) household diarrhoea outcomes, with secondary effects on storage/handling behaviours and trust in municipal water. The study takes place in two informal settlements outside Grabouw, South Africa: Siyanyanzela (n=520) and Khutsong (n=100).
The unit of randomisation, intervention, and analysis is the household. The sample includes 620 baseline households, which were surveyed in April 2024. Roll out of the video is planned for October 2025, and endline survey collection for November 2025.

There are two arms to the study:
• Control: No in-home visit during the study period.
• Treatment: One in-home visit in October 2025. Fieldworkers play an ~8-minute water-safety video on a tablet to available household members and place a small visual reminder (sticker with simple icons) on the principal drinking-water container.

Assignment to treatment is randomised using software. Treatment will take place at the household level with stratification and blocking to maintain balance.
• Strata: (i) Settlement (Siyanyanzela vs Khutsong); (ii) baseline stored-water contamination (any E. coli vs none)
• Blocking: Permuted blocks of varying size (e.g., 4, 6) within strata. Allocation is generated centrally; fieldworkers receive assignment status only.

Randomisation lists are prepared by the PI. Although all fieldworkers are residents in the study communities, treatment delivery fieldworkers are distinct from endline enumerators, who are unaware of assignment as far as possible. Intervention exposure occurs in-home only; no public/group screenings are conducted.

During the treatment visit, fieldworkers record start/stop timestamps, who viewed the video, a brief recall check, and confirmation of sticker placement. However, the analysis will be on intention to treat, and these measures are merely procedural in nature.

Contamination will be limited through in-person viewing, however spillover effects will be measured by asking all households about any exposure to the video/sticker at endline. In addition, GPS is used to compute distance to treated households.

Baseline data was collected in April 2024 and included a household survey (including questions on water collection and storage behaviour) and the collection of a stored-water sample (Colilert screening), plus roster-based health questions.

The endline survey is planned for November 2025 (based on weather data for the past 10 years, October/November is similar in temperature and precipitation compared to March/April). At endline, the survey will be repeated and will include: stored-water sample from the main drinking container; diarrhoea (30-day recall, aggregated to the household level); observed and self-reported storage/handling practices; attitudes toward municipal water; exposure checks.

Randomisation is done in an office using Stata on a laptop

Household

No

620 households

620 households

310 households treatment, 310 households control

For the primary outcome variables, using the data at baseline: The presence/absence of E. coli in the stored water samples: mean 0.1212 (sd 0.327) The proportion of households who reported someone in the household with diarrhoea in the past 30 days: mean 0.2121 (sd 0.409) Based on this, assuming a two-sided test, with alpha=0.05 and power=80%, the minimum detectable effect size for the E. coli presence/absence is approximately 7 percentage points, and for the diarrhoea it is approximately 9 percentage points. The R-squared estimated on the baseline data is low (0.02) and it is therefore not anticipated that adjustments by covariates will have much of an impact on the MDE