Informal Accountability and Cooperative Behavior: Evidence from the D.R. Congo

This study evaluates whether trusted local leaders (religious or traditional) can mobilize collective action in fragile settings by activating norms of reciprocity. In partnership with the city administration of Kananga, DRC, we embed a field experiment within a new initiative to revitalize salongo. As part of this program, residents are placed into small accountability groups whose contributions to public works—labor, materials, or money—are tracked through a new digital system. These groups meet regularly to review participation and are informed that the most active team will receive a symbolic recognition award at the end of the month. The experiment leverages this policy structure—advanced by the city administration, our implementing partner—to randomly vary two dimensions: (i) whether a religious leader, a traditional leader, or no leader attends the meetings; and (ii) whether groups are composed of participants from the same ethnic background or different ethnic groups (see PAP 1).

On the top of the field experiment, we introduce a survey experiment where we investigate how stochastic shocks and narratives shape beliefs about moral supernatural forces. Depending on a contingent state of nature, outside the control of researchers, survey payments on recurring meetings may be attached to the free-riding behavior of participants. In one state of nature, the 'Justice' state, outcomes simulate the actions of a moral god, where cooperators (those who contributed a higher share of labor and resources to community projects) receive a higher payment than free-riders. In the alternative state, the 'Neutrality' state, survey payments are the same. This cross‑randomization forms a distinct 2 × 3 factorial survey experiment—crossing the shock condition (Justice vs Neutrality) with the three leadership treatments—and allows us to identify whether trusted authorities amplify, dampen, or leave unchanged the behavioural and belief responses to moralized randomness (see PAP 2).

2025-08-11

2026-07-06

Voluntary contributions to community projects:
- Labor contributions, extensive margin per day.
- Labor contributions, intensive margin per day: hours and minutes worked.
- Cash contributions, extensive margin per day.
- Cash contributions, intensive margin per day: Congolese Francs contributed.
- Material contributions, extensive margin per day.
- Material contributions, intensive margin per day: estimated value of goods contributed.

Primary outcomes will be provided by the administrative digital tracking of contributions, which is part of a new contribution registration system advanced by our implementing partner (city administration).

Survey measures on:
- Expectations of social reciprocity.
- Beliefs in divine reciprocity.
- Self-reported community enforcement behavior.
- Dyadic measures of interactions and social expectations of with accountability group peers.

Transcription of accountability meetings recordings:
- Minutes spoken during accountability meetings.
- Frequency of key words pronounced in meetings.
- Human-based and AI based analysis of meeting content.

Endpoint surveys will be collected at 3 key moments. During accountability meeting days (2) and an endline by the end of the intervention in the neighborhood.

Accountability meetings will be recorded, transcribed, and meeting behavior will be coded into X broad categories:
- Appeals to social reciprocity/community cohesion as discourses to stimulate participation.
- Appeals to divine reciprocity as discourse to stimulate peers to contribute to neighborhood.
- Other appeals.
- Other content.

Within appeals to divine reciprocity, we will further code:
- Confirmation statements of divine reciprocity.
- Theodicies and religious coping.

We conduct a series of randomized controlled trials (RCTs) to study how informal accountability mechanisms sustain cooperation in contexts of weak state presence. Our experiments take place in Kananga, Democratic Republic of Congo, and build on partnerships with local authorities and civil society.

The umbrella project covers two complementary studies:

- Study 1: We randomize the presence of local leaders (religious and traditional leaders), the social composition of accountability groups (ethnically homogeneous or mixed), and the availability of granular information on participation to test how informal accountability channels (expectations of social reciprocity and beliefs in divine reciprocity) affect collective action. Outcomes include cooperation in community labor activities and beliefs about social and supernatural reciprocity.

- Study 2: We introduce randomized shocks to group earnings, designed to mimic real-world stochastic events (e.g., rainfall), and observe how participants and leaders interpret these shocks. By exploring variations on lottery framing, timing, and group composition, we test whether narratives of divine justice around stochastic payoffs shape cooperation in public goods provision.

The experimental design in both studies follows a factorial structure. The primary purpose is to test how informal institutions mobilize cooperative behavior and how narratives about fairness or divine justice sustain that behavior in the face of uncertainty. See Pre-Analysis Plans for more information.

Not available

Computer Randomization in R; Rain occurrence.

Accountability groups

Yes

480 accountability groups

The intervention counts with 4 participants per group, totaling 1,920 meeting slots in the main specification. -Out of the 1,920 participation slots. 320 are taken by 80 leaders that participate in 4 slots each. -1,600 citizens take part in the remaining slots. -The main analysis will be done with those 1,600 individuals 560 other citizens in control conditions (do not participate in accountability meetings). - 240 controls in pure control neighborhoods - 320 controls in treated neighborhoods Observations for individuals in treated neighborhoods will follow a panel structure. Administrative records from the community contributions tracking system will allow the analysis of 29 time periods per individual in treated neighborhoods (4 weeks with tracking from monday-saturday, 1 week with tracking from monday-friday). Thus we expect to have a panel of 58,000 individual-time observations (1,600 citizens + 80 leaders + 320 citizens tracked in control conditions = 2,000 tracked individuals in the panel of 29 time periods). Baseline/Endline surveys will count with 2,240 observations (2,000 treated neighborhood individuals + 240 pure control neighborhood individuals).

Design 1 (see PAP for details):

- Religious leader, homogeneous group, visible individual contributions: 40 groups
- Religious leader, heterogeneous group, visible individual contributions: 40 groups

- Traditional leader, homogeneous group, visible individual contributions: 40 groups
- Traditional leader, heterogeneous group, visible individual contributions: 40 groups

- No leader, homogeneous group, visible individual contributions:: 40 groups
- No leader, heterogeneous group, visible individual contributions:: 40 groups

- Religious leader, homogeneous group, individual contributions not visible: 40 groups
- Religious leader, heterogeneous group, individual contributions not visible: 40 groups

- Traditional leader, homogeneous group, individual contributions not visible: 40 groups
- Traditional leader, heterogeneous group, individual contributions not visible: 40 groups

- No leader, homogeneous group, individual contributions not visible: 40 groups
- No leader, heterogeneous group, individual contributions not visible: 40 groups


Design 2 (see PAP for details):

- Main arm -- Rain framing for shock (28 neighborhoods, 12 groups per neighborhood, 336 groups):

- Religious leader, 'divine' shock, payments before meeting: 28 groups
- Religious leader, 'divine' shock, payments after meeting: 28 groups

- Religious leader, no shock, payments before meeting: 28 groups
- Religious leader, no shock, payments after meeting: 28 groups

- Traditional leader, 'divine' shock, payments before meeting: 28 groups
- Traditional leader, 'divine' shock, payments after meeting: 28 groups

- Traditional leader, no shock, payments before meeting: 28 groups
- Traditional leader, no shock, payments after meeting: 28 groups

- No leader, 'divine' shock, payments before meeting: 28 groups
- No leader, 'divine' shock, payments after meeting: 28 groups

- No leader, no shock, payments before meeting: 28 groups
- No leader, no shock, payments after meeting: 28 groups


- Placebo arm -- Tablet framing for shock (12 neighborhoods, 12 groups per neighborhood, 144 groups):

- Religious leader, 'divine' shock, payments before meeting: 24 groups
- Religious leader, no shock, payments before meeting: 24 groups

- Traditional leader, 'divine' shock, payments before meeting: 24 groups
- Traditional leader, no shock, payments before meeting: 24 groups

- No leader, 'divine' shock, payments before meeting: 24 groups
- No leader, no shock, payments before meeting: 24 groups

Our main specification will leverage the panel structure of voluntary contributions to community projects. Design 1: Because there are few analytical frameworks accounting for serial correlation in a factorial clustered design, we follow Burlig et. al (2020) and simulate our factorial design using de-identified data of an initial wave of participation tracking. Conditional on take-up rates similar to those we observed in the pilot, simulations suggests that a sample of 40 neighborhoods, with 12 clusters of four participants each (totaling 480 clusters and 1,920 participants), will be able to detect at 85.2%, 83.6%, and 66.6% power an MDE of 20% of outcome variable’s S.D. for coefficients β_1 (Homogeneous group effect), β_4 (Religious leader effect for mixed and homogeneous groups) and β_3 (Interaction of traditional leader and homogeneous groups), respectively. This amounts to a change of approximately 15 minutes of additional labor hours per day. Design 2: To estimate statistical power for our primary hypotheses, we simulate our experimental design using real administrative data on salongo participation, following the empirical simulation methodology proposed by Burlig et al. (2020). Although the full design includes three leadership conditions (Religious Leader, Traditional Leader, and No Leader), power calculations are based on two analogous 2×2 factorial designs, each isolating a specific contrast between religious leadership and a distinct placebo condition: (i) Religious Leader vs. No Leader, and (ii) Religious Leader vs. Traditional Leader. This structure maps directly onto our theoretical framework, which seeks to identify whether religious authority uniquely shapes cooperation through supernatural narratives, relative to either the absence of leadership or non-religious authority. In practice, this design choice implies simulating 8 clusters per neighborhood—rather than 12 as required for a full 3×2 factorial—thus preserving the 2×2 symmetry and allowing for cleaner identification of coefficients in each model. Each 2×2 design maintains four clusters per cell. Simulations indicate that, with 40 treated neighborhoods, each hosting 8 clusters of 4 participants (yielding 1,280 participants), the design achieves 84% power to detect an MDE of 0.2 standard deviations for the stochastic shock effect, 86.2% for the religious leadership effect, and 59.2% for the interaction term, which captures whether religious leaders moderate belief responses to stochastic shocks. In both designs, power calculations indicated that an MDE of 0.3 of the outcome's variable S.D. is detected with over 80% power for all coefficients of interest.