All Demographic and Health Surveys receive ethical review by ICF’s institutional review board. Before being publicly released, all DHS datasets are anonymized and geographic coordinates are offset by up to 2 km in urban areas and 5–10 km in rural areas in order to prevent identification ³³ .

Data. This study utilizes data from the 2018 Demographic and Health Surveys (DHS) in Nigeria and Zambia. These countries were chosen for their availability of recent DHS data, inclusion of the men's questionnaire, representation of different geographic regions of Africa, and having different contraceptive S curve classifications. Nigeria is classified as low growth and low prevalence, while Zambia is classified as the rapid growth in the S-curve classification. ¹ The DHS employs a stratified two-stage cluster sampling design, with households randomly selected within Primary Sampling Units (PSUs) or clusters. Details of the methodology employed for each DHS survey can be found in the final reports ^{34, 35} . For the purposes of this analysis, we limited the analytic sample to non-pregnant women and men in union, as we believed that the factors influencing contraceptive use would differ among those in and not in union. In Nigeria, this resulted in a final sample of 24,822 women and 6,810 men in 1,389 clusters. In Zambia, the final sample included 6,727 women and 5,715 men in 545 clusters.

Measures. The outcome of interest in this study was modern contraceptive use. The questions used to capture information relating to modern contraceptive use were slightly different for women and men, with women being asked about current use and men being asked about use during their last sexual encounter. Methods were categorized as modern according to the DHS definition of modern method ³⁶ . Female or male condoms, contraceptive pills, emergency contraception, implants, injectables, intrauterine devices (IUDs), the lactational amenorrhea method (LAM) the standard days method, and male or female sterilization were all considered modern methods. Respondents who reported that they or their partner were using any of these methods were considered to be using modern contraception.

We examined three categories of community norms as possible influences on modern contraceptive use.

First, fertility norms were measured using the response of mean and women to the question of the number of children they would want to have if they could choose the exact number of children to have over their lifetime.

Second, gender norms were measured using the domains of the SWPER (Survey-based Women’s emPowERment index) Global index. The SWPER was originally developed for African countries as an individual-level indicator that would allow for comparison of empowerment between countries and over time ³⁷ . The measure was later improved as the SWPER Global which can applied globally ³⁸ . Since its creation it has been used in various contexts and outcomes, including reproductive and maternal health, child growth, female genital mutilation, and child vaccination ^{39– 42} . The SWPER consists of three domains—attitude toward violence, social independence, and women’s decision-making power- which have a standardized score. A score of 0 represents the average of level of the domain relative to all women in the country’s respective region (West and Central Africa for Nigeria and Southern and Eastern Africa for Zambia). Negative values then represent below average levels, and positive values represent above-average levels.

Finally, family planning norms were evaluated using two measures. The first measure assessed contraceptive decision-making through women's participation in the decision to use or not to use contraception. Women who reported that they participated in the decision-making process, either alone or jointly with their husband, were categorized as "decision-makers". On the other hand, those for whom the main decision-maker for contraceptive use or non-use was their husband or someone else were categorized as "not decision-makers". The second measure evaluated men's beliefs about contraceptive use and promiscuity. Men were asked whether they agreed or disagreed with the statement that women who use family planning (FP) may become promiscuous.

For each measure, the community-level value was calculated by taking the average value of the individual-level measures for men or women in the cluster.

Our models also included individual and community factors previously demonstrated to have associations with men or women’s modern contraceptive use. At the individual level, these included age, men’s education level, wealth quintile, number of children ever born, exposure to family planning messaging, and desire for children in the next two years. While women’s education has also been shown to be associated with modern contraceptive use, it is one of the component variables in the SWPER and was excluded as a covariate. All individual-level values of the community norms measures were also controlled for. Community control variables include distance as a problem accessing care, place of residence, and community average years of education among men.

Additional information on the calculation of each measure can be found on The DHS Program website ⁴³ .

Analysis. We estimated modern contraceptive use prevalence for each country, then analyzed cross-tabulations of men's and women's characteristics, and community-level factors by their contraceptive use status. Chi-square and t-tests determined differences in categorical and continuous variable distributions. Finally, we used multilevel logistic regression to study the relationship between contraceptive use and community-level norms for fertility, gender, and family planning.

Models were constructed using Stata 17’s melogit command with the cluster (community) as the grouping structure. The first model (model 0) had no covariates and estimated the intraclass correlation coefficient (ICC) of overall modern contraceptive use. Model 1 included only individual-level covariates, while the full model (model 2) included all individual- and community-level factors. Our analysis aimed to estimate contextual effects of the community-level variables above and beyond the individual-level effects of the variable, so we controlled for the individual-level values for the community-level aggregate covariates in model 2 ⁴⁴ . To evaluate the risk of collinearity between individual- and community-level variables, the variance inflation factor (VIF) for each variable was calculated and none had a VIF over 5. To facilitate interpretation of regression coefficients, we standardized aggregate community-level covariates to have a mean of 0 and standard deviation (SD) of 1 within each country.

We estimated multiple measures of variation for each model, including the community-level random effect, which describes the community-level variation due to unobserved covariates, as well as the ICC, which provides the proportion of the variance explained by the clustering in the population.

Individual and cluster weights were estimated and applied for all multilevel models according to DHS guidance ⁴⁵ .

All analysis ⁴⁶ was conducted using Stata 17.

Associations between modern contraceptive use and individual and community variables. Less than 15% of women and men in Nigeria use a modern contraceptive method (see Table 1). Modern contraceptive use differed by all individual and community level variables shown in Table 1 except for the opinion that contraception makes women promiscuous among men. At the individual level, modern contraceptive use was the lowest for men and women with no children and for those that want children soon.

At the community level, men and women from urban areas had higher percentage of use of modern methods. Modern contraceptive use was also significantly higher for women and men living in communities where access to care was not a barrier. The mean ideal number of children at the community level was lower among users and differed by more than one child on average among women and more than two children among men.

Table 1 also summarizes the bivariate associations between modern contraceptive use and the SWPER indices at the individual and community level. Women users had higher mean individual SWPER scores compared to non-users and the differences were significant. At the community level, users of modern contraceptive methods were consistently from communities with higher average SWPER scores. The average community SWPER score for contraceptive users was similar for men and women, except for the social independence domain, where the community average score for men who use contraception was higher than the community average score for women who use contraception.

Multilevel logistic regression results. Figure 1 and Table 2 summarize the regression results from the full model in Nigeria for women and men. Most community-level variables had a significant association with women’s modern contraceptive use. Living in a community where distance is a barrier to access and higher community-average ideal number of children were both negatively associated with women’s modern contraceptive use. Higher community levels of women’s involvement in contraceptive decisions, higher community average SWPER score in the attitude to violence domain, and more average years of men’s education at the community level were positively associated with modern contraceptive use. Men’s modern contraceptive use was significantly negatively associated with community average ideal number of children, but positively associated with increasing years of men’s education, and the percentage of men who believe FP makes women more promiscuous, with the unexpected findings that higher percentages of men in the community believing that FP makes women more promiscuous are associated with higher contraceptive use among men.

Of the individual-level covariates, only the SWPER for social independence was not statistically significantly associated with women’s modern contraceptive use. For men, age and education (not shown in figure), exposure to family planning (FP) messages, and desiring a/another child soon were significantly associated with use.

As shown in Table 2, 36% of the variance in overall modern contraceptive use among women, and 43% of the variance in overall modern contraceptive use among men in Nigeria can be accounted for by cluster membership. Among women the unexplained variance due to cluster membership is cut nearly in half, to 19%, after accounting for individual- and community-level covariates, while among men it only decreases to 31%. The random effects parameter for community was statistically significant in all models for both men and women, suggesting that our models are missing key unobserved community characteristics associated with modern contraceptive use. The random effects parameter for men was higher than that for women.

Associations between modern contraceptive use and individual and community variables. Just over half of women and men in Zambia use a modern contraceptive method (see Table 1). Bivariate analyses show both men and women had significant associations between modern contraceptive use and parity, FP message exposure, ideal number of children, and desire for another child at the individual level. Among women, there were differences in contraceptive decision-making and individual level SWPER scores for attitude to violence and decision-making between those who use modern contraception and those who do not. All the community level covariates were found to have significant associations with modern contraceptive use among women except for women’s involvement in contraceptive decision-making. Among men, our analysis showed few statistically significant findings at the community level. The only community factors found to be significantly associated with men’s modern contraceptive use were community average ideal number of children and community average SWPER scores for attitude to violence and decision-making.

Multilevel logistic regression results. Figure 2 and Table 3 summarize the multi-level regression results of modern contraceptive use for Zambian women and men. Living in a community where distance is a barrier to health care, community average ideal number of children, community proportion of women involved in contraceptive decision-making, and the community average SWPER social independence were negatively associated with women’s modern contraceptive use, while increasing community-level attitude to violence, which indicates lower acceptance of domestic violence, was positively associated with women’s modern contraceptive use.

The community-level variables of urbanicity, ideal number of children, and all three SWPER domains were significantly associated with men’s modern contraceptive use in Zambia. However, there was a negative association between use and urbanicity, and the SWPER for social independence and decision-making which was not in the expected direction.

For women, many individual-level characteristics, including ideal number of children, desire for a/nother child, whether she is a contraceptive decision-maker, and her SWPER decision-making score, were all significantly associated with modern contraceptive use. A man’s age, wealth quintile, number of living children, previous exposure to FP messages, his ideal number of children, desire for a/nother child were associated with his modern contraceptive use.

In Zambia, 20% of the variance in overall modern contraceptive use among women, and 27% of the variance in overall modern contraceptive use among men can be accounted for by cluster membership (See Table 3). Among women the unexplained variance due to cluster membership only decreased slightly to 18%, after accounting for individual- and community-level covariates, while among men it only decreases to 25%. Just as for Nigeria, the random effects parameter for community was statistically significant in all models for both men and women. However, the random effects parameters for men and women were both similar to each other in the Zambia models.