Abstract
This paper explores the correlates of violence against women in intimate relations. Using data from a large victimization survey, we assess five theoretical explanations based on patterns of association between violence and key attitudinal and socio-demographic variables: sexism, family violence, dependency, exchange and status inconsistency. We conclude that status inconsistency, understood as a complex multidimensional theory of patriarchy that integrates core elements of other theories, offers the best explanation. Women’s odds of suffering violence are higher if they have sexist partners, are disempowered or have higher statuses than their partners. Disempowered women have fewer resources to oppose the sexist culture on which violence breeds. Women with elevated statuses defy men’s dominance. Violence is intrinsic to a patriarchal system of power that fosters sexism and manifests violently when women are vulnerable or challenge men’s statuses.
Keywords
Introduction
We aim to explore the correlates of physically threatening abuse, or violence, on women in intimate relations. We assess the validity of five theories that explain it: sexism, family violence, dependency, exchange and status inconsistency. The five theories cannot be found in pure form anywhere, for they overlap, making it difficult to contrast them directly. However, by identifying their core elements and assumptions, we raise several testable hypotheses that facilitate the comparison. We test them with data from a large victimization survey and logistic regression analysis, using several indicators developed to operationalize the concepts of violence and its correlates. The results give more credence to status inconsistency, understood as a complex theory of patriarchy, than to other theories, providing a coherent model of the risk factors associated with violence against women in intimate relations.
Five theoretical approaches to explaining violence on women in intimate relations
There are five main theoretical streams on intimate partners’ violence against women. Three of them, sexism, family violence and dependency theories, identify absolute features pertaining to couples’ members or their family units as predictors of violence, while the other two, exchange and status inconsistency theories, concentrate on relational aspects of the couple. Whereas sexism is fundamentally a ‘cultural’ approach, family violence is closer to the ‘economic’ pole. Dependency, exchange and status inconsistency theories include both cultural and economic elements. Along with sexism, the three emphasize the role of patriarchy in fostering violence against women, although each conceptualizes it differently. All five theories are probabilistic: the factors identified as being associated with violence are treated as risk factors increasing the probability of observing it. Women’s lives are too complex to be completely explained by any theoretical model. The assumption is that any omitted factors would cancel each other, having no systematic effect on violence.
The sexism perspective can be traced back to Dobash and Dobash’s (1979) early feminist work on wife beating. Like other feminists, sexism scholars contend that the ultimate reason explaining women’s violence in intimate relations is asymmetric relations of power between men and women within patriarchal societies (Dobash and Dobash, 1979; Yllo, 1993). Violence is the most extreme expression of patriarchy, understood in this approach as a sexist cultural system of domination subjugating women to men: directly, through cultural norms of deference and obedience backed if necessary by the use of force; or indirectly, by shaping women’s opportunities and constraints in basic institutions such as the family and work that reinforce women’s subordination (Kim and Sung, 2000). The key variable is sexism, portrayed as a ‘quasi-necessary’, 1 if possibly non-sufficient, condition to observe violence. By quasi-necessary, we mean that sexism is seen as an exogenous causal factor in a predictive and probabilistic model that does not in any sense imply inevitability in social life. By non-sufficient, we mean that there may be other endogenous factors – women’s socio-economic dependence, family’s stressful conditions, partner’s substance abuse, etc. – that specify the conditions leading to an intimate partner’s violence, but that require sexism to operate exogenously. As Anderson (1997) states, this has led non-feminist researchers to claim that ‘feminist scholars employ single variable analyses that concentrate on patriarchy and ignore the impact of factors such as income, unemployment and age’. For these critics, feminists’ emphasis on patriarchy is one-dimensional and simplistic (Dutton, 2006; Gelles, 1985). Others regard the criticism as unfair, for it does not recognize that feminist scholars have moved away from treating sexism as the only relevant exogenous variable towards integrating it into more complex and multivariate theories of patriarchy (DeKeseredy 2011; DeKeseredy and Schwartz, 1996), as described below.
The family violence perspective argues that patriarchy is just one among a complex constellation of factors related to inequality structures creating conditions conducive to violence (Gelles, 1993; Straus et al., 1980). This ‘socio-economic’ view criticizes sexism scholars’ omission of factors such as social class that, independently or together with gender, predict violence in the family (Kasturirangan et al., 2004; Lee, 2000). Insofar as these factors specify, rather than question, the quasi-necessary role of sexism in violence, this view is indistinguishable from the sexism perspective. However, in its most common version it reverses causality, considering sexism a possible outcome of family conditions, and grants to family conditions the quasi-necessary status otherwise assigned to sexism. In this view, the multiple manifestations of violence in the family result from its location in the social structure and the stressful conditions stemming from it (Johnson, 1995).
Dependency theory belongs, with exchange theory, to a wider ‘bargaining power’ perspective that analyses the impact of women’s (not couples’, as in the family violence perspective) socio-economic resources on violence. Dependency theory stresses women’s absolute resources. It analyses how low opportunities and multiple constraints stemming from women’s positions in the socio-economic structure affect women’s control over their lives, making them dependent on their male partners, and raising the probability of experiencing violence (Harway and Hansen, 2004; Kim and Gray, 2008). It posits the existence of a patriarchal structure of power that manifests either as a cultural system of sexist domination or – this is theory’s main contribution – as a socio-economic structure of vulnerabilities, either of which can lead to women’s victimization. To put it another way, this view takes a bi-dimensional approach to patriarchy, stressing its cultural and socio-economic expressions and the need to consider both for explaining violence against women in intimate relations (Raj and Silverman, 2002). By itself, neither expression is necessary to experience violence, but when taken together, they lead to violence.
Exchange theory (Gelles, 1983) is also part of a broader ‘bargaining power’ perspective. It focuses on women’s resources relative to men’s. In its radical form, it is indifferent to gender and analyses how resources facilitate the exertion of power by whoever has more. In Gelles’s (1983: 157) words, males exert violence on women ‘because they can’. Recent views that focus on women’s violence on men belong to this symmetric approach (Straus, 2006). Exchange theory considers women’s inferiority to be ‘quasi-necessary’ for suffering violence from the dominating partner, although perhaps not sufficient, as other factors may specify the conditions that facilitate it. When exchange theory is combined with patriarchy, women’s subordination appears as its most basic element and the main exogenous cause of violence (Johnson, 1995).
The status inconsistency perspective can be traced back to work by O’Brien (1971), Rodman (1972) and Gelles (1974) on how differences in occupational and educational attainment favouring women in intimate relations disrupt traditional patriarchal roles, leading to violence on women. Some label it ‘exchange theory’ (Goode, 1971), but we prefer ‘status inconsistency’ so as not to confound it with bargaining theory’s symmetric approach. This theory, also feminist and built around patriarchy, is less vulnerable than the culturalist perspective to criticisms of one-dimensionality. Violence is not just the outcome of a sexist cultural system of domination: it is the key expression of the power system on which patriarchy is based. When the patriarchal balance of power remains unchallenged, it manifests in sexism or dependency. But when the male partner has lower socio-economic status or fewer resources than his wife, violence serves to restore the traditional order (Hoffman et al., 1994). Some scholars combine dependency and status inconsistency under a single resource theory (Fox et al., 2002; Goode, 1971). Women’s absolute resources deter violence, but when they exceed a partner’s they spark a man’s violent reaction to compensate for his status loss. A recent contribution, gendered resource theory, explicitly adds sexism to the predictive model, possibly at the expense of omitting dependency (Atkinson et al., 2005). When combined, explanations built around status inconsistency theory contribute to a vision of patriarchy as a multidimensional power structure that subordinates women through different means. None require resorting to violence, but together they do as the ultimate means to impose subordination. Violence is, firstly, an outcome of a sexist culture that denigrates women; secondly, the result of an opportunity structure that disadvantages women; and, thirdly, a means to restore subordination when it is challenged.
Objectives and hypotheses
Our main objective is to assess the validity of each of these perspectives for explaining violence on women in intimate relations (as operationalized in the next section). If the sexism perspective were correct (H1) and this factor were quasi-necessary to experience violence, it should be highly correlated with a sexist culture that denigrates and subordinates women. Since the theory does not expect sexism to be sufficient for experiencing violence, its effect on violence should weaken, but not disappear, after conditioning for socio-economic factors. However, these latter factors’ effects on violence should vanish after conditioning for sexism, proving that they are endogenous to it.
If the family violence perspective held true (H2), violence should be more common in families exposed to deprivation and stress. In addition, if sexism ensued from poor family conditions, its effect should disappear after considering such conditions. Alternatively, for the bargaining power approach to be valid (H3), violence should vary inversely to women’s personal resources, net of other variables’ effects. Two sub-hypotheses are possible, depending on whether or not sexism is additionally assumed to independently foster violence. If it is, as in dependency theory (H3.1), violence should vary inversely to women’s absolute levels of resources, and the association between sexism and violence should weaken but not disappear after conditioning for differences in such levels (reflecting the common patriarchal ground from which sexism and dependency stem, but also their independent role in fostering violence). In contrast, if as in exchange theory subordination, rather than dependency, were quasi-necessary to experience violence (H3.2), the inverse relationship between women’s resources and the odds of experiencing violence should ensue only when resources are expressed relatively to partner’s. Absolute resources and sexism should have no significant effects on violence after conditioning for relative resources, due to their being endogenous to woman’s subordination.
Finally, if the status inconsistency view were correct (H4), women’s resources relative to men’s should co-vary directly with women’s odds of experiencing violence, and the partial associations between sexism and violence (and dependency and violence) should increase when women’s resources relative to men are considered. The latter results would show that neither sexism nor dependency are quasi-necessary for experiencing violence (contrary to H3.1’s claim) and that patriarchy can act through alternative channels. Patriarchy acts normally through a cultural and socio-economic structure of domination that legitimizes it, and of which a credible threat of using force is crucial. However, when women’s resources are high enough to threaten this structure of domination, the kernel of patriarchy’s system of power emerges and men resort to violence to restore the traditional gender balance of power.
Data, variables and techniques of analysis
To test our hypotheses, we use data from the III Macro-survey of Violence on Women carried out by the Spanish Instituto de la Mujer in 2006. The sample size is very large: 32,426 women over 18 years of age residing in Spain in 2006. The sample was over-represented in some regions to have sufficiently large numbers of respondents in each. Consequently, we applied weights to balance the sample and make it representative nationally. Only women having an intimate relationship at the time of the interview were analysed: 24,131 women, or about 75 per cent of the total sample. Interviews were carried out by telephone, which might explain the high rates of response in most questions (over 95 per cent), despite their sensitive nature.
A few variables, such as family income, had higher percentages of non-response, although always lower than 10 per cent. If missing cases differed in systematic ways from valid ones, our analyses could suffer from selection bias. To minimize this, we imputed missing cases using Rubin’s Markov Chain Monte Carlo method of Multiple Imputation (Little and Rubin, 1987; Rubin, 1987). Missing values were replaced by plausible values imputed five times using a Markov Chain iterative procedure. In each imputation, the imputed value was the mean value of the fully observed cases that best matched the missing case in the values of many other variables included in a predicting equation, plus a random error. In the substantive analyses, the values generated in the five imputations were pooled to produce a single estimate. The extra variability generated by the imputation was accounted for by adding the between-imputation variance to the pooled or within-imputation variance.
Our dependent variable is a binary indicator of violence on women in intimate relations, meant to be a proxy of what others label as intimate partner’s violence. How violence is defined has immediate consequences on estimates of its prevalence and causes (Medina-Ariza and Barberet, 2003). Our definition is restrictive. It excludes attitudes and behaviours that, being diminishing and derogatory towards women, do not threaten to inflict or do not actually inflict physical damage. Conversely, we include some non-physical forms of violence excluded in other studies. These forms are severe enough to capture the possible underreporting of physical violence (their correlations with physical violence were very high). By increasing the counts of physically abused women from 195 to 510, our operationalization helps reduce problems associated with sparse data in complex cross-tabulations. We build the indicator of violence from responses to three low-level items in a battery of questions where women reported the frequency (‘never’, ‘rarely’, ‘sometimes’ or ‘often’) with which they were experiencing at the time of the interview a series of denigrating and abusive attitudes and behaviours from their partners. The three specific items asked to learn how often women: were pushed or beaten by their partners when angry; felt fear of them; and were threatened or insulted by them. We performed a Multiple Correspondence Analysis (MCA) on these items in order to identify an underlying scalar measure of violence along which women could be placed.
Like a regular Principal Component Analysis, MCA decomposes the variance of respondents’ answers to several variables into orthogonal scalar components along which respondents are distributed following normalization constraints. Often, one component is enough to capture much of the total variance. Before proceeding to extracting the component(s), MCA transforms the original categorical variables into scales, using an optimizing algorithm that assigns scores to categories in ways that maximize the correlations (loadings) between the transformed variables and the extracted principal component(s). The loadings help interpret the ‘meaning’ of the components.
Figure 1 shows the results of applying MCA to the three questions included in the analysis. The black dots indicate the values of the loadings or correlations of the transformed variables with the single extracted dimension (scale shown on the right-hand y-axis). This dimension explains 66 per cent of the variance of the responses to the three transformed variables. The loadings are high (ranging from .80 to .82), indicating that the three questions contribute highly to explaining the uncovered dimension, because they are highly inter-correlated (Cronbach’s alpha, which measures the reliability of the uncovered measurement based on transformed variables’ inter-correlations, was high: 0.7). We interpret this dimension, which we call violence, as capturing physically threatening abuse against women in intimate relations.
Indicator of violence.
Figure 1 also shows the percentages (grey bars) with positive scores in the transformed, optimally scored, variables. Positive scores mean that at around the time of the interview women experienced these forms of violence, however rarely. The prevalence of each form is low (from 0.8 to 1.3 per cent) and possibly underestimates the true prevalence, because some women may find it difficult to report abuse, especially physical, for fear of the consequences or other reasons. In assessing these figures it must be remembered that they do not express lifetime or previous 12 months’ prevalence, but prevalence at the time of the interview. The black bar in Figure 1 shows the percentage of women with positive scores in the scalar dimension uncovered by MCA, which basically coincides with the percentage experiencing any of the three forms of abuse. While there is much variation in the positive scores assigned to women on the scale by the technique (due to the different frequencies with which the three behaviours manifest), this has little leverage on the overall variability of the distribution, given the low percentage of women reporting any of these forms of abuse – 2.11 per cent of the 24,131 persons in a relationship. The main contrast is between women experiencing and not experiencing violence, and thus we dichotomized the scale uncovered by MCA into two categories of women with and without positive scores on it: experiencing or not experiencing violence in any of the three forms. This is the dependent variable.
Our analyses include three sets of independent variables. The first set seeks to capture the sexist attitudes and behaviours of male partners. The survey included a battery of questions asking respondents to report the frequency with which they experienced a list of abusive and sexist behaviours and attitudes from partners. Three of the items on the list were employed to generate the dependent variable. The remaining 21 were used to capture alternative manifestations of a partner’s sexism: from attempts at undermining woman’s abilities and limiting the woman’s decision-making power, to instances of disdain and lack of consideration for her needs and feelings.
Figure 2 shows the full list of items used to build the indicator of sexism with the same MCA technique described above. Figure 2 also shows percentages of women with positive scores in the optimally scaled variables (per cent ever experiencing the sexist attitude or behaviour). We also add the percentage of women experiencing any of the behaviours, however infrequently. This is 23 per cent, which suggests that sexism is not highly prevalent in contemporary Spain. The estimate would certainly be higher if the questionnaire had measured lifetime prevalence of sexism rather than sexism at the time of the interview.
Indicator of sexism.
The dots in Figure 2 show the loadings of the optimally transformed variables on the underlying scalar dimension of sexism uncovered by MCA. In general, women who suffer more frequently any of the behaviours listed in Figure 2 score more highly on the scale, but some behaviours predict more strongly a high score on it. For example, a woman whose partner often makes her feel guilty because she does not understand or take good care of him, or who is portrayed as clumsy and incapable of doing anything right, is strongly predicted to have a high score on the scale. Women whose partners do not want them to work or study, accuse them of flirting with other men, or insist on having sex even when they are not interested, have a high score on the scale in some cases and not in others, producing a lower average score. To simplify the interpretation, we ordered each item by the size of its loading. A clear pattern emerges showing that the scale of sexism captures more clearly attitudes and behaviours that denigrate and subordinate women than possessive behaviours related to feelings of jealously and sexual attraction. The difference is one of degree, as both forms of sexism are correlated.
A second set of independent variables classifies women according to their family or individual levels of socio-economic resources. Assessing the impact of resources on violence is crucial for testing the validity of the family violence perspective and the dependency variant of bargaining theory. We consider different types of resources, demographic, socio-economic and socio-cultural, and make a distinction between those resources that pertain to family and to personal spheres. The distinction is tentative, for family resources empower women as much as women’s resources affect the family’s structural position. The family variables considered were number of children, whether children living in the household are minors or adults, family income (markedly below, not far from, or markedly above Spain’s €900 average in 2006) 2 and whether the household is located in an urban or rural setting. Women’s personal characteristics are age, marital status (single, divorced/separated, married, other), years with current partner, employment status, education (less than secondary, secondary and more than secondary), native or immigrant status, years living in Spain, religious denomination and regular attendance to religious services. All covariates (number of children, age, years with partner and in Spain) were standardized so as to compare their effects based on changes in the same units of standard deviation.
The final set of independent variables measured women’s structural resources relative to partners’ resources. Estimating the association between women’s relative resources and violence is important for assessing the exchange variant of bargaining theory and status inconsistency theory. The variables are woman’s education, age, work status and breadwinning status relative to the partner (that is, whether her status is lower, about the same or higher than his).
To assess the association between the independent variables and violence, we employ logistic regression. To make the model linear, the dependent variable is transformed into the natural logarithm of the odds (logit) of suffering violence. The odds express the probability of experiencing violence relative to not doing it. The regression coefficients express changes in the logit per unit change in each independent variable. The exponent or number e (2.718) raised to the power of the coefficient expresses how much the odds of experiencing violence are multiplied per every unit change in the independent variables.
Results
Table 1 presents the results of regressing the logit of experiencing violence on the independent variables used to operationalize the five theories. We did this stepwise, adding variables in each consecutive model, so as to infer the inter-correlations among the independent variables and contrast them against the theories’ predictions.
Factors associated with women’s odds of experiencing violence in intimate relations.Estimates from logistic regression.
n = 24,131
Significant at the 0.01 level; ** significant at the 0.05 level
Model 1 shows the estimated impact of experiencing male partner’s sexism (as measured by our index of sexism) on women’s logits (b) and odds (eb) of suffering violence. The odds of experiencing violence are multiplied 2.88 times for every unit of standard deviation that a woman’s partner scores above the mean in the normalized index of sexism (mean = 0; standard deviation = 1).
As hypothesized by the sexism approach, partners’ sexism is strongly and significantly associated with violence. However, we cannot deduce ‘quasi-necessity’ from this association without first testing if other variables independently affect violence (net of sexism). Models 2 and 3 show that violence is indeed independently associated with family (model 2) and women’s (model 3) socio-economic resources. While sexism’s association with violence weakens after conditioning for socio-economic factors, it does so only slightly, showing that the two sets of variables affect violence mostly independently of each other. In sum, despite the indisputably key role that sexism plays on violence, the results refute the H1 expectation that sexism is a quasi-necessary condition to suffer violence and that the role of other variables is to specify its conditions of operation.
The family violence perspective and dependency variant of bargaining theory posit that resources are crucial for explaining violence against women in intimate relations. Family violence focuses on the effect of household’s low resources on violence. The effect is thought to be exogenous and operate partly through sexism. We formalized this approach in H2, and proposed to operationalize family conditions via resources/constraints that appear to pertain to the household rather than woman’s personal sphere. A key claim of family violence is that violence against women is more common in low-income households because poor living conditions generate frustration, anger and anomy among family members (Gelles, 1974; Gelles and Straus, 1988). Because households with many children are more likely to experience financial strains, they are expected to display higher rates of violence (Fagan et al., 1983; Straus et al., 1980). The strains increase when children are young due to higher childcare expenses, especially when the mother works (Lloyd and Taluc, 1999; Renner, 2009), and decrease when children reach working age. The probability of violence is expected to increase and decrease correspondingly (Grigsby, 1989; Tienda and Angel, 1982). In addition to financial constraints, a degraded ecology is also expected to raise the odds of abuse. This ecology is more frequent in urban environments where social controls are weaker and anomic conditions thrive (Fox et al., 2002; Umberson, 1987).
As shown in Model 2, only two of these effects are significant: women with more children have higher odds of experiencing violence, while women living with children older than 16 have lower odds. The main risk factors that family violence links to violence in intimate relations – low family income, presence of minors, an urban habitat – are non-significant. The absence of significant effects of low income on violence is especially challenging for the claim of family violence theory that having more children fosters violence because it increases family’s financial burden. If such were the case, the coefficient for low income should be significant, while the one for number of children should not be significant after controlling for income. 3 We observed the opposite. The number of children and low income are positively correlated (as predicted by family violence and shown in the smaller effect of number of children after entering low income), but the association is weak and explains little of why women with more children have higher odds of experiencing violence.
Similarly, if the preventative effects on violence of living with adult children were due to family’s lower financial stress, income should be positively associated with sharing a residence with adult children and the presence of adult children should weaken after controlling for income. We observe neither. The results are inconsistent with family violence predictions. The change in the association between sexism and violence after considering family conditions is also inconsistent. Compare Model 2 to Model 1. It shows that sexism and family conditions are significantly correlated, if weakly (number of children is positively correlated with sexism; adult children’s presence, negatively). This low correlation disputes H2’s claim that sexism’s effect on violence should vanish after controlling for family constraints because it is endogenous to them. In sum, the results are inconsistent with family violence theory, as we formalized it.
In contrast, the results of Model 3 are consistent with dependency theory, as formalized in H3.1, although as shown below they are inconsistent with the results in Model 4. According to H3.1, women’s socio-economic dependence and disempowerment, fostered by low personal resources is, with sexism, the means through which patriarchy expresses its violent nature. Violence could ensue either from sexism or dependency but also from both because, among other reasons, women’s low resources may thwart their capacity to challenge the culture of sexism permeating gender relations and nurturing violence. Three results give credence to this hypothesis.
Firstly, dependency theory offers a more plausible interpretation of the effects of number of children and adult children’s presence on the odds of suffering violence, and of their associations with sexism, than the one given by family violence. This alternative explanation sees these factors as conferring on women different capacities to avoid entering abusive relationships or escape from them. Dependency theorists argue that the over-representation of women with more children (especially young children) among victims of domestic violence ensues from women’s concerns about their well-being after a hypothetical separation (Acevedo, 2000; Lloyd and Taluc, 1999). Some even suggest that causation may run in the opposite direction, with large families resulting from sexual violence (Jewkes, 2002; Vives-Cases, 2009).
Secondly, several variables measuring women’s personal resources are significantly associated with sexism, but do not exhaust its impact on violence. The associations are weak (note the small change in odds of suffering violence after adding women’s resources in Model 3), but in the expected direction. This is consistent with dependency theory’s claim that low resources increase women’s vulnerability to a sexist culture that breeds violence in intimate relations. It is inconsistent with family violence theory’s claims that there should be no such association if these variables were proxies of family conditions, and that sexism’s effect on violence should disappear after conditioning for family’s poor conditions, because sexism is endogenous to them.
Finally, most of the associations between women’s personal resources and violence are as predicted by dependency theory: lower resources increase the odds of suffering violence. The results in Model 3 show that a divorced/separated status is positively and significantly associated with violence. 4 Divorced women’s higher odds of experiencing violence relative to married women have been documented in other countries (Bachman and Saltzman, 1995; Eisikovits et al., 2004). 5 The positive effect of a divorced/separated status on violence could be due to a shrink of the ‘marriage markets’ because of the older ages at which divorcees searched and found their current partners, forcing them to accept less favourable partners. Indeed, the mean age at which divorced/separated women started their current relationship was 36, as against 23 and 22 for married and single women, respectively. Other reasons could be that divorcees are socially ‘stigmatized’ or that they experience a loss in social capital after the break-up of their marriage. All interpretations share a common view that a divorced/separated status poses a handicap on women’s ability to oppose violence.
Model 3 shows that immigrant women have significantly higher odds of victimization. These higher odds have been documented extensively (Raj and Silverman, 2002), and interpreted as resulting from low social capital and isolation (Kasturirangan et al., 2004; Menjívar and Salcido, 2002). Isolation inhibits the effective operation of social controls that reinforce and internalize norms of conventional, non-violent behaviour (Umberson, 1987). It limits migrant women’s ability to live independently, increasing vulnerability to the violent manifestations of patriarchal society (Bui, 2003; Mehrotra, 1999). Finally, smaller social networks make it harder for victimized women to seek and obtain support (Raj and Silverman, 2002).
That the over-representation of immigrants among victims of violence is due to low social capital found additional support by the significantly negative association in Model 3 between attendance to Catholic religious services and violence. The few scholars who have explored the association between religion and violence have focused on the role that participation in religious activities has in extending women’s social networks of support, fostering integration and decreasing isolation (Ellison et al., 1999). Our results support this view: the odds of practicing violence for Catholics are 1/0.59, that is, 1.70 times lower than that of the non-religious.
Model 3 shows no significant impact of age on violence. This is likely due to multicollinearity: a strong, positive correlation between age and years with current partner (r = 0.90). If this latter variable is omitted, a positive and significant association between age and violence emerges, possibly because older women’s dependence on partners increases at older ages. This challenges predictions that older women should experience less violence due to the well-known criminological finding that crime and victimization decrease with age, also with respect to domestic violence (Hotaling and Sugarman, 1986; Lloyd and Taluc, 1999; Straus et al., 1980, but see Jewkes, 2002, for alternative references).
To recapitulate: Model 3 supports H3.1 in that certain socio-economic conditions (having many children, not living with an adult child, being older, divorced, or immigrant, or not practicing religious services) increase women’s odds of suffering violence by intimate partners. We find these results consistent with an explanation that stresses the quasi-necessary role played in fostering violence by a patriarchal system of power manifesting in women’s socio-economic dependence or/and sexism. The drop in sexism’s association with violence from Model 1 to Models 2 and 3 shows that there is a (weak) association between the socio-economic and cultural dimensions of patriarchy. Dependence diminishes women’s capacity to halt sexism, which in turn thrives under conditions of dependence.
Model 4 explores whether women’s ability to oppose patriarchy depends as much on their resources as on how they stand relative to their male partners’. Model 4 adds several variables, expressing differences between women and their partners in key socio-economic characteristics. As noted, two approaches focus on women’s relative resources and violence: exchange and status inconsistency theories. We formalized them into two alternative hypotheses: H3.2 and H4. Our data support more clearly H4 than H3.2: status inconsistency rather than exchange theory.
Like others before (Gelles, 1974; Hoffman et al., 1994; O’Brien, 1971), we find that women who are the main breadwinners or earn as much as their partners experience higher odds of experiencing violence, not lower as predicted in H3.2. We find a small, positive and non-significant effect of educational inconsistencies on violence, 6 regardless of whether they are measured continuously (as differences in partner’s years of study) or categorically (as a dummy variable such as the one shown in Model 4 measuring whether or not woman’s educational level is similar/higher than her partner’s). A plausible explanation for this lack of significant effect is that women’s educational attainment does not threaten men’s status because women choose fields of study that lead to jobs with less decision-making power (Hakim, 1996).
In contrast, women with higher earnings than their partner have significantly higher odds of being treated violently by him. Income inconsistencies favouring women are more significantly associated with violence than educational inconsistencies. The results in Model 4 suggest that men’s decision-making power is additionally challenged when women are older than men. The odds of suffering violence are larger when women are older than their partners than when they are not (net of woman’s age). Other researchers have similarly found that husbands who are older than their wives are less likely to hit them than if they are younger (Luke et al., 2007), perhaps because in patriarchal societies marrying a younger husband challenges man’s power to select more attractive (younger) mates. This interpretation is supported by the decline in Model 4 in the impact that a divorced/separated status had on violence in Model 3, which becomes non-significant after considering woman’s age relative to her partner. The apparently higher vulnerability of separated women to domestic violence, which we interpreted as being possibly due to shrinking marriage markets, stigma or social isolation, was indeed the result of separated women being more likely to be older than their partners, and the higher probability that older women have to experience violence.
The results of Model 4 further show that in couples experiencing status inconsistencies men display slightly less sexist attitudes than in traditional couples. The increase in Model 4 in the effect of sexism on the odds of suffering violence after adding the effect of status inconsistencies speaks of a negative association between the two factors. It was detected also when income and age inconsistencies were entered separately, suggesting that some of the starkest expressions of patriarchy may occur when its cultural, sexist system of control is less effective. The association is weak but certainly not positive (status inconsistencies are not more common in couples with sexist partners). The results of Model 4 refute H1’s expectations that sexism plays a necessary role in explaining violence against women in intimate relations, although its role is important. The results further refute H3.1, as dependency is not necessary to experience violence, although it does contribute to it. It was found to be true that women who are in precarious socio-economic conditions tend to outperform their partners (the coefficient for divorced/separated becomes smaller and insignificant, and the coefficient for practicing Catholic, larger, after considering status inconsistencies), but the impact of status inconsistencies on violence remains positive even after discounting this association. Finally, the results in Model 4 disprove H3.2 that women’s subordination to partners is at the heart of domestic violence and one of its main predictors. On the contrary, subordination reduces the odds of suffering violence, suggesting that a woman’s compliance with patriarchy’s traditional order reduces a partner’s resort to punishment, while sparking a violent reaction when challenged.
The results support H4 and hence status inconsistency theory, provided it is understood, as proposed here, as a complex theory of patriarchy whereby violence is not just the direct expression of patriarchal power but also the consequence of sexism and socio-economic dependency. Patriarchy stands on several legs. In its ordinary manifestation, it consists of sexist and diminishing ideologies that rely on using or threatening to use violence on women to be credible. Women’s empowerment can partly neutralize these ideologies and, more generally, shake patriarchy’s second leg, that of women’s socio-economic dependency, which makes them vulnerable to males’ violence. Finally, on rarer occasions patriarchy takes a starkly violent form that does not require a sexist justification or women’s disempowerment. It occurs when woman’s status is higher than men’s and challenges his superiority. Here violence aims at restoring the patriarchal order without showing much concern for legitimizing or maintaining it.
Conclusions
In this paper we aimed to assess the ability of five theories to account for women’s odds of being threatened or subject to physical violence in intimate relations. The theories overlap with each other, making it difficult to assess their validity. This has led many theorists to conclude that violence against women is a complex phenomenon that must be approached from multiple perspectives. Regrettably, this conclusion may discourage contrast of the theories, for stating that all theories work is somewhat tantamount to admitting that none do. Multidimensional explanations should not discourage the contrast of theories’ core elements and corollaries. While most theories share the same elements, they differ in others. Identifying similarities and differences helps advance theory building.
In this paper we aimed to do this by analysing the logical claims made by five theories explaining violence against women in intimate relations. This analysis allowed us to raise hypotheses about the validity of their alternative explanatory models, which we tested on a large-scale victimization survey conducted in Spain in 2006. In general, we found support for theories that stress the role of patriarchy in fostering violence, and little support for family violence perspectives that claim that stressful family conditions nurture violence.
Our analytical strategy allowed us to identify which patriarchy model, among four increasingly complex ones, best fit the data. We showed that theories that focus primarily on patriarchy-as-sexism correctly stress the key role of a sexist culture that is denigratory to women as a breeding ground for violence, but neglect the analysis of other ways in which patriarchy leads to violence without sexism’s concourse, thus failing to prove that it is necessary. We found that a more complex theory that sees patriarchy as a bi-dimensional system of control where woman’s socio-economic dependency (low personal resources) combines with sexism does a better job in predicting violence. However, we also found that dependency theory fails to show that this system of control is necessary to experience violence or, as exchange theory claims, that it is an expression of a general system of power that seeks to subordinate women by increasing their exposure to men’s violence. Instead we found that, in agreement with status inconsistency theory, subordination (a woman’s inferior status relative to her male partner) leads to lower, not higher, odds of experiencing violence. Our results support status inconsistency theory, understood as a complex theory of patriarchy based on three pillars and in which violence is key for maintaining and restoring the patriarchal balance of power. This balance is normally sustained by a cultural system of domination based on sexism, in which the use of violence makes it credible. It turns into blatant coercion when the costs of male aggression are low or traditional gender roles are disrupted.
Our results contribute to a better understanding of violence against women in intimate relations. This phenomenon is multifaceted. We identified its key aspects and how they operate within a complex model of patriarchy. The results are relevant for public policy. They show that to prevent violence against women in intimate relations it is necessary to combat the sexist culture that breeds it and to empower women with higher resources. However, the results also show that any policy that challenges the patriarchal system of power that nourishes violence may spark violent reactions from insecure men. Eroding patriarchy is possible, as shown in our results by the many couples that are not under its influence in contemporary Spain. However, authorities must be aware that women’s risks of experiencing violence are likely to increase in the transition from a patriarchal to a more egalitarian system of gender roles.
Footnotes
Acknowledgements
We thank the Spanish ‘Instituto de la Mujer’ for making available to us the data on which this research is based. Finally, we thank Elena Larrauri, the members of the Criminology and Criminal Justice Research Group of the Universitat Pompeu Fabra, the Editors of the European Journal of Criminology and two anonymous reviewers for their useful comments and suggestions.
Funding
This research was supported by the Spanish ‘Ministerio de Ciencia e Innovación’ (grant DER2009-08344 - Género y marginación: victimizacion y delincuencia).