Context
Burkina Faso is a predominantly rural country of estimated 20 million inhabitants in 2020 with a 3.1% population growth rate per year [22]. It is a primarily patriarchal country in which women’s childbearing roles and subordinate conditions are part of the social fabric [23]. In 2018, Burkina Faso was among countries with the lowest human development index and the highest gender inequality index in West Africa [24].
Data source
The data for this study came from the individual women file of the 2010 Burkina Faso Demographic and Health Survey (DHS) [25]. This is the most recent nationally-representative and complete household-based data collected on family planning, women’s empowerment, and household characteristics.
The DHS used a stratified two-stage cluster design. At the first stage, 574 Enumeration Areas (EA) were selected using the Probability Proportional to Size (PPS) sampling method based on the 2006 population census. However, one EA located in the Sahel region was not surveyed. At the second stage, a total of 14,424 households were selected using systematic random sampling based on a list of households in each EA. Within each selected household, all eligible women aged 15-49 years were interviewed. In sum, data from 573 out of 574 clusters, 14242 out of 14536 households, and 17087 out of 17363 women were produced resulting in response rates of 99.8%, 99.2% and 98.4% at cluster, household and individual levels.
Population
Our focus was on MWRA at risk of pregnancy (fecund) and who desired to limit or delay fertility by at least 2 years. We also included pregnant and postpartum amenorrhoeic women who mistimed, unplanned or did not want their current pregnancy or recent childbirth. After removal of missing values, the final unweighted study population size was 4714 MWRA, living in 573 communities, and in need of FP (Figure 2).
Outcome and exploratory variables
In the DHS recode manual, modern contraceptive methods are female and male sterilization, contraceptive pills, intrauterine contraceptive devices, injectables, implants, female and male condoms, diaphragm, contraceptive foam and jelly, lactational amenorrhea method, standard days method, two-day method, emergency contraception, cervical cap, and contraceptive sponge [25]. The respondents were asked two questions. First, “Are you currently doing something or using any method to delay or avoid getting pregnant?” If the answer was yes then we would follow up with the second question: “Which method are you using?” We coded “1” when women were using any modern contraception methods and “0” if otherwise.
Our main exposure variable is women’s empowerment measured at both community and household levels. In this study, community-level indicators were aggregated data computed for each EA using data of all women in that EA. Norms that favored violence and discrimination against women were measured by the proportions of acceptance of domestic violence, early marriage (before 18 years old), female genital mutilation, and unpaid employment. Meanwhile, norms that promote women’s rights and opportunities were measured by the proportions of women’s assets ownership, exposure to family planning messages, access to family planning health workers, years of education, and ideal number of children. Women’s agency in their households was represented by three variables: women’s attitudes toward domestic violence, participation in decision making, and freedom in seeking medical services. The three variables were constructed based on twelve questions. The categorization and codification are shown in Table 1.
Household socioeconomic characteristics (wealth and residence in rural or urban areas), and women’s sociodemographic factors (age and education) were used as control variables.
Reliability of women’s agency
The responses to the 12 questions in Table 1 were examined for internal consistency by Principal Components Analysis (PCA) and Cronbach’s alpha. Since PCA is sensitive to relative scaling, we then used original categorization from the survey and identified three constructs of women’s agency using 0.300 as the cut-off value [26]. After that, Cronbach’s alpha was applied scaled responses to assess their inter-item reliability for each specific construct; we used 0.65 as a minimum reliability cut-point [27]. The scaling was used to differentiate “empowered” from “not empowered” women. For instance, women who did not justify wife beating, those who had a say in household decisions, and those who did not find problem in getting medical help were considered as empowered. Finally, drawing from previous studies, summative scales were constructed to reflect greater women’s agency in a specific domain [28]. For attitudes toward domestic violence, a woman was given a score of 1 if she responded with 'no' to all five DHS questions (greater agency) and 0 if she responded ‘yes’ (subordination) to any question. Involvement in household decision-making was based on women’s say in family visits, own health care, or household purchases. A score of 1 was given if she ‘solely or jointly with the husband’ participated (greater agency) and 0 if ‘husband alone, someone else or other’ made the decision (exclusion). In freedom of movement to seek healthcare, we based our analysis on constraints to access healthcare and coded 1 if a women faced no constraint (autonomy) and 0 if she faced constraint (dependency) in each of the following situations: permission to go, money needed for treatment, distance to the health facility, or not wanting to go alone (Table 1).
Data analyses
We performed descriptive and bivariate analyses and fitted multilevel logistic regression models adjusted for weights accounting for cluster-effect and non-response. All analyses were conducted using Stata version 14 [29].
Firstly, we described frequencies and probability distributions of community-level indicators of gender equality, women’s agency, as well as household socioeconomic status and women’s sociodemographic characteristics. For community indicators, values higher or lower than national means were regarded as “high” or “low”. Then, we assessed the relationship between these independent variables and modern contraceptive prevalence using Chi square. Prior to the regression analysis, we ensured that there was no multicollinearity using a less than five (< 5) Variance Inflation Factor (VIF) as a threshold. In the multilevel logistic regression analysis, the null model showed an Intra-Cluster Correlation (ICC) of 20%. Three models were built to estimate predictors of modern contraception use. Model 1 analyzed the effect of women’s agency domains on contraceptive use. Model 2 added community indicators of gender equality, including those reflecting violence and discrimination against women and promotion and women’s rights and opportunities. Model 3 further added household socioeconomic status and women’s sociodemographic factors.
Ethical considerations
The proposed analysis was exempt as described in the enforcement rules issued by the Institutional Review Board of National Yang-Ming University.