We sourced data from the DHS of nineteen countries in SSA, spanning from 2015 to 2021. The data used were extracted from the DHS Program, which is available upon request [20]. We have provided the list of the countries and their survey years in Table 1. Since the inception of DHS, there have been more than 400 surveys conducted in over 90 low-and middle-income countries [21]. A cross-sectional design was used for the DHS. The respondents were sampled using a multistage sampling technique with the detailed sampling methodology highlighted in the literature [21, 22]. Our study was restricted to a weighted sample of 82,203 women in their reproductive age who were married or cohabiting. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines in writing this paper [23].

There were four outcome variables in this study. The first three were past-year experiences of physical, emotional, and sexual violence from a partner or husband. Physical, emotional, and sexual violence were derived from the modified Conflict Tactics Scale [24, 25], a list of questions used to measure the extent to which individuals in sexual relationship experience physical, emotional, and sexual violence. In the DHS, women in sexual unions: married or cohabiting were asked to indicate whether they have experienced any physical, emotional, and sexual violence in the last 12 months preceding the survey. The fourth outcome variable was created from a composite of physical, emotional, and sexual violence. This was referred to as IPV. Specific questions used to measure physical, emotional, and sexual violence are available in the literature that used the DHS dataset [26–29]. We also used the existing coding of physical, emotional, and sexual violence guided by previous studies [26–29].

We used the newly developed and validated SWPER as the key explanatory variable. It was statistically created for use in low-and middle-income countries [16]. Since its emergence, SWPER has been used to address several health and social issues, including reproductive health, maternal and child health, and other related topics [16–18]. SWPER was developed using fourteen variables. The variables consisted of (i) beating not justified if wife goes out without telling husband, (ii) beating not justified if wife neglects the children, (iii) beating not justified if wife argues with husband, (iv) beating not justified if wife refuses to have sex with husband, (v) beating not justified if wife burns the food, (vi) frequency of reading newspaper or magazine, (vii) woman education, (viii) age of respondent at cohabitation, (ix) age of respondent at first birth, (x) age difference: woman’s age minus husband’s age, (xi) education difference: woman’s minus husband’s years of schooling, (xii) who usually decides on respondent's health care, (xiii) who usually decides on large household purchases, and (xiv) who usually decides on visits to family or relatives [16]. These fourteen variables were used to create the dimensions of SWPER [16]. The three dimensions are attitude to violence, social autonomy, and decision-making. Social independence or autonomy denotes the preconditions such as the schooling attainment, information access, age at crucial life events, and spousal asset differentials that allow women to realize their goals. Decision-making on the other hand refers to the degree of the woman's involvement in household decisions, which can also be viewed as a gauge of instrumental agency. Finally, attitude to violence closely related to the concept of intrinsic agency and it is a proxy for the woman’s incorporation of gender norms-related to the acceptability of IPV [16]. We used the same coding methodology as used in the previous study conducted by Ewerling et al. [16]. For attitude to violence, the coding for each category consisted of low (score ≤ − 0.700), medium (score > − 0.700 ≤ 0.400), and high (score > 0.400). The high category denotes strong disagreement or rejection of attitude to violence (positive), whereas the low group emphasizes strong acceptance of violence (negative). Low (score ≤ − 0.559), medium (score > − 0.559 ≤ 0.293), and high (score > 0.293) were the coding and classification of the social independence dimension. Whereas, those of decision-making were low (score ≤ − 1.000), medium (score > − 1.000 ≤ 0.600), and high (score > 0.600) [16].

We included six covariates in our study. These covariates either increase or decrease women’s likelihood of experiencing IPV based on literature [26–29]. Also, the covariates were present in the DHS dataset across all the countries included in the study. The covariates were grouped into individual and contextual level variables. The individual level variables consisted of partner alcohol consumption, exposure to interparental violence, and exposure to partner controlling behavior. Likewise, household wealth index, place of residence, and geographical sub-regions were the contextual level variables.

We used Stata version 17.0 (Stata Corporation, College Station, TX, USA) to perform all the analyses. We carried out data cleaning and weighting at the country level per the DHS guidelines before appending the dataset for final analysis. To do this, the weighting variable for domestic violence module (d005) was divided by 1,000,000 to generate a new variable called “= d005_pw”. Next, we de-normalized the country level weights using the command: gen d005_pwpool = d005_pw*(total population of women; age 15–49 at the time of the survey/number of women in the resulting domestic violence subsample. Later, we appended the weighted country-level dataset for the 19 countries and used for the final analysis. We used ‘spmap’ in Stata to generate the proportion of women who experienced physical, emotional, sexual violence, and IPV in the past prior to the survey. We examined the distribution of the outcome variables across the dimensions of SWPER, and the covariates, as well as their associations using Pearson's Chi-square test. This was followed by a five-modelled multilevel binary logistic regression modelling. Prior to the regression analysis, we checked for evidence of multicollinearity among the variables using the variance inflation factor (VIF). The results showed that the minimum, maximum, and mean VIFs were 1.04, 3.48, and 2.00, respectively. Hence, there was no evidence of high collinearity among the variables. The first model had no explanatory variables or covariates, showing the variance in the outcome variables attributed to the primary sampling units (PSU). Model I was fitted to contain the three dimensions of SWPER. Model II contained the variables in Model I and the individual-level covariates. The variables in Model II and the contextual level covariates were placed in Model III. The final model (Model IV) contained the dimensions of SWPER and all the covariates. The results were presented using adjusted odds ratio (aOR) with their respective 95% confidence interval (CI). Statistical significance was set at p < 0.05.

Ethical clearance was not sought for this study since we analyzed a secondary dataset, which is already available freely to use. We obtained permission to use the dataset from the DHS program data repository before using the dataset for publication.