Evaluation of a Mass Media Family Planning Campaign on the Uptake of Contraceptive Methods in Burkina Faso

Development Media International (DMI), a non-governmental organization (NGO) that runs media campaigns to induce behavioral change in developing countries, will implement the mass media campaign in conjunction with community radio stations in rural areas of Burkina Faso. The study takes place in the coverage areas of 16 radio stations, all selected in a way to prevent overlap between coverage areas, and to have different local languages through which the campaign will be diffused. This strategy limits possible "leakages" between the treatment and the control groups. Out of the 16 clusters of radio stations, 8 will be randomly assigned to receive the media campaign, and the other 8 will be left as control.

Within each radio station cluster, only small and rural villages with no access to electricity will be sampled in order to limit access to television. The radio stations being in urban to semi-urban areas, villages that are more than 5 km away from a radio station will be excluded. Villages within radio station clusters will be randomly selected to participate in the study. A household listing survey will be conducted in all selected villages to get an exhaustive list of households and the basic information on women within each household. In addition, a village survey will be conducted with village chiefs to get a better sense of the demographics and the key figures in each village.

Based on the information provided by the listing on women at the age of reproduction, eligible women will be randomly selected to participate in the study. These women will be invited to take the baseline survey, and 3 years later, the endline survey, therefore forming a panel structure.

In addition to the listing, village, and women surveys, a clinics survey will take place to monitor the demand and supply of contraceptive methods in different areas. The clinics questionnaire will ask about contraception related behavior, including perception of family planning, usage of contraception, number of pregnancies and abortions, among other things. The investigators are also seeking administrative data from clinics within radio stations areas to examine whether distribution of contraceptive products is higher in treatment areas.

The effect of the media campaign will be analyzed based on information provided by women on how they and their partners perceive and use traditional and modern contraceptive methods, and cross-verified with information provided on demand for contraception and family planning by the clinics survey.

Qualitative research will take place prior to the launch of the campaign in order to better formulate messages that could reach the target audience.

-Data collection, reporting, and analysis:

The data collection will be conducted by Innovations for Poverty Action (IPA), and NGO specialized in conducting randomized control trials (RCTs) with presence and experience in Burkina Faso.

Data will be collected through questionnaires using Personal Digital Assistance (PDA). For the qualitative research questionnaires, answers might be recorded using paper. The surveyors might audio-record some parts of some interviews with women and health facilities using the PDAs to monitor surveying quality and performance of enumerators. The resulting datasets will be coded and clearly labeled in accordance with the information provided on the questionnaires.

Data will be collected by enumerators trained by IPA on the importance of confidentiality of the data, transparency, and accuracy of reported information. Backchecks will also be conducted to make sure that the data provided is consistent and correct. Protocols have been developed to address adverse events during data collection where enumerators go back to their supervisors and to the IPA office to address any complications. In addition, high frequency checks will be conducted whenever new data comes in to monitor the quality of data collection, and how values of different variables and missing values vary by enumerators and locations. Logic checks will be conducted to make sure that the value of variables fall within logical ranges. This allows for quick detection and correction of major outliers and high rates of attrition. Finally, all adverse events will be reported to the Massachusetts Institute of Technology's internal review board (the Committee on the Use of Humans as Experimental Subjects) as quickly as possible in accordance with the human subjects requirements.

-Sample size and minimum detectable effect:

The total sample size is approximately 8,000 women across more than 250 villages in 16 radio clusters.

To calculate this sample size, power calculations were conducted using simulations in Stata. Having two levels of clustering (region and village levels), two intracluster correlation (ICC) levels were used for power calculations, both based on the 2010 DHS data for women in Burkina Faso. The DHS data revealed a modern contraceptive prevalence rate (mCPR) of 9% among women of reproductive age (both in union and single) in rural areas and regions where the study takes place.

For these calculations, a baseline level of mCPR of 12% was assumed, therefore allowing for some increase in usage over time. Based on these numbers and calculations, the study is powered to detect a minimum of 6 percentage points increase in the mCPR.

If the assumed baseline level of mCPR is close to the actual one, power is expected to increase further as the investigators will (1) use a matched stratification on baseline levels of contraception prevalence rate at the radio station level (the level of treatment assignment), in addition to stratifying village selection on distance to health clinics, and individual selection of women on several characteristics such as education and access to a radio (2) control for baseline levels of explanatory variables and the baseline level of the outcome, and (3) use a panel structure.

It is possible, however, that the total and modern contraceptive prevalence rates are higher in the actual sample than the value assumed based on the 2010 DHS data, which could reduce power.