# Determinants of contraceptive decision making among married women in Sub-Saharan Africa from the recent Demographic and Health Survey data | BMC Women’s Health

### Source data

The study made use of pooled data from current Demographic and Health Surveys (DHS) conducted from 2010 to 2018 among 33 Sub-Saharan Africa countries.. These 33 countries were included in the study because they had current DHS data and also all the variables of interest for this study. Our study included these 33 countries under the DHS program to provide holistic and in-depth evidence of women contraceptive decision making in SSA.

The DHS Program has been working with developing countries around the world to collect data about significant health issues including fertility. This standard Demographic and Health Survey is a population-based survey, nationally representative, contains high-quality data that follow standardized data collection procedures, have consistent content over time, and is collected through uniform questionnaires.

The survey target groups were women aged 15–49 and men aged 15–59 in randomly selected households in each country with a multi-stage stratified cluster sampling design for each country. The study involved a cluster sampling process (ie enumeration areas [EAs]), followed by systematic household sampling within the selected EAs. The sample frame usually excludes nomadic and institutional groups such as prisoners and hotel occupants. Detailed information was collected on the background characteristics of the respondents including maternal health and child health [32]. The data for this study were extracted from the individual record (IR file) file from the standard DHS dataset of Sub-Saharan Africa countries with at least one survey from 2010 -2018. A total of 76,516 currently married women who are not pregnant and are current users of family planning were included from 18 low income, 11 lower middle income and 4 upper-middle-income Sub-Saharan countries (11 East African, 6 Central African, 13 West African, and 3 South African countries).

### Eligibility identification

Reproductive age group women aged (15–49 years) currently married women who are not pregnant and are current users of family planning preceding three years the survey in the selected enumeration areas in 33 Sub-Saharan African countries included for this study. Whereas, countries (Central Africa Republic, Eswatini, Sao Tome Principe, Madagascar, and Sudan) did not have a DHS survey report after the 2010/2011 survey year were excluded. As well, three Sub-Saharan Countries (Botswana, Mauritania, and Eritrea) were excluded due to the dataset was not freely available. The outcome variable of this study was the decision-maker to use contraceptives.

### Dependent variables

The dependent variable had three (3) categories namely: the women-only decision making (coded as 1), joint (mother and husband/partner) decision making (coded as 2), and husband/partner-only decision making on contraceptive utilization (coded as 3).

### Statistical analysis

The analysis began with the computation of contraceptive decision-making among married women from 33 Sub Sharan Africa countries. Secondly, we appended the dataset and this generated a total sample of 76,516. After appending, we compute v005/1,000,000 (Women’s individual sample weight/1,000,000) to develop weighted country-based and socio-demographic characteristics (Tables 1, 2).

Multinomial logistic regression (MNLR) model is generally applicable when the outcome variable is composed of polychotomous categorical having multiple choice. It is a simple extension of logistic regression that allows each category of unordered responsive variables to be compared to an arbitrary references category providing several logit regression models.

Multinomial logistic regression models are equivalent to simultaneous estimation of multiple logits where each of the categories is compared to one selected based category.

Let Y_{i1}be 1 if the ith decision-maker is manly women-only and 0 otherwise. Similarly, Yi2 be 1 if the ith decision-maker is jointly(women/husband/partner and 0 otherwise. Yi3 be 1 if the ith decision-maker is manly husband/partner and 0 otherwise.

All variables included in bivariate analysis were analyzed in the multinomial logit model. A multinomial logistic regression model was used to estimate variations in the probability of decision-makers to use contraceptives. When using multinomial logistic regression, the relative risk ratios were determined for all independent variables for each category of the dependent variable except the reference category, which is omitted from the analysis. The regression model was fitted to the data to explore the association between a set of independent variables explaining the likelihood of decision making on contraceptive a woman decision making on contraceptive as opposed to being in all other categories. The form of the equation fitted to the data was as follows:

$$mathrm{ln}frac{p({y}_{i}=m)}{p({y}_{i}=1)}=mathrm{a}+{sum }_{ k=1}^{k}{beta }_{xm }{x}_{ik}={z}_{mi}$$

(1)

A dependent variable (contraceptive decision-making) that had 3 categories, is represented by m in the equation above, and this requires the calculations for (m-1) equations, one for each category relative to the reference category to describe the likelihood of contraceptive decision making and the independent variables. For the women-only category of the dependent variable, for example, the following equation derived from the latter is then estimated:

$$mathrm{p}({y}_{i}=mathrm{m})=frac{mathrm{exp}left({z}_{mi}right)}{1+{ sum }_{n=1}^{m}{exp}^{{(z}_{hi})}}$$

(2)

In the multinomial logistic regression, the husband-only is the comparison category. The model parameter estimates and the attendant Relative Risk Ratios (RRR) for the multinomial logit model is that for a unit change in the predictor variable, the logit of outcome m relative to the reference group is expected to change by its respective parameter estimate given that the variables in the model are held constant. The RRRs can be obtained by exponentiation of the multinomial logit coefficients (({e}^{mathbf{c}mathbf{o}mathbf{e}mathbf{f}mathbf{f}mathbf{i}mathbf{c}mathbf{e}mathbf{ n}mathbf{t}})), or by specifying the **rrr** option. The alpha threshold for significant results was set at p = 0.05 (95%).