Although intimate partner violence (IPV) is inherently a relational event shaped by couple-level factors, most empirical examinations of IPV-related attitudes have used individuals as the unit of analysis. We apply a dyadic perspective to the study of attitudes about the acceptability of IPV, harnessing couple-level data from 33 countries in sub-Saharan Africa, a region characterized by particularly high levels of both the incidence and acceptance of IPV. We document considerable geographic heterogeneity in the distribution of attitudinal concordance or discordance regarding the acceptability of IPV within couples, a descriptive finding that is overlooked by studies focused on individuals as the unit of analysis. In addition, applying a dyadic perspective to the correlates of attitudinal concordance, we demonstrate that joint exposure to schooling, work, and media is more predictive of joint rejection of IPV than are singular exposures of wives or husbands. Finally, we show that distinct combinations of attitudes within couples are associated with differential likelihoods of wives reporting having recently experienced emotional, physical, or sexual IPV. In particular, when both partners reject IPV, wives are significantly less likely to report experiencing each type of IPV in the past year compared with any other combination of attitudes. Our results reveal that a dyadic perspective provides a comprehensive understanding of intracouple contexts that enhances our perspective on these important demographic outcomes.
Attitudes regarding the acceptability of intimate partner violence (IPV) vary markedly across time and space (Cools and Kotsadam 2017; Pierotti 2013; Tran et al. 2016; Uthman et al. 2009; Waltermaurer 2012). Over time, rejection of IPV—as measured by the proportion of people who say that beating one's wife is not justified under various hypothetical circumstances—has increased substantially in most countries where scholars have examined these attitudes longitudinally (Pierotti 2013; see also Cools and Kotsadam 2017; Simister and Mehta 2010). Comparing across space, countries and regions differ widely in the degree to which this type of violence is viewed as acceptable, with particularly high levels of acceptance in sub-Saharan Africa and South Asia (Sardinha and Catalán 2018; Tausch 2019; Tran et al. 2016; Uthman et al. 2009; Uthman et al. 2010; Waltermaurer 2012). This variation in attitudes is important because IPV is more likely to occur in households in which individuals believe that such violence is acceptable (Flood and Pease 2009; Speizer 2010; United Nations 2015) and because women living in areas with greater acceptance of IPV are more likely to be subjected to backlash if they depart from established gender norms (Cools and Kotsadam 2017).
To date, most empirical examinations of the correlates of attitudes toward IPV have used individuals as the unit of analysis (for an exception, see Alio et al. 2011) and have often focused on women (Cools and Kotsadam 2017; Pierotti 2013; Swindle 2018; Uthman et al. 2010; Waltermaurer 2012). Yet IPV is inherently a relational event that is shaped by couple-level dynamics and individual-level characteristics (Conroy 2014; Ghuman et al. 2006; Rodriguez and Straus 2017). Husbands and wives enter marriage with their own lived experiences and perspectives, which interact and are reshaped over time. Couples also develop a shared repertoire of experiences and information as their lives unfold jointly. Attitudes toward IPV and women's risk of experiencing IPV are likely influenced by a combination of these individual and joint experiences. For example, when both the husband and the wife are exposed to norms defining IPV as a violation of human rights, these exposures may reinforce each other and lead both members of the couple to form stronger convictions that such violence is unjustifiable (Alio et al. 2011). At the same time, individuals often enter relationships with partners similar to themselves on attributes such as education, occupation, and age, and this tendency may extend to attitudes toward IPV. In this case, a dyadic perspective may better capture the dynamics around assortative mating and the social processes through which people select into marriage with those who have similar background characteristics.
We contribute to this literature on attitudes toward IPV and intracouple dynamics by applying a dyadic perspective, focusing on couples rather than individuals. We harness couple-level data from 33 countries in sub-Saharan Africa (SSA)—a region characterized by particularly high levels of both incidence and attitudinal acceptance of IPV—to examine geographic variation in couple-level concordance (vs. discordance) in acceptance and rejection of IPV. First, we ask, what is the geographic distribution of couple-level attitudinal concordance or discordance regarding the acceptability of IPV across 33 countries in SSA? Second, we ask, what couple-level (dyadic) characteristics predict attitudinal concordance or discordance in accepting or rejecting IPV? Third, we ask, what is the association between couple-level attitudinal concordance or discordance in the acceptability of IPV and women's reports of IPV when couple-level background characteristics are controlled for? Our findings reveal that a dyadic perspective provides a comprehensive account of intracouple contexts that enhances understandings of these important demographic outcomes.
Temporal, Cross-National, and Gendered Variation in Attitudes Toward IPV
IPV is a global concern, with well-documented detrimental effects on women's and families' health and well-being (Alio et al. 2009; Durevall and Lindskog 2015; Garcia-Moreno and Watts 2011; True 2012; Watts and Mayhew 2004; World Health Organization 2013). In SSA, reports of IPV are among the highest in the world, with about one in three women reporting experience of some form of IPV in their lifetime, and about one-half of women and one-third of men reporting that IPV is acceptable under some circumstances (Cools and Kotsadam 2017). Over the past three decades, a global consensus has emerged among transnational institutions and many national-level governments that IPV is antithetical to the moral underpinnings of modern society (Merry 2009, 2011; Pierotti 2013; True 2012). Beginning with the 1993 United Nations Declaration on the Elimination of Violence Against Women, transnational organizations, local legislatures, and media campaigns have promoted the belief that violence against women is unacceptable, and 144 countries have outlawed IPV since 1993 (Abramowitz and Moran 2012; Merry 2009; Swindle 2018; True 2012).
Evidence points to considerable attitudinal change toward IPV at the individual level, particularly in low- and middle-income countries. Pierotti (2013) documented an increase in rejection of IPV among women in 23 of 26 low- and middle-income countries in Africa, Asia, the Middle East, and Latin America. In some places, this increase was dramatic; in Nigeria, for example, the percentage of women rejecting IPV rose from 33% in 2003 to 52% just five years later. Cools and Kotsadam (2017) extended this research by showing that women's acceptance of IPV declined by about 2 percentage points per year, on average, between 2003 and 2013 in 21 African countries. The prevalence of women's self-reported experience of IPV also declined significantly in SSA during this period; however, the magnitude of this decline was smaller than that of IPV acceptance, and the prevalence of self-reported IPV remains high (Cools and Kotsadam 2017).
Despite evidence of global declines in acceptance of IPV, there remains striking geographic heterogeneity in the prevalence of IPV rejection across the African countries where our study is based. For example, in Burundi in 2010, more than 70% of women expressed that wife beating was justified in at least one scenario; in Malawi that same year, only 13% of women expressed this opinion (Cools and Kotsadam 2017). Rates of rejection also differ substantially by gender in Africa; using surveys conducted between 2003 and 2007 in 17 African countries, Uthman et al. (2010) found that women were, on average, twice as likely as men to justify wife beating. The authors attributed these gender differences to gender gaps in literacy, among other factors (see also Sardinha and Catalán 2018). These gender differences also differ markedly across countries: women in Namibia were only 1.06 times as likely as men to justify wife beating, whereas women in Madagascar were 3.75 times as likely to do so.
The Need for a Dyadic Approach
Despite pronounced gender differences in IPV rejection, many empirical examinations of attitudes toward IPV have focused primarily on women (Abrahams et al. 2006). Research that has considered both men and women suggests that the correlates of IPV rejection may differ between men and women. For example, Cannonier and Mocan (2018) showed that a policy change in Sierra Leone that increased access to primary schooling resulted in an increase in women's—but not men's—rejection of IPV. Likewise, Swindle (2018) found that in Malawi, television use is associated with higher acceptance of IPV among men, whereas television use is associated with higher rejection of IPV among women. Very few studies have taken a dyadic perspective, examining the attitudes of male and female partners in conjunction (Conroy 2014; Rodriguez and Straus 2017).
Despite the absence of research applying a dyadic perspective to IPV attitudes, several recent studies have taken such an approach to attitudes about women's autonomy and household decision-making, finding that disagreement between couples is widespread and often the modal scenario (Conroy 2014; Ghuman et al. 2006; Jejeebhoy 2002; Krishnan et al. 2012; Seymour and Peterman 2017; Story and Burgard 2012; Uddin et al. 2017). This body of research has demonstrated that couple-level concordance in gender-based attitudes is positively associated with contraceptive use among couples who desire to limit their fertility (Uddin et al. 2017) and with preventive health utilization, including antenatal care (Pratley 2016; Story and Burgard 2012).
Scant research has examined couple-level concordance or discordance about gendered power dynamics and women's autonomy in the context of SSA (for exceptions, see Alio et al. 2011; Conroy 2014; Krishnan et al. 2012), and we know of only one study that examined couple-level concordance in IPV-related attitudes on a cross-national level (Alio et al. 2011). Using data collected between 2003 and 2007 from six countries in SSA, Alio et al. (2011) found that joint acceptance of IPV is associated with a higher likelihood of a woman reporting having experienced IPV. We build on this research by (1) extending the geographic reach from 6 to 33 African countries; (2) using data collected between 2008 and 2018, after the substantial increase in rejection of IPV occurred in SSA during the first decade of this century (Cools and Kotsadam 2017; Pierotti 2013); and (3) examining couple-level correlates of attitudinal concordance.
Theoretical Perspectives on What Predicts Attitudes Toward IPV
One set of explanations for individual-level variation in attitudes about IPV focuses on the role of global cultural scripts in the diffusion of new norms, such as the rejection of IPV (Pierotti 2013). This perspective draws on a large body of sociological and demographic research that has demonstrated the influence of world culture—a set of globally circulating cultural scripts providing moral guidelines for how individuals should behave—on both national-level policies and individual-level attitudes and behavior (Baker and LeTendre 2005; Boli and Thomas 1997; Elliott 2007; Meyer 2010; Meyer and Jepperson 2000). According to this literature, global cultural scripts are often disseminated through face-to-face interactions (occurring in schools, local NGOs, markets, and other social arenas) and national and international media (i.e., television, radio, newspaper, Internet), which expose people to new ideas and norms that may either explicitly or implicitly challenge existing ideas about IPV (Baker and LeTendre 2005; Boyle and Hoeschen 2001; Boyle et al. 2002; Charles 2020; Essary 2007; Krause et al. 2017; McEneaney and Meyer 2006; Pierotti 2013; Swindle 2018). In turn, this body of research has demonstrated that people exposed to potential conduits of diffusion—such as frequent media exposure and school attendance—are more likely to express that IPV is unacceptable than those who have not been similarly exposed (Friedman et al. 2016; Jensen and Oster 2009; Pierotti 2013; Uthman et al. 2009).
A second set of explanations for individual-level variation in attitudes about IPV relates to women's bargaining power in intimate relationships. This perspective posits that women's education, wage labor opportunities, and other measures of status within the family may increase their bargaining power in relationships by providing them with improved options outside marriage (Lundberg and Pollak 1996; Manser and Brown 1980). This increased bargaining power could lead some women (i.e., those who have more education and who engage in wage labor) to feel more comfortable asserting themselves by condemning practices such as IPV (Friedman et al. 2016). Men who are educated and engage in wage labor might also have different attitudes about IPV acceptability that may be further amplified by their wives' employment and education experiences. Nonetheless, across diverse contexts, including SSA, evidence suggests that wives' gains in education and employment are associated with higher rates of IPV; in Africa, wives' employment is also associated with higher acceptance of IPV (Cools and Kotsadam 2017; Hornung et al. 1981; Weitzman 2014). These findings are often interpreted as a “backlash” effect, whereby gains in women's education and employment correspond with higher reports of IPV experiences and acceptance as a way to reestablish the gender status quo.
Both of these theoretical perspectives (i.e., cultural diffusion and bargaining power) have been used by scholars to explain the global change in attitudes toward IPV and to predict individual-level variation in attitudes, and we apply our dyadic, couple-level approach to examine these perspectives. Rather than trying to adjudicate between these perspectives, we consider them as complementary insights into co-occurring and mutually reinforcing social processes through which rejection of IPV may proceed. We pay particular attention to these two theoretical explanations, given their preeminence in providing overarching theoretical explanations for variation in attitudes toward IPV. Nonetheless, factors not explicitly considered in these theories could also be important predictors of attitudes to IPV, including previous exposure to gender-based violence, family background and socioeconomic status, and attitudes regarding the broader acceptability of violence (Hayes and van Baak 2017; Vyas and Heise 2016; Willis-Esqueda and Delgado 2020).
Couple-Level Correlates of IPV Attitudes
A central aim of our study is to explore whether similarities or differences in couple's characteristics are associated with couples' joint attitudes toward IPV. We envision two possible ways that correlates of attitudes toward IPV might operate in a dyadic, couple-level framework.
First, similarities in background characteristics between both husbands and wives could be important for predicting couple-level concordance about the acceptability of IPV. Joint exposure to anti-IPV messaging (through, for example, schools and media) might have a stronger effect on attitudes than if only the wife or only the husband were exposed because the couple may be more likely to jointly internalize anti-IPV messages and discuss them with each other. At the same time, given that those with similar characteristics (i.e., education, work, age, and so on) are more likely to select into marriage with each other (Lopus and Frye 2020), married couples may also hold similar views on IPV. However, couples' similarities in characteristics such as education and employment may correspond with the backlash that might be associated with increases in IPV acceptance among one or both members of the couple (Behrman 2019).
Second, the characteristics of one member of the couple might be important for predicting attitudinal concordance in rejecting IPV. For example, a husband's formal education may lead him to share with his wife that IPV is against the law (even if she lacks education herself), or a wife may discuss with her husband a television show with anti-IPV messaging that she watched (even if the husband does not watch television). However, in discordant cases, husbands and wives may not discuss this issue openly or may not convince one another, leading their divergent views to persist. This latter possibility could be particularly likely if discordant attitudes are indicative of broader tensions and ideational differences between partners, which might be related to differences in background characteristics and life exposures.
Another goal of our study is to investigate whether couple-level perspectives on IPV acceptability are correlated with women's reports of IPV experiences. Here we draw on Alio et al. (2011) and research showing that couple-level concordance around gender-egalitarian perspectives is associated with behavioral outcomes that are more protective of women's health and well-being than are other couple-level combinations of opinions (Story and Burgard 2012; Uddin et al. 2017). According to this perspective, we would expect that couples in which both the husband and wife jointly accept IPV are more likely to report experiencing IPV; conversely, couples in which both the husband and wife jointly reject IPV are expected to be less likely to report experiencing IPV.
When husbands' and wives' perspectives on IPV differ, the case may be more complex. Consistent with research on couple-level concordance/discordance on other topics related to gender norms (Story and Burgard 2012), discordant attitudes may indicate broader tensions and ideational differences between partners. As such, both types of discordance might predict a higher likelihood of the wife reporting having experienced IPV, compared with households where both partners concur that wife beating is justifiable under some circumstances. Alternatively, women's rejection of IPV indicates that they would be more likely to seek support outside the home or ask sympathetic family members to intervene compared with women who consider IPV to be an acceptable and normal part of the marriage (McCleary-Sills et al. 2016; Overstreet and Quinn 2013). In this case, we may find that discordance created by husbands accepting but wives rejecting IPV is associated with a lower likelihood of wives reporting having experienced IPV, compared both with concordant acceptance of IPV and with discordance created by husbands rejecting and wives accepting IPV. For the discordant case in which wives accept IPV but husbands reject it, we can imagine two possible scenarios pointing in opposite directions. On the one hand, husbands may be less likely to act violently against their wives based on their anti-IPV attitudes (Eckhardt et al. 2012). On the other hand, wives in these unions may be more likely to disclose in a survey interview that they have experienced IPV if they believe it to be acceptable.
Data and Sample
We use Demographic and Health Survey (DHS) data collected between 2008 and 2018 from 33 countries in SSA (for a list of countries, years, and sample sizes, see Table A1, online appendix). The DHS data are nationally representative of reproductive-aged women (i.e., ages 15–49) and include cross-sectional information on reproductive health, education, and gender and family norms. The DHS is typically collected by one or more host-country government agencies, with technical assistance from ICF and funding from USAID. The survey methodology and instruments are standardized across countries to facilitate cross-country comparisons. We use one DHS survey round for each SSA country that has conducted a DHS survey since 2008 that included both women and men.1
All reproductive-aged women present in all selected households are interviewed, along with all eligible men in a subsample of selected households.2 From among those interviewed, DHS data allows the identification of women and men who are married to each other or living together. For these couples, the DHS provides a “couples-recode” data set that links husbands' and wives' responses. Using this file, we perform a listwise deletion and exclude 4,772 couples because of missing information on at least one variable used in our analysis—about 5% of the sample of 101,100 respondents (for more detailed information on missing values, see Table A2 in the online appendix). The resulting sample is 96,328 currently partnered couples from 33 countries throughout SSA (Table 1).
The DHS also presents a subsample of female respondents with a detailed IPV module that collects self-reports of emotional, physical, and sexual IPV with the current partner in the last 12 months.3 The DHS and the country-specific survey teams make the utmost effort to ensure that this module is asked in private, given the sensitive nature of the questions. For about 45% of our couple sample, the DHS also has information on whether the wife reported IPV in the last 12 months (see Table A1 in the online appendix for more detailed information about this subsample). We use this IPV-experience subsample for the portion of our analysis that examines the association between concordance/discordance in IPV-related attitudes and reported IPV experiences. As a supplement, we replicate our other multivariate analyses conducted on the full sample with the IPV-experience subsample (see Table A3, online appendix).
Concordance and Discordance in Husband/Wife Rejection of IPV
To capture whether there is concordance or discordance in husbands/wives' responses to questions about the acceptability of IPV, we create a four-category categorical variable from a series of questions that husbands and wives are asked separately about whether wife beating is acceptable (or “justified”) under five scenarios: (1) the wife goes out without telling the husband; (2) the wife neglects the children; (3) the wife argues with the husband; (4) the wife refuses to have sex with the husband; and (5) the wife burns the food.
Based on these questions, we create indicator variables separately for husbands and wives, coded as 1 if a husband/wife rejects wife beating (i.e., reports that it is not justified across all five scenarios) and 0 if the husband/wife accepts wife beating (i.e., reports that it is justified for one or more scenario). Consistent with the literature, we refer to rejection of wife beating as “rejection of IPV” even though it refers to only one form of IPV (i.e., physical violence) across five specific scenarios (for a similar approach, see Cools and Kotsadam 2017; Kurzman et al. 2019; Pierotti 2013; Uthman et al. 2009).4 Based on these individual-level binary indicators of IPV rejection, we create a categorical variable that considers the husband-wife responses jointly (1) concordance in rejecting IPV across all scenarios; (2) discordance in which the husband, but not the wife, rejects IPV across all scenarios; (3) discordance in which the wife, but not the husband, rejects IPV across all scenarios queried; and (4) concordance in IPV acceptance (or failure to reject IPV) in at least one of the hypothetical scenarios.5
Wives' Recent Reports of IPV
We explore three IPV outcomes asked of the IPV-experience subsample (see Table A1, online appendix). In the DHS IPV module, women are asked whether they have ever experienced one of a series of violent acts; if they respond affirmatively, they are asked whether they have experienced that act in the last 12 months.
The first IPV outcome we explore is a dichotomous indicator of whether the woman reports any form of emotional IPV in the last 12 months, as indicated by an affirmative response to any of the following questions: “In this last 12 months, has your husband ever humiliated you?/threatened you with harm?/insulted or made you feel bad?”
The second IPV outcome we explore is a dichotomous indicator of whether the woman reports any form of physical IPV in the last 12 months, as indicated by an affirmative response to any of the following questions: “In the last 12 months, has your husband ever slapped you?/twisted your arm or pulled your hair?/pushed you, shook you, or thrown something at you?/punched you with a fist or something that could hurt you?/kicked you, dragged you, or beat you up?/tried to choke you or burn you on purpose?/threatened or attacked you with a knife, gun, or any other weapon?”
The third IPV outcome we explore is a dichotomous indicator of whether the woman reports any form of sexual IPV in the last 12 months, as indicated by an affirmative response to any of the following questions: “In the last 12 months, has your husband ever physically forced you to have sexual intercourse with him even when you did not want to?/forced you to perform other sexual acts you did not want to?”
Dyadic Explanatory Variables
We apply our dyadic approach to variables hypothesized to be key correlates of IPV attitudes according to cultural diffusion and bargaining theories. We thereby assess whether joint husband-wife exposure to a given explanatory variable (such as primary schooling)—as opposed to individual exposure—is important for couple-level attitudes. Because acceptance of IPV is correlated with reports of IPV (Cools and Kotsadam 2017; Uthman et al. 2011), we include the same set of controls for models predicting attitudes as for models predicting reports of IPV (for a similar approach, see Cools and Kotsadam 2017; Speizer 2010).
First, quasi-experimental research in Africa found that schooling is associated with women's increased rejection of IPV (Friedman et al. 2016), perhaps because schooling provides exposure to cultural scripts condemning IPV (Baker and LeTendre 2005; Pierotti 2013) and increases women's bargaining power in the household (Lundberg and Pollak 1996; Manser and Brown 1980). Given that overall levels of schooling are low in many parts of SSA (Table 1), we use primary school completion as our measure of schooling attainment. We include dichotomous indicators for (1) both the husband and wife completed primary school (reference category); (2) only the husband completed primary school; (3) only the wife completed primary school; and (4) neither the husband nor the wife completed primary school.
Second, quasi-experimental research suggests that exposure to television media positively impacts women's rejection of IPV (Jensen and Oster 2009), and media exposure has been hypothesized to be important for the promulgation of global cultural scripts condemning IPV (Krause et al. 2017; Merry 2011; Pierotti 2013). We control for media with variables for whether respondents report watching television and reading the newspaper frequently (“at least once a week”).6 For both television and newspaper, we include dichotomous indicators for (1) both the husband and wife report consuming this media source frequently (reference category); (2) only the husband reports consuming this media source frequently; (3) only the wife reports consuming this media source frequently; (4) neither the husband nor the wife reports consuming this media source frequently.
Third, employment is often thought to be protective against IPV acceptance and experience by increasing household resource availability and improving women's outside options if they want to leave abusive partnerships. Nonetheless, scholars have shown that women's employment is positively associated with both acceptance and reports of IPV (especially when women out-earn their husbands), which is often interpreted as evidence of a backlash against women who challenge the status quo (Cools and Kotsadam 2017; Hornung et al. 1981; Weitzman 2014). We include dichotomous indicators for whether (1) both the husband and wife report that they have worked in the last 12 months (reference category); (2) only the husband reports work; (3) only the wife reports work; (4) and neither husband nor wife reports work. Our definition of work includes only work done for cash and is not for family members or self-employment. We do this because research suggests only remunerative employment that takes women outside the direct patriarchal authority of male relatives increases women's autonomy and bargaining power (Anderson and Eswaran 2009).
Fourth, there may be differences in intrafamilial power dynamics depending on whether the marriage is a first union. These differences may be particularly relevant in SSA, where widow remarriage is common and often stigmatized (Newton-Levinson et al. 2014). Thus, all models include dichotomous indicators for whether (1) the current marriage is the first union for both the husband and wife (reference category); (2) the current marriage is the first union for the husband only; (3) the current marriage is the first union for the wife only; and (4) the current marriage is the first union for neither partner.7
Finally, studies suggest that age differences may be another important indicator of power that is correlated with IPV, particularly when husbands are much older than wives (Volpe et al. 2013). We include indicator variables for whether (1) the husband and wife are the same age (within 3 years of each other); (2) the wife is older (4 or more years older); (3) the husband is somewhat older (4 to 10 years older); and (4)the husband is much older (11 or more years older). We use this asymmetrical coding scheme because it is much more common for husbands to be considerably older than wives in the African contexts we study.
Additional Explanatory Variables
Factors other than the dyadic explanatory variables could also be associated with attitudes and behaviors related to IPV. Thus, all multivariate models also include controls that are common risk factors for IPV.8
Some scholars have suggested that wealth is protective against acceptance and perpetuation of IPV because men with few resources use violence to assert control or cope with stress (Goode 1971; Vyas and Watts 2009) and wealth is negatively correlated with both acceptance and reports of IPV in SSA (Cools and Kotsadam 2017). Consistent with research on IPV in Africa (Cools and Kotsadam 2017; Speizer 2009; Uthman 2009), we measure wealth using the DHS wealth index, which is a composite measure of household living standards created using principal components analysis on asset ownership and living conditions. We include dichotomous indicators of the couple's wealth quintile (poorest, poorer, middle, richer, and richest). The wealth quintile is assigned to the household based on its wealth position relative to others in the same country and year.
We include a dichotomous indicator of rural residence (compared with urban residence) because norms and resources may differ between urban and rural areas, with implications for IPV (Klomegah 2008).
Polygyny has been cited as a risk factor for IPV acceptance and reported incidence (Abramsky et al. 2011; Jewkes et al. 2002; Karamagi et al. 2006), although some research suggests that such a finding might be related to the types of people who select into polygyny (Behrman 2019). To address the possible effect of polygyny, all models include a dichotomous indicator for whether the union is polygynous (reference is monogamous union).
Union type is also important. Couples in Africa who report “living together as if married” are more likely than are those who report being “married” to be educated and to live in urban areas (Calvès 2016), and their children are more likely to experience positive outcomes along various dimensions of child well-being, including nutrition, access to healthcare, and schooling (Pierce and Heaton 2019). We include a dichotomous indicator for whether the wife describes her relationship status as “living together as if married” (reference category is married).9
To help address potential reporting bias in reports of IPV acceptability, we follow Hayes and van Baak (2017) by including an indicator variable for whether the husband was present during the module assessing the wife's acceptance of IPV. As Table 1 shows, the husband was present during this module in less than 3% of cases.10
Women's childbearing status is often included in studies examining IPV in Africa (Hayes and van Baak 2017; Mann and Takyi 2009; Speizer 2010) because parity may be an important determinant of women's status within the family and could also influence her ability to leave an abusive relationship. We control for this using a continuous measure of current parity.
To capture life cycle effects that lead women to face differential risks of IPV at different states in the life course (Alio et al. 2011; Benebo et al. 2018; Speizer 2010), we include continuous measures of the husband's age and age squared (we de-mean the age variables in the analysis).11
All models also include a continuous control for survey year to control for unobserved time trends, such as the increasing prominence of international campaigns against IPV. The models also include country fixed effects (indicator variables for each country) to control for unobserved country-level factors that may influence our results.
First, we ask, what is the geographic distribution of attitudinal concordance or discordance regarding the acceptability of IPV both across and within 33 countries in SSA? This descriptive analysis highlights the importance of taking a couple-level approach—as opposed to an individual-level approach—when exploring attitudes about IPV by highlighting heterogeneous patterns of concordance and discordance. All descriptive estimates are weighted using country-specific weights provided by the DHS; following DHS protocol for couple data, we use the husband weights because nonresponse bias is a greater concern for men than for women.
Second, we ask, what couple-level characteristics predict attitudinal concordance or discordance in accepting or rejecting IPV? Using multinomial logistic regression, we explore the couple-level characteristics that predict husband-wife concordance and discordance in rejecting or accepting IPV. Here we focus on expanding our dyadic perspective to many of the common risk factors for IPV predicted by social diffusion and bargaining perspectives, such as schooling, employment, and media exposure. We are particularly interested in exploring whether joint exposure to each variable of interest is associated with significantly higher attitudinal concordance than is the sole exposure of wives or husbands. To this end, we use joint exposure as the reference category to test whether sole or no exposure is significantly different. To account for multiple couples in our analysis living in the same households (about 6.5% of the sample), we cluster robust standard errors at the household level.
Third, we ask, what is the association between couple-level attitudinal concordance or discordance in attitudes toward IPV and women's reports of recent IPV? We conduct a multivariate analysis to explore whether couples' concordance or discordance in accepting or rejecting IPV is associated with wives' reports of experiencing each type of IPV (sexual, emotional, and physical) in the last 12 months, using the IPV-experience subsample. We conduct this analysis to show that concordance/discordance in husbands' and wives' attitudes about IPV are meaningful analytical categories that are predictive of other dynamics within the family. Couples' views on wife beating might influence women's likelihood of reporting IPV, or couples' views on the acceptability of wife beating may be influenced by women's reported experiences of IPV. We cannot adjudicate between these two causal pathways. Nonetheless, this analysis provides valuable insights into whether concordance in attitudes about violence corresponds with reported experiences of violence in ways not captured by previous studies that have examined the association between reported IPV experiences and husbands' and wives' attitudes measured separately. To account for multiple couples in our analysis living in the same households, we cluster robust standard errors at the household level.
Descriptive Overview of Concordance-Discordance in Husband and Wife Rejection of IPV
We begin by presenting pooled estimates of husbands' and wives' rejection of IPV, estimated independently. The top and middle panels of Figure 1 show that 50% of the wives and 70% of the husbands in our sample report rejection of IPV in all the DHS scenarios. In the bottom panel of Figure 1, we present couple-level estimates of husbands' and wives' concordance or discordance in accepting or rejecting IPV, which reveal important differences between the more conventional individual-level assessment of the prevalence of IPV rejection and the dyadic perspective that we take here. Individual-level estimates show that the majority of both husbands and wives in our sample reject IPV, but joint estimates show the converse: the majority of couples (61%) include at least one response that a husband is sometimes justified in beating his wife. Within couples, concordant attitudes about IPV are more common than discordant ones: about 39% of couples are concordant in rejecting IPV, and 19% of couples are concordant in accepting IPV, for a total of 58% agreement. On the other hand, in about 11% of couples, only the wife rejects IPV; in about 31% of couples, only the husband rejects IPV.
In Figure 2, we disaggregate these pooled estimates by country to reveal enormous geographic heterogeneity across Africa in patterns of concordance/discordance in IPV rejection. The countries in our sample span almost the full possible range of the percentage of couples who agree that wife beating is not justified under any of the five circumstances, from 2% in Guinea to 90% in South Africa. We observe more than 50% concordance in rejection of IPV in only 11 of the 33 countries (Angola, Benin, Comoros, Ghana, Madagascar, Malawi, Mozambique, Namibia, Rwanda, São Tomé and Príncipe, and South Africa). We find more than 50% concordance in acceptance of IPV in only 1 country (Guinea). Regarding instances of discordance within couples, between 30% and 50% of husbands, but not wives, reject IPV in about one-half of the countries in our sample (16 of 33 countries), and the prevalence of husband-rejecting and wife-accepting couples is highest in Ethiopia (48%), Niger (46%), and Sierra Leone (44%). Across all countries except Lesotho, it is less common to observe wives but not husbands rejecting IPV than it is to observe the opposite. The prevalence of wife-rejecting and husband-accepting couples ranges from a low of about 4% of couples in Guinea to a high of 17% to 18% of couples in Kenya and Lesotho.
Multinomial Logistic Regression Analysis of the Couple-Level Characteristics That Predict Concordance/Discordance in Rejection of IPV
Our descriptive analysis suggests that previous research examining heterogeneity in men's and women's rejection of IPV has overlooked important differences in the extent to which wives and husbands agree or disagree with each other about the acceptability of IPV. We now explore the factors that predict concordance and discordance in husbands' and wives' rejection of IPV. Table 2 presents the results of a multinomial logistic regression model that predicts concordance/discordance in accepting or rejecting IPV, expressed as relative risk ratios. Our main explanatory variables are also coded in a dyadic perspective to help shed light on whether joint exposure—as opposed to singular exposure—to the control of interest is associated with different patterns of concordance/discordance. To better contextualize these results, Figure 3 plots the average marginal effects of each of the dyadic, couple-level variables. These average marginal effects can be interpreted as showing how much higher or lower the probability of observing each of the four categories of couple-level attitudes to IPV is for the displayed categories of explanatory variables (e.g., when only the husband, only the wife, or neither person completed primary school) compared with the omitted category (e.g., when both members of the couple completed primary school).
The most notable finding from Figure 3 is that for most of the dyadic explanatory variables, the probability of observing both partners rejecting IPV is significantly lower when only one partner or neither partner is exposed to a given variable (e.g., primary school, work, frequent newspaper reading) compared with when both partners are exposed to the same variable (demonstrated by the negative red point estimates with confidence intervals that do not cross 0). For example, compared with both partners completing primary school, having only one or neither partner complete primary school is associated with a 5–8 percentage point reduction in the probability of joint rejection of IPV. Similar findings apply for the television, newspaper, work, and number of union variables. Together, these results suggest that joint exposure to schooling, employment, and media is much more predictive of joint concordance in IPV rejection than is the sole exposure of either the husband or the wife to the variable in question, thus casting doubt on the notion that individual exposure of one member of the couple to school, work, or media might spill over to the other member and correspond with the joint rejection of IPV. The age variables also exhibit an interesting pattern: the husband being somewhat or much older than the wife is associated with a significantly lower probability of joint rejection (compared with same-age couples), although the wife being older than husband is associated with a significantly higher probability of joint rejection (compared with same-age couples).
In addition to the dyadic variables highlighted in Figure 3, the multinomial logistic regression results presented in Table 2 suggest that several other measures are important predictors of couple-level attitudes toward IPV. Table 2 documents a clear wealth gradient in concordance in the rejection of IPV: the richest wealth quintile (compared with the poorest) is associated with a lower relative risk of concordance in accepting IPV compared with concordance in rejecting IPV. Likewise, the richest wealth quintile (compared with the poorest) is associated with a lower relative risk of both types of discordance compared with concordance in rejecting IPV. Family structure also appears to be an important predictor of couple-level attitudes, with both polygyny and parity being positively associated with the relative risk of concordance in accepting IPV compared with rejecting it. On the other hand, the presence of the husband during the wife-beating module is not significantly associated with concordance, perhaps because this was not a common experience (occurring for less than 3% of couples).
Logistic Regression Analysis of the Association Between Concordance/Discordance in IPV Rejection and Wives' Reports of Recent IPV
The previous sections establish the prevalence of varied patterns of concordance and discordance in IPV rejection within couples and show how distinct combinations of couple-level characteristics predict concordance and discordance. As a next step, in Table 3, we explore the associations between attitudes and reported behaviors by examining whether husbands' and wives' attitudinal concordance or discordance is associated with wives' reports of experiencing IPV in the last 12 months, using the IPV-experience subsample. These results can be viewed graphically in Figure 4, which plots the predicted probabilities of women reporting emotional, physical, or sexual IPV for each of the four categories of couple concordance/discordance. Figure 4 shows that concordance in rejecting IPV is associated with significantly lower probabilities of reported emotional, physical, and sexual IPV compared with all other categories of concordance/discordance. At the other extreme, concordance in accepting IPV is associated with significantly higher probabilities of reported emotional, physical, and sexual IPV compared with all other patterns of concordance/discordance. These differences are largest for physical violence: wives who agree with their husbands that wife beating is justified are about twice as likely to report physical violence in the past year (25%) as wives who agree with their husband that wife beating is not justified in any scenario (13%); comparatively, about 18% to 20% of wives in marriages in which only one member of the couple rejects IPV report physical IPV in the past 12 months.
Couples with discordant attitudes about IPV also follow a fairly consistent pattern for the reported IPV outcomes. Women in couples with only the wife rejecting IPV have higher probabilities of reported emotional and sexual IPV than women in couples with both members rejecting IPV but lower probabilities of reported emotional and sexual IPV than women in the other two groups. Women in couples with only the husband rejecting IPV have higher probabilities of reported emotional and sexual IPV than women in couples with both members rejecting IPV or only wives rejecting IPV, but they have lower probabilities of reported emotional and sexual IPV than women in couples with both members accepting IPV. On the other hand, the probability of reports of physical IPV is not significantly different for the two types of discordant couples.12 Although the probability of reports of physical IPV is not significantly different for the two types of discordant couples, the probability of physical IPV in either discordant couple type is significantly lower than in the concordance-accept couples and significantly higher than in the concordance-reject couples. These results further demonstrate the importance of using a dyadic perspective to better understand the social context in which IPV occurs. If we instead were to use individual-level reports, we would obscure these meaningful differences in the configurations of IPV-related attitudes within families.
Discussion and Conclusion
This article contributes a dyadic perspective to the investigation of attitudes regarding the acceptability of IPV across Africa. When attitudes are examined only at the level of individuals, a majority of both husbands and wives reject IPV in all five hypothetical scenarios. Yet these statistics obscure the enduring acceptance of IPV within households in Africa. When we consider husbands and wives together, we find that more than 60% of all households and a majority of households in most countries include at least one member who says that IPV is justified under one or more of the five circumstances. We document striking heterogeneity in couple-level concordance or discordance in IPV-related attitudes: the proportion of couples who agree that wife beating is never justified ranges from 2% to 90%.
Next, we apply a dyadic perspective to the correlates of couple-level attitudes toward IPV, demonstrating that joint exposure to schooling, work, and media is more predictive of joint rejection of IPV than are singular exposures of wives or husbands. The fact that joint schooling and employment are predictive of joint rejection could support a bargaining perspective, which suggests that women's greater control of resources relative to their spouses increases their bargaining power in relationships and ability to advocate for the rejection of harmful practices (Friedman et al. 2016). Some of our results, however, complicate the bargaining perspective. For example, we find that the odds of discordance in which the wife but not the husband rejects IPV are greater for couples in which the wife has primary schooling but the husband does not than for couples in which both members complete school. This finding is not consistent with bargaining and could even be interpreted as a backlash against women's higher achievement (Cools and Kotsadam 2017). An alternative possibility is that joint exposures to schooling, work, and media predict a greater probability of joint rejection because joint exposure to global cultural scripts is mutually reinforcing, which is consistent with the global cultural diffusion literature (Baker and LaTendre 2005; Swindle 2018). At the same time, similarities in attitudes may be related to marriage market selection processes, whereby people prefer to match with others who hold similar opinions (Lopus and Frye 2020). Another possibility is that couples with different background characteristics—such as education or media exposure—do not share information with their partners as actively.
The final step of our analysis shows that different combinations of husband-wife attitudes are associated with unique likelihoods that the wife will report having experienced IPV in the past year—distinctions that would be missed if the attitudes of husbands and wives were measured separately. Consistent with previous research (Alio et al. 2011), we find that women living in households in which both members reject IPV have the lowest reported prevalence of recent IPV and that women living in households in which both members accept IPV have the highest reported prevalence of recent IPV. Among couples who disagree about the acceptability of IPV, we find that those in which only the wife rejects IPV have a lower likelihood of reporting experiencing IPV than do those in which only the husband rejects IPV. This finding could be due to reporting bias or to women who believe IPV to be unacceptable being more likely to have access to support from relatives and others in their community (McCleary-Sills et al. 2016; Overstreet and Quinn 2013). These results demonstrate the importance of considering husbands and wives jointly when considering the association between attitudes and behavior: couple-level attitudes look quite different than individual-level attitudes, and couple-level concordance or discordance in attitudes toward IPV are meaningful predictors of reported IPV.
These results have important policy implications. Traditionally, IPV interventions have primarily focused on empowering women and exposing them to messages about gender equality (Alexander and Welzel 2015; Pierotti et al. 2018; Wyrod 2008). Recently, programs have begun to focus more attention on challenging the gendered cultural understandings of men and prompting behavior change to reduce the prevalence of IPV (Barker et al. 2010; Casey et al. 2018; Gibbs et al. 2019; Pierotti et al. 2018). Despite some success, these male-centered initiatives have been met with resistance, particularly when they threaten to undermine the fundamental expectation that women be subordinate to their husbands (Pierotti et al. 2018). The findings discussed in this article suggest that policy or programmatic efforts to transform masculine gender norms and controvert the acceptability of IPV may be more effective if they target couples rather than individuals (Gupta et al. 2013; Krishnan et al. 2012). The exact form that a couple-oriented approach might take likely depends on the setting or cultural context; in some settings, it may be appropriate for interventions to work with both members of the couple jointly, whereas in other settings, it might be more appropriate to target husbands and wives separately.
Despite these contributions, some important limitations to our data and approach must be kept in mind when considering the implications of our findings. Like all research that uses surveys to examine attitudes toward IPV, we cannot determine whether people have really changed their attitudes, and we cannot distinguish between individuals who believe that IPV is wrong and those who report that it is wrong as a result of social desirability bias. Even if responses were influenced by social desirability bias, this nonetheless provides valuable insight on the factors that might be associated with men's and women's knowledge of the socially desirable answer to questions about wife beating, which may over time correspond with deeper attitudinal change (Hafner-Burton and Tsutsui 2005; Pierotti 2013).
An additional limitation to our approach is that we are unable to assess causal directions using cross-sectional multivariate approaches because attitudes and behaviors may occur simultaneously and influence each other. For example, couples in which one or both members report that IPV is unjustified under all five circumstances may be less likely to report the experience of IPV because of the effect of these beliefs on intrahousehold dynamics. On the other hand, if IPV occurs within a household, both the husband and the wife may be more likely to deem it acceptable through a process of ex-post rationalization. Conversely, women who view wife beating as unacceptable may be less likely to report experiencing it themselves (even if they do experience it) out of shame. We are unable to adjudicate between these different scenarios given that both questions were asked simultaneously and both were self-reported. Nonetheless, our results suggest that measuring attitudes at the couple level provides a more accurate picture of the link between attitudes and behavior.
Finally, as in all correlation analyses, unobserved factors may have influenced both partner characteristics and attitudes about IPV. Although we control for many of the known risk factors for IPV, there are some important known risk factors—such as childhood experiences of abuse among both husbands and wives—that we are not able to control for because of DHS data constraints. Furthermore, other difficult-to-observe factors—such as personality traits and psychological disposition—are likely associated both with our respondents' decisions to work, consume media, and pursue schooling and with their propensity to reject dominant norms regarding IPV. Likewise, the methods we use in our analysis do not allow us to ascertain whether people select into partnerships with those who hold similar attitudes or whether attitudes are developed over the course of the marriage as a product of interactions and experiences.
Taken together, our results demonstrate the importance of integrating a dyadic perspective into studies of IPV attitudes and behaviors. In part, this could be done by using existing data sources, such as the DHS, that allow for the creation of couple-level variables. At the same time, a dyadic approach should be integrated into longitudinal data sources. Such data could be used to provide important information about how and why within-couple level concordance and discordance in IPV attitudes and behaviors change over time and could help illuminate the issues of selection and reverse causation, which are difficult to address using cross-sectional data.
Our analysis contributes to a growing number of demographic studies that have applied a couple-level approach to key demographic outcomes, such as fertility preferences, contraceptive behaviors, and child health and well-being outcomes (Pratley 2016; Uddin et al. 2017; Yeatman and Sennott 2014). Nonetheless, most micro-oriented studies on these topics continue to focus on individual accounts of social and demographic processes in ways that may obscure important heterogeneity in couple-level attitudes and behaviors. Further application of the dyadic approach to demographic outcomes that entail a relational component—such as IPV, fertility and contraceptive use, child health, and household decision-making norms—will provide a more nuanced and comprehensive understanding of how the intracouple context is associated with these important demographic outcomes.
Authors are listed in alphabetical order; both authors contributed equally. We are grateful to Abigail Weitzman, Mónica Caudillo, Jeffrey Swindle, Doron Shiffer-Sebba, and Pilar Gonalons-Pons for helpful comments on earlier versions of the manuscript.
We use the most recent survey available (as of July 2019). Cameroon is excluded from our analysis because some of the variables included in our analysis were not collected.
The proportion of households included in this male subsample varies across our sample. In 23 countries, men were interviewed in one-half of all sampled households; in 5 countries (Chad, Malawi, Mozambique, Tanzania, and Uganda), this rate was one in three; and in 4 countries (Ethiopia, São Tomé and Principe, Zambia, and Zimbabwe), men were interviewed in all sampled households. The age range for male eligibility also varied across countries. In 23 countries, men are eligible if they are aged 15–59. The male eligibility age range is 15–49 in Nigeria and Liberia; 15–54 in Angola, Kenya, Malawi, Uganda, and Zimbabwe; and 15–64 in Benin, Mozambique, and Namibia.
In 14 countries (see Table A1, online appendix), the IPV module was not asked of women in households included in the male subsample. In Burundi and Kenya, the IPV module was administered to one randomly selected eligible individual (male or female) per household. In Ethiopia, Mali, and Rwanda, the IPV module was administered to one randomly selected eligible woman in one-half of the sampled households. In Malawi and São Tomé and Principe, the IPV module was administered to one randomly selected eligible woman in one-third of the sampled households. For the remaining 12 countries, this module was administered to one randomly selected woman (who was eligible to complete the woman’s survey) in each sampled household. Data on violence in the DHS are also collected about the most recent partner for women who are currently separated or divorced; however, because of the paper’s couple-level focus, only violence by current partners is included.
In our sample, 2,452 wives and 1,157 husbands reported that they do not know whether wife beating is justified for at least one of the scenarios; respondents who reported that they “do not know” whether wife beating is justified are coded as 0 for that scenario. Thus, the 0 category technically refers to failure to reject wife beating, although for simplicity throughout the article, we refer to rejecting versus accepting IPV. As Figure A1 (panels A and B) shows, percentages of couples’ concordance/discordance are substantively the same if respondents who report that they do not know are instead coded as missing (multivariate results in Table A4 of the online appendix are similarly robust to coding “do not know” responses as missing).
In one country in our sample (Congo Brazzaville), couples are asked only three of the questions about the acceptability of wife beating. For this country, the husband/wife rejection of IPV indicators are created from these three questions. We follow a similar strategy in the other instances in which a respondent lacks information on one of the five scenarios (i.e., the indicators are created based on IPV rejection over all scenarios answered). Both our descriptive and multivariate results are robust to excluding respondents that lack information on one or more questions on the acceptability of wife beating (see panel C of Figure A1 and Table A5, online appendix).
The DHS also includes questions about frequency of listening to radio, but we focus on television and newspaper exposure because the literature suggests they are particularly important for the diffusion of norms and are more likely to involve active attention (Jensen and Oster 2009; Westoff and Koffman 2011).
Although information about the length of the current union would be useful, this information is unavailable in the DHS for women who have been married more than once. Controls for age, however, may proxy for some dimensions of union length.
Although childhood experiences of witnessing and experiencing abuse and recent consumption of alcohol are common risk factors for IPV, data constraints limit our ability to control for these factors. The DHS includes questions related to these topics as part of the IPV module administered to select women; however, these questions are administered to husbands in only five countries, only three of which were also administered the IPV module. Thus, the sample of couples with this information for both members of the couple is very small (less than 5% of our full sample). As a robustness check, we rerun our main analyses controlling for the wife having witnessed violence in childhood, the wife having experienced violence in childhood, and husband’s consumption of alcohol (as reported by wife). Our main results are substantially robust to this inclusion and are available upon request.
Marriage in SSA, however, is a heterogeneous process, and the “married” category of this variable includes civil unions, religious weddings, traditional ceremonies, and de facto marriages with no ceremony at all, as well as combinations of these different marriage types, each of which convey unique types of legal and social recognition (Frye and Trinitapoli 2015; Meekers 1992). Thus, rather than differentiating between two distinct legal or social statuses, this measure should be interpreted as distinguishing between two different subjective definitions of the union, as reported by the wife. For clarity and parsimony, we use the terms “husband” and “wife” to describe all male and female partners in our sample, while noting that about 17% of wives describe themselves as “living together as if married.”
Our main results are also robust to excluding the 3% of cases in which the husband was present during the module assessing wife’s acceptance of wife beating (available upon request).
Because we control for the age difference between the couple and the husband’s age, we do not include controls for wife’s age.
This difference across outcomes could be explained by the fact that the IPV rejection question refers to physical IPV only; thus, it is plausible some men reject physical IPV but not emotional or sexual IPV.