Article Text
PDF
Abstract
Background There has been an ongoing debate about the extent and nature of gender differences in the experience of intimate partner violence (IPV). Disagreement about the appropriate definition of IPV is central to this debate.
Methods This study used latent class analysis (LCA) to map the patterns of physical violence, sexual coercion, psychological abuse and controlling behaviour, and examined whether LCA can better illuminate the gendered nature of this experience than conventional measures of IPV. Data from the 2004 Canadian General Social Survey were analysed, which included 8360 women and 7056 men 15 years of age and over who reported a current or ex-spouse or common-law partner.
Results Results revealed more variation in the patterns of IPV for women than for men. Six classes were found for women, whereas four classes were found for men. Women and men were equally likely to experience less severe acts of physical aggression that were not embedded in a pattern of control. However, only women experienced a severe and chronic pattern of violence and control involving high levels of fear and injury. For women and men, intermediate patterns of violence and control, and patterns describing exclusively non-physical acts of abuse were also found. The results also revealed substantial differences in the IPV subtypes for those reporting about a current versus an ex-partner.
Conclusion These results support the use of LCA in identifying meaningful patterns of IPV and provide a more nuanced understanding of the role of gender than conventional measures. Implications for sampling within IPV research are discussed.
- Family violence
- domestic violence
- spouse abuse
- Canada
- gender studies SI
- population surveys
- public health epidemiology
- social research
- violence RB
Statistics from Altmetric.com
- Family violence
- domestic violence
- spouse abuse
- Canada
- gender studies SI
- population surveys
- public health epidemiology
- social research
- violence RB
Over the past 20 years, a large body of research has documented the prevalence and health consequences of intimate partner violence (IPV). One of the most contentious issues that has emerged is the extent of women's and men's use of violence in intimate relationships and questions about gender symmetry in the experience of IPV.1–3 Evidence supporting gender symmetry comes primarily from studies involving population-based or community samples that use a narrow definition of IPV, which operationalises partner violence as any act of physical violence.4 These studies find that women are either equally likely or more likely than men to perpetrate an act of physical aggression against a partner. In contrast, feminist research often involves clinical samples of abused women and uses a broad definition of IPV that considers the multitude of ways in which a partner can threaten, intimidate or inflict harm, including the use of physical force, threats of violence, sexual coercion, verbal abuse, financial abuse and acts of coercive control.5–8 This research finds that women are more likely than men to report severe violence, such as having been beaten, choked, or strangled9; sexual assault or sexual coercion10 11; ongoing physical violence, physical injury and more serious types of injuries.11 12 Women are also more likely than men to fear injury or death,11 12 and to be killed by a partner.13 14
In response to the apparent contradictory evidence, Michael Johnson proposed a theory to reconcile these findings. He suggested that this evidence is not in fact contradictory, but rather is describing two qualitatively different patterns of IPV that are strongly conditioned by gender.15–20 The first subtype, called situational couple violence, is hypothesised to be gender symmetric, resulting from situational factors or conflicts that escalate to the use of violence by both partners. However, this aggression is not part of a broader pattern of coercive control in the relationship and is usually (although not always) characterised by less severe and less frequent violence. In contrast, intimate terrorism is hypothesised to be almost exclusively perpetrated by men against female partners. It is described as a chronic pattern of violence and control that is rooted in patriarchal gender norms. Johnson subsequently expanded the typology to include two more subtypes: violent resistance and mutual violent control. Violent resistance is also hypothesised to be gender asymmetric, describing relationships where women use violence against men who are violent and controlling. Mutual violent control is hypothesised to be gender symmetric, describing relationships where both partners are violent and controlling.
Although existing research offers some indirect evidence for Johnson's theory, to date only a few studies have directly tested the validity of the proposed typologies. Johnson's own research has provided some support for the distinction between situational couple violence and intimate terrorism in the USA,17 19 20 and others have found similar evidence in Canada21 and the UK.22 However, although these studies provide some preliminary support for Johnson's theory, more research is needed that flushes out the nature of these typologies23 and that includes both women and men.24
This study used latent class analysis (LCA) to test for gender differences in the patterns of physical violence, sexual coercion, psychological abuse and controlling behaviour. It also provides a preliminary exploration of the construct validity of these patterns by testing their associations with theoretically relevant characteristics including the frequency of violence, fear that one's life was in danger, injury, having to take time off or stay in bed as a result of the violence, and partner alcohol use during the incident(s). Finally, the study examined whether relationship status (ie, current vs ex-partner) influences the types of abuse experiences that are identified.
Methods
Sample
Data from Statistics Canada's 2004 General Social Survey on Victimisation were used, which is a geographically stratified population-based survey of 23 766 non-institutionalised individuals 15 years of age and over living in the ten provinces. One eligible person per household was randomly selected and was interviewed in either English or French via computer-assisted telephone interviewing. The response rate for the survey was 74.5%.25 This study was not conditional on human subjects approval.
This analysis includes violence data from the current or ex-spouse or common-law partner (n=15 979). For those who reported having both a current and an ex-spouse or common-law partner (n=1070), violence data from the relationship in which the respondent reported any type of abuse was selected for the analysis. For those who reported some type of abuse by both partners (n=95), the relationship involving physical or sexual violence was selected, and for those who reported physical or sexual violence by both partners (n=9), the most violent relationship was selected. Those who reported being gay, lesbian, or bisexual (n=187) were excluded given that the gender symmetry debate relates to heterosexual unions. Those with any missing IPV data were also excluded (n=376). Almost two-thirds of these excluded respondents (n=235) had missing data on all IPV items. The final analytic sample size was 15 416, which includes 8360 women and 7056 men.
Measures
IPV and abuse. The emotional and financial abuse module included seven binary items (ie, yes/no) assessing various forms of abuse and control. No timeframe was imposed for this module. For those reporting about a current partner, the items were worded in the present tense. For those reporting about an ex-partner, the items were worded in the past tense. Respondents were asked if their partner perpetrated any of the following acts:
tries (tried) to limit your contact with family or friends,
puts (put) you down or calls (called) you names to make you feel bad,
is (was) jealous and doesn't (didn't) want you to talk to other men/women,
demands (demanded) to know who you are (were) with and where you are (were) at all times,
harms (harmed), or threatens (threatened) to harm, someone close to you,
damages (damaged) or destroys (destroyed) your possessions or property,
prevents (prevented) you from knowing about or having access to the family income, even if you ask (asked).
The physical and sexual violence module included a modified version of the Conflict Tactics Scales.26 These items also had a binary response format (ie, yes/no). Respondents were asked if their partner perpetrated any of the following acts in the previous 5 years:
threatened to hit you with his/her fist or anything else that could have hurt you,
thrown anything at you that could have hurt you,
pushed, grabbed, or shoved you in a way that could have hurt you,
slapped you,
kicked you, bit you, or hit you with his/her fist,
hit you with something that could have hurt you,
beat you,
choked you,
used or threatened to use a gun or knife on you,
forced you into any unwanted sexual activity, by threatening you, holding you down, or hurting you in some way.
Violence-related characteristics. Respondents who reported at least one act of physical or sexual violence were asked whether their partner had been violent on more than one occasion and the number of times the partner had been violent, whether they were ever physically injured in any way during these incident(s), whether they ever had to stay in bed for all or most of a day as a result of the violence, whether they ever had to take time off from their everyday activities as a result of the violence, whether they ever feared that their life was in danger because of their partner's violent or threatening behaviour, and whether their partner was drinking during these incident(s).
Statistical analysis
Latent class analysis (LCA) was conducted using Mplus version 5.1.27 Beginning with a one-class model, models with one additional class were evaluated. The model for the overall sample, ignoring gender, was first identified. The presence of differential measurement of IPV by gender, or differential item functioning (DIF), within the overall model was then tested by conducting a likelihood ratio (LR) χ2 difference test, comparing the fully unrestricted model for women and men to the restricted model that forced the thresholds for the IPV items to be equal across the classes for women and men. Differential measurement of the individual IPV items was also examined by testing the direct effects between gender and each of the items. All analyses were weighted. Influential observations on the parameter estimates and the log likelihood function were also assessed and excluded.
The goodness of fit of the models was evaluated using the Bayesian Information Criteria (BIC),28 the sample size adjusted BIC,29 parametric bootstrapping of the Pearson and LR χ2 statistics,30 and cross-validation on five randomly selected halves of the dataset. The presence of local dependence was assessed using the standardised bivariate residuals. The quality of classification of the classes was examined using the average posterior probability of class membership for respondents assigned to their most likely class as well as the overall entropy statistic. Finally, the face validity of the classes was considered when selecting a model.
For the construct validity analysis, the final latent class models were re-run, treating each violence-related characteristic as an additional indicator in the model. To test for differences in these characteristics across the classes, LR χ2 difference tests were conducted, comparing the unrestricted model that allowed the thresholds for that characteristic to vary across the classes with the restricted model that forced the thresholds to be equal.
Finally, the analysis that examined relationship status (current vs ex-partner) and IPV involved running the final latent class models using the auxiliary variable function in Mplus to estimate the proportion of respondents within each class reporting a current versus an ex-partner. These estimates are obtained by randomly assigning individuals to pseudo-classes based on their posterior probability of class membership. Next, data on the prevalence of the classes, the proportion of respondents reporting a current or an ex-partner, and the sample size were used to convert the results from the auxiliary variable analysis to estimates of the prevalence of the IPV subtypes within the current and ex-partner subgroups.
Results
There were no significant gender differences in the experience of any emotional or financial abuse (p=0.36) or any physical violence (p=0.11) (table 1). However, women were significantly more likely than men to report sexual violence (p<0.001). There were also significant gender differences in the likelihood of reporting most of the individual IPV items, with women being more likely than men to report most of the items except partner jealousy and/or that the partner monitored their whereabouts.
Prevalence of violence and abuse by a current or ex-spouse or common-law partner, 2004 Canadian General Social Survey
Test for differential measurement of IPV by gender
For the overall sample, a six-class model allowing local dependence between the first and third items in one of the classes was selected as the final model (data not shown). The LR χ2 difference test comparing the fully unrestricted six-class model for women and men to the restricted six-class model that forced the thresholds for the IPV items to be equal for women and men was statistically significant (LR χ2=367.2, df=69, p<0.001), indicating the presence of differential measurement of IPV by gender. Moreover, seven out of the twelve direct effects between gender and the IPV items were statistically significant at p<0.05 (data not shown), suggesting gender differences in the likelihood of reporting these items within the classes. These results provide strong evidence that separate models are needed to describe women's and men's experiences of IPV.
Latent class models for women and men
For women, a six-class model allowing local dependence between the first and third items in one of the classes was selected (table 2). This model excludes five influential observations. For men, a four-class model allowing local dependence between the fifth and six items in one of the classes was selected (table 3). This model excludes three influential observations.
Six-class model of intimate partner violence and abuse by a current or ex-spouse or common-law partner for women (n=8355), 2004 Canadian General Social Survey
Four-class model of intimate partner violence and abuse by a current or ex-spouse or common-law partner for men (n=7053), 2004 Canadian General Social Survey
For women, the largest class represents a ‘No violence or abuse’ group and describes 85.1% of the population. The second class, called ‘Jealousy, verbal abuse’, describes 8.1% of the population and is primarily defined by the partner putting them down or calling them names and the partner being jealous or demanding to know who they're with and where they are at all times. The third class, called ‘Physical aggression', describes 2.6% of the population and is primarily defined by relatively less severe acts of physical aggression. About half of women in this class also report that their partner threatened them with violence and that he put them down or called them names. The fourth class, called ‘Severe violence, control, verbal abuse’, describes 1.8% of the population and represents the most severe pattern of violence and control. Women in this class have a high likelihood of reporting all types of violence including having been beaten or choked and threats of violence, as well as jealousy, verbal abuse, different types of controlling behaviour, and having their possessions damaged or destroyed. Over half of women in this class also report that their partner threatened to harm or actually harmed someone close to them and over 30% report actual or threatened weapon use and unwanted sexual activity. The fifth class, called ‘Physical aggression, control, verbal abuse’, describes 1.3% of the population and involves relatively less severe acts of physical aggression, verbal abuse, jealousy and limiting contact with family or friends. About half of women in this class also report that their partner damaged or destroyed their possessions or property and that he threatened them with violence. The sixth class, called ‘Control, verbal abuse’, describes 1.1% of the population and involves verbal abuse and a range of controlling behaviours including threats to harm someone close to them, destruction of their property, and limiting their access to or knowledge of the family income.
For men, the first three classes are comparable to those observed for women. The most prevalent class represents a ‘No violence or abuse’ group and describes 90.3% of the population. The second class, called ‘Jealousy, verbal abuse’, describes 5.3% of the population. The third class, called ‘Physical aggression’, describes 2.8% of the population and is comparable to the results for women in terms of the prevalence and the nature of this pattern. The fourth class, called ‘Moderate violence, control, verbal abuse’ describes 1.5% of the population and involves minor and moderate acts of violence, threats of violence, verbal abuse, controlling behaviour and having their property damaged or destroyed. About one-third of men in this class also report having been beaten or choked and 28% report actual or threatened weapon use.
Violence-related characteristics of the classes
For both women and men, there were statistically significant differences across the classes for all of the violence-related characteristics (table 4). Women in the severe violence class were most likely to report five or more episodes of violence, fearing that their life was in danger, physical injury, having to take time off or stay in bed, and partner alcohol use during the incident(s). Women in the physical aggression class were least likely to report these experiences and women in the intermediate physical aggression class tended to be intermediate for most of these experiences. Men in the moderate violence class were more likely than those in the physical aggression class to report all of these experiences.
Proportion of women and men in the latent classes reporting various violence-related experiences, 2004 Canadian General Social Survey
Although the number of episodes of violence was comparable for women and men in the physical aggression class, women in this class were more likely than men to report fearing that their life was in danger, having been injured, having to take time off or stay in bed, and partner alcohol use. The moderate violence group for men was comparable to the intermediate physical aggression group for women with respect to nearly all of these characteristics.
Relationship status and IPV
Table 5 describes the prevalence of the IPV classes for those reporting about a current versus an ex-partner. About 94.1% of women and 95.2% men in the current partner sample were classified in the no violence or abuse class. In contrast, only 51.4% of women and 66.2% of men in the ex-partner sample were classified in the no violence or abuse class. For both women and men, the more severe patterns of violence and control were only identified in a substantial way in the ex-partner sample.
Prevalence of the latent classes in the current and ex-partner samples for women and men, 2004 Canadian General Social Survey
Discussion
This study was a response to the ongoing definitional debate about the manner in which IPV should be operationalised and the important ways in which choice of definition influences understanding of the role of gender in this experience. Consistent with other population-based studies, this study found that although women and men were equally likely to experience some form of physical aggression by an intimate partner, women were more likely to report sexual coercion. However, the results of the LCA revealed a more nuanced picture of women's and men's experiences of IPV than is suggested by these aggregate-level measures. The findings support the contention that women and men are equally likely to experience relatively less severe and infrequent acts of physical aggression that are not embedded in a pattern of control. Although the nature and the frequency of this aggression were comparable for women and men, the impact was greater for women.
The results also support the contention that women disproportionately experience the most severe and chronic pattern of violence involving highly controlling and threatening behaviour. This violence was the most chronic and severe of all of the patterns that were found in this study, with 70% of women reporting five or more episodes of violence in the past 5 years, and about 80% reporting having been beaten or choked, fearing that their life was in danger, and having been injured. Women experiencing this pattern were also the most likely to report that their partner was drinking during the incident(s).
The results of this study also demonstrate that general population surveys can identify women who experience this most severe pattern of IPV. However, the more severe patterns of violence and control for both men and women were primarily identified in the ex-partner sample. The current partner sample mainly identified those who did not report any violence or abuse. These findings are consistent with a previous analysis by Johnson, Leone and Xu.31 As compared with those who experienced violence by ex-partners, those currently experiencing more severe violence and control may be less likely to agree to participate in public surveys, particularly if they are living with their partner or if their partner is nearby during the interview.32 33 Women may also be difficult to reach if they are highly mobile and excluded entirely from survey sampling frames if they are living in shelters. The higher prevalence of various forms of IPV in the ex-partner sample may also be explained by the association between partner violence, particularly more severe violence, and relationship dissolution.34 35
The study also identified intermediate patterns of violence and control for both women and men, which were comparable with respect to the violence-related characteristics. An important question that cannot be addressed in this study is the extent to which women and men who experienced the IPV subtypes are primarily the victims of abuse, are reciprocating the violence or are primarily the aggressors initiating the violence and coercive control. The survey only asked respondents about their experiences of victimisation and not their perpetration of IPV. There was also no information collected about the motives for the violence, who initiated the violence or the ‘mutuality’ of the violence. This contextual information could have helped in better defining the structure, context and meaning of these classes. Future research on IPV would benefit from collecting this information within surveys.
The study also identified patterns involving exclusively non-physical acts of abuse for women and men, including a pattern for women involving a range of highly controlling and threatening behaviour. Another possible limitation of the survey that may complicate the interpretation of these ‘non-physical’ abuse classes is the use of different timeframes to measure non-physical abuse and physical or sexual violence. No timeframe was used for the emotional and financial abuse module, whereas the physical and sexual violence module focused on occurrences in the past 5 years. Therefore, it is possible that some individuals within these ‘non-physical’ abuse classes may have experienced violence at an earlier point in the relationship.
As previously noted, these data reflect violence in the past 5 years. The survey also excluded those who reported an ex-partner, but who did not have contact with that person in the past 5 years. Therefore, it is likely that these results underestimate the prevalence of violence over the long term and overestimate the prevalence of no violence. In fact, studies of community samples have shown that desistance or cessation of partner physical aggression is common, particularly among those reporting less severe or less frequent violence.36 37
In conclusion, the results of this exploratory analysis suggest that LCA is a useful analytic tool that can describe in a more nuanced manner the full range of violence, abuse and control experiences than measures that focus on the presence or absence of a single dimension of abuse. Research is needed that further validates the patterns that were found in this study and that explores whether they are distinct from each other in terms of their aetiology, health consequences and help-seeking characteristics. Future research could also explore the extent to which the various forms of IPV that were found in this study apply to other settings, including those with different gender norms. Longitudinal research is also needed that examines the course of these IPV experiences over time and the extent to which individuals remain within a particular abuse pattern or progress across subtypes. Evidence supporting the distinction between these patterns of abuse could contribute to a better understanding of the epidemiology of IPV and could potentially better inform decision-making about resource allocation, policy, and practice in relation to IPV.
What is already known on this subject
A large body of research has documented the negative social, economic and health consequences of IPV.
However, researchers continue to debate the appropriate definition of IPV.
The definition used within IPV research is consequential for the way in which we understand women's and men's experience and perpetration of IPV.
What this study adds
This study fills in a research gap by including data on both women's and men's experiences of IPV.
The results highlight the importance of moving beyond aggregate measures of IPV that focus on the presence or absence of a single dimension of abuse.
The findings support the gender symmetry hypothesis for less severe and less frequent aggression that is not embedded in a pattern of control. However, gender asymmetry characterised the most severe and chronic pattern of violence and control, as only women were found to experience this pattern.
The results also have implications for sampling within IPV research. They demonstrate that population-based surveys can identify a range of abuse experiences, including severe violence and control, but only if respondents are asked about abuse by an ex-partner.
Acknowledgments
The first author wishes to thank members of the dissertation committee including Emily Agree, Andrea Gielen, Michelle Hindin and Daniel Webster. Thanks also to Karen Bandeen-Roche for assistance with the analysis. The first author is also grateful for the support provided by Jodi-Anne Brzozowski, Robin Fitzgerald, Anthony Matarazzo, and Roxan Vaillancourt at the Canadian Centre for Justice Statistics.
References
Supplementary materials
Web only data jech.2009.095208
Files in this Data Supplement:
Footnotes
Funding This research was supported by a doctoral research award from the Institute of Population and Public Health at the Canadian Institutes of Health Research and the Tom Symons Research Fellowship at Statistics Canada.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.