Impacts of community policing on security: evidence from Mbujimayi in the Democratic Republic of Congo

ABSTRACT

This study adopts a quasi-experimental design to estimate the impact on security of the introduction of community policing in a district of the city of Mbujimayi, Democratic Republic of Congo. The authors suggest that security is best captured by a multidimensional index from which they derive a series of outcome indicators such as the incidence of insecurity, the vulnerability to insecurity, or the severity of insecurity. The data originate from representative surveys conducted at pre- and post-treatment times in the pilot and control districts. The main impact estimates are calculated using the Difference-in-Difference (DiD) approach. The DiD estimates are cross-checked with DiD estimates obtained with propensity score matched data to control for the selection bias inherent to quasi-experimental designs. After a year of treatment, the short-term effect of community policing is found to be efficient based on the security indicators and consistently so across approaches. The study finds that community policing impacted strongly on the two dimensions of security: access to policing services and police legitimacy. The effect of community policing is found however to be significantly higher for men than for women and to have had no or only marginal impact on younger women who remained vulnerable to insecurity even after treatment.

KEYWORDS:

• Impact study
• community policing
• Democratic Republic of Congo
• quantitative study

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. DPKO, Manual on Community-Oriented Policing in United Nations Peace Operations, Ref. 2018.04, New York, UNHQ, 2018.

2. UNDP, Community Security and Social Cohesion. Towards a UNDP Approach. Bureau for Crisis Prevention and Recovery, December 2009.

3. UNODC, Training Manual on Policing Urban Space, Criminal Justice Handbook Series, New York, 2013.

4. See Berk-Seligson, Orcés, Pizzolitto, Seligson, and Wilson (2014a) and Berk-Seligson, Orcés, Pizzolitto, Seligson, and Wilson (2014b).

5. Independent Evaluation of the Security Sector Accountability and Police Reform Programme: Final Evaluation Report (see: https://goo.gl/wnujyq, last seen on 7 December 2017).

6. GMRRR, 2006. Rapport final de présentation des travaux de réflexion du Groupe mixte de réflexion sur la réforme et la réorganisation de la police nationale congolaise, Ministère de l’Intérieur, Kinshasa.

7. The training was conducted by trainers of the PNC itself with civil society members whose skills and competencies were upgraded by officers of the UN mission in DRC, the MONUSCO.

8. Some specialized units remained stationed in the district of Bipemba.

9. Perhaps the tensest moment in the city of Mbujimayi was 14 October 2016 when insurgents attacked the village of Kena Nkuna, about 20 km south-west of Mbujimayi. In response, provincial authorities declared a curfew in the city during nighttime that lasted for 2 months. Authorities appeared to have also feared recruiting activities of young sympathizers within the city itself.

10. Ethnic demographic data is not available for Mbujimayi. Piermay (1986) insists on the Luba ethnic homogeneity of the city, citing only the difference between the Western Luba, occupying the Western part of the city (which also hosted the majority of the Luba refugees from the Katanga) and the Eastern Luba occupying the eastern side of the city (Muya district). Our own data in Mbujimayi collected through representative surveys in Bipemba and Muya in 2016 and 2017 provides information on the language spoken at home by residents that indirectly reinforces the claim that the city is ethnically homogenous. In Bipemba, 94.1% and in Muya 91.7% of the population speak Tshiluba – the Luba language – at home. The second language spoken is Swahili (3.3% in Bipemba and 2.5% in Muya) or French (1.7% in Bipemba and 5% in Muya).

11. Using satellite imagery and Geographic Information System to map out areas and create sample frames is often used in humanitarian contexts (Lin & Kuwayama, 2016; Wampler, Rediske, & Molla, 2013). At the time of the research, the city of Mbujimayi had not been mapped at building level previously unlike some other Congolese cities like Lubumbashi for instance. The map of the city available at that time can be viewed under the following URL: http://fosm.org/#map/14/-6.1515/23.5985. A team member undertook to map streets and residential areas of the pilot and control areas using aerial imagery (see results under https://www.openstreetmap.org/#map=13/-6.1092/23.5846). As mapping all buildings or man-made structures was not feasible for reason of lack of time, samples of lots with building areas per neighborhood were mapped and then estimations about the number of lots for the neighborhood were made. A total of 18,477 lots’ nodes were thus mapped using this procedure. We estimated that Bipemba has 43,394 lots while Muya has 14,453. As the lots towards the center of the city were bigger with larger buildings than the lots towards the outskirts, we combined the size of residential areas and the number of lots to build our sample. We applied a 10% weight to the size of the residential areas and a 90% weight to the number of lots. This combining technique differs from other studies which usually count buildings or roof tops only as basis to create samples. There are caveats to counting roof tops in the case of Mbujimayi as some homes were found out to be empty during data collection as people had moved out. This happened in zones where underground mining activity by artisanal miners had endangered the stability of the soil.

12. To map the administrative borders of the districts and, within the districts, the neighborhoods, city officials and neighborhood chiefs were consulted by the Geospatial Information System (GIS) survey team. When conflicts over territories arose between the chefs de quartier, we relied on the bourgmestre for arbitration. Traces were collected with the application Osmtracker previously installed in smartphones and then uploaded in OpenStreetMap.

13. The use of t-test is derived from the central limit theorem that assumes a normal distribution of the outcome when the sample size is over 30 observations. Therefore, the mean comparison test, respectively the percent comparison test, follows a student distribution (as mean is normal distributed and variance χ² distributed (test = ($\frac{Normal}{{\chi }^2}$), where χ² is Normal²)).

14. We run the analyses with matched data using the second method; impact estimates were found to follow the same trend as with non-matched data.

Additional information

Notes on contributors

Dominique Wisler

Dominique Wisler (1960), PhD, has been trained at Universities in Switzerland, Germany, and the United States. A full-time consultant, founder and chair of Coginta, a Swiss-based NGO dedicated to police reform (www.coginta.org), he has managed, evaluated, or designed police and Ministry of Interior restructuring programs in many countries, including Afghanistan, Chad, Sudan, Iraq, Turkey, Mozambique, Haiti, Bosnia and Herzegovina, the Occupied Territories of Palestine, the Republic of Guinea, and the Democratic Republic of Congo. He is a standing member of security sector reform rosters of experts of the United Nations Department of Peacekeeping Operations, the United Nations Development Program, and the Swiss Foreign Office. A former professor in political sociology at the University of Geneva, he has published widely in scientific reviews. His latest books are a co-edited volume on comparative community policing with case studies in Africa, Asia, Europe and the Americas (CRC Press, London, 2009) and a book on the early democracy of Switzerland (Georg, 2008).

Silva Monti-Ohannessian

Silva Monti-Ohannessian (1974), PhD, has studied econometrics and statistics at the Universities of Geneva and Paris. As a freelance consultant with an expertise in qualitative data modeling (conjoint analysis, logit, probit) and in programming, she worked in various areas like marketing, clinical research, crime, school success, decision-making process analysis, policy impact study etc. She developed a crime index highlighting areas of high activity, innovated trade-off models used in the decision-making process (PhD) and led statistical projects in clinical research. She published her PhD work (Edilivre, Paris, 2008) and actively animated various conferences in statistics.

Rafael Avila Coya

Rafael Avila Coya (1967) is a GIS consultant who has worked for different organizations, mostly in Africa, like eHealth Africa for eradication of Polio in North Nigeria, Crowd Cover for awareness of deforestation in the Congo Basin, imports of data (schools, health facilities, WASH, road network, etc.) into OpenStreetMap for HOT US, and above all Coginta, for different security related projects in the Democratic Republic of Congo, Guinea, Chad, Burkina Faso, Ivory Coast and Romania. His skills are in base cartography, populations’ estimates, sampling, among others. He is member of the OpenStreetMap Foundation since 2012 and the Humanitarian OpenStreetMap Team (HOT) since 2014 and has volunteered in mapping for humanitarian crisis like the Ebola outbreak in Guinea, typhoon Hayjan in the Philippines, and Nepal Earthquake in Nepal among many others.