Characteristics and recidivism in relation to arrest: differentiating between partner violent perpetrator subtypes

ABSTRACT

This study aimed to describe and compare arrested and non-arrested male intimate partner violence (IPV) perpetrators, in terms of individual characteristics and variables related to the IPV incident, in a sample of 628 perpetrators reported to the Swedish police. We also explored recidivism rates in relation to arrest within the total sample, as well as among subtypes of partner violent men (i.e. generally violent [GV] and partner only [PO] violent). The perpetrators in this longitudinal study were reported to the police for male-to-female perpetrated IPV and subjected to a structured violence risk assessment between 2011 and 2014. The results showed that arrested perpetrators were more likely to be reported for severe forms of IPV, being assessed by the police post-arrest with a higher risk for recidivism, and being more likely to be prosecuted for the reported IPV incident. There was also a significant interaction effect between subtypes and arrest for IPV recidivism within 12 months post-arrest. Among those perpetrators who were arrested, GV perpetrators were more than four times as likely as PO violent perpetrators to recidivate in IPV. The results of this study highlight the importance of differentiating between subtypes when examining the impact of sanctions on IPV recidivism.

KEYWORDS:

• Arrest
• intimate partner violence
• subtypes
• recidivism
• deterrent

Prior findings of the effect of arrest on intimate partner violence (IPV) recidivism are inconclusive (Vigurs, Wire, Myhill, & Gough, 2016). Following the Minneapolis Domestic Violence Experiment (Sherman & Berk, 1984), which showed arrest to be an effective deterrent for IPV recidivism, subsequent replications have failed to replicate this result. For example, some studies have found arrests to increase recidivism among IPV perpetrators (e.g. Dunford, Huizinga, & Elliot, 1990). Moreover, despite the conclusion of Sherman and Berk (1984) that arrest likely works better for specific types of perpetrators, there has still only been one attempt to integrate the research on IPV perpetrator typologies with the arrest literature (Johnson & Goodlin-Fahncke, 2015). Therefore, this study sought to explore differences between arrested and non-arrested perpetrators in terms of demographics, case-related variables and recidivism rates, using both the total sample as well as differentiating between the two most consistently identified subtypes of partner violent men: the partner only (PO) violent subtype and the generally violent (GV) subtype (Petersson & Strand, 2018).

Over the last decades, the arrest of a perpetrator has been a commonly used legal intervention imposed by the criminal justice system in response to IPV. The arrest rates are difficult to compare, however, as countries and states have adopted different arrest policies. Mainly, such policies dictate either mandatory arrests or arrests being made based on the police officer’s discretion. In the United States, where several states have mandatory arrest policies (Xie & Lynch, 2017), 32% of the perpetrators were arrested or detained following a police report of IPV (Tjaden & Thoennes, 2000). The corresponding arrest rate in England and Wales is reportedly 51% (Her Majesty’s Inspectorate of Constabulary and Fire & Rescue Services, 2017). In Sweden, the arrest rate for IPV perpetrators is 25% (National Council for Crime Prevention [NCCP], 2008). Notably, unlike parts of the United States, Sweden does not have mandatory arrest laws for IPV perpetrators. In Sweden, the decisions to arrest are based on the police officer’s discretion.

The use of arrest as a criminal justice sanction imposed on IPV perpetrators received increased attention from scholars and policymakers following the findings of the Minneapolis Domestic Violence Experiment (Sherman & Berk, 1984). In this experiment, police officers responding to IPV incidents randomly applied one of three predetermined interventions: arresting the alleged perpetrator, temporarily separating the couple, or giving some form of advice (e.g. mediation). A 6-month follow-up, using both official recidivism data and victim reports, indicated that IPV perpetrators who had been arrested were less likely to recidivate in such violence compared to their non-arrested counterparts.

However, subsequent attempts to replicate the findings of the Sherman and Berk (1984) study have disputed the deterrent effect of arrest on IPV recidivism. For example, the Minneapolis Domestic Violence Experiment was replicated in five other American cities. These replication attempts, known collectively as the Spousal Assault Replication Program (SARP) studies, failed to uniformly replicate Sherman and Berk’s (1984) findings (e.g. Berk, Campbell, Klap, & Western, 1992; Maxwell, Garner, & Fagan, 2001; Sherman et al., 1992a). Although some studies demonstrated that arrest had a deterrent effect on recidivism (Berk et al., 1992), other studies showed that arrest increased the rates of recidivism (Dunford et al., 1990). In an effort to analyse data from all five replication attempts, results showed that arrest had a modest deterrent effect when measuring recidivism via victim reports (Maxwell et al., 2001). However, the same deterrent effect disappeared when deriving recidivism data from official police records.

Although the examination of the causal effect of arrest on recidivism requires an experimental design, such studies are associated with several legal and ethical challenges. This includes, for example, withholding arrest in those cases where needed (e.g. due to severe victim injuries: Hilton, Harris, & Rice, 2007). As such, the number of eligible cases in the Minneapolis Domestic Violence Experiment was limited due to ethical and legal considerations, such as excluding cases of severe violence where the victim had sustained serious or life-threatening injuries. The number of eligible cases was also limited as the study design required both the victim and the suspect to be present at the scene. Given the abovementioned difficulties, Hilton et al. (2007) argued that non-experimental designs are required for a full understanding of the effect of arrest on recidivism. To this end, non-experimental studies have also produced conflicting results concerning the deterrent effect of arrest on IPV recidivism. For example, studying the efficacy of arrest, protection orders or a combination of both types of interventions, recidivism was found not to differ between types of interventions (Broidy, Albright, & Denman, 2016). Moreover, using longitudinal crime victimization survey data, studies have suggested that the effect of reporting IPV to the police serves as a stronger deterrent for recidivism compared to a perpetrator’s arrest (Felson, Ackerman, & Gallagher, 2005; Xie & Lynch, 2017). However, in another study also using survey data of crime victimization, arresting the perpetrator was found to reduce the odds of recidivism by 45% (Cho & Wilke, 2010). Furthermore, studying police-reported IPV, Hilton et al. (2007) concluded that arrested perpetrators were more likely to recidivate than those perpetrators who were not arrested. However, this result was explained by the police officers’ propensity to arrest perpetrators with a higher risk for IPV recidivism (as measured by violence risk assessments). Thus, drawing on previous studies with different research designs and data, the effects of arrest on IPV recidivism has shown inconsistent results.

These conflicting results could be explained by the heterogeneity among IPV perpetrators. Detailed analyses of the data generated by the SARP studies showed that the effect of arrest on IPV recidivism varied as a function of perpetrator characteristics. For example, some studies reported that arrest was associated with lower recidivism rates among employed and married perpetrators than among unemployed and unmarried perpetrators (Berk et al., 1992; Sherman, Smith, Schmidt, & Rogan, 1992b). More specifically, Sherman et al. (1992b) noted that perpetrators who were unemployed and unmarried were more likely to recidivate in IPV following an arrest, compared to perpetrators who were unemployed and unmarried but not arrested. Conversely, arrested perpetrators who were employed and married were less likely than their non-arrested counterparts to recidivate. In support of the conclusion by Sherman et al. (1992b), other results from the SARP studies showed that arrest was more likely deterrent for perpetrators with less extensive arrest records compared to perpetrators with more extensive arrest records (Sherman, 1992). Taken together, these results indicate that there is a differential effect of arrest on IPV recidivism depending on the type of perpetrator.

Despite previous recognitions that the deterrent effect of legal interventions (e.g. arrest) most likely depends on perpetrator type, as well as the recommendation for future research to investigate this further (e.g. Sherman & Berk, 1984; Willams, 2005), only one prior study has attempted to merge these research domains. Using data from three of the SARP studies, Johnson and Goodlin-Fahncke (2015) clustered perpetrators into the four subtypes previously identified by Holtzworth-Munroe, Meehan, Herron, Rehman, and Stuart (2000: i.e. the family-only, the dysphoric/borderline, the generally violent/antisocial, and the low-level antisocial subtype). Overall, the study found that arrest had no deterrent effect on any of the subtypes (Johnson & Goodlin-Fahncke, 2015). In contrast, an arrest was found to increase IPV recidivism among the generally violent and the low-level antisocial perpetrators.

Although a step in the right direction, a growing body of research (e.g. Cantos & O’Leary, 2014; Petersson & Strand, 2018) has questioned the validity of the four-subtype typology advocated by Holtzworth-Munroe et al. (2000) and used in the Johnson and Goodlin-Fahncke (2015) study. This four-subtype solution was the result of an attempt to validate Holtzworth-Munroe and Stuart’s (1994) proposed typology consisting of three subtypes (i.e. family-only, borderline/dysphoric and generally violent/antisocial). These subtypes were hypothesized to differ on three dimensions; the severity and frequency of violence, the generality of violence, and psychopathology. The family-only subtype was suggested to use less severe forms of violence infrequently, being violent only towards their partner or family, and having no or few problems related to psychopathology. The borderline/dysphoric subtype was suggested to use severe forms of violence (e.g. sexual violence), being violent predominantly towards their partner but also occasionally towards others outside of the relationship, as well as displaying traits of borderline personality disorder or severe forms of depression. Finally, the generally violent/antisocial subtype was thought to use severe forms of violence frequently, being violent towards their partner but also against others (e.g. friends, strangers and co-workers), as well as demonstrating traits of antisocial personality disorder and psychopathy.

However, validation attempts of the three-subtype solution proposed by Holtzworth-Munroe and Stuart (1994) have produced conflicting results. This includes conflicting results in terms of the number of subtypes and their characteristics. For example, the identification of the low-level antisocial subtype was not in line with the researchers’ originally proposed three subtype-solution, nor has this subtype consistently been identified in other studies (e.g. Cunha & Gonçalves, 2013; Graña, Redondo, Muñoz-Rivas, & Cantos, 2014). Moreover, studies have failed to differentiate between the borderline/dysphoric subtype and the generally violent/antisocial subtype in terms of borderline personality disorder traits (e.g. Delsol, Margolin, & John, 2003; Holtzworth-Munroe, Meehan, Herron, Rehman, & Stuart, 2003; Huss & Ralston, 2008). Furthermore, the three originally proposed subtypes have also proven difficult for clinicians to accurately identify (Langhinrichsen-Rohling, Huss, & Ramsey, 2000; Lohr, Bonge, Witte, Hamberger, & Langhinrichsen-Rohling, 2005). For example, when asked to sort IPV perpetrators into these subtypes, mental health professionals agreed on the subtype placement for only 26.6% of the perpetrators in the sample (Langhinrichsen-Rohling et al., 2000).

In light of this, a recent systematic review concluded that the two most consistently identified subtypes of IPV perpetrators across typology research are the PO violent subtype (resembling the family-only subtype) and the GV subtype (resembling the generally violent/antisocial subtype: Petersson & Strand, 2018). In their review, these two subtypes were found in all 30 studies that were included. Given the consistency of which these two subtypes appear this seems to be the most solid IPV perpetrator typology to build upon. Furthermore, several studies have found that it is sufficient to categorize the PO violent and GV subtypes using only the generality of violence dimension proposed by Holtzworth-Munroe and Stuart (1994: e.g. Boyle et al., 2008; Cantos, Goldstein, Brenner, O’Leary, & Verborg 2015; Herrero, Torres, Fernández-Suárez, & Rodríguez-Díaz, 2016; Petersson et al., 2019). These studies have also found that the resulting PO violent subtype and the GV subtype closely resembles the family-only and the generally violent/antisocial subtypes, respectively, on the other dimensions (i.e. severity/frequency of violence and psychopathology) described by Holtzworth-Munroe and Stuart (1994). As such, these two subtypes differ in several behavioural and psychosocial characteristics. For example, compared to the GV subtype, the PO violent subtype has been found to display lower levels of psychological IPV (e.g. threats: Boyle, O’Leary, Rosenbaum, & Hassett-Walker, 2008; Petersson, Strand, & Selenius, 2019), aggression (Boyle et al., 2008), violent attitudes and sexism (Herrero, Torres, Fernández-Suárez, & Rodríguez-Díaz, 2016; Petersson et al., 2019) and substance abuse (e.g. Cantos, Goldstein, Brenner, O’Leary, & Verborg, 2015; Herrero et al., 2016). Moreover, the PO violent subtype has been described as being more socially well adjusted outside of the relationship compared to the GV subtype. For example, the PO violent subtype has demonstrated higher levels of community integration and participation (Herrero et al., 2016), having higher levels of non-marital social skills (Johnson et al., 2006), as well as demonstrating higher levels of pro-social personality traits (e.g. agreeableness and conscientiousness: Walsh et al., 2010). Noteworthy, some studies have reported non-significant differences between the PO violent perpetrators and non-violent control groups. These non-significant differences include levels of non-marital and marital skills, marital satisfaction (as reported by both victim and perpetrator), attitudes and beliefs about IPV, substance abuse, and negative life stress (e.g. Delsol et al., 2003; Holtzworth-Munroe et al., 2000).

Studying the effect of arrest on recidivism among the PO violent and the GV subtypes holds important practical implications for reducing IPV. For example, studies have found that arrest can deter or increase the risk for future IPV depending on perpetrator type. It is therefore important to be able to identify and differentiate between such perpetrators so that this form of legal sanction is used efficiently. As such, perpetrators who are deterred by arrest might not need any further legal sanctions to prevent them from recidivating after their release. In contrast, this does not imply that perpetrators for whom arrest is non-deterrent or provocative should not be arrested. Rather, where possible, this would imply the need for other follow-on actions such as pretrial detention, prosecution, conviction, supervision or treatment for such perpetrators (Johnson & Goodlin-Fahncke, 2015). However, such follow-on actions are not always possible and in such cases, the criminal justice system, together with other victim support organizations, should focus on the victim. As such, identification of perpetrators who are most likely non-deterred by arrest would also help to identify the cases where, during the perpetrator’s arrest, victim-oriented protective measures should be implemented (e.g. restraining orders and protective living). Thus, in such cases, an arrest could serve as a temporary protective intervention, enabling attention to be directed towards victim safety planning. Also, arresting the perpetrator would in such cases create what Kelly et al. (2013) called ‘space for action and reflection’ for the victim to consider their options and receive support, as well as provide an opportunity to empower the victim through coordinated community responses (Johnson & Goodlin-Fahncke, 2015; Williams, 2005).

The aim of the present study was two-fold. First, we sought to describe and compare arrested and non-arrested IPV perpetrators in terms of demographic characteristics and variables related to the reported IPV incident. Second, we sought to explore and compare recidivism rates in relation to arrest within the total sample, as well as among the PO violent subtype and the GV subtype.

Method

Study design

The present study was of a longitudinal design and was conducted within the frame of a 7-year prospective research project. This larger research project aimed to implement and evaluate the use of structured violence risk assessments for various forms of family violence within one rural and one remote police district in Sweden. By the time of the start of the project in 2009, these two police districts did not perform structured violence risk assessments for family violence as required by the National Police Board. Thus, the two police districts were willing to co-operate with the research team in implementing such risk assessments within their organizations.

Sample

The sample consisted of 628 male alleged perpetrators reported to the police in the two participating districts for an IPV related crime against a female intimate partner, between 2011 and 2014. The perpetrators were between the ages of 16 and 86 (M = 39.1, SD = 13.0). In terms of immigrant background, 144 (22.9%) perpetrators were born in another country than Sweden. The majority of perpetrators (n = 498, 79.6%) were cohabitating with their female victims. The most common reported IPV incident was physical assault (45.4%), followed by severe violation of a woman’s integrity (33.0%), illegal threats (11.8%), other crimes (5.7%: e.g. stalking, arson, and non-physical violation of restraining orders), sexual assault (2.4%) and attempted murder or manslaughter (0.3%).

Data

The material in this study consisted of information on male-to-female perpetrated IPV retrieved from structured risk assessments for such violence, as well as from a separate coding sheet that the police were requested to fill in for the overall longitudinal research project.

Risk assessments

The risk assessments were performed by Swedish police officers using the Swedish version of the Brief Spousal Assault Form for the Evaluation of Risk (B-SAFER: Kropp, Hart, & Belfrage, 2008). The B-SAFER contains a total of 10 perpetrator risk factors for IPV recidivism and five victim vulnerability factors. These risk and victim vulnerability factors in the B-SAFER are rated as either present, partially present, or absent, based on several sources of information. These sources include interviews with the victim, interviews with the alleged suspect, and a review of register data (e.g. previous convictions). Based on the presence and absence of risk and victim vulnerability factors, two summary risk ratings are made: one for acute IPV (i.e. recidivism within a month) and one for severe or lethal IPV. These summary risk ratings are estimated as low risk, moderate risk or high risk, based on the presence and relevance of risk and victim vulnerability factors (Kropp et al., 2008). In this study, the B-SAFER summary risk ratings for both acute and severe or lethal IPV were dichotomized as elevated risk (i.e. combining ratings of moderate and high risk) and low risk. The B-SAFER has been reported to have good to excellent interrater reliability and predictive validity (e.g. Gerbrandij, Rosenfeld, Nijdam-Jones, & Galietta, 2018; Serie, van Tilburg, van Dam, & de Ruiter, 2015), as well as shown valid and reliable results for the Swedish police (Belfrage & Strand, 2008; Storey, Kropp, Hart, Belfrage, & Strand, 2014).

From the B-SAFER assessments, we used information on the risk factor measuring the perpetrator’s involvement in general criminality and violence to categorize the perpetrators into the PO violent and the GV subtypes (see the Procedure section). Moreover, we also used the two summary risk ratings for IPV recidivism.

Case-related variables

A separate coding sheet was developed by the authors of this study and was used to collect case-related information regarding the perpetrator and the reported IPV incident. Information used in this study consisted of demographics, details of the reported IPV incident, if the perpetrator was arrested, prosecuted and sentenced to prison or forensic psychiatric care, the level of risk management strategies initiated by the police in each case, and if a restraining order was initiated in each case. Demographics of the perpetrators included age, immigrant background (0 = no; 1 = yes), marital status (0 = unmarried; 1 = married), children younger than 18 years in common with the victim living at home (0 = no; 1 = yes) and if the perpetrator was cohabitating with the victim by the time of the reported IPV (0 = no; 1 = yes).

The other case-related variables concerned the reported IPV incident. Of specific relevance for the current study, we had access to information concerning a perpetrator’s arrest for the reported IPV incident (0 = no; 1 = yes). We did not have data on the length of the arrest, but in Sweden, this can vary between a couple of hours to a maximum of three days. Moreover, the IPV incidents were categorized as either non-physically (0: e.g. threats and harassment) or physically violent (1: e.g. assault, aggravated assault and attempted homicide), where physically violent IPV was considered as more severe due to its potential to cause physical injuries. Other case-related variables included whether the perpetrator was prosecuted for the reported incident (0 = no; 1 = yes), and if the perpetrator received a prison sentence or forensic psychiatric care for the reported IPV incident (0 = no; 1 = yes). The level of risk management strategies initiated by the police was measured as standard (0) or more than standard (1). A standard level of risk management included passive protective actions such as holding a security talk with the victim or making referrals to other victim support organizations (e.g. women’s shelters). More than standard levels of risk management included protective actions such as restraining orders, protected living or receiving an alarm package. Finally, we examined whether a restraining order was initiated in each case (0 = no; 1 = yes).

Procedure

The data for this study were collected in the two police districts involved in the research project. The B-SAFER assessments were carried out by the police between 2011 and 2014. In Sweden, the National Police Board’s guidelines state that all reported IPV crimes must be subjected to an initial, unstructured, risk assessment to assess the risk for future violence (The Swedish National Police Board, 2010). This risk assessment is unstructured as the police officer has to make a subjective assessment of the risk for further violence towards the victim, based on his or her knowledge and experience. Those IPV cases where the risk for future violence is assessed as present must, in turn, be subjected to a structured violence risk assessment using the B-SAFER. Thus, cases where the police officer deems the risk for further IPV as very low or absent are not subjected to a structured risk assessment. This means that the B-SAFER assessments concern more severe forms of violence reported to the police.

Since the decision to arrest a perpetrator is made by the police at the time of the police report, a possible arrest precedes the B-SAFER assessment. The police stored the B-SAFER assessments in paper and/or digital form and these assessments were accessible only within each district’s police headquarters.

Subtype classification

Following previous studies (e.g. Boyle et al., 2008; Cantos et al., 2015; Herrero et al., 2016; Petersson et al., 2019), the perpetrators in this study were categorized as either GV or PO violent based on their use of general criminality and violence. To this end, the B-SAFER requires the assessor to collect information about a perpetrator’s previous and current use of general criminality and violence, using both official records and interviews with the victim and the perpetrator. As this risk factor is rated in terms of present, partially present, or absent, we dichotomized this risk factor as either present (i.e. combining ratings of present and partially present) or absent. Perpetrators who were assessed by the police as generally criminal/violent were designated as GV (n = 327). The remaining perpetrators (i.e. assessed as not being generally criminal/violent) were classified as PO violent (n = 301). As such, in contrast to the GV perpetrators, the PO violent perpetrators in this study had never been convicted nor accused by their partner to be generally violent/criminal.

Recidivism and follow-up

Recidivism data consisted of official records accessed via the Swedish national police crime register. This register contains information about convictions as well as if a person is subject to a pending criminal investigation. In the present study, we examined the overall recidivism as well as IPV recidivism. Overall recidivism was defined as being convicted or suspected for any type of crime (including IPV). IPV recidivism was defined as having been either convicted or suspected for an IPV related crime (either against the same female partner as in the initially reported incident or towards a new intimate female partner).

As the perpetrators in this sample were reported to the police between 2011 and 2014, the follow-up period differed between subjects. However, the mean follow-up time in this study was 28 months (SD = 11.0, range = 5–50 months). Although using various time-periods for follow-up, the majority of previous studies examining the effect of arrest have used a 6-month follow-up to measure recidivism (e.g. Johnson & Goodlin-Fahncke, 2015; Sherman & Berk, 1984). In this study, we used three follow-up time-periods. First, we examined recidivism within the first 6 months following the police report of the IPV incident. Second, we examined recidivism within 7–12 months following the reported IPV incident. Finally, we also analysed recidivism rates combining the 6-month and the 7–12 month follow-up to an overall follow-up time period. The rationale for also analysing recidivism rates for a longer follow-up period than the first 6 months stems from findings that the first year post-intervention is a critical period where the majority of IPV perpetrators recidivate (e.g. Goldstein, Cantos, Brenner, Verborg, & Kosson, 2016; Petersson & Strand, 2017).

Statistical analysis

Statistical comparisons were carried out using chi-square tests and odds ratios (OR) with 95% confidence intervals (CI) for categorical, dichotomous, variables. Comparisons of continuous variables were made using independent sample t tests. Effect sizes were calculated using phi for chi-square tests, and Cohen’s d for the independent t tests. We also carried out direct binary logistic regressions, using subtype and arrest as predictors for different measures of recidivism (i.e. the dependent variables). For these analyses, subtype was coded as 1 = ‘generally violent’ and 0 = ‘partner only violent’, whereas arrest was coded as 1 = ‘arrested by the time of the police report’ and 0 = ‘not arrested by the time of the police report’. The dependent dichotomous variables of recidivism were coded as 1 = ‘recidivated’ and 0 = ‘did not recidivate’. Before running the logistic regression analyses, the multicollinearity assumption was examined and was found not to be violated.

The significance level used in this study was p < .05. All statistical analyses were computed using IBM SPSS (version 26). This study received ethical approval from the Swedish Ethical Review Board.

Results

Comparison between arrested and non-arrested perpetrators

In the overall sample, 26.1% (n = 164) of the perpetrators were arrested following the police reported IPV incident. Before any further analyses were carried out we excluded those perpetrators in the sample who were sentenced to prison (n = 60) or forensic psychiatric care (n = 6) for the reported IPV incident. The rationale for this exclusion was that these perpetrators’ time at risk (for recidivating) was limited due to them being incarcerated either for parts of the follow-up period or the full follow-up period. The number of arrested perpetrators eligible for the remaining analyses of this study was reduced to 124, as 40 perpetrators were arrested by the time of the police report and later sentenced to prison or forensic psychiatric care. Additionally, 26 perpetrators in the sample were not arrested by the time of the police report but later sentenced to prison or forensic psychiatric care.

For the first set of analyses, we compared arrested and non-arrested IPV perpetrators in terms of demographic and incident-related variables (see Table 1). A significant difference was found for having children younger than 18 years together with the victim. As such, arrested perpetrators were less likely to have children together with the victim. Moreover, arrested and non-arrested perpetrators differed on all but two variables related to the reported IPV incident. Arrested perpetrators were more than three times as likely to be reported for physical IPV, as well as prosecuted for the IPV incident. Arrested perpetrators were also nearly three times more likely to be assessed by the police with elevated summary risk ratings for acute IPV recidivism and severe or lethal such recidivism. However, there were no differences between arrested and non-arrested perpetrators in terms of the level of risk management strategies initiated by the police and whether a restraining order was initiated or not.

Table 1. Differences between arrested and non-arrested IPV perpetrators (n = 550)

	Arrested (n = 124)		Non-arrested (n = 426)
	M	SD	M	SD	t	df	p	d
Demographics
Age	39.4	12.5	39.4	13.1	−0.48	548	.962	0.0
	n	%	n	%	Chi-2	df	p	OR	95% CI	phi
Immigrant background	31	25.0	95	22.4	0.38	1	.537	1.2	[0.7, 1.8]	.03
Married	29	23.4	106	24.9	0.12	1	.724	1.1	[0.7, 1.7]	.02
Children (< 18 years) with victim*	49	39.5	211	50.2	4.41	1	.036	0.6	[0.4, 1.0]	−.09
Cohabitating with victim	101	81.5	343	80.7	0.04	1	.853	1.1	[0.6, 1.8]	.01
IPV incident related variables
Physical violence*	114	91.9	328	78.5	11.53	1	.001	3.1	[1.6, 6.2]	.15
Elevated risk for IPV Recidivism*	79	65.8	168	41.6	21.83	1	.000	2.7	[1.8, 4.1]	.20
Elevated risk for severe/lethal IPV recidivism*	52	43.3	90	22.3	20.76	1	.000	2.7	[1.7, 4.1]	.20
Prosecuted for IPV incident*	47	38.2	69	16.8	25.55	1	.000	3.1	[2.0, 4.8]	.22
Standard level of risk management^a	53	54.1	201	51.0	0.30	1	.587	0.9	[0.6, 1.4]	−.03
Restraining orders initiated	36	32.7	214	30.8	0.15	1	.694	1.1	[0.7, 1.7]	.02

IPV = Intimate partner violence; OR = odds ratios; CI = confidence intervals.

^aThis includes ‘passive’ protective measures such as security talks with the victim and referrals to crime victim support organizations.

*Indicates statistically significant results

Arrest and recidivism in the total sample

Since perpetrators in this sample had varying follow-up periods, as they were reported to the police between 2011 and 2014, initial analysis on mean follow-up time between arrested and non-arrested perpetrators was carried out. An independent samples t test revealed a significant difference in follow-up time in months between arrested (M = 25.9; SD = 10.8) and non-arrested (M = 28.1; SD = 10.7) perpetrators, t(548) = 1.99, p = .047, eta = .01. However, this difference was not considered problematic given that both groups were followed-up, on average, for a longer time than the longest follow-up period used in this study (i.e. 12 months). As can be seen in Table 2, there were no significant differences between arrested and non-arrested perpetrators in terms of recidivism.

Table 2. Recidivism rates among arrested and non-arrested IPV perpetrators (n = 550)

	Arrested (n = 124)		Non-arrested (n = 426)
	n	%	n	%	Chi-2	df	p	OR	95% CI	Phi
Six-month follow-up
IPV recidivism	9	7.3	39	9.2	0.43	1	.510	0.8	[0.4, 1.6]	−.03
Overall recidivism	21	16.9	83	19.5	0.41	1	.524	0.8	[0.5, 1.4]	−.03
7–12 month follow-up
IPV recidivism	12	9.7	24	5.6	2.57	1	.109	1.8	[0.9, 3.7]	.07
Overall recidivism	26	21.0	59	13.8	3.72	1	.054	1.6	[0.9, 2.7]	.08
Overall follow-up
IPV recidivism	21	16.9	63	14.8	0.34	1	.559	1.2	[0.7, 2.0]	.02
Overall recidivism	42	33.9	125	29.3	0.93	1	.334	1.2	[0.8, 1.9]	.04

IPV = intimate partner violence; OR = odds ratios; CI = confidence intervals.

Main and interaction effects of arrest and subtype on recidivism

Next, we conducted a series of binary logistic regressions to examine the main and interaction effects of arrest and offender subtype on the various measures of recidivism. As can be seen in Table 3, there was a main effect for subtype on the following recidivism measures in the logistic regression models: overall recidivism within 6 months, overall recidivism within 7–12 months, and overall recidivism within 12 months. These results demonstrate that GV perpetrators were more likely than PO violent perpetrators to recidivate in overall criminality (including IPV) within the previously mentioned periods.

Table 3. Binary logistic regression analyses of the main effects and interaction effect (*) of subtype and arrest on recidivism (n = 550)

	β	Wald	p	OR	95% CI
IPV recidivism within 6 months^a
Subtype	0.6	2.7	.099
Arrest	−18.6	0.0	.997
Subtype*arrest	18.8	0.0	.997
Constant	−2.6	94.8	.000
Overall recidivism within 6 months^b
Subtype	1.2	19.2	.000	3.2	[1.9, 5.4]
Arrest	−0.8	1.6	.205
Subtype*arrest	0.7	1.1	.304
Constant	−2.1	93.3	.000
IPV recidivism 7–12 months^c
Subtype	0.2	0.3	.591
Arrest	0.0	0.0	.947
Subtype*arrest	0.8	1.0	.324
Constant	−2.9	90.0	.000
Overall recidivism 7–12 months^d
Subtype	0.7	5.2	.023	1.9	[1.1, 3.4]
Arrest	−0.2	0.1	.766
Subtype*arrest	0.9	2.2	.139
Constant	−2.2	94.6	.000
Overall IPV recidivism^e
Subtype	0.5	2.9	.090
Arrest	−0.9	2.0	.157
Subtype*arrest	1.4	4.0	.046	4.1	[1.0, 16.6]
Constant	−2.0	91.5	.000
Overall recidivism^f
Subtype	0.9	17.6	.000	2.5	[1.6, 3.9]
Arrest	−0.4	1.1	.294
Subtype*arrest	0.9	3.3	.069
Constant	−1.4	66.4	.000

IPV = Intimate partner violence; OR = odds ratios; CI = confidence interval.

^aOmnibus test = χ2 (3) = 14.92, p = .002. Cox & Snell R square = .027. Nagelkerke R square = .060

^bOmnibus test = χ2 (3) = 32.66, p = .000. Cox & Snell R square = .058. Nagelkerke R square = .093

^cOmnibus test = χ2 (3) = 5.12, p = .163. Cox & Snell R square = .009. Nagelkerke R square = .024

^dOmnibus test = χ2 (3) = 18.98, p = .000. Cox & Snell R square = .034. Nagelkerke R square = .059

^eOmnibus test = χ2 (3) = 14.57, p = .002. Cox & Snell R square = .026. Nagelkerke R square = .045

^fOmnibus test = χ2 (3) = 38.83, p = .000. Cox & Snell R square = .068. Nagelkerke R square = .096

More importantly, although there were no significant main effects for either arrest or subtype on IPV recidivism, we found a significant interaction effect in the binary logistic regression analyses between subtype and arrest for IPV recidivism within 12 months (see Table 3). Drawing on the results from the logistic regression, GV perpetrators who were arrested by the time of the police report were more than four times as likely as arrested PO violent perpetrators to recidivate in IPV within the first 12 months post-arrest. In an attempt to explain this interaction effect we repeated the logistic regression analysis, this time also controlling for violence severity (1 = initially reported for physical IPV, 0 = initially reported for non-physical IPV) as this variable was previously shown to differ between arrested and non-arrested perpetrators in this sample (see Table 1). However, the results remained the same, demonstrating no main effects but a significant interaction effect between arrest and subtype on IPV recidivism (β = 1.5, p = .040, OR = 4.3; 95% CI = 1.1–17.4).

Arrest and recidivism within subtypes

Finally, we also studied the recidivism rates among arrested and non-arrested perpetrators separately for the two subtypes (i.e. within-groups analysis). As can be seen in Table 4, a significant result was found within the PO violent subtype for IPV recidivism within the first 6 months following the police report. PO violent perpetrators who were not arrested were more likely to recidivate in IPV compared to those PO violent perpetrators who were arrested. However, no other differences in recidivism rates were found between arrested and non-arrested PO violent perpetrators.

Table 4. Recidivism rates among arrested and non-arrested PO violent perpetrators (n = 275)

	Arrested (n = 57)		Non-arrested (n = 218)
	n	%	n	%	Chi-2	df	p	OR	95% CI	Phi
Six-month follow-up
IPV recidivism*	0	0.0	15	6.9	4.15	1	.042	1.3	[1.2, 1.4]	−.12
Overall recidivism	3	5.3	24	11.0	1.68	1	.194	0.4	[0.1, 1.5]	−.08
7–12 month follow-up
IPV recidivism	3	5.3	11	5.0	0.00	1	.947	1.0	[0.3, 3.9]	.00
Overall recidivism	5	8.8	22	10.1	0.89	1	.766	0.9	[0.3, 2.4]	−.02
Overall follow-up
IPV recidivism	3	5.3	26	11.9	2.13	1	.145	0.4	[0.1, 1.4]	−.09
Overall recidivism	8	14.0	44	20.2	1.11	1	.291	0.7	[0.3, 1.5]	−.06

PO = partner only; IPV = intimate partner violence; OR = odds ratios; CI = confidence intervals. The figures under ‘overall follow-up’ represents unique perpetrators who recidivated at any point in time during the follow-up. The figures for the other two follow-up periods represent the number of recidivism crimes. As such, some perpetrators recidivated in more than one crime.

*Indicates statistically significant results

As can be seen in Table 5, we also found a significant difference within the GV subtype. This result implied that arrested GV perpetrators were more likely to recidivate in any type of offence within the 7–12 month follow-up period, compared to non-arrested GV perpetrators. In an attempt to explain these results, and since we did not have data on perpetrators’ arrest for the recidivism crimes, we controlled for available data regarding the subsequent prosecution of GV perpetrators who recidivated within the first 6 months, as this could have affected their possibility to recidivate during the 7–12 month follow-up (e.g. due to being arrested). The analysis showed no difference in proportions between arrested (60.0%) and non-arrested (65.5%) GV perpetrators who were prosecuted for the recidivism crime that occurred within the first 6 months, χ²(1, 70) = 0.15, p = .696 (OR = 0.8; 95% CI = 0.2, 2.6), phi = −.05.

Table 5. Recidivism rates among arrested and non-arrested GV perpetrators (n = 275)

	Arrested (n = 67)		Non-arrested (n = 208)
	n	%	n	%	Chi-2	df	p	OR	95% CI	Phi
Six-month follow-up
IPV recidivism	9	13.4	24	11.5	0.17	1	.678	1.2	[0.5, 2.7]	.03
Overall recidivism	18	26.9	59	28.4	0.06	1	.812	0.9	[0.5, 1.7]	−.01
7–12 month follow-up
IPV recidivism	9	13.4	13	6.3	3.55	1	.059	2.3	[0.9, 5.7]	.11
Overall recidivism*	21	31.3	37	17.8	6.00	1	.018	2.1	[1.1, 4.0]	.14
Overall follow-up
IPV recidivism	18	26.9	37	17.8	2.61	1	.106	1.7	[0.9, 3.2]	.10
Overall recidivism	34	50.7	81	38.9	2.90	1	.088	1.6	[0.9, 2.8]	.10

GV = generally violent; IPV = intimate partner violence; OR = odds ratios; CI = confidence intervals. The figures under ‘overall follow-up’ represents unique perpetrators who recidivated at any point in time during the follow-up. The figures for the other two follow-up periods represent the number of recidivism crimes. As such, several perpetrators recidivated in more than one crime.

*Indicates statistically significant results.

Discussion

This study aimed to describe and compare arrested and non-arrested perpetrators on demographic and incident-related variables, as well as to explore and compare recidivism rates in relation to arrest in the total sample but specifically for the PO violent subtype and the GV subtype.

The overall arrest rate in this sample, 26.1%, was in line with previous Swedish studies (NCCP, 2008). The findings show that, compared to the non-arrested perpetrators, arrested perpetrators were less likely to have children together with the victim, more likely to be reported for severe IPV (i.e. physical violence), being assessed by the police post-arrest with a higher risk for IPV recidivism, as well as being more likely to be prosecuted for the reported IPV incident. However, the level of risk management initiated by the police and the initiation of a restraining order did not differ between cases with arrested and non-arrested perpetrators. Furthermore, we found no differences between arrested and non-arrested perpetrators in the total sample in terms of recidivism rates. However, we did find an interaction effect between subtype and arrest on IPV recidivism within 12 months post-arrest. Among those perpetrators who were arrested by the time of the police report, GV perpetrators were more than four times as likely as the PO violent perpetrators to recidivate. Moreover, although based on a small number of observations, the within-groups analyses suggested a possible deterrent and provocative effect of arrest on recidivism for the PO violent and the GV subtype, respectively.

Overall, this study demonstrates that the police had a propensity to arrest high-risk perpetrators who had committed more severe forms of IPV (i.e. physical violence). These results mirror the findings of Hilton et al. (2007) who reported that arrest was positively associated with a higher assessed risk for recidivism and greater incident severity. Noteworthy, the risk assessment for recidivism made by the police in this study was carried out after the perpetrator’s arrest and did therefore not guide decisions to arrest. In addition, our results are similar to previous research that demonstrated that police are more prone to arrest IPV perpetrators who have physically assaulted the victim (e.g. O’Neal & Spohn, 2017). This suggests that the police are more likely to arrest alleged perpetrators in those cases where the victim has sustained greater levels of injury (e.g. visible bruises or lacerations), thus leading the police to have probable cause to suspect that a crime has been committed.

The main finding of this paper was the usefulness of incorporating IPV perpetrator subtypes into the analyses when studying the effect of arrest on recidivism. Thus, our results support the suggestions that the effect of arrest on recidivism varies as a function of perpetrator type (e.g. Sherman & Berk, 1984; Williams, 2005). As such, the initial results showed no difference in recidivism rates when comparing arrested with non-arrested perpetrators in the total sample. However, when we accounted for the subtype classification in the analyses (i.e. the PO violent and the GV perpetrators), important results emerged. Among those perpetrators who were arrested by the time of the reported IPV incident, GV perpetrators were more than four times as likely as the PO violent perpetrators to recidivate in IPV.

There are several possible explanations for the significant interaction effect found in this study. Previous studies have established that the GV perpetrators constitute a more violent, high-risk, subtype compared to the PO violent perpetrators. For example, the GV perpetrators have demonstrated higher levels of severe physical and psychological violence, a higher propensity to escalate their violent behaviour, as well as displaying violent attitudes (Cavanaugh & Gelles, 2005; Holtzworth-Munroe et al., 2000; Petersson & Strand, 2018). Furthermore, the GV perpetrators have also been found to display more problems related to psychosocial adjustment than the PO violent perpetrators, including more problems with substance use, employment and mental health (Petersson et al., 2019). Unsurprisingly, previous studies have also demonstrated that GV perpetrators are more likely to recidivate than PO violent perpetrators (Goldstein et al., 2016; Huss & Ralston, 2008; Petersson & Strand, 2017). Given the GV perpetrators’ more violent disposition and psychosocial problems, it is highly likely that these perpetrators simply do not care about being arrested. Thus, drawing on their antisocial lifestyle, the prospect of punishment is of little concern to them. Relatedly, as opposed to the PO violent perpetrators, GV perpetrators have been found to have a more extensive arrest history (e.g. Cantos et al., 2015; Theobald, Farrington, Coid, & Piquero, 2015). Although we did not have access to such data in this study, the GV perpetrators in our sample may have previously been arrested for other crimes (including IPV). In turn, this could have sensitized them and, thus, diminished the deterrent effect of being arrested.

It is also possible to explain the differences in recidivism between the arrested PO violent perpetrators and the arrested GV perpetrators from a theoretical viewpoint, using the stake in conformity theory (Toby, 1957). As outlined in Toby’s (1957) theory of stake in conformity, individuals who have strong ties to society (e.g. by being employed and married) risk to lose more by being arrested for criminal conduct than those individuals who lack such stakes in conformity (i.e. individuals who are unemployed and unmarried). Expectedly, the PO violent perpetrators and the GV perpetrators have been found to differ on many characteristics related to the level of stake in conformity. As such, the lower levels of recidivism among the PO violent perpetrators found in this study could be related to findings that the PO violent subtype is more socially well adjusted outside of their relationship. For example, this subtype has a higher degree of community participation and integration than the GV subtype (Herrero et al., 2016). The PO violent subtype has also been described as being similar to non-violent control groups on traits such as levels of non-marital and marital skills, marital satisfaction, attitudes and beliefs about IPV, substance abuse, and negative life stress (e.g. Delsol et al., 2003; Holtzworth-Munroe et al., 2000). Coupled with findings that the PO violent subtype has a higher stake in conformity, in terms of employment rates, education levels and is more likely to be married to their partner (e.g. Cantos et al., 2015; Goldstein et al., 2016; Mach, Cantos, Weber, & Kosson, 2017), arrest may have a deterrent effect on recidivism for this subtype. For the PO violent perpetrator, an arrest could entail losing his employment, relationship, and his social status and position. Being more socially well adjusted outside of their relationship and having higher levels of community integration and participation (Herrero et al., 2016), the anticipated reactions of the surrounding society to their arrest may act as a strong specific deterrent for recidivism. Moreover, given that the PO violent subtype can be considered as having a high stake in conformity, the results in this study are similar to the results presented in the SARP studies. As such, Sherman et al. (1992b) concluded that four of the SARP studies demonstrated that arrest had a deterrent effect for those perpetrators who had a high stake in conformity.

On the other hand, the GV perpetrators are less socially well adjusted outside of their relationship and have more problems related to psychosocial adjustments such as mental health problems, substance abuse and employment issues (e.g. Petersson et al., 2019). The GV subtype has also been described in previous studies as having a lower stake in conformity (e.g. Cantos et al., 2015; Goldstein et al., 2016; Mach et al., 2017). From a theoretical perspective, these perpetrators do not risk losing their employment, social status or social position to the same extent as the PO violent perpetrators, as the GV perpetrators often already lack these features. Thus, the lack of strong ties to conventional society, mainly through their poor psychosocial adjustment, could lower their threshold for violating the law. Relatedly, due to their low degree of integration with conventional society and their poor psychosocial adjustment, perpetrators with a low stake in conformity may be more likely to feel wronged by the victim and the criminal justice system for being arrested for the reported IPV. Therefore, they may also feel less obliged to suppress their anger at the victim after being arrested and thus be more inclined to recidivate (Sherman et al., 1992b).

Interestingly, despite the results showing that arrested perpetrators in this sample had committed more severe forms of IPV, and were assessed by the police as being more likely to recidivate in both acute and severe forms of IPV, arrested PO violent perpetrators were less likely to recidivate in IPV. This indicates that violence severity and the assessed risk for future IPV is not predictive for recidivism for this subtype. In contrast, the GV perpetrators who were arrested were more likely to recidivate. Thus, violence severity and assessed risk for future IPV seem to be predictive for this subtype. These results are identical to previous research that concluded that violence severity was predictive for the GV subtype, but not for the PO violent subtype (Goldstein et al., 2016). According to Goldstein et al. (2016), these results were not surprising for the GV perpetrators as they, due to their antisocial and violent disposition, were hypothesized to be more likely to recidivate regardless of initial violence severity. A possible explanation as to why violence severity was not predictive for the PO violent perpetrators could be that victims of such perpetrators, who are usually exposed to less severe forms of IPV (Petersson & Strand, 2018), may wait with reporting their partner to the police until a more severe incident occurs. Thus, although being reported for severe forms of IPV, given their otherwise less violent disposition and their higher stake in conformity, an arrest could have a deterrent effect making PO violent perpetrators reluctant to recidivate.

Finally, the within-groups analyses carried out in this study indicate that there might be within-subtype differences in recidivism between arrested and non-arrested perpetrators. These results demonstrated that arrested PO violent perpetrators were less likely than non-arrested PO violent perpetrators to recidivate in IPV for the first 6 months of follow-up. Furthermore, arrested GV perpetrators were more likely than non-arrested GV perpetrators to recidivate in overall criminality (including IPV) within the 7–12 months follow-up. These results are partially in line with the only previous study to examine the effect of arrest on recidivism among subtypes of partner violent men (Johnson & Goodlin-Fahncke, 2015). In their analyses of data retrieved from an experimental study, Johnson and Goodlin-Fahncke (2015) reported that arrest had no deterrent effect for neither the PO violent nor the GV subtypes. However, arrest seemed to increase recidivism rates among the GV subtype – indicating a provocative effect of arrest on recidivism for this subtype. However, two major methodological limitations preclude us from drawing any conclusions related to the deterrent and provocative effects of arrest on recidivism in the present study. First, this study was not conducted with an experimental design. Second, the within-groups analyses were based on a small number of observations and need to be replicated using a larger sample. However, the within-group analyses suggest that future research on this topic also should account for within-subtype variation to receive a full understanding of the effect of arrest on recidivism.

Limitations

This paper is not without limitations. First, this study was non-experimental, meaning that we cannot make any statements of the causal effects of arrest on recidivism. In this longitudinal study, however, we could control for several variables previously found to pose as possible confounders, including the severity of IPV, the perpetrator’s risk for recidivism, and several other demographical variables (e.g. Hilton et al., 2007). Moreover, we also controlled for additional possible confounders that could have affected recidivism such as incarceration, level of risk management strategies initiated by the police as well as if the case proceeded to prosecution. Nevertheless, the current study and its results are exploratory and should be seen as a starting point for more methodologically rigorous studies to replicate.

Other limitations of the present study concerned the nature of the data. First, relying on official police records for recidivism data likely underestimates the amount of actual recidivism (e.g. Goldstein et al., 2016). Although we did not have access to such data in the present study, the use of victim reports to measure recidivism would have likely demonstrated higher rates of recidivism. Second, since this study was based on cases of police-reported IPV that the police deemed serious enough to warrant a structured violence risk assessment, the type of IPV examined in this study most likely constitutes the most severe forms of IPV. As such, generalizing the results of this study to less severe forms of IPV is not recommended. Related to the discussion of the generalizability of the study’s results, the current study was based on data from one rural and one remote police district in Sweden. Therefore, it is uncertain if the results can be generalized to an urban setting in Sweden. However, the results concerning the differences between the subtypes in this study are partially in line with the results from the SARP studies, which were mainly based in urban settings. Albeit, drawing on the exploratory nature of this study, replication attempts should be made in urban settings as well. Moreover, it can also be contested that physical IPV is to be considered as more severe than psychological IPV. As such, it has been reported that psychological IPV can have an as deleterious and negative impact on the victims as physical IPV (O’Leary, 1999). Still, we chose to consider physical IPV as more severe, mainly due to the heightened risk for serious bodily injuries to victims.

Practical implications

Drawing on the results of this study, the effect of arrest on recidivism seems to differ depending on the subtype. The results indicate that arrest is not a sufficient legal action to reduce IPV for GV perpetrators, whereas relatively few arrested PO violent perpetrators recidivated. Therefore, identifying the arrested IPV perpetrator as either PO violent or GV, in connection to the arrest, could guide decisions related to the need for further legal sanctions or protective measures to prevent recidivism.

For GV perpetrators, an arrest should be combined with other interventions, including (if possible) pre-trial detention, conviction, sentencing, or supervision (Johnson & Goodlin-Fahncke, 2015). When arresting GV perpetrators who are reported for severe forms of IPV, the results of this study demonstrate that the Swedish police need to work more diligently together with attorneys to extend the arrest into pre-trial detention. In Sweden, a suspect can be arrested for a maximum of 72 hours. During this time, the attorney is responsible for issuing a remand (i.e. pre-trial detention) order to the court, which decides whether or not to place the suspect on remand. There is no limit for time on remand in Sweden, although the suspect on remand will have his or her case tried at a new remand hearing every 2 weeks. However, many IPV cases do not proceed to court (Tjaden & Thoennes, 2000), partially due to the victim’s refusal to cooperate in the investigation. Data from the research project that this study was based on showed that 27% of the women victimized by IPV did not want to cooperate in the police investigation. Therefore, many of the suggested follow-on interventions are not always applicable as cases are dropped. Thus, during the arrest of a GV perpetrator, the focus should be directed towards risk management aiming to protect and empower the victim. The empowerment of victims can include motivating the victim to participate in the police investigation, as this could increase the chance of charging the perpetrator for the crime, as well as motivating the victim to participate in the risk management strategies provided by the police and other community support agencies. Encouraging the victims of an arrested GV perpetrator to participate in the police investigation is therefore crucial. For example, according to O’Neal and Spohn (2017) prosecutors were nearly 13 times more likely to file charges against the perpetrator if the victim cooperated in the investigation.

In the case of PO violent perpetrators, an arrest could act as a deterrent for recidivism. The need for additional interventions might therefore not be as important. Important to bear in mind, however, is the fact that there were only three (5.3%) arrested PO violent perpetrators who recidivated in IPV in this study. As such, the conclusion about the need for fewer interventions is based on a relatively small number of observations. Regardless, research has shown that PO violent perpetrators are more likely to respond positively to treatment interventions than GV perpetrators. For example, some studies have demonstrated that PO violent perpetrators are more likely to attend and complete treatment (e.g. Huss & Ralston, 2008; Mach et al., 2017). As such, referring such perpetrators to treatment following their release from arrest could have favourable results in reducing IPV recidivism. Furthermore, given this subtype’s higher stake in conformity and therefore greater respect for the criminal justice system, a proactive talk by the police aiming to inform the perpetrator of the consequences of violent behaviour could also help these men desist from using violence towards their female partner following the release from arrest. In Sweden, such proactive talks are held by the police with IPV perpetrators who are deemed by the police as appropriate for such an intervention (NCCP, 2017). This appraisal is based on the perpetrator’s psychosocial functioning and motivation to end their violent behaviour, as the proactive talk aims to make the perpetrator understand the negative outcomes associated with using violence towards an intimate partner.

Conclusions

The results of this study support the usefulness of incorporating IPV perpetrator typologies into the analyses when studying the effect of legal sanctions on recidivism. As such, this study further cements the GV perpetrators as a high-risk subtype (e.g. Cavanaugh & Gelles, 2005; Petersson et al., 2019), seemingly not deterred by arrest (Johnson & Goodlin-Fahncke, 2015). Besides the need to replicate the within-group results of this study with a larger sample, future studies could also favourably examine time to recidivism, using for example survival analysis, to offer a more detailed description of the effect of arrest for the two subtypes.

Disclosure statement

No potential conflict of interest was reported by the authors.