Prevalence and Predictors of Bidirectional Violence in Survivors of Intimate Partner Violence Residing at Shelters

Abstract

There has been a long-standing debate regarding whether or not there is gender symmetry in intimate partner violence (IPV); however, shelter samples have been understudied thus far. This study investigates the prevalence and predictors of IPV perpetration in a sample of 227 women in battered women's shelters. Participants were asked to complete a number of measures assessing demographics, Diagnostic and Statistical Manual of Mental Disorders (4th ed., text rev.; DSM-IV-TR) diagnoses, traumatic life events, and perpetration and victimization of IPV. Although the vast majority of women in this sample (93%) report perpetrating some form of IPV, few women endorsed violence that was not mutual (5.3%). Furthermore, for every type of IPV assessed, women were victimized significantly more than they perpetrated. Results also indicate that women's perpetration of IPV, and predictors of such perpetration, varied across type, severity, and measurement of violence. However, most IPV outcome variables were predicted by women's experience of victimization. Taken as a whole, these results support the assertion that context matters when examining the relative rates of perpetration as well as its predictors.

The question of gender symmetry in intimate partner violence (IPV) has long existed, with some family violence researchers asserting that there are equal rates of IPV among men and women (e.g., Gelles, 1974; Straus, 1971), and many feminist researchers positing that men perpetrate a disproportionately higher rate of IPV relative to their female counterparts (e.g., Martin, 1981). This debate continues today as research has demonstrated support for both gender symmetry (e.g., Straus & Gelles, 1986; Teten, Sherman, & Han, 2009) and gender disproportion (e.g., Mooney, 1994; Selic, Pesjak, & Kersnik, 2011). Regardless of whether there is symmetry, research does suggest that women are perpetrating IPV to some degree (Archer, 2000). Furthermore, women's use of violence in retaliation to ongoing abuse is associated with an escalation of violence and increased safety concerns for abused women and their children (Archer, 2000; Hamberger & Larsen, 2015). Thus, it is clearly important that we better understand the factors that predict women's perpetration of IPV. The current study seeks to extend this research by investigating the prevalence of bidirectional IPV in women in shelters as well as exploring what predicts women's perpetration of IPV. For the current study, bidirectional violence is defined as mutual engagement in violence by both intimate partners.

Limitations of Research on Women's Perpetration of IPV

Variation in Sample Type

Researchers propose multiple explanations for the inconsistency in research findings regarding the degree women perpetrate IPV. The source of data, or sample type, has been proposed as a primary reason for inconsistent findings. Archer (2000) conducted a meta-analysis of IPV prevalence across a variety of sample types. Results demonstrated that when all sample types (e.g., students, community, treatment, refuge) were collapsed in a single analysis, women were significantly, albeit only slightly, more likely to perpetrate an act of physical aggression than their male counterparts. When frequencies of perpetration were broken down by sample type, student samples, which comprised the majority of samples in the meta-analysis, had effect sizes in the female direction. However, community samples had effect sizes in the male direction or had no sex differences, and refuge (i.e., battered women's shelters) samples and samples in which couples were undergoing counseling for relationship problems each had effect sizes in the male direction. This highlights the significant influence of sample type as well as the role oversampling students may play in misrepresenting women's use of violence in intimate relationships.

Notably, in contrast to the above literature, which mixed clinical and student samples, a recent meta-analysis conducted by Hamberger and Larsen included only peer-reviewed studies that recruited from a clinical or service setting (Hamberger & Larsen, 2015; Larsen & Hamberger, 2015). Based on the identified studies, the authors conclude that although there seems to be gender symmetry in terms of the perpetration of IPV, there are important differences in the motivation and consequences of IPV. Specifically, women's use of physical violence is often in retaliation to violence they experience. In addition, men more often use tactics that are life threatening, causing women to be more often injured and fearful, and to suffer greater consequences to their sense of autonomy within the IPV relationship. In addition, men were more likely to engage in sexually abusive behavior (Hamberger & Larsen, 2015).

Directionality of IPV

In addition to oversampling students, another limitation of research on women's perpetration of IPV is that the samples do not provide context regarding whether the violence being examined is unidirectional or bidirectional in nature. A review of the literature considering the mutuality of IPV found that 57.5% of reported IPV was bidirectional (Langhinrichsen-Rohling, Misra, Selwyn, & Rohling, 2012). Although the overall rate of IPV endorsed varied by sample type, the proportion of the reported IPV that was bidirectional did not. Somewhat consistent with the results of the aforementioned meta-analytic review (Archer, 2000), the majority of the unidirectional violence in select samples (e.g., military treatment/legal samples) was perpetrated by men (Langhinrichsen-Rohling et al., 2012). However, in epidemiological, community, and student samples, women perpetrated more of the unidirectional violence, and in female treatment-seeking samples, the rates were equal. It is worth noting that no battered women's shelter samples were included. Swan and Snow (2002) conducted a study including women residing at battered women's shelters. They found that 50% of participants endorsed bidirectional violence, with 34% and 12% reporting unidirectional violence perpetrated by men and unidirectional violence perpetrated by women, respectively. However, it should be noted that in addition to women residing in battered women's shelters, their sample included women recruited through a large inner-city health clinic and family court and women who had been arrested for perpetrating IPV. Consequently, their results do not shed light on the unique experiences of women residing at shelters specifically.

Measure of IPV

The various means of measuring IPV is another potential explanation for inconsistent findings related to women's perpetration of IPV. Although the most widely used self-report assessment of IPV is the Conflict Tactics Scale (CTS; Straus, Hamby, Boney-McCoy, & Sugarman, 1996), other measures have been implemented as well. Examples include the Sexual Experiences Survey (SES; Koss, Gidycz, & Wisniewski, 1987), the Psychological Maltreatment of Women Inventory (PMWI; Tolman, 1989), as well as single-item measures (Langhinrichsen-Rohling et al., 2012; Swan & Snow, 2002). Even among studies that have implemented the CTS, there is considerable variability. One potential reason for this variability is the existence of multiple versions of the CTS. The current version of the CTS, the Revised Conflict Tactics Scale (CTS-2; Straus et al., 1996), revised the original scales (i.e., Physical Assault, Psychological Aggression, and Negotiation) and added two new scales, Injury and Sexual Coercion. In addition, the total number of items increased from 19 to 39, therefore capturing a greater variety of behaviors in its newest version (Straus et al., 1996). Consequently, studies that used the original version, such as those sampled for Archer's (2000) meta-analysis, were solely assessing physical aggression. Furthermore, the CTS can be scored in a multitude of ways, including prevalence (i.e., whether or not the participant endorsed IPV) and chronicity (i.e., the number of times acts of aggression occurred), and can assess across different time frames. There is support for the assertion that different methods of scoring result in different conclusions. Specifically, in the aforementioned meta-analysis, it was found that although, across samples, women were more likely to perpetrate at least one physical act of aggression, men were more likely to inflict injury (Archer, 2000). Thus, reports of gender symmetry in IPV may over-simplify our understanding of women's perpetration of IPV. As has been demonstrated, sample type, directionality of IPV, and measures used to assess IPV significantly affect results regarding prevalence rates of IPV perpetrated by men and women, and, consequently, convolute any effort to assess the gender symmetry debate. Thus, future research needs to be more intentional and provide a more specific context in which to examine the relative rates of IPV victimization and perpetration in women.

Predictors of Violence Perpetrated by Women

As previously mentioned, despite the disagreement regarding the extent to which women perpetrate IPV, there is evidence that women are, to some degree, using violence in intimate relationships. Consequently, parties from both sides of this debate have a vested interest in ascertaining what factors put women at risk for doing so. It is critical that researchers explore predictors of women's perpetration of IPV to contextualize previous findings, and ultimately reduce women's risk of experiencing and perpetrating violence. To date, the literature (e.g., Iritani et al., 2013; Mappin, Dawson, Gresswell, & Beckley, 2013; Shorey et al., 2012) has explored a wide variety of predictors of women's use of IPV, including (a) demographic variables, (b) previous and/or concurrent victimization, and (c) psychological disorders.

Demographic Variables

Previous research has identified race/ethnicity as one predictor of women's perpetration of IPV. Some studies have found that Latina, African American, and Native American women perpetrate IPV at higher rates than White or Asian American women (e.g., Iritani et al., 2013). However, other studies have found race to be unhelpful in predicting IPV perpetration (Mennicke & Wilke, 2015), and have instead noted that other factors such as low socioeconomic status and education level (Buttell, Wong, & Powers, 2012) may be risk factors. Age has been found to be a significant predictor such that younger women endorse perpetrating at higher rates (Iritani et al., 2013), though other authors have noted that women's perpetration of IPV may increase from adolescence through young adulthood (Johnson, Giordano, Manning, & Longmore, 2015). In addition, other authors have found no significant effects related to age (Mennicke & Wilke, 2015). Another demographic variable that has been examined is relationship status. Specifically, Iritani and colleagues (2013) as well as Mennicke and Wilke (2015) found that women who were married to their partners perpetrated greater levels of IPV.

Findings that some women of color, women of low socioeconomic status, and women with lower education levels may perpetrate IPV at relatively higher rates is consistent with power and control theories (Straus, Gelles, & Steinmetz, 1980), which posit that lack of resources and/or perceived control may result in increased perpetration as a means of regaining a sense of power. Although this theory was originally applied to men's perpetration, more recently authors have posited that “the same issues of gender-related power and status can explain how young women, as well as young men engage in behaviors (e.g., [IPV]) in order to gain power and status in their relationship” (Dardis, Dixon, Edwards, & Turchik, 2015, p. 141). Consequently, it is possible that women of marginalized identities, who may understandably perceive themselves to have less power and control, could perpetrate at higher rates.

Victimization

There is empirical research to suggest that both childhood abuse (e.g., Iritani et al., 2013; Mappin et al., 2013) and IPV victimization (e.g., Graves, Sechrist, White, & Paradise, 2005; Schumm, O'Farrell, Murphy, Murphy, & Muchowski, 2011) may be predictors of women's IPV perpetration. This is consistent with social learning theories, which posit that witnessing and experiencing abuse results in victims learning to enact these same behaviors (Bandura, 1977; Curtis, 1963). However, upon closer inspection, it is evident that the relationships may be complex. For example, using a community sample of women known to have perpetrated IPV, Caldwell, Swan, Allen, Sullivan, and Snow (2009) controlled for various experiences of IPV victimization and found that only physical IPV, not sexual, psychological, or coercive control, predicted unique variance in IPV perpetration above and beyond the others. Thus, the type of IPV (i.e., physical vs. psychological) may play an important role in understanding the variability within the current literature.

Psychological Disorders

Within the general category of psychological disorders, alcohol and other substance use disorders have received considerable attention when exploring women's use of violence (e.g., Crane, Oberleitner, Devine, & Easton, 2014; Iritani et al., 2013; Larsen & Hamberger, 2015). With regard to alcohol use, it has been demonstrated that women who drink excessively, either occasionally or frequently, and women who have an alcohol use disorder perpetrate IPV at higher rates than women who do not (e.g., Crane et al., 2014; Iritani et al., 2013). However, other studies have demonstrated no such relationship (e.g., Shorey et al., 2012). To address mixed results in the literature, Foran and O'Leary (2008) conducted a meta-analysis, and found that the positive association between alcohol use and IPV perpetration is significant, but the effect size is small. That is to say that although women's alcohol use is predictive of their IPV perpetration, it is not sufficient in explaining their perpetration. There is also considerable support for drug use disorder as a risk factor for women perpetrating IPV (Schumm et al., 2011; Shorey et al., 2012). However, based on their more recent meta-analysis, Larsen and Hamberger (2015) conclude that the relationship between alcohol, drugs, and IPV remains unclear due to inconsistent findings and methodological problems, including unclear and inconsistent definitions of “use” or “abuse.”

In addition to the risk substance use disorders pose for women's IPV perpetration, there is support for the role of other psychological disorders. Shorey and colleagues (2012) conducted a study using a sample of women court mandated to batterer intervention programs that examined a wide variety of mental health diagnoses. They found that probable diagnoses of major depressive disorder (MDD), posttraumatic stress disorder (PTSD), generalized anxiety disorder (GAD), social phobia, borderline personality disorder (BPD), and antisocial personality disorder were all associated with higher levels of IPV perpetration. No such effect was found for panic disorder. Similarly, symptoms of MDD, bipolar disorder, and BPD were found to predict perpetration in a sample of 281 women who had been convicted of IPV against an opposite-sex partner (Henning, Jones, & Holdford, 2003). Furthermore, the positive association between depression and IPV perpetration was also found in a study utilizing a sample of largely low-income African American women. However, in the final model, which included victimization history, PTSD, and various anger styles, the relationship disappeared (Swan, Gambone, Fields, Sullivan, & Snow, 2005). These results demonstrate the complexity of exploring predictors of IPV perpetrated by women and the importance of continued research.

Addressing Limitations of the Existing Literature Within the Current Study

Arguably, the most compelling conclusion from the existing literature on both rates and predictors of IPV perpetrated by women is that the context is immeasurably important. Although “context” can refer to situational context of IPV (e.g., setting, who else is present, who initiates, use of drugs and alcohol), it can also refer to methodological context (e.g., sample type, the way in which IPV is operationalized and measured, etc.). The latter is what the authors are referring to when discussing “context” throughout this article. Specifically, there is evidence to suggest that sample type significantly affects results regarding women's perpetration of IPV. One population that has not received adequate attention, with regards to ascertaining the prevalence and predictors of women's perpetration, is residents of battered women's shelters. Although samples of battered women's shelter residents have been used in a few studies to date, it should be noted that their perpetration of IPV has rarely been studied in a clear context. Women in shelter constitute a unique population due to the high level of severe violence they have commonly experienced, their ongoing safety concerns, and their high risk for returning to the abusive relationship (D. M. Johnson & Zlotnick, 2009). Consequently, the rates at which shelter women perpetrate violence, and their reasons and risk factors for doing so, may differ from other populations (e.g., students, women mandated to batterer intervention programming).

As previously established, another important contextual variable is the way in which IPV perpetration is measured. Researchers have suggested that future studies work to expand the definitions of IPV to account for concerns related to power, control, and ongoing patterns of coercion (McCloskey, 2007), and have highlighted the importance of examining the contextual nature of violence perpetration (Kaukinen, Gover, & Hartman, 2012). Consequently, future studies should assess various forms of IPV and, ideally, examine it in multiple ways. Furthermore, as research has demonstrated variable results based on whether IPV is unidirectional versus bidirectional (e.g., Melander, Noel, & Tyler, 2010; Testa, Hoffman, & Leonard, 2011), directionality of IPV may be another important contextual variable to explore in future research.

The current study addresses several limitations and gaps in the previous literature by examining both the rates and predictors of women's perpetration of IPV in a specifically defined context, a sample of women residing at battered women's shelters. This is an essential population to study not only because they are a unique sample that has been understudied historically but also because they are at an acute risk and their own perpetration of IPV may result in increased victimization (Archer, 2000). Furthermore, because these women are all in shelter for their experiences of IPV, any IPV perpetrated by them is inherently bidirectional. Finally, the current study sought to assess IPV in a relatively comprehensive way by using the CTS-2 to assess various forms of IPV perpetration (physical, sexual, psychological), different severities (minor and severe), and multiple scoring methods (prevalence, chronicity). By examining women's perpetration in a single sample type, in which all of the women's perpetration is bidirectional in nature and in which IPV is measured in a multitude of ways, the results of this study are located in a clearer context, unencumbered by qualitative differences that convolute much of this body of literature. Specifically, this study ascertained (a) the frequency of various types of violence perpetrated by women in shelter within the context of their relationship, which resulted in their shelter stay; (b) whether there were significant differences between women's perpetration and victimization on various types of violence; and (c) which variables (i.e., demographic, victimization, psychological disorders) significantly predicted different types of violence perpetrated by the women who comprised this sample.

Method

Participants

Data were collected from 227 women residing in two battered women's shelters in a medium-sized city in the Midwest. Ages ranged from 18.00 to 64.00 years, with a mean age of 35.00 and a median age of 34.00 years. The average length of IPV relationship was 68.40 months, with a standard deviation of 84.08 months and a median length of 36.00 months. A majority (79.7%) of women had at least some of their children in shelter with them. Furthermore, within the current sample, the majority of women met criteria for several current Diagnostic and Statistical Manual of Mental Disorders (4th ed., text rev.; DSM-IV-TR; American Psychiatric Association [APA], 2000) diagnoses, including IPV-related PTSD (68.3%), MDD (47.1%), GAD (14.5%), social phobia (11.9%), alcohol use disorder (8.8%), drug use disorder (7.0%), and BPD (25.1%). Number of DSM-IV-TR diagnoses ranged from 0 to 6 with a mean of 1.80 and standard deviation of 1.20. Additional demographic information can be found in Table 1.

Table 1.

Sample Characteristics (N = 227).

Race/ethnicity
African American	111 (48.9)
Caucasian	93 (41.0)
More than one race	14 (6.2)
Other	9 (4.0)
Hispanic/Latino	19 (8.4)
Sexual orientation
Exclusively heterosexual	209 (92.1)
Sexual minority	18 (7.9)
Highest education obtained
Less than high school	56 (24.7)
Graduated from high school/GED	171 (75.3)
Relationship status
Single, never married	67 (29.5)
Currently/previously married	115 (50.7)
Witnessed IPV as child	132 (58.1)
Childhood physical abuse	84 (37.0)
Childhood sexual abuse	78 (34.4)
Prior lifetime IPV	164 (72.2)

Note. Values are either reported as n (%); race/ethnicity Other: Asian, Native American. GED = General Education Diploma/High School Equivalency Certificate; IPV = intimate partner violence.

Procedure

The relevant Institutional Review Board approved study procedures, and participants were provided informed consent. Recruitment at the shelters was conducted with posted flyers and brochures containing information regarding the research. Participants who reported experiencing IPV in the month prior to entering shelter were scheduled for an interview with research staff. Interviews were conducted in a private space within the shelter by trained research staff under the supervision of a licensed psychologist.

Measures

Demographics

A semi-structured verbal interview was utilized to collect demographic information including age, race/ethnicity, education, marital status, prior IPV, and witnessing of parental/caregiver abuse in childhood.

IPV

The CTS-2 (Straus et al., 1996) is a 78-item paper-and-pencil self-report measure assessing perpetration and victimization of IPV. Within the current study, the CTS-2 was used to assess perpetration and victimization within the month prior to shelter as well as across the relationship that resulted in the necessity for shelter. The CTS-2 consists of acts of physical, psychological, and sexual violence. Sample items include “My partner insulted or swore at me,” “My partner threw something at me that could hurt,” and “My partner used force to make me have oral or anal sex.” Straus and colleagues (1996) reported ranges between .79 and .95 for internal consistency and moderate rates for construct validity. Cronbach's alphas for the current sample are adequate (α = .69-.79). A benefit of the CTS-2 is the variety of scoring options, including chronicity of violence and violence prevalence.

IPV chronicity

For the current study, chronicity of violence was defined as the number of times each behavior occurred during the month prior to shelter. Given the severe nature of the current sample, the time frame of 1 month prior to shelter was utilized to avoid a ceiling effect and increase accuracy in reporting. Furthermore, as previously mentioned, there was significant variable in length of relationship (SD = 84.08), and as such, utilizing the time frame of 1 month prior to shelter allows for standardization of time. Chronicity of violence was created for both perpetration and victimization of IPV. Separate chronicity scores were created for severe psychological violence, minor psychological violence, severe physical violence, minor physical violence, severe sexual violence, and minor sexual violence. Finally, a total score for violence chronicity (victimization) was calculated by summing severe and minor psychological, physical, and sexual victimization to explore the predictive role of violence chronicity on perpetration prevalence (defined below).

IPV prevalence

Finally, within the current study, prevalence of violence was defined as whether or not the participant endorsed perpetrating and/or experiencing IPV across the course of the relationship that brought them to shelter. Consistent with chronicity, separate prevalence scores were created for perpetration and victimization of severe psychological violence, minor psychological violence, severe physical violence, minor physical violence, severe sexual violence, and minor sexual violence.

IPV-related PTSD

The Clinician-Administered PTSD Scale (CAPS; Blake et al., 1995) is a verbal semi-structured interview to assess for probable PTSD. The CAPS utilizes the 17 diagnostic criteria for PTSD from the Diagnostic and Statistical Manual of Mental Disorders (4th ed.; DSM-IV; APA, 1994). Using a 5-point Likert-type scale (0 = never to 4 = daily or almost every day; 0 = absent to 4 = extreme), participants were asked items about the frequency and intensity of reexperiencing, avoidance, and arousal PTSD symptoms in the last month related to the current IPV that brought them to shelter. Combined frequency and intensity scores of three or greater were utilized to determine PTSD criterion. Sample items include “In the last month, have you ever had unwanted memories of the abuse,” “In the last month, have you ever tried to avoid thoughts or feelings about the abuse,” and “In the last month, have you had any problems falling or staying asleep.” The CAPS is a measure with well-established internal reliability (α = .73 to .85) and concurrent validity, with other empirically validated measures of PTSD (Blake et al., 1995). Interrater reliability was assessed in this study for 21 randomly selected CAPS interviews and was found to be strong (κ = .83). The total CAPS scores also exhibited strong internal consistency in the present data set (α = .93).

Psychological disorders

The Mood, Anxiety, and Substance Modules of the Structured Clinical Interview for DSM-IV Axis I Disorders, Patient Edition (SCID-I/P; First, Gibbon, Spitzer, Williams, & Benjamin, 2002) were verbally administered to assess DSM-IV diagnoses of MDD, social phobia, GAD, and substance use disorders. The BPD Module of the Structured Clinical Interview for DSM Disorders–II (SCID-II; First, Gibbon, & Spitzer, 1997) was verbally administered to assess presence or absence of BPD. Number of DSM-IV-TR diagnoses was calculated by creating a count of diagnoses in the SCID-I, SCID-II, and the CAPS. Lobbestael, Leurgans, and Arntz (2011) found adequate rates of interrater reliability for all modules utilized within the current study (.65 to .95). Interrater reliability for the current sample was established via exploration into 16 randomly selected interviews for the most frequent SCID-I/II diagnosis, MDD (κ = .87).

Trauma history

The Trauma History Questionnaire (THQ; Green, 1996) is a paper-and-pencil self-report measure to assess for traumatic events that have occurred in a participant's lifetime. The THQ is comprised of 24 items that ask participants to specify whether an event has occurred in their lifetime, the number of times the event has occurred in their lifetime, and the initial age the event occurred. The items assess for the areas of sexual assault, physical assault, crime, and general trauma. Sample items include “Has anyone ever tried to take something directly from you by using force, threat of force, such as a stick up or mugging?” and “Has anyone, including family members or friends, ever attacked you with a gun, knife, or some other weapon?” The THQ was used in the present study to assess for participants' experience of childhood physical abuse and childhood sexual abuse. Prior research has shown good test–retest reliability for the THQ (Green, 1996).

Statistical Procedures

Given the inconclusive nature of the extant literature as well as the lack of sufficient exploration into the population of battered women's shelter residents, an exploratory approach, evaluating a comprehensive set of variables previous research has identified to be relevant to women's perpetration of violence, is warranted. To further explore gender symmetry in IPV, descriptive statistics were run investigating the mean, standard deviation, and prevalence of women's perpetration and victimization of physical, sexual, and psychological violence reported within the relationship that brought them to shelter via the CTS-2. Furthermore, paired t tests were run comparing the participants' reported perpetration of violence against their partner in the month prior to shelter with their experience of violence by their partner in the month prior to shelter. To explore predictors of IPV perpetrated by women across the relationship that necessitated shelter, categorical prevalence variables were utilized, and a series of theoretically and empirically relevant bivariate correlations were conducted. Following identification of significant correlates, all significantly related variables were added into a series of logistic regressions predicting the prevalence of women's IPV perpetration. All categorical variables were dummy coded prior to analyses.

Results

Descriptive statistics of women's IPV perpetration and victimization can be found in Table 2. Descriptive statistics demonstrate that although the majority of women (93.0%) endorsed perpetrating some form of violence, it was very rare for a woman to endorse perpetrating violence that was not mutual (physical: n = 8, 3.5%; sexual: n = 3, 1.3%; psychological: n = 1, 0.4%). However, it was common for women to endorse victimization of violence that was not mutual (physical: n = 168, 74.0%; sexual: n = 91, 40.1%; psychological: n = 13, 5.7%). Mutuality was defined as the endorsement of both perpetration of violence against their partner and victimization of violence by their partner.

Table 2.

Descriptive Statistics of Violence by Perpetration and Victimization (N = 227).

Type	M (SD)	Prevalence	t
Physical minor			10.01***
Perpetration	6.33 (12.87)	55.5%
Victimization	21.67 (28.25)	87.7%
Physical severe			7.21***
Perpetration	2.29 (10.66)	32.6%
Victimization	14.73 (6.00)	78.0%
Sexual minor			6.68***
Perpetration	1.85 (6.10)	19.3%
Victimization	7.48 (13.66)	52.8%
Sexual severe			6.04***
Perpetration	0.49 (2.73)	6.6%
Victimization	4.93 (11.50)	37.6%
Psychological minor			11.47***
Perpetration	22.16 (20.78)	90.9%
Victimization	37.83 (25.80)	95.4%
Psychological severe			9.80***
Perpetration	5.45 (10.44)	52.3%
Victimization	17.39 (21.37)	84.3%

Note. Reported means (M), standard deviations (SD), and ts are the results of paired t tests comparing victimization and perpetration during the month prior to shelter; prevalence rates are reported as endorsement across the length of the relationship.

^***p < .001.

Results of the paired t tests comparing perpetration with victimization of IPV in the month prior to shelter demonstrated significant mean differences for each category: severe psychological violence, minor psychological violence, severe physical violence, minor physical violence, severe sexual violence, and minor sexual violence (see Table 2 for results). For each category of IPV, women endorsed significantly greater victimization than perpetration. Thus, within the current sample, women were far more likely to experience IPV than commit IPV against their partner in the month prior to shelter.

Results of the series of bivariate correlates explored can be found in Table 3. All significantly related variables were included in subsequent logistic regressions. Logistic regressions were not completed for women's perpetration of severe physical violence, and women's perpetration of minor psychological violence as only violence chronicity (victimization) significantly correlated with severe physical violence, and no variables were significantly correlated with minor psychological violence.

Table 3.

Bivariate Correlations With Prevalence of Women's Perpetration of Violence (N = 227).

	Physical Minor	Physical Severe	Sexual Minor	Sexual Severe	Psychological Minor	Psychological Severe
Age	−.16*	−.12	−.04	.06	−.08	−.15*
Race	.10	−.04	−.11	−.26***	−.03	.00
Relationship status	.14*	.11	.05	−.07	.03	.13*
Education	−.04	−.06	−.09	.06	−.03	−.14*
Childhood IPV	.01	−.09	.02	.07	.02	.07
CPA	.11	.04	.02	−.01	.01	.04
CSA	.03	.05	.05	.07	−.00	.11
Lifetime IPV	−.06	.09	.16*	.14*	.05	.18**
PTSD (IPV)	.00	.04	.05	−.03	.09	.06
MDD	−.13	−.09	.00	−.06	−.05	−.13
GAD	.00	.00	.02	−.16*	.04	.12
Social phobia	.08	.08	.04	−.06	.07	−.00
Alcohol use disorder	.02	.07	.11	−.02	.04	.08
Drug use disorder	−.06	.01	.08	.05	.03	.09
BPD	.06	.07	.16*	-.07	.03	.17**
Violence chronicity	.10	.28***	.15*	.21**	−.09	−.01

Note. Childhood IPV refers to witnessing of IPV between caregivers during childhood; lifetime IPV refers to whether this was the participants' first violent relationship. IPV = intimate partner violence; CPA = childhood physical abuse; CSA = childhood sexual abuse; PTSD = posttraumatic stress disorder; MDD = major depressive disorder; GAD = generalized anxiety disorder; BPD = borderline personality disorder.

^*p < .05.

^**p < .01.

^***p < .001.

A logistic regression analysis was conducted to predict perpetration of minor physical violence using age and relationship status as predictors. A test of the full model against a constant-only model was statistically significant, indicating that the predictors as a set reliably distinguished between women who endorsed perpetrating minor physical violence against their partner and those who did not (see Table 4 for results). The Wald criterion demonstrated that only age made a significant contribution to prediction with younger women more likely to endorse perpetration of minor physical violence. Relationship status was not a significant predictor.

Table 4.

Summary of Logistic Regressions Predicting Women's Perpetration of Violence.

Outcome	Predictor(s)	Nagelkerke R2	B	χ2	df
Physical minor	Age	.05	−0.03*	8.62**	2
	Relationship status		−0.58
Sexual minor	Current BPD	.10	0.70*	15.87**	3
	Lifetime IPV		0.90*
	Violence chronicity		0.05*
Sexual severe	Race	.32	2.36**	41.09***	4
	Current GAD		19.24
	Lifetime IPV		1.11
	Violence chronicity		0.10**
Psych severe	Age	.15	−0.03	26.71***	5
	Relationship status		−0.56
	Education		0.59
	Current BPD		−0.66*
	Lifetime IPV		1.05**

Note. Lifetime IPV refers to whether this was the participants' first violent relationship; education refers to whether the individual graduated from high school. BPD = borderline personality disorder; IPV = intimate partner violence; GAD = generalized anxiety disorder.

^*p < .05.

^**p < .01.

^***p < .001.

A logistic regression analysis was completed to predict perpetration of minor sexual violence using BPD, lifetime IPV, and total violence chronicity (victimization) as predictors. The full model was significant; furthermore, the Wald criterion demonstrated that all variables significantly contributed to the prediction, with women diagnosed with BPD, a prior history of IPV, and total violence chronicity by their partner in the month prior to shelter more likely to endorse perpetration of minor sexual violence (see Table 4 for results).

Next, a logistical regression was conducted predicting perpetration of severe sexual violence exploring race, GAD, lifetime IPV, and total violence chronicity (victimization) as predictors. Results demonstrated that the full model was significant; however, lifetime IPV and GAD were not significant predictors of severe sexual violence. The Wald criterion identified only race and total violence chronicity as significant predictors, with White women and those experiencing more victimization of IPV in the month prior to shelter more likely to endorse perpetration of severe sexual violence (see Table 4 for results).

Finally, a logistic regression was completed to predict perpetration of severe psychological violence using age, relationship status, education status, BPD, and lifetime IPV as predictors. Consistent with the previous analysis, the full model was significant; however, BPD and lifetime IPV were the only significant predictors of perpetrating severe psychological violence (see Table 4 for results). Women without a diagnosis of BPD and with a prior history of IPV were more likely to endorse perpetrating severe psychological violence.

Discussion

The current study adds to the extant literature on women's perpetration of IPV by exploring rates and predictors of various types of violence (i.e., physical, sexual, and psychological) in a sample of residents of battered women's shelters. The results, as a whole, support the assertion that methodological context (e.g., sample type, IPV measurement) is invaluable when examining rates and predictors of IPV perpetration. Although the vast majority of women in this sample report perpetrating some form of IPV (93.0%), it was rare for women to endorse violence that was not mutual (5.3%). Furthermore, for every type of IPV assessed, women were victimized significantly more than they perpetrated. These results are consistent with previous literature that established samples selected for higher levels of violence (e.g., couples in counseling for relationship difficulties, residents of battered women's shelters) yield results in which male partners are perpetrating IPV at higher rates than their female counterparts (e.g., Archer, 2000).

Consistent with rates of perpetration and victimization, predictors varied considerably based on the type of violence assessed (see Table 4). In fact, no two forms of violence had the same set of predictors in the final model. Taken together, the results of the current study can be interpreted as further evidence that sample type and the way in which IPV is measured are relevant contextual factors in understanding women's use of IPV.

Although significant differences were identified in variables that predict women's use of the six IPV outcomes used in the current study, several themes emerged. Most notably, most IPV outcome variables (i.e., minor physical, severe psychological, severe physical, and severe sexual) were predicted by some measure of women's experiences of victimization. This is consistent with previous studies demonstrating IPV victimization as a risk factor for women's perpetration (e.g., Graves et al., 2005; Schumm et al., 2011; Swan et al., 2005) as well as assertions that women's use of IPV may be in response to men's perpetration, as a means of protecting themselves (e.g., Allen, Swan, & Raghavan, 2009).

In addition to variables associated with IPV victimization, age was associated with minor physical violence, and race was associated with severe sexual violence. Specifically, women who were younger and White reported higher rates of perpetration. The results regarding age are consistent with some of the previous literature (e.g., Iritani et al., 2013); however, the results pertaining to race contradict previous findings and theoretical applications. Specifically, previous studies found that either, consistent with applications of power and control theories, racial minority groups endorse higher rates of IPV perpetration (e.g., Iritani et al., 2013) or there were no racial differences (e.g., Buttell et al., 2012).

BPD was interestingly found to be a risk factor for women's perpetration of minor sexual IPV and a protective factor against women's perpetration of severe psychological violence. The former is consistent with previous studies that have demonstrated BPD to be a risk factor for perpetration (e.g., Henning et al., 2003; Shorey et al., 2012) as well as theoretical literature that posits that borderline personality traits may be part of a profile that puts individuals at risk for perpetrating (Dardis et al., 2015). The latter, however, is inconsistent with literature. These inconsistent findings regarding BPD provide further support for the assertion that methodological context matters. It is not possible to extrapolate from this single study the mechanisms that may explain these divergences in findings. Future research should further explore this phenomenon.

The findings from this study have important implications for the literature on women's IPV perpetration as well as practical implications for the prevention and treatment of bidirectional violence. This study is the first, to our knowledge, that comprehensively examines rates of IPV victimization and perpetration as well as predictors in a sample of residents of battered women's shelters. Due to the importance of contextual factors (i.e., sample type, measurement of IPV), results from one study with a particular sample should not be generalized to other populations. Consequently, examining these questions in this particular sample is both novel and essential. Women residing at battered women's shelters commonly experience higher rates of severe abuse and ongoing safety concerns compared with other samples (D. M. Johnson & Zlotnick, 2009). The current study sought to gain information that would contribute to our knowledge of how to best serve women residing at shelters. Notably, hardly any of the perpetration occurred in the absence of victimization and further, experiences of victimization predicted four of the six types of IPV perpetration. These results lend credence to the idea that IPV may occur in a cyclical fashion, in which women respond to IPV victimization by fighting back, which in turn may result in an increased risk for future victimization. While this cannot be concluded, given the cross-sectional design, this preliminary evidence warrants further examination into this possibility. Consequently, the results of this study highlight the need to address women's use of violence and how it may serve to escalate the situation, when safety planning with women residing at shelters. Furthermore, these results suggest a need for therapeutic interventions that address not only women's victimization but also their perpetration to more thoroughly address factors that contribute to women's safety concerns.

The results of the current study should be understood in context of its strengths and limitations. A primary strength of the current study is the focus on a highly relevant (i.e., high levels of violence, ongoing safety concerns, risk of returning to abuser), problematically understudied sample of women residing in battered women's shelters (Johnson & Zlotnick, 2009). This study also utilized multiple methods of data collection (i.e., semi-structured interviews and paper-and-pencil self-report). Although this is largely a strength of the study's design, it can also be interpreted as a limitation because some IPV-related information (history of a prior IPV relationship, witnessing IPV in childhood) was obtained verbally, whereas others (data from CTS-2) were obtained via paper-and-pencil self-report. In addition, assessing a variety of forms of IPV allowed for a more contextual understanding of women's IPV victimization and perpetration. In this way, the use of the CTS-2 can be considered a strength as it allows researchers to examine IPV in a number of ways; however, the exclusive use of the CTS-2 is also a limitation to the current study. Specifically, the CTS-2 does not sufficiently contextualize the violence it measures, and results obtained through using it often differ from data obtained from other sources (e.g., police incident reports, help-seeking statistics, etc.; Dobash, Dobash, Wilson, & Daly, 1992). Furthermore, the current study assessed IPV chronicity in the month prior to shelter, which is a deviation from the 1-year time frame the CTS-2 traditionally assesses. Although this decision was an intentional one given the nature of this particular sample, it does make it more difficult to compare the results of this study with others. Finally, the current study did not utilize collateral reports of violence from participants' partners, and data are cross-sectional, which does not allow for conclusions regarding causation.

Future research should address these limitations by examining this population through a prospective longitudinal design and the use of measures that allow for greater contextualization of the IPV being measured (i.e., use of qualitative or mixed methods studies). One way to accomplish the latter may be through the use of a “calendar method” such as the Timeline Followback Spousal Violence interview (TLFB-SV; Fals-Stewart, Birchler, & Kelley, 2003). The TLFB-SV can improve accuracy of participants' reports of IPV as it uses a calendar as well as various occasions (e.g., birthdays, holidays) to help improve participants' accuracy. In addition, it is administered as a semi-structured interview, and thus allows for the collection of both quantitative and qualitative data, including contextual details participants may offer. Future research might also replicate the level of specificity in the current study with regard to how IPV was measured. Furthermore, specific exploration into other samples of interest (e.g., treatment, community, and student samples) is warranted, as well as including and comparing various sample types within a single study. Finally, studies that may help to elucidate the mechanisms through which experiences of IPV victimization are related to rates of IPV perpetration are needed.

In conclusion, relative rates of IPV perpetration and victimization differed from results of studies using other sample types (e.g., student samples; e.g., Archer, 2000). Specifically, women were victimized significantly more than they perpetrated for all of the ways in which IPV was measured. In addition, the results of the study demonstrate the importance of considering methodological context when examining women's use of violence. Both rates of perpetration and the variables that predicted women's use of violence varied based on the way in which IPV was measured. Notably, however, previous experience of victimization was a significant predictor in nearly all models, suggesting that for women residing at shelters, prior victimization seems to be an especially salient risk factor

Biographies

Samantha C. Holmes, MA, is a doctoral candidate at The University of Akron's Collaborative Program in Counseling Psychology. Her research includes the impact of trauma, broadly, and sexual trauma, specifically, as well as the etiology and presentation of eating disorders.

Nicole L. Johnson, PhD, is an assistant professor at Lehigh University. Her research is dedicated to the exploration and implementation of prevention programming (primary, secondary, and tertiary) for gender-based violence (e.g., rape, intimate partner violence [IPV], trans* violence). Her professional interests include the implementation/evaluation of culturally sensitive treatment, advocacy, outreach, feminist theory and practice (including teaching, supervision, and mentorship), and mixed method research.

Elsa E. Rojas-Ashe, MEd, is a doctoral candidate at The University of Akron's Collaborative Program in Counseling Psychology. She has dedicated much of her graduate training to working with survivors of trauma, including sexual assault, military sexual trauma, and IPV. Her research interests include the impact of trauma, specifically sexual trauma and experiences of trauma in childbirth.

Taylor L. Ceroni, MA, is a doctoral student at The University of Akron's Collaborative Program in Counseling Psychology. Her research interests include trauma and abuse, IPV, posttraumatic stress disorder (PTSD), and feminist issues, specifically in the areas of sexism and empowerment.

Katherine M. Fedele, MA, is a doctoral candidate in counseling psychology at The University of Akron. Her research interests include prevention and treatment of PTSD in diverse trauma victims, vicarious traumatization in crisis workers, posttraumatic growth, and the psychology of oppression.

Dawn M. Johnson, PhD, is a licensed psychologist and an associate professor in the Department of Psychology at The University of Akron. She is also core faculty in the Collaborative Program in Counseling Psychology. A majority of her research has focused on the treatment of PTSD in victims of IPV as well as developing interventions to promote women's sexual health and empowerment, and prevent violence against women.