Early diversion and empowerment policing: evaluating an adult female offender triage project

ABSTRACT This paper provides an evaluation of a police pilot early-diversion scheme for adult females who were arrested for low-severity offences using a natural experiment design. The intervention is novel in that it diverts arrestees to a women’s centre for assistance to address their criminogenic needs rather than process them through the criminal justice system. The intervention is timely and attractive given its rehabilitative features and its potential for reducing demand on the criminal justice system through community resolution. The study found a promising effect of the intervention on rates of rearrest and daily risk of rearrest over a 12-month follow-up period, but a higher frequency of rearrest among those of the intervention group who were rearrested. The findings are discussed in relation to the political context, theoretical background and police performance and the gendered dynamics of offending.


Introduction
This paper describes and tests the effect on rearrest of a police pre-charge diversion scheme for female adult offenders arrested for low-severity offences. The projectthe Adult Female Triage Projectis unique for adult female arrestees in that it sought to divert them out of police custody towards a supporting organisation, Together Women Project (TWP), before and instead of being charged or cautioned. While early-diversion schemes are an expedient and increasingly popular mechanism for managing low-severity arrestees their effectiveness is unknown. This study provides the first evaluation of such a scheme for adult females and presents the political, theoretical and empirical background to this increasingly popular police approach. The project was undertaken in a region of the UK, but its outcomes will be relevant to police officers and researchers around the globe who are exploring new ways of improving police effectiveness and efficiency in an era of shrinking resources.
The primary method described in this evaluation is a natural experiment of the effectiveness of the intervention in reducing rates of rearrest, time-to-rearrest and number of subsequent rearrests within a one-year follow-up period. The paper is also informed by a series of interviews with police officers, youth justice and Drug Intervention Programme (DIP) workers delivering the intervention in the custody suite; observations within the custody suite; interviews with staff at TWP and attendance at the intervention management meetings which were undertaken to fully understand the process of developing and delivering the intervention and the underlying drivers that led to its introduction.

Background to the intervention
In December 2012 Humberside Police launched an innovative diversionary scheme from police custody for adult female offenders. Building on the well-established arrest referral scheme introduced under the Police and Criminal Evidence Act (PACE) 1984, which later became the DIP in the Drugs Act 2005, a screening assessment was conducted with women detained in the police custody suite to determine their eligibility for pre-charge diversion. This screening assessment was based upon a triage model developed by Hull Youth Justice Service as part of a national strategy to develop informal alternatives to criminal justice sanctions for young offenders. The youth justice intervention was designed to identify low-level, less serious and first-time young offenders (Bateman 2012) who could be appropriately dealt with outside the criminal justice system. For young offenders this has typically involved an expansion of the warning and reprimand scheme introduced under the Crime and Disorder Act 1998 to include: letters of apology, restorative meetings and reparation (Home Office 2012a). Using similar criteria, Humberside Police introduced a triage assessment tool to assess suitability of female arrestees for diversion to TWP instead of cautioning or charging those who admitted their offence.
Since the Corston Report into women in the criminal justice system (Corston 2007) there has been a growing recognition by policy-makers and criminal justice practitioners in the UK that existing provision for low-severity female offenders has been patchwork and potentially counterproductive in reducing the risk of reoffending. In particular, a succession of governmental reports (Corston 2007(Corston , 2011, House of Common Justice Committee 2013) highlighted a combination of domestic, personal and socio-economic vulnerabilities that increase the risk of female offending and that should be addressed separately from male offending. In her report, Baroness Corston suggested that this requires: A radical new approach, treating women both holistically and individuallya woman-centred approach. I have concluded that there needs to be a fundamental re-thinking about the way in which services for this group of vulnerable women, particularly for mental health and substance misuse in the community are provided and accessed; there needs to be an extension of the network of women's community centres to support women who offend or are at risk of offending and to direct young women out of pathways that lead into crime. (2007, p. 2) One of the main outcomes of this recommendation was the establishment of TWP which is a regional charity with centres in Bradford, Hull, Leeds, Sheffield and HMP New Hall adult female prison. TWP was established with funding from the Ministry of Justice 'to develop and test a new gender-specific community approach to women offenders and women at risk of offending ' (Togetherwomen.org, 2014). The ethos of TWP is to provide a safe women-only space in which women can access support and services through TWP and other similar agencies (Heidensohn and Silvestri 2012). This provides a 'one-stop-shop' for women who can then address all of their needs in one location (Together Women 2012). As a charity TWP is therefore well positioned to work with both female offenders (or those at risk of offending) and statutory criminal justice agencies like police and probation. With a heavy emphasis on early intervention and diversion TWP seeks to work with women to address the underlying problems that it is believed often lead women to offend (Corston 2007).
Historically in the UK, the options available for handling low-severity female offenders have been no different to the options available for low-severity male offenders and these operate at two levels: in the police station and through the Magistrates' Court. In the Magistrates' Court the two main sentencing options are a fine or a community order. Whilst both types of sentence leave the offender at liberty in the community and can be understood as types of diversion, these diversions occur at the post-conviction stage and operate at a later and different point in the criminal justice system than the intervention described here. In the police station the police have traditionally had three options for dealing with suspects: charge, caution and no further action. Charging a suspect represents the point at which the police pass responsibility for prosecution over to the Courts and Crown Prosecution Service. Cautioning a suspect is an official warning in circumstances where there is enough evidence to charge but the circumstances and severity of the offence does not warrant a prosecution. 'No further action' is the release of the suspect when there is insufficient evidence to charge or caution. In addition to these options an arrest referral scheme was introduced into the police custody suite in the late 1990s and was consolidated under the mantle of the DIP in 2005. The DIP is not administered by the police but by a separate agency working in partnership with the police in the custody suite. This arrest referral scheme involves identifying drug and alcohol related offending behaviour and referring the suspects for support and monitoring. Low-severity suspects are usually given a period of unconditional police bail and referred to the DIP programme for drug and alcohol rehabilitation. For more serious offences the referral can be made as well as, rather than instead of, formally charging the suspect.
There are therefore three important progenitors for this intervention. The first is a growing awareness since the Corston report (2007) of the need for holistic empowerment and support services for female offenders and women at risk of offending. The second is the existence of the DIP arrest referral scheme operating within the police custody suite and the third is the growth of local youth triage schemes designed to avoid criminalisation and reduce the risk of reoffending amongst low-severity young offenders. These three progenitors combine to provide the necessary skills, partnership and rationale for the adult female offender triage intervention.
While diversion from custody schemes have been implemented widely, schemes that seek to divert from the criminal justice system prior to charge are rare for adults and robust empirical evidence of their effectiveness rarer still. Empowerment support work, such as provided by TWP, has also existed for many years. However, to our knowledge, this two-pronged intervention for adult females is the first of its kind to be evaluated. Consequently, there is a limited evidence base on which to predict the success of this intervention in reducing or delaying reoffending. Early-diversion schemes are more common for juvenile offenders, but a meta-analysis failed to find a general effect of these schemes on reoffending (Schwalbe et al. 2012), although some support for the effectiveness of restorative justice and family interventions was demonstrated. Adult drug or alcohol arrest referral (AAR) schemes in England and Wales also lack a rigorous evaluation of their effectiveness and a Home Office evaluation of the AAR Pilots found that they had no effect on re-arrest rates (McCracken et al. 2012). A meta-analysis of diversion and referral schemes for Class A Drug-users concluded there was no reliable evidence that the DIP was either effective or cost effective (Hayhurst et al. 2015). Furthermore, no comprehensive review of the effectiveness of post-charge diversion schemes for adult females exists. An evaluation of diversion after charge to TWP found no effect of the intervention on reoffending of female arrestees at 12-month follow-up, compared to a matched sample of women serving a community order (Jolliffe et al. 2011). The Adult Female Triage Project differs from that evaluated by Jolliffe et al. (2011) in the timing of the diversion prior to the receipt of a criminal record, which is arguably an important factor in desistance and one that informs the hypotheses of this research, which are detailed below.
The political dynamics of austerity, empowerment and police performance There are several key policy drivers behind the introduction of the project. The first of these can be crudely summarised as an austerity driver. The police, like the rest of the UK public sector, have had to make significant savings as their budgets have been cut to help reduce the national debt. The Adult Female Triage Project is motivated in part by the desire to reduce the burden on both the police and criminal justice system by diverting low-severity female offenders from the system whilst also reducing reoffending.
The second policy driver is the current vogue towards early-intervention initiatives. Whilst this may also have a root in public spending cuts the wider political and policy context can be located within the Conservative's 'Breakdown Britain' report (Social Justice Policy Group 2006) and the Coalition Government's Troubled Families Programme with their emphasis on a collapsing welfare system and associated social and economic problems that include: family breakdown, educational failure, economic dependence, indebtedness and addiction (Department for Communities and Local Government 2012). The policy infers that these problems lead to increasing dependency, institutionalisation and stigmatisation which can lock people into socially excluded and criminogenic conditions. Without wishing to be drawn into the wider critical debates about the Troubled Families Programme (for good examples see Levitas 2012, Hayden andJenkins 2014) the Adult Female Triage Project has taken inspiration from the early intervention, cost-saving and family support aspects of this policy agenda. By offering arrestees support and advice instead of a criminal caution or conviction the intention of the Adult Female Triage Project is that these women are empowered to improve their lives and the lives of their families. This approach was further bolstered by the timely presentation of the Swift and Sure Justice Government Plan (Ministry of Justice 2009) that emphasised a need to innovate to improve efficiencies and promote public confidence in criminal justice by diversifying the supply of services to help rehabilitate offenders. The project also resonates with the 'Big Society' agenda and its underlying themes of more voluntary, third sector and commercial partnerships with the public sector whilst simultaneously trying to enhance accountability and responsibility in civic society (Jordan 2010, Bochel 2011, Maguire 2012.
A final, crucial, policy driver has been the introduction of new police detection outcomes. From April 2013 a new, broader range of 'crime' outcomesexpanded further in April 2014were introduced by the Home Office to include the additional category of 'community resolutions' (Home Office 2013, 2014. This allowed the new possibility of formally recording diversionary schemes like the Adult Female Triage Project on the same terms as a detection, thus improving performance statistics by including a wide range of successful policing outcomes.

Theoretical framework of the intervention: labelling and empowerment
During the course of our interviews with staff involved in the intervention two theoretical drivers were routinely alluded to in their justification of the intervention: labelling and empowerment. Both of these were closely tied to the primary purpose of the intervention which was to reduce reoffending rates amongst low-severity adult female offenders.
Whilst not the language explicitly used by staff, the first theoretical basis for the intervention is the labelling perspective (Lemert 1951, Becker 1963, Goffman 1968). This theory suggests that the early diversion of low-severity female offenders away from charge and prosecution will move women away from a criminal label and towards assistance through TWP. Rather than stigmatise, alienate and criminalise women TWP offers support and safety to reduce the risks associated with female offending. While this could be viewed with a degree of cynicism when set alongside the political dynamics discussed earlier, officers and staff working in the custody suite reported that having the option to divert some low-severity female offenders prior to charge was a far fairer, more benevolent and less depressing option than a caution that offered little or no support.
The second way staff understood the purpose of the intervention was to support and empower, rather than punish, women who were offending primarily because of social circumstances rather than any underlying malicious or inherently criminal mindset. This links to a wider criminological debate about a culture of chivalry within criminal justice towards certain categories of female offenders (Walklate 2009). Evidence that this attitude exists is mixed and generally stronger for Magistrates' Court outcomes than the police (Eaton 1986, Carlen 1988, Hedderman and Gelsthrope 1997. In the Magistrates' Court women appear to be sometimes judged in terms of how closely they conform to a feminine role rather than the offence they have committed. Hedderman and Gelsthorpe's (1997) research suggests that female defendants are distinguished as either 'troubled' or 'troublesome'. 'Troubled' women are those in need of help whereas 'troublesome' women are those who transgress gender expectations through the crimes they commit or how they present themselves.
The Adult Female Triage Project is definitely aimed at women who are 'troubled' and in need of support from TWP. Whether or not this represents a positive support for women or a form of misplaced paternalism is not a question we can answer. However, these gendered dynamics are integral to the theoretical framework underpinning this project. This is because it provides an implicit explanation for low-severity female offending that also justifies the form of diversionary activity undertaken by TWP. Gelsthorpe et al. (2009) explore these dynamics in relation to the types of provision that promote desistance from crime for women and conclude that these amount to precisely the sort of help offered by TWP. These types of support can be summarised as part of an empowerment project designed to provide women with a safe space away from potentially abusive partners in which to access individually tailored support and develop independent skills (Heidensohn and Silvestri 2012). Consequently, the belief that low-severity female offending should be understood as being caused by a combination of desperation, oppression and abuse is at the heart of the project. In terms of what types of women are deemed appropriate for diversion and the type of empowerment project they are diverted to, there is a clear feminist-inspired theoretical framework. Previous studies of Together Women's Project (Hedderman et al. 2008, Jolliffe et al. 2011 further reinforce this dynamic and demonstrate an ongoing commitment by TWP to addressing criminogenic needs amongst women at risk of offending. Yet despite this framework there is also a dilemma at the heart of this approach which we shall return to in our Discussion when we consider the implications of reifying gendered explanations of the causes of offending in the criminal justice system and how this might impact on the future expansion of the project to include adult male offenders.

Intervention
The project aims to assess the suitability of every adult female brought into police custody. The triage process consists of three stages. Stages one and two take place in the custody suite and stage three takes place at TWP's premises.
Stage 1: A woman is arrested and brought to the custody suite where she is booked in by a custody sergeant. She is provided with a leaflet explaining the triage scheme and the types of support she can access through TWP. This is normally followed by the standard interview under caution and if the woman admits the offence and shows some remorse then she passes the first eligibility criterion for triage referral. At the next convenient point either a youth justice worker or a DIP worker conducts an initial assessment of needs to help determine if the woman is suitable for triage rather than another more established intervention such as drug treatment. This is usually after the woman is interviewed under caution but can be affected by factors such as whether the woman is intoxicated, aggressive, meeting with her legal representative or when there are several woman brought into custody within a short time frame which then creates a queue. The main aim of the assessment is to determine the woman's offending history, her previous or ongoing involvement in the criminal justice system and if she has any personal circumstances that require support. Examples of such personal circumstances include: prostitution, drug and alcohol misuse, abuse and violence, mental health, housing problems, homelessness, poverty and children in care or adopted.
Stage 2: Before a triage referral can be approved checks against the Offender Unique Reference Number (URN) and Police National Computer (PNC) are made to confirm details about offending history. Arrestees with a caution in the previous two years or any conviction are not eligible for the intervention. The Home Office Gravity Score for Adults (Association of Chief Police Officers 2013) is used to determine eligibility for referral. The triage scheme is intended only for women whose offences have a gravity score of 1 or 2, indicating a less serious offence. Examples of offences with gravity scores of 1 or 2 include common assault, drunk and disorderly and shoplifting (up to £100 in value). If the woman meets these eligibility criteria she is referred to TWP, given a date to attend an initial meeting at the TWP premises, recorded as a diversion and released. If a woman is not approved she is charged and processed as usual.
Stage 3: Funding to support the Adult Female Triage Project was obtained via the Rank Foundation's Community Development Fund which allowed TWP to employ two key workers to undertake the one-to-one initial assessments with the women who are referred by the police. These initial appointments involve a much fuller assessment of the woman's history and needs including details about the woman's physical and mental health, social services involvement, social integration, dependents, co-habitants, alcohol and drugs issues, accommodation, money matters, personal safety, employment, training and education. This information is then used to generate a risk score that informs a plan of work undertaken with the woman over a fixed period of time. The types of support offered to women referred to TWP include a range of courses designed to help women with skills such as: parenting, relationships, confidence building, IT and basic skills, anger management and domestic abuse awareness. Alongside these courses TWP offers various advice and support services including counselling, access to a drug and alcohol specialist, mentoring, debt and money advice and tenancy support. If a woman does not attend her initial meeting TWP follow-up with a phone call but if she fails to engage with the support offered this information is then reported back to the police and this precludes the female offender from future eligibility for the intervention. The lack of any further sanction by the police for non-compliance effectively makes attendance at TWP appointments voluntary.
This paper tests the effectiveness of the intervention for adult females on rates of rearrest, the length of time to rearrest and number of subsequent arrests within a 12-month follow-up period. It was hypothesised that females who were eligible and referred to the diversion scheme would be (i) less likely to reoffend, (ii) would take longer to reoffend and (iii) would do so less frequently than a control group of female offenders who were not diverted to TWP.

Study design and cohort allocation
The impact of the intervention was evaluated through a natural experiment design. This study capitalised on an uneven roll-out of the intervention. For several months, issues relating to organisational resources within the assessment teamreflecting the pilot nature of the initiativelimited the number of arrestees who could be assessed. As a result of this, many arrestees who met two of the three eligibility criteria 1 for the diversion scheme were not assessed and, consequently, not referred to the diversion scheme. This created an opportunistic control condition, which permitted a natural experiment. To account for potential systematic bias between the groups, comparison statistics are detailed below. Arrestees in the control group were processed through the Criminal Justice System as usual.

Ethics
The study was submitted and approved through the University's Ethics Committee. The primary source of information is crime data which was provided to the researchers under a data sharing protocol with Humberside Police. Observations and interviews were conducted for fact finding and information gathering purposes and no data from these has been directly reported. This was made clear to all participants involved in the study.

Sample and context
The time frame for the intervention was 1 December 2012-15 July 2013. Females arrested during this time were followed up for one year, resulting in a follow-up time frame from 3 December 2012 to 15 July 2014. Of the 643 female offenders arrested between 1 December and 15 July 2013, 44 were seen by the triage team and were referred to TWP and represented the intervention group. Among the remaining arrestees, 114 were eligible for the intervention, but were not seen by the triage team and so could not be referred. These arrestees were processed as usual and represent the control group.
The study took place in the city of Hull, which is an administrative division of Humberside Police force area in the North of England. Hull has a population of approximately 250,000 and covers approximately 70 km 2 . All arrestees in Hull are processed at the same custody suite, which is the only custody suite in the city. Six hundred and forty-three female offenders were arrested between 1 December 2012 and 15 July 2013 in the intervention area.

Inclusion criteria/eligibility
Eligibility for the study was dependent on the arrestee index offence being of a gravity score 2 or below (Association of Chief Police Officers 2013), admitting to the offence, having no cautions in the previous two years and having no convictions.

Measures
The outcome variables, described below, were used to determine the effectiveness of the intervention. The predictor variables, also described below, were used to compare the characteristics of the intervention and control groups. More explanation of this can be found in the Statistical methods section below.

Outcome variables
Rearrest: Participants were coded as re-arrested (1) if they appeared in the police records for a separate offence between 2 and 365 days from their initial arrest. A 2-day 'grace' period was included to avoid mistaking rearrest for reoffending when participants were rearrested for administrative purposes. For example, occasionally, arrestees with dependent children are released to care for their children and then rearrested soon afterwards when police wish to proceed with interviews. It is important to note that if a person was arrested outside Humberside the incident would not be included in the data set. Information on the geographical breadth of offending by this group was not available, but we expect instances of offending outside Humberside to be rare.
Time to rearrest: The time between initial arrest and rearrest was measured in days. Participants who were not rearrested in the 12-month follow-up period were recorded as censored data. Using this technique resulted in all participant having equal follow-up periods.
Number of rearrests: Number of rearrests per participant was recorded as a continuous variable, beginning at zero for those participants who were not rearrested.

Predictor variables
Age: Participant age at time of first arrest was calculated by subtracting the date of the arrest from their date of birth.
Index of multiple deprivation: Using participant post code at time of first arrest, Lower Super Output Areas (LSOAs) were identified for each participant. The index of multiple deprivation score (IMD; McLennan et al. 2011) for 2010 that corresponded to each LSOA was obtained from the Office for National Statistics and served as a proxy for individual deprivation. Indices of multiple deprivation are a composite of seven domain variables that reflect neighbourhood deprivation in the UK: income, employment, health and disability, education skills and training, barriers to housing and services, living environment and crime. Scores for the 124 LSOAs in the study sample ranged between 0.53 and 87.8 (Median 45.4). 2 The IMD metric does not reflect actual deprivation but the level of deprivation relative to other areas in England. Higher scores indicate greater levels of deprivation.
Crime deprivation: Using a similar technique to that described above, LSOA codes were identified for each participant and an index of area crime rate for 2010 was obtained for each participant and served as a proxy for individual exposure to crime. Indices of crime rates are a domain of overall deprivation and are calculated based on the rate of four crime types recorded by police in that area: violence, burglary, theft and criminal damage. As with IMD, higher scores indicate greater crime deprivation.
The IMD and Crime Deprivation score were not independent of each other as the IMD score, being a composite of a number of social indicators, incorporated measures of area crime. This composite is weighted so that Crime deprivation represents 9.3% of the overall deprivation (Department for Communities and Local Government 2011). The Spearman rank correlation between the two variables for the study sample was 0.73, indicating a high correlation between crime deprivation and overall deprivation within neighbourhoods. Nonetheless, these two variables tested different phenomena and their inclusion here allowed exposure to crimepotentially reflecting attitudes towards crime as well as exposure to crime and criminal behaviourto be tested separately from more general deprivation.
Ethnicity: Based on self-reported ethnicity according to five Home Office categories of ethnicity, participants were coded as White (1) or Non-white (0).
Offence type: Arrest offence type was coded as one of five categories using an abbreviation of the Home Office Counting Rules for Notifiable Offences (Home Office 2012b): violence, criminal damage, public order, theft or drugs.
Group allocation: Participants who were eligible for the intervention and were referred to the women's project were coded as being in the intervention group (1). All female arrestees who were not seen by the assessment team were potential members of the control group. From this sample, it was possible to retrospectively test their eligibility in terms of previous convictions or cautions and the gravity of their current offence and to generate a sample based on these criteria. However, it was not possible to test for their admission of guilt and so this remains a potential confounding variable, albeit one that is lacking in empirical evidence. Respondents who met the eligibility criteria but were not seen by the assessment team, and consequently not referred to the women's project, were coded as being in the control group (0).

Statistical methods
As noted above, the intervention and control groups were created opportunistically which leaves the internal validity of the study design susceptible to the effect of confounding variables. For example, bias regarding ethnicity, offence type, offender's appearance or offence type could all have affected the arrestee's likelihood of receiving or not receiving the intervention. A series of chi-square tests and Mann-Whitney U-tests were undertaken to test for differences between the groups. Groups were compared on age, overall deprivation of home area, crime-related deprivation of home area, ethnicity and offence type.
The groups were compared on the proportion of arrestees who were re-arrested within one year using a log-rank test for equality of survivor estimates. 3 Cox regression survival analysis was used to compare the time without rearrest between the groups and these differences were visualised using Kaplan-Meier survival estimates. Number of subsequent rearrests between the groups were compared using a Mann-Whitney test. The low base rate of reoffending and the limited available sample size prohibited the use of null hypothesis statistical testing to compare the groups. In situations of low statistical power, as is frequent in pilot criminal justice programmes, effect sizes and confidence intervals provide more valuable information than binary null hypothesis testing (Schneider and Darcy 1984) and prevent potentially efficacious interventions from being disregarded simply due to unavoidably low statistical power (Hoenig and Heisey 2001). Confidence intervals are preferable to p-values in this context as they provide a range of likely effects for future iterations of this intervention rather than a binary indicator that is based on an arbitrary cut-off point, has little practical relevance and is highly dependent on sample size.

Intention-to-treat analysis
The effectiveness of the interventionpre-charge diversion to TWPwas tested on an intention-totreat basis. Intention-to-treat analysis aims to reflect the way in which interventions work in the 'real world', wherein patients sometimes fail to take their prescribed medication, service users fail to attend counselling sessions or, in this case, arrestees fail to attend their appointments at a women's centre. In this analysis, all of those referred were regarded as having received the intervention, regardless of whether they attended their initial appointment at TWP. However, in order to further explore the impact of TWP attendance, a post hoc, per protocol analysis is included below. Figure 1 describes the path followed by the population of arrestees during the timeframe of the pilot study. Descriptive statistics and comparison between the intervention and control groups on five comparison variables are described in Table 1.

Results
The descriptive statistics demonstrated that the intervention and control groups did not differ in ethnicity, deprivation, exposure to crime or arrest offence type. The intervention group were younger than the control group, which was taken into consideration in all subsequent analyses.
Rearrest: Of the 158 arrestees, 35 (22.2%) were rearrested within the follow-up interval, representing a 0.07% daily failure incident rate. The median time to rearrest was 128 days (Inter-quartile range (IQR) 54-219 days). A smaller proportion of arrestees in the intervention group (13.64%; n = 6) were rearrested than in the control group (25.44%; n = 29). The difference in risk of rearrest, including 95% confidence intervals between the two groups was −0.12 (−0.25 to 0.01), representing a 46% improvement (relative risk reduction) in rearrest rates in the intervention group compared to the control group (Risk ratio 0.54 (−0.24 to 1.20)).
Time to rearrest: The Cox proportional hazard ratio of intervention to control group was 0.52 (95% CI 0.22-1.25), indicating that respondents who received the intervention spent longer without rearrest than those in the control group. Controlling for the difference in age that existed between the groups had little effect on the hazard ratio (0.53, 95% CI 0.22-1.29). Of those rearrested, the median time to rearrest was 128 days (IQR 54-219). Of those who were rearrested, the median time to rearrest was shorter for those who received the intervention (Mdn 37.50; IQR 31-128; n = 6) than those who did not (Mdn 156; IQR 81-219; n = 29). Therefore, in these cases, the intervention was associated with faster rearrest (Hazard ratio 1.20, 95% CI 0.45-3.21).
Small number or rounded percentage not reported to avoid identification of participants. b Age was not associated with any of the three main outcome variables.

Per protocol analysis
Of the 44 arrestees who received the intervention, all were diverted from custody and did not receive convictions for their offences and all were referred to a women's centre. Of those referred, 41% (n = 18) attended their initial appointment which was used as the cut-off point in the evaluation to determine those women who received both parts of the intervention. Compared to those who did not receive the intervention (n = 114), those who attended their initial appointment were less likely to be rearrested (9.09% intervention group vs. 25.44% control group; 17% risk difference (95% CI −31.8% to −0.03%) spent longer without rearrest (Hazard ratio 0.33, 95% CI 0.08-1.38) and had fewer rearrests (U = 1.37, p = .17). However, as this per protocol analysis is based on a very small sample with only two failure events, the results should be treated with caution. Post hoc analysis did not detect any difference between the attending and non-attending groups on any of the predictor variables.

Discussion
The results of this evaluation suggest some potential for the intervention as a means of reducing reoffending among low-risk female offenders. The effect size estimates of this pilot intervention indicate promising effects on rearrest compared with the control group over a 12-month follow-up period but the estimate parameters indicate that caution should be maintained. In this discussion our aim is to try to understand this result in the context of the drivers discussed earlier. In particular, we consider the theoretical explanations for the project's results, the gender dynamics of the project and the potential issues involved in extending the model to adult male offenders. The low rate of rearrest, the small number of treated arrestees and the pilot nature of this study prohibited the meaningful use of null hypothesis significance testing to make a categorical determination of the effectiveness of this intervention: practically meaningful effects could be inadvertently dismissed as a result of low statistical power (Hoenig and Heisey 2001). In cases such as this, it is more appropriate to use an effect size estimate and the 95% confidence intervals that surround this figure to make an estimate of the true effect size of the intervention. The 95% confidence intervals give an indication of the range of possible effect sizes within which the true effect is likely to lie. While it should not be stated without reservation that the intervention was a success, the results indicate desirable effects of the intervention on rearrest and length of time to rearrest and very promising effect sizes. However, the parameters around these effect sizes suggest that there is some possibility that the true effect of the intervention is in the undesirable direction. Therefore, caution should be exercised in rolling-out this intervention and, if possible, a larger scale study of this intervention should be undertaken to corroborate these initial findings.
In general, those arrestees who received the intervention were less likely to be rearrested than those in the control group although there was some variability in this. However, contrary to hypothesis three, among those study participants who were rearrested, those in the intervention group were arrested more frequently than those who did not receive the intervention. It is important to reemphasise the small sample size of this study and the small number of rearrests in the intervention group (n = 6), which limits more enthusiastic inferences being drawn from this result.
The rate of rearrest among both intervention and control groups was low: only one-in-five arrestees were arrested again in the 12-month follow-up period. However, the intervention group were half as likely to be rearrested as the control group, suggesting that the intervention contributed to further reductions in risk of rearrest. Similarly, while those arrestees who received the intervention were, on average, half as likely to be arrested on any day, the confidence intervals suggest a small possibility that the true effect of the intervention may not be so desirable. It is noteworthy that the quickest rearrest (after five days), was for a member of the intervention group. Therefore, while the intervention appears to have been successful for the majority of the intervention group, it is likely that the intervention was not suitable for some. This is well illustrated by the shape of the Kaplan-Meier survival estimates for the intervention group in Figure 2. While the slope of the line for the control group is steady, the line representing the intervention group drops quickly before stabilising, with several rearrests around 35 days. This finding is further supported by the evidence relating to hypothesis three. The analysis revealed that, among arrestees who 'failed', that is, were rearrested within the follow-up period, the median number of rearrests was greater for those who received the intervention than for those who were processed as usual through the criminal justice system. The small number of failures among the intervention group (n = 6) limits the generalisability of the findings, but it suggests that, for some individuals, the intervention was not suitable and may have had a more negative effect on their likelihood and rate of rearrest than treatment as usual. Post hoc analyses 4 revealed no difference in demographics or characteristics of first arrest between those from the intervention group who did and did not 'fail'. The small sample size means that we cannot rule out this group of 'failures' as a statistical anomaly. However, if it is the case that some arrestees are more likely to fail, fail earlier and subsequently be arrested more frequently, it is imperative that predictors of this failure (e.g. motivation to desist or severe substance abuse problems) be identified quickly and incorporated into the eligibility criteria for triage disposal. While TWP has many potentially beneficial components, some arrestees may be beyond their help or that the problems faced by these individuals may be better managed through the criminal justice system.
The intervention comprised of two partsa diversion from the criminal justice system and a referral to TWP. It should be recalled that attendance at TWP was voluntary with failure to attend only precluding a future triage disposal. Although 44 arrestees received these two parts of the intervention, the intention-to-treat approach taken in the analysis did not account for actual attendance at TWP, which was low (41% attended their appointments). While this was an intentional component of the study design as it incorporated arrestee compliance into the delivery of the interventionreflecting how such an intervention would be applied in practiceit limits the extent to which the different mechanisms of the intervention can be isolated. Rather than work as theorised,that is, the diversion from custody reduces the harms associated with labelling and TWP empowers women by helping them to address criminogenic risk factors such as substance misuse or financial problems, it is possible that just one of these components is necessary to reduce risk of rearrest. As all arrestees in the intervention group received a triage disposal, a study design that could help to address this question was not possible. Therefore, the effect of either component could not be isolated. However, the per protocol analysisdespite low statistical powersuggests that those who complied fully with the intervention were slightly less likely to be rearrested. While practitioners applying such an intervention may be more concerned with success rates than with knowing the 'active ingredient', isolating the effects of the twin parts of the intervention has value for understanding how the criminal justice system affects low-risk female offenders and for determining the value of women's centre like TWP. Furthermore, while we have theorised the components of the intervention as working independently, it is possible the avoidance of a conviction interacts positively with directed access to targeted services like TWP to increase the likelihood of desistance from crime.
Understanding whether and how the intervention impacted on the behaviour of the women is crucial to making sense of how and why it can achieve positive results. It was very difficult to investigate exactly what work is undertaken with specific women at TWP, how resource-intensive it is or against what goals this work is set because the charity provides ongoing, holistic and therefore clientled support. Further, some support activities are farmed out to other providers who despite often using the TWP premises are not part of the organisation, which creates barriers to consistent information recording and sharing. As an ethos towards supporting and empowering women this may be an excellent approach but it creates difficulties for evidencing what criminal justice outcomes TWP are achieving. Furthermore, within a Risk-Needs-Responsivity framework, it is arguable that in times of economic austerity, the 'Need' of this low-severity offending group for intervention is low relative to more serious offenders (Bonta and Andrews 2007). This debate is now reasonably well rehearsed (see e.g. Jolliffe et al. 2011, Gelsthorpe andHedderman 2012) but nevertheless raises some important questions for how an organisation like TWP can demonstrate its impact on offending behaviour. Whilst we agree with the conclusions reached by Gelsthorpe and Hedderman (2012) that the premature introduction of a payment by results model into organisations like TWP would be counterproductive it does not seem sustainable for TWP to continue without a clear evidence base from which to measure the impact of its activities. For instance, at the time of writing we believe our study is the first experimental-designed evaluation that begins to demonstrate that the work undertaken by TWP may have a beneficial impact on rates of reoffending. Yet we do not know precisely what is working with which women. At present the empowerment ethos employed by TWP provides a reasonable defence against the imposition of external, potentially clumsy or cumbersome performance measures but this position may be difficult to maintain in the current economic climate. So unlike Gelsthorpe and Hedderman (2012) we think this question requires a faster and more robust response if the activities of TWP are to continue along the same lines that they currently do. The question then becomes how can performance measures compatible with the ethos of TWP be developed? One possible solution could be to look at the current government roll-out of new outcome measures for therapeutic interventions under the auspices of Improving Access to Psychological Therapies (IAPT). These outcome measures are based on a 'stepped care' model of therapeutic engagement where clinicians work collaboratively with patients to develop realistic, reviewable goals within certain timescales. The intention of this model is to be as unrestrictive as possible whilst also striving to evidence outcomes, manage patient expectations and strengthen understanding of the therapeutic relationship (National Health Service 2011). Exploring the potential of this type of collaborative model could offer TWP a method of evidencing outcomes that is compatible with its wider goals of empowering women and providing holistic services and support.
Humberside Police is currently in the process of extending this project to low-severity adult male offenders. This will operate in a similar fashion to the female intervention in the police custody suite but rather than being referred to a single organisation like TWP for empowerment support work a separate supporting and mentoring charity will provide case management and referral routes to appropriate offender-focused support organisations. This raises some interesting questions about the provision of support services for male offenders and whether they should be equally committed to the empowerment of male offenders. However, empowerment support work for women is justified by a feminist emancipatory agenda that has successfully demonstrated the unequal and insensitive treatment of women by men. On this basis there is no justification for a male empowerment project. Yet low socio-economic status correlates with low-severity offending by both men and women and a counter argument would be that it is social and economic empowerment of the disadvantaged that is actually what is required. Despite this, there are clearly some gender-specific issues that influence likelihood of reoffending and rehabilitation support services. These have been reasonably well articulated and absorbed by the criminal justice system and associated tertiary voluntary and third sector agencies like TWP who now offer bespoke services to help meet these needs. In particular, there is an acknowledgement that women usually have specific problems in terms of childcare, domestic abuse and parenting (Corston 2007, House of Commons Justice Committee 2013. Yet an equivalent body of research into whether there are specific problems men face and services to address these problems is less well known or understood. This is arguably because the overwhelming majority and focus of criminological research and criminal justice practice is either consciously or unconsciously geared towards male offenders (Smart 1977, Walklate 2004 and whilst this may well be true it does mean an important question has not been systematically asked or addressed. Are there specific gendered problems that lead men to offend and what services are required to address them? Whilst most current statutory and voluntary services are geared towards male offenders this has evolved over time as a consequence of both the number of male offenders compared to female offenders and the associated interest of predominantly male policy-makers and researchers who have historically tended not to think about male offending as a gendered issue (Cain 1989, Newburn and Stanko 1995, Walklate 2004. Whilst the focus on female offending and experience of criminal justice is entirely warranted the lacunae of research into men and masculinities is increasingly acknowledged. Influential studies into street and gang violence (Hallsworth 2005, Cobbina et al. 2010, the night-time economy (Hobbs et al. 2003) and male victimisation (Stanko 1990, Gadd et al. 2003) have begun to point to some of the social and cultural dimensions that help explain the relationship between masculinity and crime. What has not yet happened is a male equivalent of the Corston report (2007) which brings together some of the issues explored in this research to consider in what ways male-specific needs and services should be provided within the criminal justice sphere of activity. Certainly there is currently no equivalent provision for low-severity male offenders of the sort provided by TWP. It may be that that no equivalent is required but without a similar coherent evidence-based strategy for an early diversion and support programme for male offenders there is important difference between how the scheme we have evaluated works for women and how a scheme for men might function.
Returning to the policy initiatives driving this intervention the expansion of this and other earlydiversion schemes seem likely. When combined with the associated policy logics of early intervention, marketisation of the public sector and payment by results the need to reduce demand on the criminal justice system and target resources more and more effectively is only likely to intensify. With these changes comes a change in priorities. The necessity to demonstrate cost-effectiveness, consumer satisfaction and profit margins will continue to drive innovation and change. The introduction of new crime outcomes for police can create new forms of partnership and community engagement such as this early-diversion scheme. But expressed differently, they also incentivise 'outsourcing' and the rationalisation of criminal justice functions. This is a new environment where both partnership and rationalisation appear to sit comfortably next to each other but without a full or clear understanding of the implications of increasingly organising criminal justice as a type of industry (Taylor 1999, Maguire 2012. This study suffers from a number of limitations that should be considered when interpreting the findings. Firstly, the use of police data to follow up on arrestees is limited as a source of censored data because it is based on the absence of the arrestee from police records. As police force records only count arrests in that region, an arrestee moving to a new residence outside the study police force area or even dying during the follow-up period (without this information becoming apparent to the police or the research team) could create a false negative for rearrest. Secondly, although the study identified a 46% difference in reoffending and a 48% lower daily likelihood of rearrest between intervention and control groups, the small sample size and low base rate of rearrest mean that the study could not conclusively report that the intervention was effective. Small sample sizes and low base rates of rearrest are common limitations of fledgling criminal justice interventions but the effect sizes of these results should still offer encouragement for future interventions. Finally, while the descriptive statistics found little difference between the control and intervention groups, the possible existence of confounding variables should not be overlooked. As the evaluation began after the intervention commenced, it was not feasible to randomly allocate eligible arrestees to experimental condition. Notably, it was not possible to determine an admission of guilt among the control group of arrestees. As noted, no empirical evidence could be found to suggest a relationship between admission of guilt and likelihood of reoffending among this population of offenders, so we are confident that this did not affect the comparability of the groups. Furthermore, only limited data was available to assess the comparability of the groups, which, combined with the small sample size, prohibited the more rigorous propensity score matching (PSM) method of comparison. Taken alongside the promising results, a further evaluation, using random allocation and a larger sample size that reflects the low base rate of reoffending among these offenders should be undertaken.

Conclusion
The study presents a natural experiment that compared reoffending rates amongst low-severity female offenders referred to an early-diversion pilot project with reoffending by a similar group of women who were processed as usual through the criminal justice system. The overall finding from this study is that the intervention was associated with lower likelihood of rearrest but that the promising results need to be replicated using a larger sample. The intervention is innovative at two levels. Firstly, it offers an early diversion scheme that provides a route out of the criminal justice system before the woman is charged with an offence. Most diversionary activity has historically worked at the post-conviction stages and therefore inevitably attracted a criminal label whereas this early diversion scheme mirrors the early intervention logic that currently drives much of the preventative and cost-saving innovations across criminal justice. Secondly, the intervention diverts women to a charity that provides a range of holistic empowerment support work. Rather than criminalise and sanction women this intervention seeks to address the difficulties that often drive low-severity female offending. The intervention's ethos is inspired by both theoretical and empirical evidence about female offending and relies on a strong partnership approach for effective delivery. Despite the existence of similar pre-charge triage projects in youth justice as far as we are aware this is the first evaluation to provide any promising empirical results about pre-charge diversion for adult female offenders. However, a final cautionary note compels us to conclude that extending this scheme to adult male offenders introduces a different set of both gender dynamics and practical problems and the promising results reported here should not be extrapolated to males. Further, whilst our study tentatively suggests the work of TWP is having a positive impact on reoffending rates a more rigorous basis for assessing the impact of empowerment support work undertaken with either men or women is required to sustain this and other similar intervention into the future. Notes 1. It was not possible determine an admission of guilt among the control group. This presents a possible confounding factor in the analysis. However, an extensive review of literature found no empirical evidence that admission of guilt or denial of offending are related to likelihood of reoffending among a low-severity female sample. 2. IMD scores for England have an approximately log-normal distribution with median 17.4 (Interquartile range, 9.7-30.1; Min. 0.5, max. 92.6).
3. This technique was preferred over PSM because there was a very limited number of available variables on which to match the groups and the already small sample size would have been further reduced by PSM. 4. Available upon request from the corresponding author.