Resources and Intimate Partner Violence in Sub-Saharan Africa

— Combining DHS data for 580,000 women from 30 different countries in Sub-Saharan Africa, we analyze how both the incidence and the acceptance of intimate partner violence vary across time and space, in a region with record high levels of violence against women. We review the existing literature regarding the impact of resources on intimate partner violence, extracting testable and often conflicting hypotheses at the micro and macro level, and on the interaction across levels. We propose to extend existing theory to take into account attitudes at the community level. In the empirical analysis, we find no evidence that resources protect against abuse at the individual level, although resources are associated with lower acceptance. We find that resource inequality, both within the household and at the aggregate level, is associated with more abuse. Finally, we find that employed women face greater risk of abuse in communities with relatively higher acceptance of wife-beating. 2017 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


INTRODUCTION
Violence against women is a widespread form of human rights violation, and intimate partner violence is by far its most important component (Devries et al., 2013;Fried, 2003;Heise, 2011). Intimate partner violence is associated with a wide array of negative outcomes for the women who are abused, including pregnancy loss and sexually transmitted infections (Durevall & Lindskog, 2015;Krishnan, 2005). It also has considerable negative externalities, affecting more individuals than the ones who are abused, through fear of abuse and psychological stress from witnessing violence (Jewkes, 2002;True, 2012).
Intimate partner violence is prevalent in all societies, but the level and the degree to which it is considered acceptable vary greatly. The countries of Sub-Saharan Africa (SSA) have very high levels of violence against women (Garća-Moreno, Ellsberg, & Watts, 2005;Devries et al., 2013). Also, 14 out of the 15 countries with the highest share of women who deem wife-beating justifiable are found in SSA (World Bank, 2011). There is nevertheless a lot of variation across the region and over time, suggesting that intimate partner violence also depends on factors at the contextual level (Jewkes, 2002;Johnson, Ollus, & Nevala, 2008;True, 2012). In the present paper, we explore this variation in order to shed light on a wide range of hypotheses regarding the relationship between resources and abuse. 1 We first give a broad description of the spatial and temporal variation in the prevalence of abuse in SSA, and in the degree to which it is considered justifiable. By extending the analysis to women's actual experience with abuse, this description complements Pierotti (2013), who documents a recent declining trend in women's acceptance of intimate partner violence. Our analysis suggests that both acceptance and the incidence of intimate partner violence declined in SSA over the period of analysis. Importantly, the data show a great deal of variation across time and space, suggesting the need for taking specific contextual factors into account. We move on to investigating the relationship between abuse and resources at different levels of analysis. Resources are measured in terms of wealth, education, and employment. We use exceptionally ample micro data from the Demographic and Health Surveys (DHS), containing information on attitudes regarding wifebeating for 586,255 women and the experience of abuse for 156,929 women, located in 30 SSA countries and interviewed over the years 2003-13. There are three general versions of theories on how resources affect the prevalence of intimate partner violence. The standard resource theory (e.g., Goode, 1971) posits that women with few resources are more at risk of abuse, and that men with fewer resources are more likely to be abusive. More recent scholarship, including bargaining theories of the household, tends to focus on relative resources. On the one hand, relatively less female resources could lead to more abuse due to marital dependency (e.g., Vyas & Watts, 2009). On the other hand, relatively more female resources could increase violence due to the stress induced by status inconsistencies (e.g., Hornung, McCullough, & Sugimoto, 1981). A refinement of the relative resource theory is the gendered resource theory (Atkinson, Greenstein, & Lang, 2005), proposing that the effect of relative resources on abuse will depend on husbands' gender ideologies, where the degree to which men hold breadwinner ideals is crucial. These different theories operate at different analytical levels. The simplest form of resource theory is concerned only with absolute resources at the micro level. Relative resource theory, on the other hand, is concerned with relative resources within the couple, as is the gendered relative resource theory. Very few theories explicitly address the macro level when analyzing intimate partner violence, but the importance of this level is often implied. Moreover, it may interact with the other levels. We propose to expand existing theory by explicitly taking into account the macro level, both directly and as a moderator. Specifically, we believe that the existence of a violence back-* The authors thank Rannveig Kaldager Hart, Lori Heise, Gö ran Holmqvist, Niklas Jakobsson, Annika Lindskog, Kalle Moene, and several seminar participants for helpful comments and suggestions. This paper is part of the research activities at the center of Equality, Social Organization, and Performance (ESOP)  lash, a situation where female resources increase the risk of abuse, hinges on prevailing acceptance of wife-beating in the surrounding community. In this framework we propose a contextual acceptance employment hypothesis, whereby female employment leads to a relatively higher risk of abuse in communities where wife-beating is considered acceptable.
Acknowledging that violence is a ''multifaceted phenomenon grounded in an interplay among personal, situational, and sociocultural factors" (Heise, 1998, p. 263 (2011)). We explicitly model the correlation of intimate partner violence with resources at the individual, the couple, and the community level, and we allow for interactions across these levels.
With respect to the standard resource theory, our findings show that household wealth, though strongly negatively linked to the acceptance of wife-beating, is only weakly negatively correlated with the risk of actual abuse in the household. Furthermore, richer areas are not less violent-prone than poorer areas. Societies with a high degree of economic inequality, however, have higher levels of abuse. We also find little support for the simplest economic bargaining model in our data, as for most women, their level of education-allegedly a source of bargaining power-is positively correlated with abuse risk. Moreover, female employment is consistently associated with higher risk of abuse. With respect to the relative resource theory, we find that intra-household inequality in education, regardless of which spouse has more years of schooling, is associated with more violence. Female resources at the macro level in terms of improved outside options do not seem to imply lower levels of abuse either, as areas with higher levels of female employment and education are relatively more violent prone. Consistent with the contextual acceptance employment hypothesis, we find that being employed is correlated with an even higher risk of abuse in areas with a high level of acceptance of wife-beating.
This paper adds to the literature in several ways: First, we provide a thorough review of the literature as it relates to income, education and employment and women's experience of intimate partner violence. Second, from this literature we extract a number of conflicting hypotheses and we investigate how they fit with the data. Some of these hypotheses have not been tested in the literature that we cover. Third, despite the large amount of previously existing hypotheses, we identify gaps in the literature and we propose new hypotheses that we also investigate empirically. Fourth, we combine data from a whole region, which results in a large data set with substantial variation across communities. This allows us to test hypotheses at different levels of analysis.
Our findings underscore the need for high-quality data and careful consideration of analytical level when exploring the relationship between resources and intimate partner violence. For one, information on attitudes towards wife-beating cannot replace data on actual abuse in analyses of how resources relate to the incidence of abuse. Although being accepting of wife-beating is positively correlated with the probability of actual abuse on part of both victim and perpetrator, it is monotonically negatively correlated with wealth and education, thus masking the more complicated relationship between resources and abuse. Similarly, employment is only weakly related to acceptance but strongly, and positively, related to abuse. The interaction between macro-level acceptance and micro-level employment is highly relevant for predicting levels of abuse. Moreover, our findings show that aggregates such as household wealth, or the total level of resources in a society, are insufficient predictors of abuse: It is important to consider also the distribution of resources; between household members in particular, but also at the macro level and in interactions across different levels.

CONCEPTUAL FRAMEWORK AND HYPOTHESES
The relationship between resources and intimate partner violence is given considerable attention in social science research. We give a comprehensive review of this literature, with particular focus on the empirical hypotheses regarding how economic resources influence the risk of abuse. The resources given most attention in the literature are income (or wealth), employment, and education, which will also be the focus of this paper. 2 (a) Micro-level resources: The role of absolute and relative resources in the household At the individual level, resources are often argued to be empowering and protective against intimate partner violence (Jewkes, 2002). However, the relationship between resources and violence need not be linear, and some even argue for a backlash, whereby increased resources lead to more abuse for women (True, 2012). We here discuss the reasoning behind these opposing predictions.
In the psychological literature, poverty is associated with stress, which is thought to influence the degree of abuse (Jewkes, 2002). The frustration-aggression hypothesis is the most common psychological theory on the link between poverty and aggression (Barlett & Anderson, 2013). From a social perspective, resource theory regards the family as a power system and suggests that men with few other resources may use violence to maintain dominance within the family (Goode, 1971;Vyas & Watts, 2009). In both cases, the prediction is that income or wealth lowers the incidence of abuse.
Standard resource theory further predicts that employment and education are also protective against abuse, beyond their effect on income and wealth. This is in line with the view of the World Health Organization, which argues that female employment should be strongly supported as a means to reduce intimate partner violence (Garcia-Moreno et al., 2005). High educational attainment is negatively associated with being both a victim and a perpetrator of abuse across a wide range of different studies (see Jewkes, 2002 for an overview).
Education is also assumed to affect behavior via identity and learning about the normative foundations of society, and it may expand horizons as well as increase exposure to global discourses rejecting partner violence (Pierotti, 2013). Friedman, Kremer, Miguel, and Thornton (2011) exploit a randomized scholarship program that raised girls' secondary schooling in Kenya and find a reduction in acceptance of wife-beating. Mocan and Cannonier (2012) use a policy reform in Sierra Leone in 2001 and find that increased primary schooling also reduces women's acceptance of wife-beating.
An additional avenue for the protective role of individual resources is evoked in bargaining theories of the household, where women's outside options-usually considered to be the utility level in case of divorce-are crucial in determining the outcome of the bargain (Eswaran & Malhotra, 2011;Farmer & Tiefenthaler, 1997;Lundberg & Pollak, 1996;Manser & Brown, 1980;McElroy & Horney, 1981;Pollak, 2005). In