Statistical analysis

Data analysis was performed using Stata statistical software (Release 14, StataCorp). Sampling weights were used to obtain a national representation of the survey results. Data for married females were extracted from all four surveys to build a graphical trend of evolution for CEB. Weights were calculated separately for each sampling stage and each cluster, using probability sampling. R statistical software (R V.3.4.1 and R Studio V.1.3.959) was used to plot the time series trend for the mean number of CEB as presented in figure 1. The unit of analysis was married females between the ages of 15–49 years and the number of CEB. For the primary analysis, that is, assessing the association of socio-economic factors with the number of CEB, data from the 2017 GMHS was used. The demographic characteristics of the study respondents including place/region/zone of residence, wealth index, age, media exposure, level of education, age at marriage, number of abortions and age at first sex were presented as weighted frequencies and percentages. The mean and SD of CEB were also reported by the demographic variables and differences tested using the χ² test. Three separate models were used to assess the association of socioeconomic characteristics with CEB. First, a linear regression model was considered using CEB as a continuous response (model 1). Subsequently, two response variables were computed from the continuous response variable by grouping the number of CEB as, model 2: no births (ie, 0 CEB) vs at least one birth (ie, CEB≥1), model 3: among respondents with at least one birth, we compared those with 1 or 2 CEB (mid-size) vs ≥3 CEB (large size). In model 2, logistic regression was used to assess the association of maternal socioeconomic characteristics with having at least one birth versus no birth. Among respondents with at least one birth (model 3), logistic regression was used to assess the association of maternal socioeconomic characteristics with having 3 or more births versus having less than 3 births.

(Main): mean CEB for different time points by age groups. CEB, children ever born.

In all models, bivariate and multivariate techniques were used to assess the association of demographic variables with CEB. Based on literature review, all the selected sociodemographic variables were included in the multivariate model whether significant in bivariate analysis or not. For ease and clarity of interpretation, zone was used in all bivariate and multivariate models and not region of residence. The collinearity of the socioeconomic variables was checked using the variance inflation factor (VIF), where values >10 indicate a high correlation.²⁰ Parameter estimates were reported with their 95% CIs and variables with p<0.05 in the bivariate/multivariate analyses were considered statistically significant. In the linear regression model, a histogram of the residuals and a quantile-quantile (q-q) plot were used to assess the distributional assumption.