Community Policing and Public Trust: A Field Experiment in Uganda
Last registered on November 10, 2017


Trial Information
General Information
Community Policing and Public Trust: A Field Experiment in Uganda
Initial registration date
November 10, 2017
Last updated
November 10, 2017 4:41 PM EST

This section is unavailable to the public. Use the button below to request access to this information.

Request Information
Primary Investigator
Brown University
Other Primary Investigator(s)
PI Affiliation
Office of the Prime Minister of Uganda
PI Affiliation
University of Pennsylvania
Additional Trial Information
In development
Start date
End date
Secondary IDs
The ability of the police and other state security institutions to enforce the law depends on the trust and cooperation of the policed. Our study is designed to address the challenge of building trust and cooperation between police and citizens in Uganda, using the “Muyenga model” of community policing. The Muyenga model is explicitly designed to create opportunities for more positive, mutually respectful interactions between civilians and police officers by allowing UPF officers to respond more proactively to the needs of citizens and communities; by providing mechanisms to report acts of corruption and abuse; and by encouraging citizens to rely on state security and justice sector institutions when crimes are committed or violence occurs. The model has been piloted successfully in a small number of communities, and is now scaled up sufficiently to allow a rigorous study of its effectiveness. Our study is one of six field experiments in the Evidence in Governance and Politics (EGAP) network’s Metaketa IV initiative.
External Link(s)
Registration Citation
Blair, Robert, Guy Grossman and Benjamin Kachero. 2017. "Community Policing and Public Trust: A Field Experiment in Uganda." AEA RCT Registry. November 10.
Experimental Details
Relations between civilians and the Ugandan Police Force (UPF) remain strained, with correspondingly relatively low levels of citizen trust, cooperation and crime reporting. This study aims to test whether the above problems could be addressed via a homegrown model of community policing, which has already been successfully piloted (on a small scale) in Uganda. Known as the “Muyenga model”, this approach to community policing is designed to create opportunities for more positive, mutually respectful interactions between civilians and police officers by allowing officers to respond more proactively to the needs of citizens and communities; by providing mechanisms to report acts of corruption and abuse; and by encouraging citizens to rely on state security and justice sector institutions when crimes are committed or violence occurs. The Muyenga model comprises three components:

1. Frequent motor cycle and especially foot patrols by police officers. The goal of the “meet and greet” patrols is to increase “police presence” but also to form ties and increase trust between the local officers and residents.
2. Frequent “town hall” meetings with citizens and local leaders. The primary purpose of the meetings is to allow residents, local leaders, other community stakeholders, and police officers to discuss chronic crime and disorder problems, and to formulate local solutions. Town hall meetings are also designed to provide information to citizens about police procedures and rules of conduct (for example, whom to contact in case of officers’ misconduct).
3. Training and support to community watch teams (CWTs). Community watch teams allow greater security presence (in the face of shortfalls in police manpower), but also create a cadre of residents that better understand police procedures and resources. Specifically, CWTs act as liaisons between the community and UPF officers at the beat and station levels. The idea behind NWTs is to create a group of civilians with a better understanding of the law and of UPF’s roles and responsibilities, who can then serve as points of contact between community members and the police. Among other responsibilities, watch teams are specifically trained to monitor the incidence of gender-based violence (GBV) in their communities.

Through these three mechanisms, police officers learn about local security needs and are then expected to address them with resources and decision-making authority decentralized to the local level.

In this study, we also test whether community policing could be strengthened using a second intervention, which offers training in prevention of GBV for community leaders and volunteers from villages in the study area. These community volunteers will then serve as focal points within their villages, disseminating knowledge of GBV and delivering strategies for preventing and mitigating it. Training will also be delivered to police officers to raise awareness on the seriousness of GBV and to teach them how to handle cases reported to them. The training will be organized at the station level. This intervention was selected since focus groups discussions the research team conducted at the “seed” stage, suggested GBV as both pervasive and major unreported crime.
Intervention Start Date
Intervention End Date
Primary Outcomes
Primary Outcomes (end points)
We will measure three core families of outcomes: (1) trust in the police and the state; (2) rates of crime and violence; and (3) citizen cooperation with the police. To measure trust in the police and the state, we will conduct a household survey of all 288 villages in the sample. (As described above, we will randomly select two villages per parish to participate in the evaluation. We anticipate that villages will have an adult population of approximately 300-350 individuals on average.) Twelve residents will be randomly sampled from the consenting adult population of each village, for a sample size of approximately 3450 respondents. Endline data collection will begin 8 months after implementation begins. We anticipate the survey, which will be conducted in local languages using IPA Uganda, will last approximately 1.5 hours. If we secure sufficient funding, we would also implement a series of behavioral games to measure trust and cooperation.

We will measure rates of crime and violence using two sources: (1) the household survey described above, and (2) administrative data from the UPF. For the household survey, we will implement a version of the National Crime Victimization Survey (NCVS), modified for the Ugandan context. For respondents who report being a victim of crime, we will also ask if and with whom they filed a complaint. Blair and colleagues successfully adapted the NCVS to measure crime and violence in rural Liberia, and we expect to be able to do the same in Uganda. Here we are especially interested in testing whether women increase reporting of domestic violence and other forms of gender based violence. For the administrative data, we plan on digitizing records of all criminal complaints lodged by residents of any of the villages in our sample between January 1, 2017 and 6 months after the end of implementation.

Finally, to measure citizen cooperation with the police, we will request that UPF officers at each post record the date and location of three categories of citizen cooperation: (1) tips about specific crimes and about more general crime “hot spots;” (2) denunciations; and (3) requests for assistance. A short survey will also be conducted with police officers working in stations and posts in study areas. The survey will last about 45 minutes and will focus on officers' perception of local problems and dynamics, their relationships with citizens, and police behavior.
Primary Outcomes (explanation)
Secondary Outcomes
Secondary Outcomes (end points)
Secondary Outcomes (explanation)
Experimental Design
Experimental Design
We will evaluate these two interventions using a randomized controlled trial. The unit of randomization is the police station. We will implement a 2x2 factorial design in which stations are randomly assigned to one of four groups:

1. Community policing
2. GBV training
3. Community policing and GBV training
4. Neither (pure control)

Randomization will proceed in three stages. First, we will identify 72 police stations serving rural communities from among the stations in the study districts. We began collecting information on these stations (for example, number of officers, distance to district capital, number of villages the station covers) to allow us to place stations with similar characteristics in blocks (using the blockTools R package). Second, we will randomly assign each of these stations to one of the four treatment groups, within blocks. For purposes of data collection, we will randomly select 1 police unit within the jurisdiction of each police station, if a station doesn't have any post we pick the station itself, and chose the parish where the police unit is located for our sample, within each parish, four villages. Third, within each village we will select (using a random walk) 12 survey respondents, 7 women and 5 men.
Experimental Design Details
Not available
Randomization Method
Randomization done in office using the blockTools package in R.
Randomization Unit
Police stations.
Was the treatment clustered?
Experiment Characteristics
Sample size: planned number of clusters
72 police stations
Sample size: planned number of observations
3,456 survey respondents
Sample size (or number of clusters) by treatment arms
18 police stations community policing, 18 police stations GBV training, 18 police stations community policing and GBV training, 18 police stations control
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
We will estimate treatment effects using standard OLS estimators, with weighted indices of each family of outcomes as our dependent variables. For individual-level outcomes (e.g. trust in the police, cooperation with the police and crime reporting), we have power of 0.8 to identify an effect size of 0.22 standard deviations with 72 police stations and 48 villagers per station (sampled from four villages located in a single parish serviced by a station), assuming a conservative interclass correlation of 0.1. On the other hand, this sample size will only allow us to identify large effects for station level outcomes (around 0.4 SD). Because our core outcomes are operationalized at the individual level, our study is sufficiently powered for nearly all of the analyses we plan to run.
IRB Name
University of Pennsylvania Institutional Review Board
IRB Approval Date
IRB Approval Number