Drivers of desire for more children among childbearing women in sub-Saharan Africa: implications for fertility control

We pooled data from the DHS of 32 sub-Saharan African countries. Specifically, we used data from the women’s file of the various countries. The DHS focuses on essential maternal and child health markers, including fertility preference [27]. The DHS employs a two-stage stratified sampling technique, which makes the survey data nationally representative [28]. A total of 232,784 married and cohabiting women with birth history who had complete information on desire for more children made up the sample for the study. Details of the methodology adopted by the DHS have been reported elsewhere [28]. Table 1 gives a detailed description of the study sample.

Desire for more children was the outcome variable. This was derived from the question “Would you like to have a (another) child with your husband/partner, or would you prefer not to have any more children with him?” It had five responses: “want a (another) child,” “want no more,” “cannot get pregnant,” “undecided,” and “don’t know.” Our outcome variable was computed from two of these responses, namely “want a (another) child,” coded as 1 and “want no more,” coded as 0. Hence, women who responded that they want another child were considered as having a desire for more children while those who responded that they want no more were considered as not having a desire for more children. Women who provided any other response (“cannot get pregnant,” “undecided,” and “don’t know”) were excluded because their responses were unclear about their fertility preference.

The study used eleven independent variables, grouped into individual level and contextual level factors. The individual level factors included age, highest educational level, partner’s highest educational level, parity, current use of contraceptives, exposure to media (radio, television and newspaper/magazine), ideal number of children, decision making autonomy (decision on healthcare, decision on large household purchase and decision on visits to family or relatives), and number of living children. The contextual level factors were place of residence and wealth status. These variables were considered because of their statistically significant relationships with desire for more children in previous studies [2, 29, 30]. Details of how each of these variables were coded can be found in Table 2. Based on the findings of previous studies [2, 12, 21‐26], we hypothesized that older women would be less likely to desire for more children compared to younger women; women with higher levels of education would be less likely to desire for more children compared to those with no formal education; women whose partners have higher levels of education would have lower odds of desiring for more children compared to those whose partners have no formal education. Other hypotheses that guided the analysis and results of the study were that the odds of desire for more children would decrease with increasing parity, wealth quintile, higher number of living children, contraceptive use and exposure to media. Women who consider 6 + as the ideal number of children, those who do not take decisions alone, and those who live in rural areas would be more likely to desire for more children.

Stata version 14.0 was used to process and analyse the data. The analysis began with a computation of the prevalence of desire for more children in SSA using bar chart. After this, we pooled the datasets and calculated the proportions of desire for more children for each of the explanatory variables. We then used a bivariate logistic regression to assess the association between the explanatory variables and desire for more children. This was done to identify significant explanatory variables for the next part of the analysis, which involved multilevel logistic regression. For the multilevel logistic regression, a two-stage approach was employed, where women were nested within clusters and clusters were considered as random effects to cater for the unexplained variability at the contextual level [31]. Four models were generated from the multilevel modelling, consisting of the empty model (Model 0), Model I, Model II, and Model III. Model 0 showed the variance in desire for more children attributed to the distribution of the primary sampling units (PSUs) in the absence of the explanatory variables. Model I had the individual level factors and desire for more children while Model II contained the contextual level factors and desire for more children. The final model (Model III) was the complete model that had the individual and contextual level factors and desire for more children. Model comparison was done using the log-likelihood ratio (LLR) and Akaike’s Information Criterion (AIC) tests. Odds ratio and associated 95% confidence intervals (CIs) were presented for all the models apart from Model 0. To ensure non-existence of correlation between the significant explanatory variables, we ran a multicollinearity test, using the variance inflation factor (VIF), and the results showed no evidence of collinearity among the explanatory variables (Mean VIF = 1.71, Maximum VIF = 2.93 and Minimum VIF = 1.03). Statistical significance was declared at p < 0.05. Sample weight (v005/1,000,000) was applied to correct for over- and under-sampling while the SVY command was used to account for the complex survey design and generalizability of the findings. According to Hatt and Waters [32], pooling data can reveal broader results that are ‘‘often obscured by the noise of individual data sets.’’ To calculate the pooled values, an additional adjustment is needed to account for the variability in the number of individuals sampled in each country. This is accomplished using the weighting factor 1/(A*n_c/n_t), where A is the number of countries asked a particular question, n_c is the number of respondents for the country c, and n_t is the total number of respondents over all countries asked the question [33].

The DHSs obtained ethical clearance from the Ethics Committee of ORC Macro Inc. as well as Ethics Boards of partner organisations of the various countries such as the Ministries of Health. During each of the surveys, either written or verbal consent was provided by the women. This was a secondary analysis of data and, therefore, we did not need further approval for this study since the data is available in the public domain. However, we sought permission from MEASURE DHS website and access to the data was provided after our intent for the request was assessed and approved on 3rd April, 2019. Further information about the DHS data usage and ethical standards is available at http://​goo.​gl/​ny8T6X

The socio-demographic characteristics of the women who participated in the study are presented in Table 2. Table 2 further presents the proportion of women who desire for more children and the unadjusted odds results of the association between the socio-demographic characteristics and desire for more children. Desire for more children was high among women aged 15–19 (95.3%), those with no formal education (69.3%), those whose partners had no formal education (70.9%), women with one birth (94.4%), women who were not current contraceptive users (69.3%), and those who were not exposed to media (68.3%). High prevalence of desire for more children was also found among women who perceived six or more children as the ideal number of children (72.2%), those who do not take decisions alone (66.6%), those who had no living children (96.6%), women who lived in rural areas (66.6%), and poorest women. Results from the bivariate analysis showed that all the independent variables had statistically significant association with desire for more children (see Table 2).

Table 3 presents results of the multilevel logistic regression analysis of the predictors of desire for children among married and cohabiting women in SSA. With the fixed effects results, results of the individual level predictors showed that women aged 45–49 [AOR = 0.04, CI = 0.03–0.05], those with higher education [AOR = 0.80, CI = 0.74–0.87], and those whose partners had higher education [AOR = 0.88; CI = 0.83–0.94] had lower odds to desire for more children, compared to women aged 15–19, those with no formal education, and those whose partners had no formal education, respectively. Furthermore, the likelihood of desire for more children decreased among women with four or more births [AOR = 0.10, CI = 0.09–0.11], those who were using contraceptives [AOR = 0.68, CI = 0.66–0.70], and those who had four or more living children [AOR = 0.09 CI = 0.07–0.12], compared to those with one child, those who were not using contraceptives, and those who had no living children, respectively. On the other hand, the odds of desire for more children was high among women who considered six or more children as the ideal number of children [AOR = 16.74, CI = 16.06–17.45] and women who did not take decisions alone [AOR = 1.58, CI = 1.51–1.65], compared to those who consider 0–3 children as the ideal number of children and those who took decisions alone, respectively (see Table 3, Model III). With the contextual factors, the odds of desire for more children was high among women who lived in rural areas compared to urban areas [AOR = 1.07, CI = 1.04–1.13].

In terms of the random effects results, in the empty model, there were substantial variations in the likelihood of desire for more children across the clustering of the PSUs (σ2 = 0.03, 95% CI 0.02–0.04). The empty model showed that 3% of the total variance in desire for more children was attributed to between-cluster variation of characteristics (ICC = 0.03). The between-cluster variations showed a decrease from 3–2% from the empty model to the individual-level only model (Model I). From Model I, the ICC increased to 3% (ICC = 0.03) in the contextual level only model but decreased to 2% in the complete model (Model III), which had both the individual and contextual level factors. This explains that the variations in the likelihood of desire for more children could be attributed to the differences in the contextual level factors (see Table 3, Model III).

The findings from this study may be instructive to the current fertility control interventions and initiatives across SSA. The study has indicated the traits of women with desire for more children in SSA. Admittedly, nearly all countries in SSA have instituted measures to regulate fertility by moderating both demand and supply factors [60]. Our findings have provided evidence from current data sets that may aid in benchmarking parameters for evaluating and reviewing existing policies and interventions in order to render them more sensitive to the current population. To contextualise our findings and derive much benefit at the country level, governments of SSA and partner organisations should be sensitive to the in-country nuances driven by cultural, geographical, and socioeconomic factors. This may be beneficial in ensuring that the number of children desired by women falls within national estimations and expectations.

The use of nationally representative datasets of 32 countries in SSA and the multi-stage sampling technique to select the respondents is a major strength of this study. These make it feasible to generalise the findings to all women in sexual unions in SSA. The relatively large sample size also aided in fitting robust logistic regression models to model the factors associated with desire for more children while controlling for confounders. Despite these strengths, it is impossible to establish temporality of sequence, and the possibility of social desirability biases cannot be overruled.