Methodology

Research Design and Sample

This report is based on a cross-sectional survey of adult residents in five administrative regions of the Central African Republic: the capital city Bangui, and the prefectures of Lobaye, Ombella M’Poko, Ouham, and Ouham Pende. The four prefectures and Bangui were selected to reflect the range of experience of conflicts in the CAR and to provide geographic representation from North to South. Together, they account for over half the population of CAR (52%).

Within each region, we used a systematic random selection method to sample villages proportionately to the population size, then households, and finally individuals, for interview. Villages were randomly selected from a comprehensive list of villages for each selected prefectures using the 2003 general population census data.[1] A total of 117 villages were selected, and 6 villages had to be replaced by the nearest accessible village because of local instability and/or inaccessibility.

Within the villages, interviewers were assigned to zones where they selected every other household in a randomly chosen direction, starting from the center of the zone. A household was defined as a group of people normally sleeping under the same roof and eating together. In each household, interviewers randomly selected one adult to be interviewed from a list of all eligible adults. Three attempts were made to contact a household or individual before replacing them with another.

The minimum sample size for each of the prefectures and Bangui was 352 individuals, for a total sample size of 1,760. Out of 2,192 households approached for interviews, 1,879 agreed to participate (86% participation rate). Within these 1,879 households, a total of 1,969 individuals were approached, and 1,879 participated in the interviews (95% participation rate, one individual selected per participating household).

The Committee for the Protection of Human Subjects at the University of California, Berkeley, reviewed and approved the study protocol in October 2009. In CAR, where no similar review board exists, the protocol was reviewed and approved by the Ministry of Economic, Planning, and International Cooperation. Approval to conduct interviews was also obtained from the local authorities at each survey site. Informed consent was sought for each selected participant; neither monetary nor material incentives were offered for participation.

Research Instruments

Interviews were conducted using a standardized semi-structured questionnaire covering (1) respondents’ demographics, (2) priorities and services, (3) health, (4) social cohesion and community relations, (5) security and conflict resolution, (6) domestic violence, (7) peace, (8) justice and accountability, (9) the International Criminal Court, (10) non-judicial measures for victims and reparations, (11) exposure to violence, and (12)  psychological impact of the conflicts. The questionnaire was developed by experts in the field and after consultation with local experts. Response options based on pilot interviews were provided to the interviewer for coding, but were never read to study participants with the exception of questions employing a scaling format (e.g., Likert scale). An open-ended field was always available to record complete responses. 

The questionnaire and consent documents were first developed in French and then independently translated into Sango, the primary local language spoken throughout the country. An independent back-translation and pilot surveys were used to finalize and validate the instruments. Once the questionnaire was finalized, it was programmed into a Personal Digital Assistant (PDA) using KoBo, our custom data collection package.[2] The use of PDAs enables enumerators to enter the data directly as they conduct interviews. The forms contain a built-in verification system that reduces the risk of skipping questions or entering erroneous values, resulting in higher quality of the data. Daily synchronization with a central computer allows the lead researchers to check data for consistency and outliers during data collection.

Figure 1: Digital Data Collection

Data Collection and Analysis

Data collection took place during six weeks between November and December 2009. Five teams of two men and two women, for a total of twenty interviewers, implemented the study under the guidance of the lead researchers. The interviewers were university students or professionals with research experience. Prior to data collection, they participated in an eight-day training covering the study objectives and content, survey and interview technique, the use of the PDA, and trouble shooting and solving technical problems.[3] The training included mock-interviews and pilot-testing with randomly selected individuals at non-sampled sites.  

At the survey sites, each interviewer was expected to conduct four interviews per day, each lasting an average of one hour. One-on-one interviews were conducted anonymously in a confidential setting. Due to the sensitivity of some of the questions, the interviewers were assigned to same-sex respondents. Upon selection of study participants, oral rather than written informed consent was obtained because of the high illiteracy rate in CAR. After data collection, the data were imported for analysis with the Statistical Package for Social Science (SPSS). All the results presented here accounted for the complex sampling methodology and weight factors.

Limitations

There are several limitations inherent to the method and context of the study. The sample was designed to be representative of the region under study, not for the whole of CAR, since only part of the country was included in the survey. Some villages, households, and individuals had to be replaced, and how the replaced individuals differ from those interviewed is unknown. In addition, it is possible that responses were influenced by inaccurate recall, social desirability and concerns over safety in areas affected by armed conflict. The training, use of a consent form, anonymous interviews, confidentiality, supervision, and quality control were all designed to reduce biases and errors. Constructs and terminology used for this study were not defined or explained to the participants to avoid influencing them. As a result, they were free to interpret those concepts based on their own understanding. To address this limitation, we asked respondents to define key concepts (e.g., peace, justice), and throughout the questionnaire, we carefully chose phrasing and translation that would avoid misunderstanding.

[1] 3ème Recensement Général de la Population et de l'Habitation de 2003 (RGPH03)

[2] Since 2007, the Human Rights Center has been developing tools to facilitate electronic data collection.

[3] The PDAs are designed to be used by surveyors with no or little computer experience.