Performances of bayesian structured additive regression and logistic models in assessing factors associated with modern contraceptive use among ever-in-union women in Nigeria

Abstract

Background

Modern contraceptive use plays a key role in reducing unintended pregnancies, improving maternal health, and curbing rapid population growth. In Nigeria, despite growing awareness and government-led interventions, usage remains low and unevenly distributed across regions. While most studies focus on individual-level determinants, limited attention has been paid to how geographic and spatial differences influence contraceptive behavior. This study compared the performance of Bayesian structured additive regression with frequentist logistic regression in identifying factors associated with modern contraceptive use among ever-in-union women in Nigeria.

Methods

This cross-sectional study used data from the 2018 Nigeria Demographic and Health Survey (NDHS) of ever-in-union women aged 15–49 years. Bayesian structured additive regression and frequentist logistic regression models were applied. The deviance information criterion (DIC) was used for model evaluation.

Results

Modern contraceptive prevalence was 11.5% among 31,152 ever-in-union women. Strong north-south disparities were identified, with higher uptake in urban and southern regions. A nonlinear relationship was observed between age at first marriage and contraceptive use, with 20.9% of women marrying before age 15years. Women with more than three children alive had 48% higher odds of contraceptive use (aOR = 1.48; 95% CrI: 1.24–1.66). Working women (aOR = 1.91; 95% CrI: 1.89–2.23) and those knowledgeable about modern methods (aOR = 1.78; 95% CrI: 1.66–2.25) also had higher odds.

Background

Modern contraceptive use refers to scientifically approved methods such as oral pills, condoms, intrauterine devices (IUDs), injectables, implants, and sterilization for preventing unintended pregnancies and managing family size. These strategies benefit maternal health, minimize adolescent births, lower infant mortality, and slow population increase [1].

Globally, modern contraceptive prevalence has increased minimally, but Sub-Saharan Africa has continued to face challenges. In this case, modern contraceptive prevalence in Africa rose from 23.6% in 2008 to 28.5% in 2015 [2], while the region accounted for 21% of the global burden of low use [3]. In Nigeria, the modern contraceptive prevalence rate (mCPR) among currently married women was 11.1% in 2013 and increased only slightly to 17% by 2018 [1], despite numerous interventions. Despite years of initiatives, including the Nigerian government’s alignment with the global Family Planning 2020 (FP2020) initiative, modern contraceptive use remains low, contributing to Nigeria’s continued annual population growth rate of 3.2% [4–6]. The FP2020 initiative, which aimed to reach 120 million additional women and girls worldwide with modern contraception by 2020, made significant progress but fell short of its primary goal. Recognizing this, Nigeria has renewed its commitment under FP2030 to expand equitable access to family planning, particularly for adolescents, youth, and underserved populations.

Recent spatial analyses have revealed ongoing regional differences, with greater uptake in urban and southern regions as compared to rural and northern areas [7, 8]. These findings show the importance of investigating how spatial factors interact with sociodemographic and socioeconomic characteristics to impact modern contraceptive usage.

Most studies on the use of modern contraceptives in Nigeria did not consider the spatial effect with regard to the use of modern contraceptives [9–13]. Most of these studies used frequentist methods to identify age, region, education, age at first marriage, and number of children ever born as factors associated with modern contraceptive use [14–17]. Although frequentist methods have helped us understand factors at the individual level, they frequently overlook the bigger picture, particularly how a person’s location can influence their access to and usage of contraceptives. Bayesian techniques not only provide more accurate estimates, but they also assist in identifying geographic patterns and clusters with particularly low rates of contraceptive use. This deeper insight is crucial for shaping policy strategies and designing evidence-based interventions tailored to meet the unique needs of different regions across Nigeria [18, 19].

This study aimed to compare the performances of Bayesian structural additive regression with frequentist logistic regression in assessing the factors associated with modern contraceptive use among ever-in-union women in Nigeria. Specifically, it addresses the following questions: [1] Which statistical modeling approach between the Bayesian structured additive and the Frequentist logistic regression offers better model performance? [2] What individual, demographic, and socioeconomic factors are associated with modern contraceptive use in Nigeria? [3] What are the spatial variations in modern contraceptive use across Nigeria’s six geopolitical zones?

This research contributes to the existing body of literature by highlighting the added value of spatial modeling using Bayesian methods and offering insights into region-specific disparities that can inform policy and programming for increased contraceptive uptake in Nigeria.

Methods

Study setting and design

The study population is ever-in-union women aged 15 to 49 years in Nigeria. Nigeria has 36 states and the Federal Capital Territory (Fig. 1).

Fig. 1.

Map of Nigeria showing 6 geopolitical zones, 36 states and the Federal Capital Territory

This study is a secondary analytical cross-sectional study that used nationally representative secondary data from the 2018 Nigeria Demographic Health Survey (NDHS). The questionnaire for data collection consisted of modules on primary information about households, the sexual and reproductive health of the respondents, and the birth history of the respondents.

Sampling

The DHS data in Nigeria is collected via a two-stage, stratified sampling design in all 36 states in Nigeria and the FCT with the clusters as the primary sampling units. While clusters are selected at the first stage, households are selected at the second stage. All eligible women of reproductive age (15–49 years) living in the selected households were interviewed.

Data management

We extracted data from 31,152 women of reproductive age 15–49 years who were ever-in-union. For data collection, the NDHS used computer-assisted personal interviewing (CAPI) in 2018, with more details published elsewhere [1]. IBM SPSS (version 20) was used for descriptive analysis. All Bayesian structured additive regression (STAR) analysis were implemented in R (version 4.0.3) using the BayesX R interface (R2BayesX package, version 0.3–1.3). While variations in categorical data were presented with frequency tables and percentages. We included sampling weights in the analysis to account for the unequal probability of selection at the cluster levels and to account for nonresponse since the samples are not self-weighted.

Variables

The outcome of interest in this study is modern contraceptive use, represented as a binary response variable classified as either Yes or No. Any respondent who uses at least one of the different types of modern contraceptives within 12 months preceding the survey date was classified as a user (Yes) or nonuser (No). The explanatory variables are the respondent’s current age, respondent’s age at first cohabitation/marriage, respondent’s age at first birth, partners’ age, religion, ethnicity, number of children alive, number of children had, family size intention, residential type, wealth index, level of education, partners’ level of education, geopolitical zone, working status, knowledge of family planning, and access to media.

Data analysis

Descriptive statistics and a comparison of frequentist logistic regression with Bayesian structured additive regression were used to evaluate which method best fits the model. Modern contraceptive usage was modeled using the structured additive regression model.

For the frequentist logistic regression, the structure of the model is given by equation:

where is the probability of the occurrence of an event that is the expected value of the response variable (such that if the respondent currently uses any modern contraceptive method and , if otherwise), represents the explanatory variables, β is the regression coefficient, and is the index of individual records at 5% level of significance [20].

Assumptions of binary logistic regression

1. The response variable is binary.

2. Observations are independent.

3. There is no multicollinearity among predictors.

For the structured additive regression models (STAR), compared to classical regression models, there is a replacement of the linear predictor by a structured additive predictor [21];

= + () +.. +.

is the smoothing function that describes the nonlinear relationship between the explanatory variables () and logit probability of the response variable. These functions will be estimated using Bayesian cubic penalized regression spline, is the nonlinear spatial effect (s) of zone , and is the regression coefficient for the categorical covariates (v). The spatial effects may be further split into spatial correlation (structured) and uncorrelated (unstructured) effects such as:

Assumptions of the structured additive regression model

1. It does not require linear relationships between predictors and the outcome.

2. The outcome variable does not need to follow a normal distribution.

Model Building

We fitted a series of spatial generalized additive models which included spatially structured random effects and nonlinear terms to account for regional variations and nonlinearity in age, respectively.

Models fitted

Model 1: = ().

Model 2: = () + f (age of respondents).

Model 3: = () + f (age of respondents) + geopolitical zones.

Model 4: = () + f (age of respondents) + geopolitical zones + demographic characteristics.

Model 5: = () + f (age of respondents) + geopolitical zones + demographic characteristics + knowledge indicators.

Model 6: = () + f (age of respondents) + geopolitical zones + demographic characteristics + knowledge indicators + socio-economic status.

Model selection and diagnostics

Model performance was evaluated using the Deviance Information Criterion (DIC). DIC is a Bayesian model selection criterion that assesses model fit while penalizing complexity. It was chosen because it is specifically suited for Bayesian models estimated via Markov Chain Monte Carlo (MCMC), unlike likelihood-based criteria such as AIC or BIC which rely on maximum likelihood estimation (MLE). Further details of prior information and model specification are shown in Appendix 1.

Results

Description of study participants

Out of 31,152 eligible ever-in-union women, referring to those who were married, living with a partner, widowed, divorced, or separated, 3,582 (11.5%) were users of modern contraceptive methods, while 27,570 (88.5%) were nonusers (Fig. 1). The overall arithmetic mean age of the respondents was 32.28 years (standard deviation = 8.76).

Among this population of ever-in-union women, a large majority were Hausa (37.1%) and practiced Islam (55.0%). Most reported a family size intention of approximately 5–9 children (50.1%) (Table 1). In terms of residential and educational status, most of these women lived in rural areas (63.1%) and had no formal education (42.8%). Although 70.9% were currently employed, most still fell within the poorer (21.6%) and poorest (21.5%) wealth index categories. Furthermore, a high proportion (66.6%) lacked knowledge of family planning (Tables 1 and 2).

Table 1.

Distribution of the studied ever-in-union women by their sociodemographic characteristics

Variable	Frequency	Percentage	MC Prevalence
Age (years) (Mean + SD)	32.28 + 8.76
15–19	1866	6.0	2.6
20–24	4515	14.5	8.2
25–29	6205	19.9	11.6
30–39	10,791	34.6	14.7
>= 40	7775	25.0	11.0
Age at first marriage (Mean + SD)	18.53 + 5.10
< 15	6514	20.9	7.0
15–19	13,955	44.8	10.5
20–24	6624	21.3	17.1
25–29	2872	9.2	14.2
>=30	1187	3.8	9.8
Age at first birth (years) (Mean + SD)	19.68 + 4.50
< 15	2096	7.2	8.8
15–19	14,382	49.5	9.8
20–24	8422	29.0	15.8
25–29	3109	10.7	15.9
>=30	1071	3.7	12.1
Partner’s Age (years) (Mean + SD)	42.29 + 11.20
15–19	26	0.1	9.6
20–24	584	2.0	13.7
25–29	2347	8.1	9.8
>=30	25,931	89.8	13.1
Religion
Christians	13,750	44.1	17.0
Islam	17,144	55.0	7.2
Traditionalist	128	0.4	3.1
Others	130	0.4	5.4
Ethnicity
Hausa	11,563	37.1	5.5
Igbo	4402	14.1	12.9
Yoruba	3752	12.1	23.7
Others	11,410	36.7	13.1
Number of children alive
<= 3	17,025	54.7	9.6
> 3	14,127	45.3	13.7
Number of children ever had
<= 3	15,085	48.4	9.8
> 3	16,061	51.6	13.1
Family size intention
<= 4	8334	27.6	17.2
5–9	15,141	50.1	11.2
>=10	6763	22.4	4.3
Residential type
Urban	11,504	36.9	16.5
Rural	19,648	63.1	8.6
Wealth index
Poorest	6687	21.5	4.5
Poorer	6724	21.6	7.2
Middle	6476	20.8	11.3
Richer	6093	19.6	16.3
Richest	5172	16.6	20.7
Level of education
No education	13,328	42.8	4.5
Primary	5448	17.5	13.1
Secondary	9584	30.8	17.5
Higher	2792	9.0	21.1
Partner’s level of education
No education	9803	33.9	4.0
Primary	4436	15.4	11.3
Secondary	9769	33.8	16.3
Higher	4487	15.5	19.3
Work Status
No	9052	29.1	7.4
Yes	22,100	70.9	13.2
Knowledge of family planning
No	20,739	66.6	9.2
Yes	10,413	33.4	16.0
Total	31,152	100	11.5

Table 2.

Model diagnostics comparing bayesian structured additive regression and frequentist logistic regression models for modern contraceptive use

	Frequentist	Bayesian
Models	Log-likelihood	pD	DIC
Model 1	22230.788	4.519	187.181
Model 2	21875.883	9.985	160.712
Model 3	21154.842	19.973	159.932
Model 4	17769.462	25.192	139.973
Model 5	17516.771	31.761	119.985
Model 6	17007.148	46.781	110.152

Comparison of the logistic regression and STAR model

Based on the principles of model selection, deviance reflects how much unexplained information remains after a model has been fitted [21]; therefore, a lower value indicates a better fit. As shown in Table 2, model performance improved progressively as additional covariates were included. For the frequentist logistic regression, log-likelihood values ranged from 22230.788 to 17007.148, while in the Bayesian structured additive regression, DIC values ranged from 187.181 to 110.152. These results suggest that the Bayesian approach provided a more flexible and better-fitting model for this dataset, likely due to its ability to account for nonlinear relationships and spatial effects.

Model 6, which achieved the lowest log-likelihood and DIC values in their respective approaches, was selected as the best-fitting model. DIC was particularly suited for assessing the Bayesian models because it balances fit and complexity and is derived from posterior distributions estimated via Markov Chain Monte Carlo (MCMC).

Logistic regression of modern contraceptive use among the studied women ever-in-union

Based on Model 6, respondents aged 30–39 years have higher odds (OR = 1.49, p-value = 0.036) of using modern contraceptives than those in the age category of 15–19 years. Additionally, those whose partners are between the ages of 25–29 have lower odds (OR = 0.090, p-value = 0.001) of using it compared to the reference category (Table 3). Respondents who are traditionalists are less likely (OR = 0.186, p-value = 0.021) to use modern contraceptives compared to Christians. Based on ethnicity, Yorubas had higher odds (OR = 1.433, p-value = 0.001) of using modern methods compared to Hausas, and respondents with more than three children alive had higher odds (OR = 1.720, p-value = 0.001) of using modern methods of contraception compared to those who had approximately three children or fewer. Those whose family size intention is equal to or above ten have lower odds (OR = 0.502, p-value = 0.001) of using modern methods compared to those whose intention is for a family size of at most four. Respondents in the Southeast have lower odds (OR = 0.461, p-value = 0.001) of using modern methods of contraception compared to those in the North Central region of Nigeria. Urban residents seem to be more likely to use modern contraception, as they have higher odds of (OR = 1.188, p-value = 0.001). Respondents’ level of education is seen to have an impact on the use of these modern methods, as those with the highest level of education have higher odds (OR = 2.237, p-value = 0.001) than those without education, and the same is true for their partner with higher odds (OR = 1.632, p-value = 0.001) at the highest level of education. Ever-in-union women who are in the richest level of wealth index have higher odds (OR = 1.988, p-value = 0.001) of using these methods compared to those who fall in the poorest category. It is important to note also that those who know about family planning through a method or more have higher odds (OR = 1.280, p-value = 0.001) of using these modern methods compared to those who do not know at all.

Table 3.

Adjusted odds ratios (aOR) and 95% confidence intervals from frequentist logistic regression for predictors of modern contraceptive use

Variables	aOR	95% CI	p-value
Age (years)
15–19 (Ref)
20–24	1.429	1.003–2.037	0.048
25–29	1.467	1.02–2.109	0.039
30–39	1.49	1.027–2.162	0.036
>= 40	1.099	0.749–1.612	0.629
Age at first marriage
< 15(Ref)
15–19	1.108	0.963–1.276	0.152
20–24	1.124	0.945–1.338	0.186
25–29	0.856	0.686–1.069	0.17
>= 30	0.637	0.469–0.863	0.004
Age at first birth (years)
< 15(Ref)
15–19	1.042	0.855–1.27	0.682
20–24	1.115	0.897–1.386	0.325
25–29	1.011	0.783–1.305	0.932
>= 30	1.069	0.763–1.498	0.698
Partner’s age (years)
15–19 (Ref)
20–24	0.089	0.026–0.31	0.001
25–29	0.09	0.027–0.3.027.3	0.001
>= 30	0.085	0.025–0.284	0.001
Religion
Christians (Ref)
Islam	0.557	0.495–0.628	0.001
Traditionalist	0.186	0.045–0.776	0.021
Others	0.343	0.157–0.749	0.007
Ethnicity
Hausa (Ref)
Igbo	1.014	0.787–1.306	0.916
Yoruba	1.433	1.156–1.776	0.001
Other	1.217	1.034–1.433	0.018
Number of children alive
<=3 (Ref)
> 3	1.72	1.426–2.075	0.001
Number of children ever had
<=3 (Ref)
> 3	1.358	1.117–1.651	0.071
Family size intention
< 4 (Ref)
5–9	0.809	0.736–0.889	0.001
>= 10	0.502	0.425–0.593	0.001
Geopolitical Zone
North Central(Ref)
North East	1.22	1.052–1.416	0.009
North West	0.993	0.836–1.178	0.932
South East	0.461	0.367–0.578	0.001
South South	0.515	0.446–0.596	0.001
South West	0.897	0.76–1.058	0.196
Residential type
Rural (Ref)
Urban	1.188	1.081–1.306	0.001
Wealth Index
Poorest (Ref)
Poorer	1.176	0.993–1.393	0.061
Middle	1.428	1.205–1.691	0.001
Richer	1.753	1.466–2.097	0.001
Richest	1.988	1.637–2.416	0.001
Level of Education
No education (Ref)
Primary	1.737	1.503–2.008	0.001
Secondary	1.954	1.685–2.267	0.001
Higher Institution	2.237	1.846–2.711	0.001
Partner’s Level of Education
No education (Ref)
Primary	1.475	1.249–1.743	0.001
Secondary	1.638	1.403–1.912	0.001
Higher Institution	1.632	1.37–1.944	0.001
Work Status
No (Ref)
Yes	1.066	0.96–1.184	0.23
Access to Media
No (Ref)
Yes	1.113	0.991–1.25	0.071
Knowledge of Family planning
No (Ref)
Yes	1.28	1.172–1.398	0.001

Generalized additive regression model with Spatial effect

This section presents the results of the generalized additive regression framework with spatial effects, examining factors associated with modern contraceptive use in Nigeria. Estimation was carried out using a fully Bayesian approach via Markov Chain Monte Carlo (MCMC) simulation, with a burn-in of 2,000 iterations and 12,000 posterior samples retained for inference. Convergence diagnostics indicated stable estimation, supporting the reliability of posterior draws. The Bayesian framework is advantageous as it allows direct evaluation of the full posterior distribution without relying on large-sample Gaussian approximations. The main findings are summarized in Table 4.

Table 4.

Posterior odds ratios (POR) and 95% credible intervals from the bayesian structured additive regression model for predictors of modern contraceptive use

Variables	Posterior Mean	POR	95% CrI
Intercept	−6.562	0.001	0–0.008.008
Age at first birth	−0.533	0.587	0.535–0.882
Religion
Christians(Ref)		1.000
Islam	−7.299	0.001	0–0.004.004
Traditionalist	−8.793	0.000	0–0.001.001
Others	−2.318	0.374	0.238–0.512
Ethnicity
Hausa (Ref)		1.000
Igbo	0.322	1.488	1.303–1.522
Yoruba	0.717	1.724	1.648–2.391
Other	0.710	1.492	1.396–2.185
Number of Children Alive
<= 3 (Ref)		1.000
> 3	0.393	1.481	1.239–1.662
Number of Children ever had
<= 3 (Ref)		1.000
> 3	0.390	1.677	1.561–1.791
Family size intention
<= 4 (Ref)		1.000
5–9	−1.550	0.736	0.173–0.817
>= 10	−0.229	0.257	0.131–0.379
Residential type
Rural (Ref)		1.000
Urban	0.015	1.015	1.012–1.366
Wealth Index
Poorest (Ref)		1.000
Poorer	0.914	1.401	1.357–2.271
Middle	0.415	1.661	1.587–2.132
Richer	0.253	1.776	1.696–2.301
Richest	0.226	1.798	1.675–2.173
Level of Education
No education (Ref)		1.000
Primary	0.350	1.259	0.169–1.318
Secondary	0.552	1.576	1.471–2.262
Higher Institution	0.798	1.743	1.659–2.25
Partner’s Level of Education
No education (Ref)		1.000
Primary	1.290	1.275	1.157–1.704
Secondary	0.641	1.527	1.481–2.003
Higher Institution	0.609	1.544	1.47–1.947
Work Status
No (Ref)		1.000
Yes	0.904	1.910	1.886–2.234
Access to Media		1.000
No (Ref)
Yes	0.629	1.196	1.154–2.321
Knowledge of Family Planning		1.000
No (Ref)
Yes	1.110	1.788	1.661–2.251

Structured additive regression model of modern contraceptive use among the studied ever-in-union women

The table suggests that ever-in-union women who were from urban places of residence have 4.2% higher posterior odds of using modern contraceptives compared to those from rural places of residence. Additionally, ever-in-union women whose highest educational level was primary, secondary, and higher institution had 5.3%, 9.5%, and 31.8%, respectively, higher odds of using modern contraceptives compared to those who had no education. Islamists and Traditionalists had 79.59% and 39.64% lower odds of using modern contraceptives, respectively, compared to those women surveyed who were Christians. Igbo women and other ethnic groups had 39.42% and 93.76% lower odds of using modern contraception relative to Hausa women, respectively, but Yoruba women had 10.18% higher odds of using modern contraceptives relative to Hausa women.

Furthermore, ever-in-union women who were currently working had 23.4% higher odds of using modern contraceptives compared to those who were not currently working. Ever-in-union women who had access to media had 32.1% higher odds of using modern contraceptives compared to those who did not have access to media. Those who have knowledge of family planning have 27.9% higher odds of using modern contraceptives than those who do not have knowledge of family planning (Table 4).

Spatial effect of modern contraceptive usage by the states of the studied women ever-in-union

Spatial effects

The spatial effect maps illustrate how geographic location independently influences modern contraceptive use after adjusting for socio-demographic and contextual factors. In several Northern states, particularly those in the North Central and parts of the Northeast, positive spatial effects were observed. This indicates that, despite socio-economic and infrastructural disadvantages, contraceptive uptake in these areas is higher than the model predicted based on individual and household characteristics alone. Conversely, many Southern states, including Lagos and parts of the Southwest, exhibited negative spatial effects, suggesting that contraceptive use in these regions is lower than expected given their generally favorable socio-demographic profiles.

These insights further highlight the strength of the Bayesian structured additive regression model in detecting unobserved geographic patterns that the frequentist logistic regression could not account for [23]. While the frequentist model identified important individual-level predictors, it assumes independence across observations and overlooks spatial dependencies. By incorporating spatially structured random effects, the Bayesian approach not only provided a better model fit but also revealed region-specific influences critical for understanding and addressing disparities in contraceptive use. This demonstrates the added value of Bayesian methods for designing targeted, location-specific interventions (Fig. 2).

Fig. 2.

Posterior spatial effects of modern contraceptive use across Nigerian states from the Bayesian structured additive regression model

Nonlinear effects of modern contraceptive use in Nigeria

The smoothed charts show the posterior means of the nonlinear effects of respondent’s age, age at first cohabitation or marriage, and partner’s age. Lower levels of modern contraceptive use are associated with respondents aged 15–24 years (Fig. 3a), followed by higher use among those aged 25–40 years, and a lower use from age 41 and above. Modern contraceptive use is also negatively associated with age at first cohabitation/marriage (Fig. 3b), while a slight positive association is observed with partner’s age (Fig. 3c).

Fig. 3.

Nonlinear effects of (a) respondent’s age, (b) partner’s age, and (c) age at first cohabitation/ marriage on modern contraceptive use

These nonlinear relationships highlight the strength of the Bayesian structured additive regression model, which captures complex patterns in the data without assuming a strictly linear association between predictors and the outcome [31]. Unlike frequentist logistic regression, which would require explicit transformation or categorization of continuous variables to account for such effects, the Bayesian approach naturally models these relationships using smooth functions. This provides deeper insights into how key predictors are associated with contraceptive use across their entire range, supporting more precise policy and programmatic interventions.

Discussion

Recognizing the factors that contribute to the use of modern methods of contraception serves a significant role in designing successful strategies that can lead to increased use and subsequently to a reduction in the rate of population growth [7, 8]. Additionally, it will assist policymakers in making important decisions on resource distribution. This study found a substantial difference in the usage of modern contraceptive techniques, showing that most ever-in-union women are nonusers of these modern techniques compared to those in Nigeria who use them. It was discovered that those who are currently within 30–39 years old use modern contraceptives more, and this can be based on women trying to stop birthing at this age after they might have given birth to their desired number of children [8, 25, 26].

Further analysis was conducted to determine what the major concerns were, and it was found that those who got married between the ages of 20–24 years used modern methods most compared to other age categories [8, 25]. Those above this age range used it less, and it could be as a result of wanting children during those years of reproduction. This study showed that those who got married between the ages of 15–19 used modern contraceptives less than those within 20–24 years, which could be due to factors such as lack of maturity, low level of education, or none [1].

The use of these modern methods was found to be more common among women with more children either currently alive or ever had and among those whose family size intention was less than or equal to four. This suggests that the more children a married woman has, the more likely she is to seek ways to prevent further pregnancies. Conversely, women who intend to have larger families tend to use modern contraceptive methods less frequently. Similar patterns have been observed in previous studies [7, 8, 25, 29].

Ethnicity and religion were also major factors contributing to the use of modern contraceptive methods. It was observed that Hausa’s were the least users of these modern methods, which could be linked to their belief systems about polygamy, especially among Muslims and traditionalists [7, 8]. The respondent’s level of education and that of their partner were seen to have a major effect on the use of modern contraceptives; the more educated these women were, the more they knew the importance and benefits of these methods, which led to their increased usage. This study also showed that those in urban areas used it more, and those who were down the ladder of the wealth index used it less compared to those up the ladder and currently working. These findings have also been reported in earlier studies [27, 31].

These regional differences align with the spatial distribution patterns previously discussed and further emphasize the need to understand individual- and household-level predictors that vary across regions [1, 7, 8, 25]. Further analysis revealed that knowledge of modern contraceptives and exposure to media significantly influenced their usage. However, women in the northern regions had the lowest levels of exposure to both, which may partly explain the lower rates of contraceptive use observed in these areas. The results showed that it was not only a factor of these two things but also a number of others, such as their level of education, wealth index, age of the women at their first birth and first marriage, their partner’s age and level of education, religion, ethnicity, family size intention, number of children ever born and presently alive, work status and residential type. All these factors contributed to the reason for their lack of use or low use of modern contraceptives, which has remained so from past studies [1, 7, 8, 28].

Modern contraceptives have been studied over the years on different grounds of implementation [22, 34], but it is interesting to note that many of these studies used the frequentist approach in solving the challenges faced. Unfortunately, after comparing the frequentist approach with the Bayesian approach in this study, the Bayesian approach was found to provide a better fitting model for factors associated with modern contraceptive use than the frequentist approach. This method allowed us to specifically discover the major states where MC use was higher or lower than expected, after adjusting for socio-demographic and contextual factors.

Conclusion and recommendation

This study showed that, among other social-economic, and sociodemographic, knowledge indicators, the location of a woman has a great impact on the use of modern contraceptives. With our results, we can see that even with several interventions that had gone forward in the past years regarding this subject, there has still not been a decline in the rate of growth in the country. This can be subjected to the wrong channeling of these interventions. Additionally, the STAR approach yielded a better, more reliable result compared to the binary logistic approach. As deviance is a measure of overall fit of a model, the model with the least deviance shows the best fit, we found that the Bayesian approach provided a better model fit in this study, and this can be attributed to effective number of parameters () that was added to the posterior mean deviance, giving a measure of fit that is penalized for complexity [7, 30].

With this result, the following recommendations are given, researchers may consider working more with the Bayesian approach, particularly when spatial or nonlinear effects are of interest, as it can offer additional insights alongside conventional methods. Also, the government or nongovernmental organizations are advised to channel every form of modern contraceptive initiative to the locations that lack it the most, most especially the northern region of Nigeria. There should not only be a free supply of these modern contraceptives but there should be a proper education system put in place for the women in those regions because research shows that the more educated these women get, the more they are open-minded to use these modern methods for their health preservation, that of their children and to help the population at large [8, 28]. Policymakers and program implementers may use spatial data to guide the distribution of family planning resources and interventions. Specific actions include increasing funding and outreach in low contraceptive use areas, developing regionally adapted education campaigns, expanding mobile and community-based services, and institutionalizing spatial data use in national and state family planning strategies. By focusing these efforts on regions with the lowest contraceptive use, particularly in the northern states, Nigeria can make more efficient progress toward its FP2030 goals and ensure that all women and families have equitable access to family planning services [1].

Strength and limitation

This study demonstrates the strength of Bayesian structured additive regression in modeling nonlinear and spatial effects, providing a better fit than frequentist methods for understanding factors influencing modern contraceptive use. By uncovering geographic patterns and contextual influences, it offers policy-relevant insights that can guide targeted interventions, particularly in regions such as northern Nigeria where uptake remains low.

However, as the analysis is based on a single cross-sectional dataset, it limits generalizability beyond Nigeria. Additionally, the focus on spatial and socio-demographic factors means the study does not evaluate the effectiveness of previous family planning interventions, limiting conclusions about whether low adoption is due to implementation gaps or persistent socio-cultural barriers.

Furthermore, residual confounding cannot be ruled out, as unmeasured supply-side factors (such as contraceptive availability and health system capacity) and cultural norms may still influence uptake. Finally, while penalized splines help to control complexity, the inclusion of multiple smooth terms carries some risk of overfitting, which could affect the precision of individual estimates.

Abbreviations

MC

Modern contraceptive

CI

Confidence interval

CrI

Credible interval

OR

Odds Ratio

POR

Posterior odds ratio

NDHS

Nigeria demographic and health survey

EIUW

Ever-in-union women

STAR

Structured additive regression

pD

Effective number of parameters

DIC

The deviance information criterion

Appendix 1

Prior assumptions

Due to the ease of analysis and the probability of measuring a posterior odds ratio (which is a calculation of the degree of correlation in logistic regression) directly from the performance of Markov chain Monte Carlo (MCMC), the Bayesian structured additive regression of the logit model is chosen. The fitting of Bayesian models using MCMC includes treating all parameters as spread randomly according to some prior distribution. The posterior distribution is intractable, and MCMC algorithms are used to produce samples from this prior distribution so that the parameters can be measured and inferred [32]. Following the settings reported in the section on Generalized Additive Regression with Spatial Effect, a burn-in of 2,000 iterations was used and 12,000 posterior draws were retained for inference. Independent chains were run, and convergence was checked using R-hat statistics, effective sample size (ESS), and trace plots. We ran 4 independent chains using the BayesX R interface.

The term P-spline (penalized B-spline) refers to the use of the representation B-spline (basis spline), where the coefficients are calculated partly by the data to be equipped and partly by an additional penalty feature that aims to enforce smoothness to prevent overfitting. By combining the different penalties with their stochastic analogs, a Bayesian P-spline is obtained, e.g., Gaussian (intrinsic) random walk priors acting as smooth priors for the undefined coefficients in regression. Further clarification on this can be obtained from Lang and Brezger (2004) and Brezger and Lang (2006) [33].

All the regression coefficients and smooth functions are considered random variables in Bayesian analysis and are given prior distributions. Specifically, for each linear coefficient we used a Normal prior consistent with weakly informative assumptions in structured additive regression [34]. The intercept was given a flat prior, and smooth/spatial components were centered for identifiability.

The unknown smooth functions are assigned Bayesian penalized spline priors [33]. The functions are assumed to be approximated by a polynomial of degree , defined over a set of equally spaced knots of the form:

In this study, we used cubic B-splines knots, placed at equally spaced quantiles of the covariate range.

The spline is expressed as a linear combination of B-spline basis functions. The spline coefficients follow a second-order random walk prior of the form:

The density of the Inverse-Gamma prior is:

This means that the spline coefficients follow a second-order Gaussian random walk, where increments are normally distributed with variance ​. The variance parameter itself is given a weakly informative Inverse-Gamma prior.

The spatial random terms are also fitted as splines, particularly as a two-dimensional tensor product. The unknown are assigned priors of the general form:

where is the penalty matrix and is the variance parameter that controls the tradeoff between flexibility and smoothness. For the spatial effect, we follow the Besag–York–Mollié (BYM) specification [35], which decomposes the spatial effect into a structured and an unstructured component:

The structured component follows an intrinsic conditional autoregressive (ICAR) prior,

Here, is the adjacency matrix with if region and share a border and otherwise, while is the diagonal degree matrix with equal to the number of neighbors of region . The precision matrix is defined as with adjacency defined by shared borders of geo-political zones. The unstructured component is i.i.d. Both variance parameters were given weakly informative priors,

Model fit was assessed using the Deviance Information Criterion (DIC):The spatial additive Bayesian model achieved a lower DIC compared to non-spatial and purely linear specifications. Convergence diagnostics supported reliability, with all monitored parameters showing  and effective sample size greater than 1000.

Authors’ contributions

OAA & AFF conceived the study, contributed to the study design, and conducted the statistical analysis and interpretation of results and the drafting of the manuscript. Both authors reviewed the statistical analysis and revised the manuscript draft. All the authors have read and approved the final manuscript.

Funding

The authors received no funding for this study.

Data availability

The datasets analyzed during the current study are publicly available at https://dhsprogram.com.

Declarations

Ethics approval and consent to participate

The original owner of the data obtained ethical approval after the survey protocol was reviewed and approved by the National Health Research Ethics Committee of Nigeria (NHREC) and the ICF Institutional Review Board. No further approval is needed for this study. All participants and their legal guardians/next of kin provided written informed consent to participate in this study. All methods were performed in accordance with the relevant guidelines and regulations of the ethical approving authority.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.