Bringing Citizens and Government Together for Improved Water Access

How to cite

Sansonetti, Jérôme; Coville, Aidan; Rogger, Daniel. 2023. "Bringing Citizens and Government Together for Improved Water Access." AEA RCT Registry. AEARCTR-0011267. https://doi.org/10.1257/rct.11267-3.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Jérôme Sansonetti* , Aidan Coville , Daniel Rogger

* Primary investigator

Registry ID
AEARCTR-0011267
DOI
https://doi.org/10.1257/rct.11267-3.0
First registered
Last updated
Status
Completed
Trial dates
2018-10-01 – 2023-12-31
Countries
Tanzania, United Republic of ()
JEL codes
O18, O13, Q25
Registry record
http://www.socialscienceregistry.org/trials/11267

Abstract

This study evaluates the impact of a governance intervention that aims to improve the sustainability of communal water points in Tanzania. Responsibility for the maintenance of communal infrastructure in low and middle income countries is often ambiguous, leading to inefficiency and dysfunction in public and private investments. Roughly half of the communal water points in Tanzania do not produce water. To overcome inefficiencies in the maintenance of communal infrastructure, we study an intervention that aims to strengthen coproduction between the district governments and village water community organisations, which are jointly responsible for maintaining communal water points. The intervention consists of repeated `action-learning' consultations led by an independent facilitator that encourages information sharing and the resolution of ambiguities around responsibilities for maintenance between the two parties. We evaluate its impact through a cluster-randomized controlled trial and measure impacts on the functionality of communal water points, maintenance practices, and measures of coordination between the parties.

Intervention

Though public and private actors have invested in building water systems over the past four decades, water infrastructure in Sub-Saharan Africa displays high levels of disrepair, thereby translating into limited access to safe water. In Tanzania, formative research shows that while communal water points experience high levels of dysfunctionality, the vast majority of them would require relatively small investments to be brought back into service. This disconnect, between widespread dysfunction of water infrastructure and a lack of institutional support for repair, suggests an important line of investigation: what are the institutional arrangements that support maintenance of local infrastructure in low and middle income countries?

Most of the existing evidence on maintenance explores direct provision by either government or local communities. However, both approaches miss important complementarities which exist in many Sub-Saharan contexts: on the one hand, governments are best placed to source and develop expertise and undertake major capital investments. On the other hand, local communities can directly observe breakdowns at low cost, and process day-to-day repairs efficiently. Coproduction between governments and local communities in the maintenance of "common-pool" resources (such as communal water points) capitalizes on their distinct comparative advantages. Little evidence exists on effective incentives and schemes for local governments and communities to coordinate effectively for public action.

This study evaluates the impact of "Maji Endelevu" ("Sustainable Water" in Swahili) a governance intervention implemented in 156 villages spread across 40 districts of mainland Tanzania, from 2018 to 2023. The intervention aims to improve coproduction between the two parties which are de jure jointly responsible for maintaining Tanzania's communal water points: district governments and village water community groups. Rather than attempting to write ex-ante contracts between them for all possible contingencies, the intervention helps build their relationships so that when breakdowns arise, they can assign responsibilities among each other ex-post.

Specifically, the intervention consists of repeated "action-learning" consultations between the two parties. The consultations are led by an external facilitator who was trained to identify policy "grey areas", i.e. ones where the two parties show different understandings of their respective responsibilities. The core of the intervention lies in leading the two parties to agree mutual responsibilities, at two levels: to repair the specific breakdown at hand (leading to a registry of action points) and to repair breakdowns more generally. Consultations are repeated quarterly, thereby allowing parties to keep each other accountable on past agreements, and build a relationship over time. Each location receives a total of four consultations, generally one in-person lasting a day, and three over the phone lasting an hour. This approach was inspired by bottom-up interventions which create an interface between the service provider and end-users, but adapted to a setting where both parties jointly act as the service provider.

The consultations between district and community representatives were intended to be "semi-structured discussions" rather than a formal lecture. Facilitators opened the consultation by asking village representatives to describe breakdowns in their areas, and explain why they remain unrepaired. Upon hearing the presentation, district government representatives shared their view of how the breakdowns should be repaired. Facilitators pointed out any differences of opinion between the two parties, and related the disagreements to wider grey area topics. While they recall official policy when relevant, the core of the intervention lies in leading the two parties to agree to mutual responsibilities.

Facilitators were recruited based on two types of skills: hard skills, i.e. their knowledge of Tanzania's water policy and sector, and soft skills, i.e. their ability to drive a semi-structured discussion. 15 facilitators were hired, including two senior ones, and received several forms of dedicated training. This included a two-day kick-off training organized in Dodoma in March 2019 before the at-scale pilot, and another one in Dodoma in February 2020 before roll-out. The training mixed classroom training and learning-by-doing -- this allowed the implementing partner which hired the facilitators to ensure all of them were sufficient skilled to undertake the consultations. During roll-out, facilitators received on-the-job training from the two senior facilitators, who shadowed their consultations and provided feedback. Finally, remedial workshops were organized in October 2021 and April 2022. During these, the project team shared feedback from the implementation data filled out after each consultation. The feedback was shared in aggregate form, as well as individually when specific misunderstandings were found.

Consultations were repeated quarterly, and a total of four rounds were conducted. The first round was held in-person and lasted a day, serving to set objectives. In each treated district, the facilitators gathered relevant district government staff working in the water sector and representatives from treated villages in the district. In each treated district, generally four villages were selected into treatment. For each treated village, the water community organization representing the village was selected using the following protocol: i) If the village has a registered Community-Based Water Supply Organization (CBWSO), the CBWSO is selected for participation. CBWSOs are the most institutionalized form of water community organization, and they normally supervize all water points in the village; ii) If the village does not have a registered CBWSO, and only has one community organization managing all water points, that organization is selected for participation; iii) If the village has several community organizations supervizing different subsets of water points, one of these organizations is randomly drawn to participate. To do so, a water point from the baseline sample frame was drawn from the village at hand, and the community organization managing that water point was selected to participate. The district government had three representatives, namely the District Manager and two lower-ranked team members: a technical staff (e.g. water technician) and a non-technical one (e.g. community development officer). The Regional Manager corresponding with the district also attended. For each of the four treated villages, the community organization had two representatives: the highest-ranking official (generally called "Chairperson"), and one with operational responsibilities (e.g. Secretary General, Treasurer). This mix was selected to ensure decisions made during the consultation could be owned by top and medium hierarchy.

This in-person round was held not in the district capital, but in one of the treated villages, where all participants travelled. Specifically, it was held in a public location of the village which provided privacy and comfort, such as a school, health center or other community location. Participants received compensation for their travel costs through a per diem, and were provided with lunch and refreshments during the day. Aside from the main participants cited above, the in-person round also convened local dignitaries specifically for the opening session of the consultation: Ward Councilor, Village Chairperson and Village Executive Officer. This ensured there was local political support for the exercise.

The second, third and fourth rounds were conducted over the phone, and lasted one hour each. In this remote format, each treated village had their one-hour consultation with the district team separately from the other treated villages. As such, each district government had four consultations in a given round, while each village had one only. The consultation followed the same structure and contents as in Round 1, with the exception that from Round 2 onwards, facilitators also sought a status update on the action points agreed in previous round. To ensure that the phone calls were manageable, the number of participants on a given call was reduced compared with the in-person format: up to two participants from the district government's side (as opposed to three in Round 1, and the corresponding Regional Manager), and each village separately with up to two representatives (as opposed to all treated villages jointly in Round 1). In each remote format consultation, the community organization participating received a small grant of USD47 to cover the cost of time and reimburse costs associated with participation. Due to complications arising from the start of the COVID pandemic, nine of the 40 treated districts also received their first consultation round in the above-described remote format, as opposed to the in-person one.

Experimental design

The impact of "Maji Endelevu" is evaluated through a cluster-randomized controlled trial, following a two-stage randomization process: first drawing districts into treatment, and then drawing villages into treatment. This design helps assess whether treatment generates spillovers between village. Such spillovers could theoretically occur because there is a one-to-many ratio between districts and villages, and therefore treatment could divert the attention of District Managers towards treated villages, away from control ones. It is unclear whether this would bias the results (in this case, overstating them). On the one hand, district governments face resource limitations which could lead to spillovers. On the other hand, "Maji Endelevu" only mobilizes their teams for a few hours per quarter.

The empirical strategy helps assess whether such spillovers occur in practice, and then determine accordingly which model to use to identify causal impact. We start by assessing whether spillovers exist between treated villages and within-district controls (i.e., non-treated villages in treated districts). To do this, we draw on an endline survey of District Managers, which is conducted ahead of the other endline components. We compare within-district control villages and cross-district control villages.

If spillovers are found to be statistically significant, we will rely on district-level randomization to calculate treatment effects, and we will collect the other endline surveys across all 99 districts in the randomization frame (treated and non-treated districts). If spillovers are not found to be statistically significant, we will rely on village-level randomization to calculate treatment effects. To do so, we will collect the remaining three endline components only in the 40 treated districts, not surveying the 59 non-treated districts in the randomization frame.

To build the randomization frame, we started from the baseline sample, for which the sampling frame was an existing survey of communal water points conducted in 2017 as part of the UK's FCDO "Payment-by-Results" project. The survey covered 129 Local Government Authorities (LGAs). For this study's baseline sample, we randomly drew by computer eight villages per LGA. In LGAs which had eight or fewer village observations in them, all of those were sampled. We then conducted adjustments ex-ante and in replacements in the field. This resulted in a baseline sample of 962 villages spread across 129 LGAs, and those LGAs fell under 108 districts.

From the baseline sample, we excluded the following units, to reflect the intervention's nature. First, we excluded all villages located in LGAs classified as urban. The reason is that urban areas often have a different governance structure for water provision, such as that of the "water authority" where coproduction between district governments and village water community groups does not apply. For the same reason, we also excluded villages located in the region of Dar es Salaam, Tanzania's economic capital, even if their LGA was not classified as urban. The reason is that Dar es Salaam region is largely urban de facto. Second, we excluded districts which only had three villages or fewer in the randomization frame, to allow drawing villages in the second randomization stage described below. This yielded a randomization frame of 803 villages spread across 129 LGAs, and those LGAs fell under 99 districts.

To assign treatment, we implemented the following cluster-randomization process, in two stages. The first stage consisted of randomly drawing districts into treatment. We select districts, as opposed to LGAs, because District Managers manage all LGAs in their districts. As such, if in a given district some LGAs had been treated and not the others, the risk of spillovers would be greater. Out of the 99 districts, 40 were randomly drawn into treatment by a lottery operated by government (i.e. "treated districts"). The second randomization stage consisted of drawing villages into treatment (i.e. "treated villages"), and this was operated by computer. In each of the 40 treated districts, we randomly drew villages into treatment in the following way. If the district had over eight villages in the randomization frame, we drew four treated villages. If the district had eight villages or fewer, we drew half of them into treatment, with the caveat that if the district had an odd number of villages, the split village had a 50% chance of being drawn into treatment. This yielded a total of 156 treated villages. In the 40 treated districts, the villages which were not drawn into treatment constitute the subgroup of "within-district controls". There are 183 of them, which is more than 156 because some districts had more than eight villages in the randomization frame. In the 59 non-treated districts, all villages in the randomization frame constitute the "cross-district controls", and there are 464 of them.

Design details

Randomization method
Districts were drawn into treatment through a lottery conducted by officials from Tanzania's Rural Water Supply and Sanitation Agency (RUWASA). Villages were drawn into treatment through a computer-based procedure.
Randomization unit
In the first stage of the randomization process, the unit is the district. In the second phase of the randomization process, the unit is the village.
Clusters
In the first stage of the randomization process, 40 districts are drawn into treatment.
Observations
If spillovers are found and we rely on district-level randomization, we will collect endline data in 803 villages. If spillovers are not found and we rely on village-level randomization, we will collect data in 339 villages.
Arms
Following the clustered randomization design, there are 156 treated villages, 183 within-district control villages and 464 cross-district control villages.
Primary outcomes
Water Point Functionality
Secondary outcomes
Village Maintenance Index; District Maintenance Index

Cited by

No citation data available for this registration yet.

Keywords