Direct Inclusion in Municipal Decision-Making

Evidence from a Field Experiment in Lebanon

Nora Chirikure, Lara Azzam, Macartan Humphreys

Motivation

  • Participation is widely promoted in development practice
  • Yet evidence on its effects is mixed, often inconsistent with claims
  • Goal: generate rigorous, policy-relevant evidence for TDA programming in fragile settings
  • Learning goal: GIZ learning about RIE

Logic

Key idea:

  • Citizens disconnected from local government
  • Disconnect \(\rightarrow\) distrust, inability to benefit from local government

Conversely:

  • Exposure \(\rightarrow\) belief updating, learning, trust

Effects expected to be direct, individual level.

  • Strongest for the most alientated citizens living in effective municipalities
  • Weakest (negative?) for the least marginalized citizens living in ineffective municipalities

Participation Evidence

Reasons for Caution

  • Rerouting of resources to cohesion building activities
  • Often creating parallel institutions
  • Theoretical linkages to government trust unclear

Question

Does simple direct inclusion in municipal decision-making affect trust and behavior towards the municipality?

Expectations

  • Direct participation is expected to increase trust in local government
  • Effects may vary depending on initial trust levels:
    • Positive updating: largest gains expected for low-trust participants in high-capacity municipalities
    • Negative updating: effects may backfire for high-trust participants in weak-capacity municipalities
  • Effects expected to be strongest among women and minority groups

Results (preview)

Observed:

  • Attendance among randomly assigned participants was low; compliance was limited

Estimated effects:

  • Trust index: direct inclusion did not increase trust
  • Behavioral measure: treatment significantly increased the likelihood of allocating funds to the municipality
  • Effects do not vary by gender, minority status, or corruption levels

Design

Main design elements

  • Two-arm trial
  • 24 municipalities, each with three 15-member task forces
  • Recruitment for participation through baseline (conditional on interest)
  • Each task force consists of:
    • 10 appointed members (experts, municipality board members, community leaders)
    • 5 randomly selected residents
  • Control group: residents who said they would be interested in participating but were not randomly assigned to a task force

Study scheme

Treatment

  • Training on participatory approaches
  • Needs assessment and prioritisation
  • Initiatives development and proposal writing
  • Further capacity development

Participatory discussion

Examples of initiatives

  • Equipping town halls, public spaces, sport clubs, social centers, libraries and parks for community engagement
  • Garbage collection and waste management
  • Local markets and olive festivals
  • Youth engagement (music bands and waste management)

Project locations

Context: War in Lebanon (2024–2026)

  • As of April 2026, over 2,500 people killed, including in project areas (Beqaa and Baalbek governorates)
  • Over 1.2 million people displaced, including 134,439 IDPs in 636 collective shelters
  • Activity adaptation to respond to people’s urgent and immediate needs
    • Capacity-building activity for the working group focused on psychosocial support and mental health topics

Religious composition by municipality

Municipal satisfaction

Descriptives

Trust index

Trust index

Interest in taking part

Not so high! This is a treatment that half the population is not interested in.

Interest in taking part

Interest Female Age Corruption perceptions Municipality allocation N
yes, certainly 0.46 44.48 0.28 0.16 1562
yes, likely 0.46 44.32 0.28 0.14 735
maybe 0.43 46.21 0.29 0.08 658
probably not 0.49 46.31 0.26 0.15 738
definitely not 0.56 46.44 0.23 0.12 1462
  • “Willing” set is somewhat more male and somewhat younger (\(p<0.01\)).
  • Those claiming corruption marginally more likely to volunteer (\(p < 0.05\))
  • Those providing a larger allocation more likely to volunteer \(p<0.01\)).
  • But differences not huge.

Attendance

Attendance

  • Attendance among randomly assigned participants was low — compliance was limited
  • People have other things going on:
    • Increased tensions since 7 October 2023
    • War: September 2024 – November 2025
    • Municipal elections in May 2025

Main results

  • Attitudes: Survey items
  • Behavior: Allocation decision

Attitudinal measures

To what extent do you agree with the following statements:

  • I trust that my local government acts in the interests of its citizens.
  • I believe that the resource-distribution by the municipal board is fair.
  • Corruption is widespread in local government.

Attitudinal measures

Behavioral measures

Behavioral measures

Primary results: ITT

Behavioral outcomes
  Amount More Some
treated 67.64* 0.01 0.10**
  (28.27) (0.04) (0.03)
Control mean 549.50 0.441 0.666
R2 0.27 0.24 0.25
Adj. R2 0.24 0.21 0.22
Num. obs. 620 620 620
RMSE 345.53 0.44 0.40
***p < 0.001; **p < 0.01; *p < 0.05. All models include municipal fixed effects.

Primary results: LATE

Behavioral outcomes
  Amount More Some
attended 169.16** 0.03 0.27***
  (62.06) (0.08) (0.07)
Control mean 549.50 0.441 0.666
R2 0.27 0.24 0.25
Adj. R2 0.24 0.21 0.22
Num. obs. 531 531 531
RMSE 345.02 0.44 0.40
***p < 0.001; **p < 0.01; *p < 0.05. IV robust models. All models include municipal fixed effects.

Heterogeneity results: (ITT, behavioral)

Behavioral outcomes
  Amount More Some
Treated 51.61 0.01 0.08*
  (31.45) (0.04) (0.04)
male -1.90 -0.01 -0.01
  (44.96) (0.06) (0.05)
Locally marginalized -148.77** -0.11 -0.19**
  (55.04) (0.07) (0.07)
Treated × male -14.11 0.03 -0.03
  (56.43) (0.07) (0.07)
Treated × Locally marginalized 49.46 0.00 0.08
  (70.74) (0.09) (0.09)
Control mean 549.50 0.441 0.666
R2 0.28 0.24 0.26
Adj. R2 0.25 0.21 0.23
Num. obs. 620 620 620
RMSE 343.61 0.44 0.40
***p < 0.001; **p < 0.01; *p < 0.05. All models include municipal fixed effects.

Heterogeneity results: (ITT, behavioral)

  • Heterogeneity in the right direction for gender and marginalization
  • Substantively large for marginalization
  • But not significant

[Marginalized give less also; gender differences minimal]

Mixed-methods strategy

  • Experimental assignments combined with a qualitative component
  • Helps us understand the mechanisms underlying treatment effects, mapped to the DAG nodes
  • Interviews are being conducted with project participants and stakeholders

A DAG

Sub DAG

Update model given flat priors and (currently) survey data.

Query updated model

Use invitation as an instrument and assess participation effect conditional on specified confounder. Here example of a “causes of effects” estimand: the probability that inclusion is due to participation for someone that participated and reported inclusion.

Note: warnings passed from rstan during updating:

Model 1 warnings:
Bulk Effective Samples Size (ESS) is too low
Tail Effective Samples Size (ESS) is too low

Conclusions

Methodological conclusions

  • Cooperation: High levels of cooperation with German development agencies possible: commitment to a learning agenda and putting ideas to the test
  • Context: Compliance and contextual shocks (war, elections) shape what is feasible; design relatively robust
  • Measurement: Direct inclusion did not move attitudinal trust, but it did shift behavior
  • Approach: RIE and more: Mixed-methods evidence will help unpack mechanisms and inform future TDA programming

Substantive conclusions

Two and a half cheers for participation:

  • Many do not want to participate
  • Many that say they want to participate in fact do not
  • But among those that do views shift towards supporting local government, or at least away from high levels of distrust

This intervention is quite intense at the individual level and not scaleable in current form. However the findings justify a permissive argument for participation: If you do* plan to include this likely brings benefits.*