Does Decentralized Provision Crowd In or Crowd Out Support for Collective Water Infrastructure? Experimental Evidence from a Clean Water RCT

Last registered on August 20, 2026

Pre-Trial

Trial Information

General Information

Title
Does Decentralized Provision Crowd In or Crowd Out Support for Collective Water Infrastructure? Experimental Evidence from a Clean Water RCT
RCT ID
AEARCTR-0019305
Initial registration date
August 17, 2026

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
August 20, 2026, 9:28 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

Affiliation
Norwegian University of Science and Technology

Other Primary Investigator(s)

PI Affiliation
Norwegian University of Science and Technology
PI Affiliation
University of Copenhagen
PI Affiliation
Norwegian University of Science and Technology
PI Affiliation
Ghent University

Additional Trial Information

Status
In development
Start date
2026-08-18
End date
2026-12-10
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
In many low-income settings, non-state organizations compensate for limited state capacity by providing services that would otherwise be expected from the government. Such interventions can generate important welfare gains, but their consequences for the future provision of public services are theoretically ambiguous. When households gain access to a reliable decentralized technology, the marginal value of centralized provision may fall, reducing their willingness to support and finance collective infrastructure. Conversely, experiencing the benefits of reliable service delivery may raise households’ expectations, reveal the returns to improved provision, and increase demand for a collective solution.

The present study investigates the effect of randomized household access to a decentralized water-cleaning technology on willingness-to-pay for collective water provision as well as an incentivized contribution to a concrete communal water project. It thereby identifies whether personally experiencing a decentralized alternative changes revealed demand for collective provision. We investigate this question through a household-level randomized controlled trial in six villages in Uyui District, Tabora Region, Tanzania, where access to safe drinking water remains severely constrained. A total of 900 households are randomly assigned to one of three conditions: 360 households receive a household water-treatment device (Solvatten), provided free of charge by World Vision Tanzania, a humanitarian organization; 180 households receive an unconditional cash transfer designed to distinguish the effect of the water technology from the income or resource-transfer effect of receiving a valuable product; and 360 households form the control group. Two adult members of each household are interviewed separately at baseline and endline, yielding approximately 1,800 respondents and 3,600 person-wave observations.

The primary outcomes capture demand for a collective, village-level solution to unsafe drinking water. At baseline and endline, respondents complete a survey-based elicitation measuring their willingness to support an upgrade of a communal dug well at different household costs. At endline, respondents also make a real-stakes allocation from a monetary endowment, choosing how much to retain for themselves and how much to contribute toward upgrading a communal well with a hand pump and chlorination system. Contributions are matched by the project, providing an incentivized measure of households’ willingness to sacrifice private resources for collective water provision. Secondary outcomes capture the priority assigned to collective water provision and intended engagement in efforts to improve communal water infrastructure.

The household-level design identifies whether receiving a decentralized water technology changes the recipient’s own demand for collective provision. Interference may nevertheless arise if untreated households share the device or treated water or learn from the experiences of treated neighbors. We account for this possibility in four ways. First, the technology has limited capacity and is designed to serve a single household, restricting the scope for sharing. Second, households residing in a compound containing at least one treated household will be classified as compound-exposed. Third, spillovers beyond the compound will be estimated using spatial exposure mapping that captures each household’s distance-decay-weighted exposure to nearby treatment assignments, with own assignment entering separately. Conley spatial-HAC standard errors will additionally account for residual spatial correlation. Fourth, one village in which no treatment is distributed provides a treatment-free benchmark, useful for the models with repeated baseline and endline measures.

We relate primarily to two broad bodies of literature. First, our empirical approach builds on a large experimental literature studying the adoption, valuation, and welfare effects of water technologies (Ashraf et al. 2010; Kremer et al. 2011; Burlig et al. 2026). This literature provides the closest methodological and sectoral precedent for the present study but stops short of examining the downstream political consequences of decentralized water provision. Second, and more closely related conceptually, we add crucial experimental evidence to the literature on how non-state service provision affects the role of the state. One strand of this literature examines the supply side, demonstrating that NGO provision can alter government capacity and service production (Deserranno et al. 2025). Another strand examines the demand side, asking whether NGO and donor provision changes citizens’ preferences over service providers, perceptions of government legitimacy, and political engagement (Brass 2012; Dietrich and Winters 2015; Springman 2022). Existing approaches, however, have important limitations in identifying how experienced non-state provision affects demand for a public alternative. Brass (2012) relies on observational evidence, while Dietrich and Winters (2015) experimentally vary information about foreign funding. Springman (2022) examines the long-term effects of NGO health provision on preferences for NGO versus government service delivery through an ancillary evaluation of an earlier village-level RCT in Uganda.

In the present study, we combine the methodological rigor of the experimental water literature with the political-economy questions raised by research on non-state service provision. Rather than examine preferences over service providers and political engagement as the key outcomes, we treat them as potential mechanisms through which non-state provision may alter demand for a public alternative. We measure that demand directly through repeated willingness-to-pay elicitations and an incentivized, financially consequential contribution to a concrete communal water investment. The household-level design allows us to test with precision whether decentralized provision leads to private exit or strengthens collective voice.
External Link(s)

Registration Citation

Citation
Agneman, Gustav et al. 2026. "Does Decentralized Provision Crowd In or Crowd Out Support for Collective Water Infrastructure? Experimental Evidence from a Clean Water RCT." AEA RCT Registry. August 20. https://doi.org/10.1257/rct.19305-1.0
Sponsors & Partners

Sponsors

Partner

Type
private_company
Experimental Details

Interventions

Intervention(s)
The intervention includes three experimental conditions. In the water-technology arm, households receive a Solvatten water-treatment device free of charge from World Vision Tanzania. The device is designed to provide safe water for one household because of limited treatment capacity. In the monetary-comparison arm, households receive an unconditional cash transfer intended to compensate for the income and resource transfer effect of receiving a valuable product. Households assigned to the control arm receive neither the water-treatment technology nor the cash transfer during the study period. Interventions are implemented after the baseline survey and before the endline survey.
Intervention Start Date
2026-09-05
Intervention End Date
2026-12-10

Primary Outcomes

Primary Outcomes (end points)
Incentivized contribution to collective water provision.
At endline, the respondent allocates a TZS 5,000 endowment between private consumption and a contribution to a concrete communal dug-well upgrade. The outcome is the amount contributed, measured from TZS 0 to TZS 5,000 in TZS 1,000 increments.

Willingness to pay for collective water provision.
At baseline and endline, the respondent completes a WTP elicitation concerning a one-time household payment for upgrading a communal dug well with a hand pump and chlorination system. The primary repeated outcome is the respondent's maximum accepted payment on the pre-specified ladder.
Primary Outcomes (explanation)
For the incentivized contribution, each respondent receives TZS 5,000 and privately chooses how much to keep and how much to place in an envelope for the communal well upgrade. Permitted contributions are TZS 0, 1,000, 2,000, 3,000, 4,000, and 5,000. For every TZS 1 contributed by the respondent, the project contributes an additional TZS 2, so the amount credited toward collective water provision equals three times the respondent's contribution. Positive contribution will be reported as a secondary transformation.

For willingness to pay, all respondents begin with a one-time household price of TZS 2,000. A yes response leads to successively higher prices according to the pre-specified upward ladder; a no response leads to successively lower prices according to the pre-specified downward ladder. The main constructed measure is the highest accepted price. The lowest rejected price will also be retained so that willingness to pay can be treated as interval-censored. The same project description and price ladder will be used at baseline and endline. The main repeated-outcome specification will estimate endline willingness to pay conditional on baseline willingness to pay; a within-respondent change specification will be reported as a robustness analysis.

Secondary Outcomes

Secondary Outcomes (end points)
Five mechanism-based secondary outcome families will be examined:

Social connectedness and collective orientation toward other village residents.
Expected marginal personal benefits from collective water provision.
Expected village-level benefits from collective water provision.
Beliefs about the respective responsibilities of households, NGOs, and government for providing safe water.
Perceived feasibility of collective provision, encompassing confidence in implementation and maintenance, as well as expectations about other households’ financial contributions.

We will also measure three alternative outcomes of demand for collective provision:
Public-finance allocation: the share of a fixed village budget that the respondent allocates to communal water provision rather than other public services.
Intended costly participation: the participant’s stated likelihood of attending a one-hour village meeting about improving the shared well within the next two weeks, measured on a seven-point scale from 1 (“extremely unlikely”) to 7 (“extremely likely”).
Willingness to contact the responsible authority: the participant’s stated willingness to spend time contacting a village or water-authority representative to request improvements to the shared well, measured on a seven-point scale from 1 (“extremely unwilling”) to 7 (“extremely willing”).

Other intermediate outcomes will capture treatment compliance and effects: e.g., adoption and use of household water treatment, reliance on and frequency of visits to communal water sources, and time, health, and monetary measures.
Secondary Outcomes (explanation)
Mechanism outcomes are organized into five pre-specified families and measured using identical questions at baseline and endline. Four families are covered by three survey questions (the responsibilities family is instead one question with three outcomes) designed to capture a mechanism that may explain crowding in or crowding out.

For the four mechanism families with three questions, component questions will be standardized and combined into an equally weighted index. An index will be constructed when at least two of its three component questions are observed. As a robustness exercise, we will construct alternative indices using principal-component measures.

Experimental Design

Experimental Design
The study uses a baseline–endline randomized controlled design in six villages. A total of 900 geocoded residential building footprints are assigned through R-code to one of three conditions: a household water-treatment technology, a monetary-comparison treatment (50/50 into 3 or 12 USD), or a control condition. Randomization is conducted at the building-footprint level and stratified by village. Two adult household members are surveyed separately at baseline and endline, yielding up to 1,800 respondents per wave.

The main analysis estimates intention-to-treat effects. The primary comparisons evaluate the effect of receiving the household water-treatment technology relative to receiving no intervention and a pooled no-device counterfactual. The latter comparison hinges on no statistically detectable income or resource transfer effects, which will be estimated separately.
Experimental Design Details
Not available
Randomization Method
We conducted the randomization in R using geocoded building footprints as the units of assignment while stratifying by village. Within each of the five mixed-assignment villages, complete randomization assigned exactly 72 footprints to Solvatten, 52 to control, and 26 to the monetary comparison. Within the pre-designated device-free village, complete randomization assigned 100 footprints to control and 50 to the monetary comparison. Both adult respondents associated with a building footprint get the building's assignment. If the building does not contain two eligible adults, the assignment is passed on to the next residential building following a pre-specified protocol.
Randomization Unit
The unit of randomization is the geocoded building footprint corresponding to a sampled household dwelling. Outcomes are measured for two adult respondents within each randomized household. Randomization is stratified by village.
Was the treatment clustered?
Yes

Experiment Characteristics

Sample size: planned number of clusters
900 randomized household clusters: 150 in each of six villages.
Sample size: planned number of observations
1,800 adult respondents per survey wave, nested within 900 randomized building-footprint/household clusters. With baseline and endline surveys, the design yields up to 3,600 respondent-wave observations. The endline contribution outcome has a planned maximum of 1,800 respondent observations before attrition.
Sample size (or number of clusters) by treatment arms
Solvatten household water-treatment device: 360 building-footprint/household clusters, 720 adult respondents per wave.

Control: 360 building-footprint/household clusters, 720 adult respondents per wave.

Monetary comparison: 180 building-footprint/household clusters, 360 adult respondents per wave, 50/50 divided into income or resource transfer compensation.

In each of the five mixed-assignment villages, the allocation is 72 Solvatten, 52 control, and 26 monetary-comparison households. In the device-free village, the allocation is 100 control and 50 monetary-comparison households.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
For brevity, we here present the most conservative main-model power calculation, the incentivized contribution outcome, measured from TZS 0 to TZS 5,000 in increments of TZS 1,000. Assuming a discrete uniform distribution, the outcome has a mean of TZS 2,500 and a standard deviation of TZS 1,708. Requiring 80 percent power and a two-sided 5 percent significance level, and assumin two respondents per household and a within-household intracluster correlation of 0.50, no attrition, and no precision gain from baseline covariates, the minimum detectable effects are: TZS 309 (0.181 standard deviations; 12.4 percent of the mean) for Solvatten versus control, TZS 378 (0.221 standard deviations; 15.1 percent) for Solvatten versus the monetary comparison, and TZS 282 (0.165 standard deviations; 11.3 percent) for Solvatten versus the pooled control and monetary-comparison arms. The attached R script reproduces these calculations and provides sensitivity analyses for alternative within-household correlations.
Supporting Documents and Materials

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information
IRB

Institutional Review Boards (IRBs)

IRB Name
Research Ethics Committee at the Department of Economics at University of Copenhagen
IRB Approval Date
2026-03-09
IRB Approval Number
N/A