Evaluation of the Drive Intervention for High-Harm Domestic Abuse Perpetrators in England and Wales Using a Quasi-Experimental Approach

Abstract

High-risk and high-harm domestic abuse perpetrators pose the greatest risk of serious harm to (ex-)partners and family members. Robust evidence of perpetrator intervention is lacking for this group. The Drive perpetrator intervention in the United Kingdom focuses specifically on high-harm perpetrators to reduce abuse and increase safety of victims, using individual case management via support and disruption, co-ordinated multi-agency response and targeted support to victims. This paper reports on evaluation of the Drive pilot. A high quality quasi-experimental approach with individually randomised intervention and controls was applied in three areas in England and Wales. High-risk perpetrators (N = 2627) were allocated randomly to either Drive or usual care (support to victim). Change in safety and risk for Drive-associated victims was compared with the usual care victim group. Longer-term outcomes were assessed before, during and 12 months following the intervention for Drive and control group perpetrators at one site. The Drive intervention reduced abusive perpetrator behaviours and increased safety for victims compared to the control. The reduction in perpetration by both serial and repeat perpetrators sustained for a year after the end of the intervention and was statistically significant (p < 0.001) for Drive perpetrators compared to the control.

1. Introduction

Since the 1980s, work with domestic abuse perpetrators to change their behaviour has been increasingly recognised as an element of domestic abuse support services in Western countries alongside work to support victims. Behaviour change approaches have tended to be group programmes aimed mainly at intimate partner violence (IPV) perpetrators, who tend to be male, and aimed mainly at those using moderate or low levels of domestic violence and abuse able to engage in a group setting and with some motivation to change. Such programmes have been influenced by the development in work with offenders that draws on theories of both punishment and psychology (Hester and Newman 2020). This has included social learning (Bandura et al. 1961), which emphasises that behaviour is learnt and it is thus possible to ‘unlearn’ and change perpetrators’ abusive behaviour and encourage rehabilitation, systems theory that helps us understand the importance of local agencies acting in tandem as a multi-agency system to back up and enforce the work carried out in programmes, as well as understanding of patriarchal contexts (Pence and Paymar 1993; Gondolf and Jones 2001). Increasingly, emphasis on psychological aspects have influenced programme approaches, with focus on individual characteristics, attachment issues and links to earlier trauma, rather the more exclusive focus on socio-political re-education of the earlier (Duluth) models (Hester and Newman 2020).

Perpetrator interventions may explore with perpetrators the consequences of their abusive behaviour in terms of its impact on their partner (and children) with the aim of increasing empathy, accountability and motivation to change, and challenging gender stereotypes and hostile attitudes towards women, and the group setting is used to enhance such engagement. Programmes use a combination of ‘treatment’ methods (Lilley-Walker et al. 2018), with education, counselling and psychotherapy tending to be the main approaches used, and psychodynamic, psychopathological and bio-psychosocial based methods also applied. In the UK programmes usually combine gendered psycho-educational content to unpack and challenge harmful beliefs and expectations around gender and masculinity; cognitive behavioural and/or mindfulness-based approaches aimed at cultivating attention to ‘triggers’, ‘signals’ and at managing impulsivity and strong emotions; trauma-informed components addressing men’s own experiences and how these impact on current behaviour; and more recently, Motivational Interviewing techniques, aimed at developing motivation to change by exposing the discrepancy between current behaviour and men’s own positive aspirations (Iwi and Newman 2015). Iwi and Newman (2015) argue that perpetrator programmes need robust processes to assess and manage risk, inter-agency working protocols with other organisations which address issues such as information sharing and joint working arrangements, and regular information sharing with women’s support workers. Once a man is referred, or refers himself to a programme, then the work should be adjusted to fit the nature of the risk and the man’s own situation, needs and goals. In the UK this is reflected in the Respect guidance which sets the national standard for safe and accountable perpetrator work (What we do|Respect, https://www.respect.org.uk/pages/what-we-do, accessed on 13 December 2024).

In this article we focus on a particular perpetrator behaviour change intervention in the UK, called Drive, which is not a group programme but an individualised approach. Drive is specifically aimed at high-risk, high-harm, repeat and serial perpetrators who carry the greatest risk of causing serious harm to their (ex-)partners or other family members (Robinson 2016). Serial perpetrators are those who perpetrate abuse with different victims whereas repeat perpetrators are those who perpetrate abuse with the same victim. High-harm perpetrators are known to be challenging to engage in services and less likely to participate willingly in perpetrator programmes (Gondolf and Jones 2001), which is why Drive is not a group programme but uses an individual case management approach. Although the criminal justice system intervenes with high-harm perpetrators, there is a lack of dedicated perpetrator behaviour change interventions for this group which is why Drive was established. The Drive intervention is unique in the UK (and possibly elsewhere) in focusing specifically on such high-risk, high-harm perpetrators, yet led and delivered by non-governmental not for profit organisations rather than criminal justice agencies. The main aim of the Drive intervention is to reduce the number of child and adult victims of domestic abuse by deterring abusive perpetrator behaviour, with the related objective of reducing repeat and serial perpetration. As in the case of group programmes, Drive is underpinned by the expectation that perpetrators can change their behaviour, but also a recognition that such change may prove particularly difficult with this group of perpetrators. There is therefore a specific focus on alleviating psychological or social ‘needs’ that may interfere with change or safety for victims, as well as more punitive approaches to contain or constrain the individuals concerned. An integral part of this DRIVE approach is particular emphasis on agency ‘systems’ in the localities acting in a co-ordinated manner to deliver support with needs or provide disruption, as a ‘co-ordinated community response’ (Pence and Paymar 1993) to ensure both victim safety and perpetrator accountability (UN Women 2010).

Reflecting the broader legal definition of domestic abuse in England and Wales (Domestic Abuse Act 2021), Drive is aimed at perpetrators of both intimate partner abuse and abuse of other family members. As indicated, Drive uses intensive individual case management of perpetrators to focus on individual need alongside tackling further abusive behaviour. Activation of a coordinated multi-agency response is key, whereby the Drive case managers work closely with Independent Domestic Violence Advocates (IDVAs) who support victims (intimate partners or family members), and involve the police, probation, children’s social services, housing, substance misuse and mental health teams to provide a mixture of support and disruption. For instance, support may involve access to mental health support or ensuring that the perpetrator has accommodation separately to the victim. Where it is difficult to engage with a perpetrator and/or their perpetration continues, the Drive case manager initiates strategies to disrupt the perpetration which may involve the police immediately acting to remove and arrest the perpetrator, or children’s services making it clear that the perpetrator can no longer see his children. At the same time Drive case managers also work directly with perpetrators where possible, using similar approaches to group programmes in developing empathy, accountability and motivation to change, but as a bespoke offer delivered to each service user rather than a standard programme. What also differs from group programmes is that case managers may in some instances be unable to engage directly with the perpetrator concerned and instead focus on indirect work, enlisting other agencies to support and disrupt. Drive is thus a complex individual-focused intervention that connects existing services and thereby complements and enhances existing interventions and services to ensure that perpetrators change their behaviour and victims are safer.

There has been extensive debate as to whether perpetrator interventions actually reduce perpetrators’ abusive behaviour or make the lives of victims–survivors and their children safer. Meta-analyses of perpetrator programmes have tended to be inconclusive on what works in these programmes (Akoensi et al. 2013; Babcock et al. 2004; Lilley-Walker et al. 2018). Recent evidence from some randomised trials have reported reductions in abusive behaviour (Nesset et al. 2020; Lila et al. 2018), and evaluations without randomisation have also reported some positive outcomes regarding behaviour (McGinn et al. 2020; O’Connor et al. 2021), although some studies have found only small reductions in abusive behaviour (Haggard et al. 2017), or no such reduction (Herman et al. 2014). A major reason for the range of results is due to lack of consistency in methods or data collection, and consequent lack of comparability. Measuring and defining the success of perpetrator programmes is challenging because populations and interventions may vary, outcomes are limited or poorly measured, with lack of or only short follow-up periods, and insufficient attention to participant attrition or context of the interventions. Few studies have used experimental or other designs involving control groups (Turner et al. 2023). Some studies have expanded understanding of success of programmes to include stronger emphasis on (ex) partner testimony, ascertaining impact on victim safety and wellbeing alongside any perpetrator assessment of behaviour change (Kelly and Westmarland 2015; Vall et al. 2021). Given the complex landscape regarding evaluation methods and outcomes, a more relevant question thus appears to be under what conditions perpetrator programmes work and for whom, rather than whether such programmes work at all (Babcock et al. 2016).

The Drive intervention discussed in this paper was launched in April 2016 by a partnership of the NGOs Respect, SafeLives and Social Finance, and piloted for three years in three areas across England and Wales (Essex, South Wales and West Sussex). The Drive pilot intervention was funded from a combination of local government, police, Home Office and philanthropic sources. This pilot phase is the focus of the evaluation reported here. The three Drive pilot sites shared the core Drive model but also varied in approaches to delivery, management of caseload, associated administrative work and in the wider multi-agency ecosystems that they were situated within (Hester et al. 2020).

This article reports on the three-year evaluation of the Drive pilot. To date this is the largest evaluation of a perpetrator intervention in the UK to include a randomised control approach. To create more reliable and comparable measures of domestic abuse severity and harm, it has been argued that evaluations should draw on both quantitative and qualitative measures, be clearer about programme samples including detail regarding drop out, about the intervention approach, programme facilitation and facilitators, measure outcomes for at least 12 months following intervention, use a clear definition of recidivism, use of control groups, include (ex)partner/victim reports and take into account a wider range of abusive behaviour and impacts on victims (Lilley-Walker et al. 2018; Vall et al. 2021). In the evaluation of Drive, we endeavoured to apply such an approach by including the randomisation of intervention and control groups, quantitative measurement of outcomes at least 12 months following completion, qualitative interviews with perpetrators, victims–survivors and a range of professionals, and analysis of case management and other contextual data. In this paper, we focus on the key quantitative outcomes of the intervention, showing that the Drive intervention led to a substantial reduction in abusive behaviours by perpetrators and reports of increased safety for victims and children to a moderately greater degree than in control cases where merely support to the victim was provided. The evaluation shows a significant reduction in in the longer term in repeat perpetration against individual victims–survivors and in serial perpetration against more than one victim, thus indicating that there is a more sustainable impact on safety for victims when Drive is present.

Qualitative research carried out in the wider evaluation informs the discussion in this paper but is not reported in detail here. A cost analysis was also carried out but is not reported here (see Hester et al. 2020).

2. Materials and Methods

Pragmatic randomised control approach: The evaluation involved a high-quality quasi-experimental design drawing on a pragmatic randomised control approach. Given the complex and real-world nature of the Drive intervention, we decided that a pragmatic evaluation with individually randomised intervention and control groups would be the most appropriate to ascertain whether the intervention was effective and achieved its aims.

While traditional clinical ‘explanatory’ trials generally measure efficacy of a treatment under ideal conditions and using homogenous samples to avoid bias (Roland and Torgerson 1998; Patsopoulos 2011; Gamerman et al. 2019), pragmatic trials aim for heterogeneity in participants, treatments and settings with the aim to optimise the generalisability of the trial results (Patsopoulos 2011; Gamerman et al. 2019). Unlike explanatory trials, pragmatic trials could have different treatments/actions in one intervention (Roland and Torgerson 1998), and this is because they focus on person-centred outcomes (e.g., behaviour change at individual level) rather than the treatment-centred outcomes (Weinfurt et al. 2017). Nevertheless, heterogeneity in pragmatic trials might lead to dilution of the intervention effect. To overcome heterogeneity, pragmatic trials must have a large sample size. The Drive evaluation complies with these elements by focusing on a large heterogenous sample, where participants received individualised and targeted intervention with person-focused outcomes.

Sampling and randomisation: In the UK, most localities have a Multi-Agency Risk Assessment Conference (MARAC—see Learn more about the Marac–SafeLives (https://safelives.org.uk/about-domestic-abuse/domestic-abuse-response-in-the-uk/what-is-a-marac/, accessed on 13 December 2024). It should be noted that MARAC is victim-focused with emphasis on safety planning and support to victims. A MARAC is a meeting attended by representatives of local police, probation, health, child protection, housing practitioners, IDVAs and other specialists from the statutory and voluntary sectors. The role of the MARAC is to facilitate, monitor and evaluate effective information sharing in order to develop an action plan to ensure safety for highest-risk victims. All the perpetrators participating in the evaluation were those identified at MARAC as posing high harm and high risk to their victims.

Perpetrators associated with victims identified at a MARAC as at high-risk of further abuse (N = 2627) were randomly allocated to either Drive or usual MARAC care (involving IDVA support to victim) to assess behaviour change. Computer-assisted randomisation was carried out on a rolling basis by the SafeLives Research, Evaluation and Analysis team. The University of Bristol research team were provided with only anonymised data regarding the perpetrators and associated victims–survivors allocated to either the Drive intervention or control groups. The number of perpetrators allocated to the Drive arm was defined a priori, based on the capacity of the Drive case managers in the three sites. Therefore, a much greater proportion (80%) of perpetrators were allocated to the control arm.

Given that a key objective of Drive was to ensure safety and wellbeing for victims, we used data from victims–survivors to assess the effectiveness of Drive. We established two groups of victims–survivors (using IDVA data for both groups to ensure comparability):

• A comparison group to assess whether victims associated with perpetrators on Drive (the Drive victim–survivor group) experienced a reduction in abuse and risk to the same extent as reported by Drive perpetrators;
• A control group to assess whether the experiences of the victims–survivors associated with perpetrators on Drive (the Drive victim–survivor group) differed from the experiences of victims–survivors who did not have associated perpetrators on Drive (the control victim–survivor group);
• In addition, we established control groups of perpetrators in one site (Site Two), to assess longer-term outcomes and recidivism.

The intervention and control group are detailed further below.

As can be seen from Figure A1 and Figure A2 (in Appendix A), 3273 perpetrators were initially identified via victim data at the MARACs from the three sites as eligible for randomisation into intervention and control arms. Following the sifting out of repeat cases, this resulted in 2627 perpetrators, of whom 530 were randomised into the Drive intervention and 2085 into the control (usual victim care) group. This resulted in final samples of 506 perpetrators in the Drive intervention group and 603 perpetrators in the control group, where IDVA data were also available for the victim–survivor. Once allocated to the Drive intervention, due to their identification at MARAC as posing a high risk of high harm to others, perpetrators were not able to drop out. Drop out only occurred in relation to three perpetrators due to death or a move to another area.

Drive intervention cohort: All closed and completed cases allocated to Drive over the three years were included in the evaluation. A bespoke Drive case management system was used by all case managers to record all details regarding the cases they worked with, any intervention and related activity carried out, as well as start and closure dates. The case management system included a wide range of often detailed information about the needs of the individual perpetrators, the interventions used, their referrals between services and risks and behaviours of the perpetrators concerned. Domestic abuse behaviours were recorded as physical abuse, harassment and stalking (H&S), jealousy and controlling (J&C), and sexual violence. The case management data were exported to an Excel workbook, anonymised, password-protected and shared with the research team. The statistics reported in this article use the 506 completed and closed perpetrator cases from the Drive case management system.

Once in the Drive arm, participants were allocated a ‘case manager’ for about 10 months, who would attempt to make contact and engage them 1:1 work where safe to do so (known as ‘direct work’) as well as being responsible for leading on information sharing and multi-agency actions around that participant (known as ‘indirect work’). Although Drive cases varied between and within the three sites, the general process followed by case managers was as follows: the perpetrator was allocated to a case manager who then carried out background research about the individual. This would be followed by a mixture of indirect and direct work by case managers as appropriate. ‘Indirect work’ was generally much more common than ‘direct work’ and involved considerable multi-agency contacts and information sharing. Analysis of case managers’ recorded actions showed that 84% of case managers’ activities was indirect work related to perpetrators and direct work accounted for 16% overall, although they made direct contact with 65% (n = 330) of the perpetrators. A key aspect was liaison with the IDVAs to share information of importance to victim safety (and keep victims informed), including notification to the case manager if the perpetrator was creating risk and notification to the IDVA about work being carried out by case manager and other agencies to mitigate risk. When it was not possible for case managers to carry out direct work with a perpetrator, or where they needed to carry out additional checks, gather further information about perpetrator behaviour, engage levers or access expertise, Drive case managers worked with a range of local services to carry out indirect work around the perpetrator. Such multi-agency work therefore accounted for 26% of indirect work, with information sharing accounting for 60% and background research and information gathering accounting for 13%. Where case managers were able to engage perpetrators in direct work, this included a variety of face-to-face, often detailed and highly skilled work. Direct work included staying in contact with the perpetrators (68%) using emails, letters, text messages, telephone calls and home visits. Another type of direct work was the provision of direct support (21%), involving one-to-one meetings with case managers or other agencies and the perpetrators. Specific behaviour change work was less common, with only 36 service users (11%) receiving such intervention.

The Drive case managers used the Drive-DASH (Domestic Abuse, Stalking and ‘Honour’-Based Violence) risk indicator checklist to ascertain risk of significant harm from further domestic abuse posed from the perpetrator to the victim–survivor. The case manager compiled information for the Drive-DASH including information from perpetrators themselves, from police, IDVAs, etc. A Drive-DASH risk score was used to create risk profiles for each perpetrator and the research team were provided with the risk score data.

Victim–survivor cohorts: Outcome data for victims–survivors whose perpetrators were on Drive and who were themselves engaging with an IDVA were used to assess whether outcomes for Drive perpetrators were reflected in outcomes for Drive associated victims–survivors (see Figure A2 in Appendix A). There was available data for 104 victims–survivors of a possible 506, whose entry and exit forms were completed on the IDVA case management system (the ‘Insights’ data system, comparable to data recorded for the perpetrators). The smaller dataset for available for victims–survivors included only those in contact with an IDVA, while others did not access IDVA support possibly because they were no longer in contact with the perpetrator. IDVAs recorded a wide range of often detailed information regarding the victim–survivor on Insights, and victims–survivors completed an exit questionnaire.1 The research team were provided with anonymised Insights IDVA data and exit data for the associated victims–survivors.

A wider control group consisting of victims–survivors whose perpetrators had been randomly allocated to the control group rather than the Drive intervention, was also established. This victim–survivor control group consisted of 610 victims–survivors engaging with IDVAs and where Insights data were therefore available of a possible 2085 associated victims–survivors allocated to the control (Figure A2 in Appendix A). The research team were provided with anonymised IDVA data for these 610 associated victims–survivors, who thus constituted the victim–survivor control group.

Data regarding the support interventions applied for the Drive and control victim–survivor groups by IDVAs were analysed to ascertain comparability. Across the three sites results using an independent sample t-test indicated that the number of supports mobilised was similar, but slightly higher, for the Drive victim–survivor group (average = 5.4, SD = 1.7) than the control victim–survivor group (average = 4.6, SD = 1.8) (p > 0.05). the type of support the victims–survivors received did not differ between Drive and control victim–survivor groups, with most receiving safety planning (95%), 61% MARAC referral, 52% received health/wellbeing support and 47% received support from the criminal court process.

Recidivism and outcome data from police and MARACs: Both MARAC and police data were used to analyse for recidivism in the 12 months after perpetrators had completed Drive, specifically to ascertain if they had domestically abused the same or another victim. Analysis of MARAC data allowed for the assessment over time of the entire cohort, from the Drive and control groups. It also enabled comparison of Drive outcomes and ‘usual care’, as victims–survivors and perpetrators identified through MARAC and allocated to the control group also received multi-agency activities to ensure safety for victims. To assess post-Drive re-perpetration and involvement by the criminal justice system the research team were provided with anonymised longitudinal MARAC and police data for Drive intervention and control group perpetrators from one of the Drive sites (Site Two), covering before, during and up to at least 12 months after intervention by Drive. This resulted in a longitudinal MARAC cohort and longitudinal police cohort with intervention and control groups. It was not possible to obtain MARAC and police data for all three sites.

The longitudinal MARAC cohort included a total of 1323 in the one site, with 184 in the Drive intervention group and 1139 perpetrators in the control group. The research team were provided with anonymised MARAC data collated by SafeLives regarding the number of times the same perpetrators were identified at MARAC and could thus be deemed to have re-perpetrated.

The longitudinal police cohort included all the perpetrators in the Drive intervention group across the three sites with police involvement. A random subsample from the control group was compiled in order to reduce the workload for the police data team extracting the data. Stata version 14 was used to generate a random sample of control cases proportional to the number of Drive perpetrators (n = 506) and was also proportional to the number of cases per site. The research team provided SafeLives with the anonymous profiles of the perpetrators associated with Drive and a control victim–survivor sample, and SafeLives shared the actual profiles of perpetrators with the police. Using this approach, the police identified incident and crime data (domestic abuse-related and non-domestic abuse) for 149 Drive perpetrators relating to the six months before, during and in the 12 months after completion of Drive and 173 control perpetrators, which was provided in anonymised and password encrypted format to the research team for analysis.

Analysis and statistical methods: The main outcome measures reported here relate to domestic abuse behaviours by perpetrators and risk for victims–survivors. Descriptive statistics were used throughout to present frequencies regarding the samples, including demographics, interventions, behaviour profiles, and risk profiles. To identify the difference between changes in domestic abuse behaviour between the control and intervention group, Difference in Difference (DD) regression was used to assess the difference in average outcomes in the intervention group before and after the intervention minus the difference in average outcomes in the control group before and after the intervention. As we wanted to assess whether the changes in domestic abuse were due to Drive and not due to other factors, the DD regression was controlled for victim–perpetrator living arrangement, victim–perpetrator relationship status, perpetrator criminal and civil justice (CCJ) and children and young people services (CYPS) involvement. Four regression models were run for each of the domestic abuse behaviours recorded by case managers (physical abuse, sexual abuse, harassment and stalking, and jealousy and control).

Random-effect Poisson regression was conducted to assess how the risk score changed over time, accounting for other variables such as site. By using this method, we assume that the unobserved within-subject variation is uncorrelated or independent of the explanatory variables in all time periods (Wooldridge 2013). All analysis were carried using STATA version 14.

We considered the possibility of carrying out intention to treat analysis (ITT), as such analysis may indicate whether findings are overoptimistic. However, we were unable to obtain access to the behavioural data for 1482 perpetrators (see Figure A1 in Appendix A) and would therefore have only missing data if they were included in the sample size.

3. Results

Table 1. Descriptive demographics of Drive perpetrators at intake.

Perpetrators (n = 509)	n/Mean	%/SD
Age (years)	32	13.8
Men	476	93.5%
Women	33	6.5%
Ethnicity
White British	226	44.5%
White Other	8	1.6%
Asian/Asian British	6	1.2%
Arab	1	0.2%
Black/Black British	6	1.2%
Mixed race	6	1.2%
Prefer not to say	1	0.2%
Unknown ethnicity	254	50%
Missing values	1	0.2%
Perpetrator relationship to primary VS
Current intimate partner	65	12.8%
Ex-intimate partner	251	49.3%
Family member	36	7.1%
Another known person/associate	2	0.4%
Don’t know	29	5.7%
Missing values	126	24.8%
Living arrangements with the victim
Living together	44	8.6%
Living together intermittently	6	1.2%
Not living together	302	59.3%
Don’t know	31	6.1%
Missing values	126	24.8%

VS: victims–survivors; SD: standard deviation.

outlines demographic data for the 509 perpetrators allocated to Drive at baseline. Drive perpetrators ranged in age from 17 to 81 years, with an average age of 32 years (SD = 13.8). Most were men (94%), and where ethnicity was known, most (92%) identified as White British/White Other.

Demographic data for victims–survivors in Drive and control groups were available via IDVA Insights data. Most victims–survivors were women (97%), and where ethnicity was known, 91% of Drive and 93% of the control were identified as White British. The majority of victims–survivors (78% of Drive and 75% of control) reported no longer living with the perpetrator and being ex-intimate partners (64% of Drive and 59% of control) (see

Table 2. Descriptive demographics of victims–survivors at baseline.

	Drive Victim–Survivor Group (n = 196)		Control Victim–Survivor Group (n = 610)
	n/Mean	%/SD	n/Mean	%/SD
Age (years)	33	12.5	34	12.3
Men	5	3%	21	3.45%
Women	192	97.4%	588	96.5%
Ethnicity
White British	178	90.8%	568	93.4%
White Other	6	3%	11	1.8%
Asian British/Asian	11	5.6%	18	2.9%
Black British/Black	1	0.5%	8	1.3%
Other ethnic background	0	0%	3	0.4%
Sexual orientation
Heterosexual	191	97.9%	585	96.6%
LGB	2	1%	10	1.5%
No disclosure	2	1%	10	1.6%
Living situation
Living together	32	16.2%	117	19.2%
Living together intermittently	12	6%	35	5.7%
Not living together	153	77.6%	455	74.7%
Don’t know	0	0%	2	0.3%
Relationship to perpetrator
Intimate partner	54	27.4%	192	31.5%
Ex-Intimate partner	127	64.4%	357	58.6%
Intermittent intimate partner	4	2%	14	2.3%
Family member	11	5.8%	44	7.2%
Other	1	0.5%	2	0.3%
Children
Children present	123	62.4%	368	60.4%
Children not present	74	37.5%	239	39.2%
Don’t know	0	0%	2	0.3%

).

3.1. Change in Domestic Abuse Behaviour During Drive Intervention

Data from the Drive case management system allowed for the calculation of the percentages of perpetrators displaying abusive behaviours at different points in time during the Drive intervention: at intake, at midpoint and at completion of Drive. The number of Drive service users using each type of domestic abuse behaviour reduced substantially. Use of physical abuse reduced by 82% (67% to 12%), sexual abuse reduced by 88% (17% to 2%), harassment and stalking behaviours reduced by 75% (44% to 11%); and jealous and controlling behaviours reduced by 73% (45% to 12%). Figure 1 shows the change in severity for each of the types of domestic abuse for those allocated to Drive, whether rated high, moderate, standard or none. A major decrease in use of domestic abuse was observed for those perpetrators who reported high levels of domestic abuse, in particular for physical abuse, followed by harassment and stalking (H&S) and jealousy and controlling (J&C) behaviours.

The trends observed in the reduction in abusive behaviours experienced by the Drive victims–survivors and control victims–survivors are similar (see Figure 2). We applied a McNemar’s test ensure that the difference in proportions from intake to exit were statistically significant. By running the test separately for each of the four types of domestic abuse and for each arm it was shown that changes in proportions from intake to exit were statistically significant for both groups (p < 0.001).

While the trends are similar, the reduction in domestic abuse behaviours for the Drive victims–survivors arm was 8% higher for physical abuse, 9% for harassment and stalking and 4% for jealousy and control than the reductions in the control victims–survivors group. The exception was for sexual abuse for which the control group show a slightly greater reduction.

Given the moderate difference from the descriptive analysis between the Drive victim–survivor group and Control victim–survivor group we also applied Difference in Difference regression to assess statistical significance. Changes of the four domestic abuse behaviours from intake to exit were found not to be statistically different between the Drive and control victim–survivor groups as indicated by p-values (

Table 3. Difference in Difference regression results for all four DVA behaviours for the intervention (n = 196) and control group (n = 610).

	Physical Abuse				Sexual Abuse				Harassment and Stalking				Jealousy and Control
	OR	p Value	95% Conf.		OR	p Value	95% Conf.		OR	p Value	95% Conf.		OR	p Value	95% Conf.
			Interval				Interval				Interval				Interval
Time#allocation	0.69	0.45	0.26	1.81	0.68	0.64	0.13	3.44	0.56	0.15	0.26	1.22	0.67	0.33	0.3	1.49
Time (ref. intake)	0.03	0.00	0.02	0.05	0.1	0.00	0.05	0.2	0.08	0.00	0.05	0.11	0.03	0.00	0.02	0.05
Allocation (ref. control group)	1.14	0.57	0.72	1.82	0.98	0.93	0.56	1.7	1.35	0.2	0.85	2.15	1.27	0.33	0.79	2.04
Living situation (ref. Not living together)
Living together	1.23	0.47	0.71	2.12	1.46	0.24	0.78	2.75	0.49	0.01	0.29	0.83	0.83	0.48	0.49	1.39
Living together intermittently	0.81	0.64	0.34	1.93	0.8	0.69	0.28	2.33	0.58	0.2	0.25	1.33	0.62	0.27	0.27	1.45
Current relationship (ref. Current intimate partner)
Ex-intimate partner	0.69	0.1	0.44	1.07	0.76	0.33	0.45	1.31	1.19	0.39	0.8	1.76	1.09	0.67	0.73	1.64
Intermittent intimate partner	2.24	0.17	0.72	6.98	2.24	0.16	0.73	6.89	1.16	0.78	0.41	3.29	0.65	0.41	0.24	1.81
Family member	0.65	0.27	0.31	1.39	0.28	0.05	0.08	0.99	1.47	0.27	0.75	2.89	0.25	0.00	0.11	0.55
Another person	0.53	0.57	0.06	4.92	0	0.98	0	0	0.21	0.2	0.02	2.28	2.22	0.47	0.25	19.61
Criminal record for DV (ref. No)	0.98	0.92	0.65	1.48	0.87	0.59	0.51	1.47	1.3	0.17	0.89	1.89	1.42	0.07	0.97	2.09
CYPS involvement (ref. Yes)	0.85	0.38	0.6	1.22	1.15	0.54	0.74	1.79	1.71	0.00	1.23	2.36	1.05	0.78	0.76	1.46

).

3.2. Change in Risk Scores

3.2.1. Case Manager Assessment of Risk During Drive Intervention

Case managers completed the Drive-DASH at intake, midpoint and case closure. After matching questions as closely as possible, 20 questions were included in the analysis we present here. A total of 503 perpetrators had completed Drive-DASH scores across the three time points. For each time point the ‘yes’ responses were totalled to construct the Drive-DASH score (ranging from 0 to 20, the higher the number the higher the risk).

Nonetheless, there were a significant amount of ‘not applicable/not known’ responses across time points, and 49% of perpetrators had 11 or more ‘not applicable/not known’ responses in their score and therefore were not included in the analysis (23% intake, 14% midpoint and 12% case closure).

The results showed a reduction in the average Drive-DASH score from 8.3 at intake, to 7.4 at midpoint of intervention, to 6.6 at case closure. Results from the random-effect Poisson regression indicated that the reduction in Drive-DASH scores for perpetrators from intake to case closure was statistically significant (p-value < 0.001) (

Table 4. Results from random-effect Poisson regression assess changes in risk score over time via Drive-DASH score.

N = 506	IRR	p-Value	[95% Conf. Interval]
Time point (ref. Intake)
Middle	0.91	<0.001	[0.86, 0.95]
Case closure	0.82	<0.001	[0.78, 0.87]
Site (ref. Site 1)
Site 2	0.95	0.27	[0.86, 1.04]
Site 3	1.07	0.17	[0.97, 1.18]

IRR: incidence rate ratio.

).

3.2.2. IDVA Assessment of Change in Risk for Drive Comparison and Control Group Victims–Survivors

Following their intervention with victims–survivors, at exit, IDVAs recorded on Insights the extent to which they thought the risk posed to victims–survivors had changed since intake for both Drive and control victims–survivors’ risk. Overall, there was a significant or moderate reduction in risk for both groups, although the Drive victim–survivor group was more likely to experience what IDVAs saw as a significant reduction (45.1%) than the control group (40.1%). Also, the control group were more likely to have had only limited or no reduction in risk (14.6%) compared to the Drive comparison group (8.6%) (Figure 3). The IDVAs perception of whether risk reduction was sustained showed that a higher proportion of Drive victims–survivors had risk permanently eliminated (11.2%) than the control group (4%). Thus, while IDVA support was effective for both the Drive and control victims–survivors, the Drive intervention showed a further reduction in risk.

3.3. Recidivism and Sustaining Behaviour Change Following the Drive Intervention

We wanted to assess whether behaviour change by the perpetrators receiving the Drive intervention continued to be apparent at least 6 months and 12 months after the completion of Drive, and to compare this to the wider MARAC population. Using data from Site Two of Drive, we therefore ascertained whether perpetrators were reported again to MARAC and/or to the police in the 6 months and 12 months after Drive case closure and compared this to re-appearance of control group perpetrators at MARAC or in the police records across the same time periods.

3.3.1. MARAC Data Regarding Re-Perpetration Post-Intervention

Out of the 506 closed Drive cases, 184 perpetrators (37%) were from Site Two. Analysis showed a definite reduction in the appearance of Drive perpetrators at MARAC both during Drive and following the completion of Drive. Seventy-one percent of the perpetrators did not appear again in MARAC during Drive, and there was a further reduction following the intervention. Only 15 perpetrators (8%) re-appeared at MARAC in the 6-month following completion of Drive, and most of these individuals were recorded only once at MARAC while two individuals re-appeared three and four times respectively.

Regarding the 12 months after case closure, data were available for 64% of Drive perpetrators (n = 117). Only 12 (6%) of these individuals re-appeared at MARAC at 12 months post-completion of Drive and thus showed an overall reduction in re-appearance by Drive perpetrators at MARAC during this longer period. By comparison, there was an increase in the re-appearance of perpetrators from the wider MARAC control group. The data from Site Two showed that between the start of Drive in February 2016 and the latest available data from MARAC in September 2019 the perpetrators in the control group re-appeared slightly more frequently at MARAC (mean = 3.3 times) than the Drive perpetrators (mean = 2.7 times), and this difference was statistically significant (p < 0.001).

While this analysis was limited to a subsample of the Drive cohort, only from Site Two, it indicates that for this particular cohort, Drive contributed positively to reduce high-risk re-perpetration, with the effect sustained for a year after the cases were closed.

3.3.2. Police MARAC Data Regarding Re-Perpetration Post-Intervention

As outlined in Section 2, a subsample of perpetrators in the control group were selected for the police forces for the three sites to extract the data required for analysis. The analysis of police data shows that the reduction in domestic abuse behaviours by the perpetrators in the Drive intervention group continued after completion of Drive. The police records indicated a consistent and continuing reduction in domestic abuse-related incidents during and for more than 12 months following completion of Drive in relation to Drive perpetrators generally, serial and repeat. The police also recorded reductions in domestic abuse-related incidents for the control groups, but not to the same extent nor consistency as was found regarding Drive perpetrators.

The police at site Two recorded a total of 2256 domestic abuse incidents within the Drive evaluation period for the whole population group, and perpetrators may be recorded multiple times. Out of the 184 perpetrators from Site Two, 149 Drive perpetrators (81%) had police information of at least one police-recorded incident in a period of 4 years. From the randomly selected control group for Site Two (n = 184 control perpetrators) 173 cases (94%) had police data. The number of police domestic abuse incidents per Drive perpetrator ranged from 0 to 41 in a four-year period (mean = 9.4, SD = 8.0), whereas for the control group the number of incidents ranged from 0 to 134 (mean = 24.4, SD = 33.4).

When looking at the total number of domestic abuse-related incidents distributed by periods it is possible to observe a reduction in the number of domestic abuse-related incidents during and after Drive for both groups, and a slightly greater reduction was observed for those who received the Drive intervention (Figure 4). For Drive perpetrators the reduction in incidents from the period before Drive to 12 months after was 80%, and for the control group was 77% for a similar period.

3.4. Serial and Repeat Perpetration Post-Intervention

3.4.1. MARAC Data Regarding Serial Perpetration Post-Intervention

A key objective of the Drive intervention was to reduce serial perpetration, where perpetrators use domestic abuse against another partner or family member. The MARAC data from Site Two were analysed for changes in serial perpetration during and after Drive. There were a total of 1323 Drive and control group cases in the MARAC data for Site Two with 385 cases considered to be serial perpetrators (29%). We identified serial perpetrators as cases where the victim–survivor was a different individual on the different occasions the perpetrator appeared in MARAC, using victim ID as the identifier. Of the 184 Drive perpetrators in the MARAC data for Site Two, 42 (22%) were identified in this way as serial perpetrators.

The analysis showed a definite reduction in the proportion of serial perpetrators re-appearing at MARAC by both 6 months and 12 months after completion of Drive (Figure 5). Comparison of serial perpetrators in the Drive and MARAC control groups indicated that the control group serial perpetrators re-appeared more times (mean = 1.5 times) than Drive serial perpetrators (mean = 0.8 times) and that the difference was statistically significant (p < 0.001).

While more of the Drive cases appeared in MARAC during the Drive intervention (50%) than the control cases (35%), at 6 months following completion of MARAC the percentage had become similar in both groups (17% Drive and 16% control). However, by 12 months post-intervention and later, the Drive serial perpetrators showed a greater reduction in re-appearance at MARAC than the control serial perpetrators. Thus, Drive was probably more effective than ‘business as usual’ at reducing high-risk domestic abuse behaviours among serial perpetrators in Site Two.

3.4.2. Police Recorded Incidents—Serial and Repeat Perpetrators

To ascertain further whether the reduction in domestically abusive behaviours was sustained over time for serial and repeat perpetrators, we also analysed repeat and serial perpetrators in the police data. There were 322 total cases in both the MARAC and police data, and of these, 70 cases were identified as serial perpetrators (22%) and 260 cases as repeat perpetrators (81%).

The 29 serial perpetrators recorded by the police were found to have committed 643 incidents in a period of four years, with a mean number of domestic abuse-related incidents per serial perpetrator of 5.4 (SD = 11.4). For Drive serial perpetrators the number of incidents ranged from 0 to 19 incidents per case (mean = 1.5, SD = 3.5), and for control serial cases the number of incidents ranged from 0 to 59 incidents (mean = 7.5, SD = 13.5). The trends identified earlier were also observed in the police data, with a large drop in the percentage of Drive perpetrators committing domestic abuse-related incidents of 82% (from 80% to 8%) and a smaller drop for control serial perpetrators of 56% (from 100% to 44%) following completion of Drive (Figure 6).

Police data were also used to identify repeat perpetrators, that is individuals recorded committing two or more domestic abuse-related incidents against the same victim. As indicated above, there were a total of 322 total cases with both MARAC and police data in Site Two. Of these, 260 cases were recorded by the police as repeat perpetrators (81%). The police recorded a total of 2204 domestic abuse-related incidents committed by this group of perpetrators in a period of four years, with a mean number of domestic abuse-related incidents per repeat perpetrator of 21.6 (SD = 30.7). Drive repeat perpetrators had committed incidents ranging from 2 to 41 incidents per case (mean = 10.9, SD = 8.8), while control repeat perpetrators committed between 2 to 134 incidents (mean = 16.9, SD = 34.4) (Figure 7) Echoing the findings regarding serial perpetrators, there was a greater reduction in repeat perpetration over the longer time by the Drive perpetrators as compared to the control, but with a more complex picture. The repeat Drive perpetrators had a greater reduction in the number of domestic abuse-related incidents up to 6 months following completion of Drive, but the control group reduced their number of incidents to a greater degree to 6 months. Figure 7 shows the reduction in repeat perpetrators committing domestic abuse-related incidents across the different time points. The reduction for Drive perpetrators is greatest up to 6 months post-intervention and continues after that, if less sharply. There was also a reduction for the control repeat perpetrators, but less than for the Drive perpetrators, and with an increase rather than a reduction after 12 months.

4. Discussion

The aim of the Drive intervention was to reduce the number of victims of domestic abuse and increase their safety by deterring perpetrators from using abusive behaviours, in particular physical violence, sexual violence, harassment and stalking, and jealousy and control. A related aim was reducing repeat and serial perpetration. In this paper we have outlined the quantitative findings from a high-quality quasi-experimental evaluation assessing the effectiveness of Drive regarding these aims over a three-year pilot period. It should be noted that the Drive intervention was complex, with individualised and targeted support and disruption for each perpetrator. Moreover, the work of Drive case managers to develop multi-agency co-operation and links is likely to have influenced the wider work in the three localities and may thus also have affected the support to victims–survivors in the control group. The results from the evaluation, which are modest, should be seen in this context.

Previous evaluations have been criticised for focusing on physical violence and recidivism as main measurements of effectiveness. Others have found that reductions in physical violence may be accompanied by increases in emotional or other forms of domestic abuse as the perpetrator adopts other means of controlling his partner (see Kelly and Westmarland 2015, for overview). Also, basing ‘success’ on levels of officially reported or recorded incidents of physical violence is problematic because police-recorded incidents may actually increase in the immediate term as victims–survivors feel more empowered to report (Gondolf and Jones 2001; Hester and Westmarland 2005). Moving beyond previous evaluations, we were able to look not just at changes in physical violence, but also other forms of abuse such as harassment and controlling behaviour. As our results show, the Drive intervention showed a modest effect in reducing domestic abuse behaviours by the perpetrators in the intervention group, and this reduction was echoed by data from the associated victims–survivors. Moreover, Drive was especially effective in reducing domestic abuse behaviours for those perpetrators who reported a high severity of physical abuse, harassment and stalking, and jealousy and controlling behaviour. Echoing the reduction in abusive behaviours, there was also a significant reduction in the risk associated with the Drive perpetrators across the intervention period, again echoed by the associated victims–survivors.

Repeat and serial perpetration have previously been identified as signifiers of particularly harmful and high-risk domestic abuse (Robinson 2016). Regarding the reduction in repeat and serial perpetration, Drive also achieved its aims, with significant decreases in domestic abuse repeat and serial perpetration across the period of the Drive intervention and in the twelve months following completion.

However, what is also notable from our results is that the control group reported similar patterns of outcomes to the Drive cohort regarding the reduction in domestic abuse perpetration, reduction in risk of further domestic abuse behaviour and reduction in repeat and serial perpetration. Both Drive and control perpetrator cohorts were found to sustain the reduction in domestic abuse in the longer term, to 12 months following the intervention period, although police data showed the beginning of an increase in the control cohort of perpetrators reported to the police after 12 months after the intervention period. The control group of victims–survivors also experienced a substantial reduction in domestic abuse from the support they obtained as ‘usual care’. Overall, the Drive intervention was seemingly more effective at reducing domestic abuse perpetrator behaviours than ‘usual care’, although the difference was only significant regarding the reduction in risk, and repeat and serial perpetration. So how might we interpret these results? This brings to the fore some key questions about existing ‘usual care’, as well as the nature of the control we used for comparison with Drive.

What our findings appear to show is that the context is an important consideration. Two elements need to be considered in this respect: the wider multi-agency ecosystem, and the specialist IDVA support to victims–survivors. Pence and Paymar (1993) in their ground-breaking work on domestic abuse in Duluth, Minesota, talk about the importance of situating work with perpetrators in a ‘co-ordinated community response’, where the work with perpetrators is merely one aspect and safety and empowerment work with victims–survivors is another within a wider multi-agency context. This is also akin to an ecological systems model where individual work is linked to community work and wider belief systems (at micro, meso and macro levels—see Bronfenbrenner 1977). In many respects we can see this approach echoed in the Drive model. A central feature of the Drive model is the case manager approach situated in a responsive multi-agency ecosystem, with the IDVA support to the associated victims–survivors. As indicated in the introduction to this paper, much of the work carried out in relation to the perpetrators on the Drive intervention comprised indirect activities with and by other agencies in the multi-agency system (e.g., housing agency support or disruption by the police), and only 36 of the 506 perpetrators on Drive received the type of direct one-to-one perpetrator behaviour change work that might be found in general group programmes. Thus, the wider multi-agency context played an important part in the work carried out to produce a reduction in domestic abuse for the Drive cohort. However, such multi-agency work may also have provided many of the same benefits and features for the control cohort, who were based in the same localities, and thus helps to explain the reduction in domestic abuse that we saw for the control group as well. Both the Drive and control cohorts of perpetrators and associated victims–survivors were situated in and subject to the same local multi-agency systems of MARAC and IDVA victim support services. The existence of Drive, and Case Managers’ contacts with different agencies, which included elements of institutional advocacy, may also have resulted in greater efficacy of the local multi-agency systems more generally and links across agencies in the three sites (Hester et al. 2020).

The use of female partner reports has long been proposed as a valid and reliable measure of outcome or ‘success’ (e.g., Gondolf and Jones 2001). However, it is important to note that female partner or other associated victim–survivor accounts of perpetrator behaviour or attitudinal change can only be seen as reliable or valid if those partners or associated victims–survivors are actually in a position to reliably assess change, that is, are still in a relationship or have regular contact with the perpetrator (Kelly and Westmarland 2015). The associated victims–survivors included in our Drive and control groups were only those engaged with IDVAs and therefore constituted only a proportion of the victims–survivors directly affected by the Drive or control perpetrators. We do not know if the victims–survivors not engaging with IDVAs were experiencing more or less domestic abuse, but we do know that their cases were also reported to MARAC due to concern that they might experience high risk of ongoing domestic abuse. The Drive intervention, through the ‘support and disruption’ approach, works by reducing the opportunity for abusive contact with survivors from perpetrators. Baseline data indicate that victims–survivors in the intervention group (77.6%) were more likely to be living apart from perpetrators than victims–survivors in the control group (74.7%) and the Drive victims may perhaps have had an advantage in this respect from the onset which could be reflected in the results. However, their involvement with IDVAs suggest that they were still experiencing problems from the perpetrators, and indeed the wider literature indicates many victims experience further abuse post-separation (Robinson and Howarth 2012).

Involving only victims–survivors who were engaging with IDVAs meant that we introduced a very stringent test in the comparison between Drive victims–survivors and the control group of victims–survivors. Other research has shown that IDVA intervention can have a positive effect for victims–survivors (Howarth et al. 2009), and we would suggest that there is also a strong ‘IDVA effect’ in this evaluation, which supported the reduction in domestic abuse for both Drive and control cohorts. It should be noted that our quantitative exploration of IDVA support to victims–survivors showed that the Drive victim–survivor group received a longer period and greater number of contacts from IDVAs than did those in the control group, but similar number and types of support. Thus, the Drive victims–survivors received only marginally more intense support than the control group. The reduction in domestic abuse for victims–survivors attributed to Drive are probably smaller when compared to this particular control group than if we had used a control group comprised of victims–survivors without similar support.

5. Conclusions

The conclusion we draw is that Drive ‘works’ in the sense of providing a modest reduction in domestic abuse perpetration and leading to less risk for victims–survivors. Moreover, the wider multi-agency system is not only crucial to the implementation of Drive but may also in itself lead to a reduction in domestic abuse perpetration without the inclusion of the Drive Case Manager and associated model, i.e., ‘usual care’ also has a positive effect. Overall, however, Drive is marginally more effective than merely ‘usual care’ in reducing domestic abuse and in sustaining the reduction in the re-perpetration of high-risk, and especially high-harm serial perpetrators, and Drive may also be resulting in enhanced ‘usual care’ in the localities where it is provided.

6. Strengths and Limitations

Among the limitations of this study, missing data in the case management system, especially regarding the risk scores for perpetrators, may have distorted the results. However, case managers appeared to approach record keeping in a ‘positivist’ sense, with greater likelihood of recording items where these were deemed important.

Using IDVA data as a comparison and control to assess the effectiveness of Drive has limitations as not all victims–survivors engaged with an IDVA, and in the randomisation process fewer victims–survivors were allocated to the Drive victims–survivors sample than to the control victims–survivors sample. Moreover, some IDVAs failed to submit forms to Insights, resulting in a smaller sample than the overall number of victims–survivors engaging with an IDVA. Also, not all victims–survivors were in contact with an IDVA throughout the intervention period leading to some missing data. Nonetheless, using IDVA data was the most robust and viable option to obtain data regarding domestic abuse behaviours experienced for either group of victims–survivors. Moreover, using this approach overcame the problems faced by previous evaluations (see Kelly and Westmarland 2015), while only including victims–survivors engaging with IDVAs also meant that we introduced a very stringent test in the comparison between the Drive victims–survivors and the control victims–survivors.

Another limitation of this study refers to the intervention follow-up findings. Post-intervention MARAC and police data were only available for one of the three sites, albeit the largest site, thus limiting our analysis of the whole population. This affects the generalisability of the findings. We obtained police data from a further site, which showed similar results; however, it is unknown whether the follow-up findings sustain for this and the third site.

Finally, as indicated earlier, we considered the possibility of carrying out intention to treat analysis (ITT) to investigate whether our findings are overoptimistic but were unable to do so given the missing behavioural data for 1482 perpetrators (see Figure A1 in Appendix A). While the denominators for the control group would be different, this is unlikely to affect the effects of the intervention. Considering that the difference in difference regression analysis that we carried out showed no significant effect by allocation arm, we decided that including ITT analysis would be unlikely to have improved our findings.

A key attribute of pragmatic trials is that they are designed to answer questions more relevant to policy makers rather and elucidating biological or social mechanisms (Weinfurt et al. 2017). The results from this evaluation were used to inform UK government policy and led to an increase in the number of sites where it was delivered, and an overall increase in the number of perpetrators subject to the Drive intervention. Unfortunately, the continuation of Drive since our evaluation was completed has not included randomisation and it has therefore not been possible to move to a full trial of the intervention. We are currently carrying out a follow-up study to examine whether Drive continues to show the promising, if modest, results found in the evaluation reported here.