Introduction

A criminal record has dire consequences for those who have one. Being officially branded as a ‘criminal’ carries both formal sanctions, such as prison sentences or other penalties, and informal punishments, such as blocked access to employment opportunities, banking, guilds and voting. Moreover, it tarnishes one’s reputation. Leaving aside the issue of just desserts, a criminal record is a powerful feature of the criminal justice system with consequences above and beyond the ascribed sanction for wrongdoing.

Labelling theory predicts that the criminal record—being called a ‘bad element’ (Becker, 1963; Lemert, 1951; Schur, 1965)—has additional adverse consequences in the form of subsequent criminal behaviour. For some, contact with the criminal justice system can spiral into further involvement in delinquency, as the individual embraces a master status of an offender (Brewer et al., 2019). While this is not true for everyone who has such interactions with law enforcement, the evidence nevertheless shows the strong predictive power of the stigma for self-image and, consequently, subsequent criminality (Halsey et al., 2017).

However, studying the effect of the criminal record on subsequent criminal behaviour has always been challenging because the label interacts with and is moderated by the effect of the sanction (Paternoster & Iovanni, 1989; Restivo and Lanier, 2015) . An offender apprehended for a crime is likely to receive some punishment, such as a fine or prison sentence. There are known effects of these sanctions on the propensity to commit crimes in the future (Nagin 2013). Hence, while labelling theory predicts a non-negligible effect on criminal behaviour, deterrence theory predicts that sanctions will also have a non-negligible effect on criminal behaviour and disentangling the two is difficult, as whenever there is punitive response to a crime, the label is automatically assigned as well.

Moreover, we are unaware of systematic attempts by law enforcement to capitalise on the criminal record as an intervention without a sanction—at least as adult offenders are concerned (see also Farrington et al., 2014) . If the record is so important for people, why not use it as a deterrent more robustly? Could the threat of a criminal record be used in lieu of a sanction, and if so, as an intervention, would it be effective against subsequent recidivism? There is no clear answer to these questions because once guilt is admitted and/or incriminating evidence is accumulated against a suspect, the law does not have room for counterfactual conditions: a response to the crime must occur, and the crime cannot go unrecorded, which limits the ability to conduct tests in which the criminal record alone is the tested stimulus.

Instead, the criminal justice system can use it as leverage, or as a ‘Sword of Damocles’ (Sherman, 2011): if you stop committing crimes, we will expunge the incident from the record. In terms of the Sword’s application, suspended sentences, deferred prosecution, out of court disposals and probation are all examples. However, these responses bring us back to our original Gordian knot: the criminal tag stays on the offender and overshadows the sanction. This is the theoretical and practical conflation we must unscramble (Bernburg et al., 2006). Until we separate the deferred sanction and the criminal record, the effectiveness of the Sword remains unclear. We therefore need settings in which the criminal record alone is compared to an alternative, without additional conditions or punishments—deferred or immediate.

These special circumstances exist with simple cautions used by police in England and Wales. A simple caution occurs when the police formally register a crime the offender has admitted to committing but do nothing else with the case. Unlike incidents of ‘no further action’, the simple caution generates a formal criminal record. Although it is not a conviction, it does show up on criminal record checks. There are no additional sanctions associated with it, and there are no conditions the offender must respect: it is purely a criminal record, which we may interpret as a proxy variable for a label.

Thus, we can explore alternatives to the simple caution (i.e. the label) in isolation from the sanction that cannot be explored using a charge because in any counterfactual group (to an intervention), the subject is exposed not only to labelling by criminal record but also to a formal court process and, potentially, time at His Majesty’s pleasure. Any of these factors may cause the deterrent effect. For a simple caution, it can only be the criminal record label although the mechanism by which the label works is far more complex. As we have noted, labelling theory concerns stigmatisation and this can occur on multiple levels. On one hand, the stigma may be at the community level; friends, family, neighbours, colleagues being aware of the label. On another hand, the stigma may be entirely self-generated by the individual based on their own perceptions and self-image. We cannot untangle this mechanism easily.

Ideally, we would test the effect of a simple caution against nil response by the police to an admitted crime; however, as we noted, this scenario is impossible once the incident has been detected and registered in the police database. Instead, we can compare the assignment of the label versus its expungement once certain conditions are met. For example, should the offender participate in a rehabilitative programme, the police would erase the criminal record and give the offender a clean slate; and should the offender wish to discontinue their participation in the program, they would have to choose between dropping out and keeping the criminal branding, or staying the course and having their criminal record deleted.

In this article, we describe a randomised controlled trial designed to test the relative effectiveness of simple cautions versus participation in a rehabilitative program called Pathfinder, in which completion of the programme leads to expungement of the criminal record. The scheme incorporates the pathways framework (Sampson & Laub, 2003), whereby offenders are held accountable for their actions while addressing the underlying needs that are linked to their involvement in crime. We measure both crime counts as well as crime harm, within 2 years post-random assignment, and report the results herein.

Literature review

Criminal records and labelling of offenders

Research has repeatedly shown the adverse consequences of labelling (see review in Barrick, 2014; see also Bernburg et al., 2006; Klein, 1986; Krohn et al., 2014; Ray & Downs, 1986; Smith & Broege, 2020). Although the effect of the tag may be conflated by the sanction (Restivo & Lanier, 2015), the stigmatisation associated with being branded a ‘criminal’ seems to be a powerful antecedent to a host of problems, including further involvement in crime, psychological pathologies, social ostracisation and a generally negative self-image (Abrah, 2019; see also review in Cullen, Chouhy and Jonson 2020). As such, the criminal record is a negative turning point for the labelled individual (Pager, 2003) that forces them ‘to take on the label as a key aspect of their identity’ (Lemert, 1951). The negative label then becomes a self-fulfilling prophecy in that those who are labelled begin to identify with the label and become what they have been labelled (Chenane et al., 2021: 496). Hence, contact with law enforcement increases the likelihood of recidivism (Becker, 1963; Farrington et al., 2014) , and as the offender moves through the criminal justice system (i.e. police > prosecution > courts > sanctions), the effect of the label intensifies (see Smith & Paternoster, 1990, but cf. Kurlychek et al., 2006, who show a possible decayed effect over time).

Consequently, the criminal record is a potent element in the identity, social status and prospects of the many millions of people who have it. Although debate over the utility of the criminal record continues (Burton et al., 2021), while it remains an integral consequence of and inherent feature of criminal behaviour (Ispa-Landa and Loeffler, 2016) , it is interesting how little research exists on the ways in which the state can use it more robustly as a stand-alone response to crime. Feingold (2021) reviewed the literature on the stigma of incarceration; Basto-Pereira et al. (2015) reviewed the available evidence on the criminal record as an outcome; Visher et al. (2006) synthesised the data on the effect of non-custodial employment programmes on recidivism rates of ex-offenders and included the criminal record as a selection criterion in their influential systematic review; Martin et al. (2020) showed the effect of the criminal justice stigma on health outcomes. However, the stigma interacted with the sanction, in all these cases, and few studies have rigorously examined the criminal record as the independent variable, at least as adult offenders are concerned. As a standalone stimulus for offenders, its effect on crime—or any other outcomes, for that matter—is largely missing from the debate.

Out of court disposals

Processing offenders through the court and prison systems is expensive (Neyroud, 2018a, 2018b), may not work (Durlauf & Nagin, 2011) and does not fit all offence types. Prosecuting common, low-harm offences, even when admission of guilt and evidence have been secured, is often not in the public’s interest. Therefore, out of court disposals have been considered a viable alternative to the traditional justice route of charge–court–sanction (Gaffney et al., 2021; Petrosino et al., 2010; Nagin et al., 2009) . Overall, the research is strongly supportive of these pre-trial interventions (Gaffney et al., 2021; Petrosino et al. 2019; Wilson et al., 2018; Wilson and Hoge 2013). ‘The observed effect size of 0.144 corresponds to a decrease in reoffending of approximately 13%’ (Gaffney et al., 2021: 3), which can be contextualised in terms of both the rehabilitative nature of the alternative offered to the young offenders and the non-labelling framework they attempt to introduce. The essence of these initiatives is that police officers divert offenders away from the traditional criminal justice system towards rehabilitative programmes.

Out of court disposals are the closest scenario to a standalone intervention composed of only a criminal record, specifically for juvenile delinquents (Wilson et al., 2018; Wilson and Hoge 2013), but they are not quite that. Once a case has been introduced to the prosecution and becomes public knowledge, the labelling has probably occurred. Even if the offender’s immediate contacts or the general public are unaware of the information, the potential is there for this to happen and so it is not possible to control for this eventuality in experimental conditions with any realistic measure. Even in studies such as Chiricos et al. (2007), which found significantly less recidivism in men and women whom had their criminal records withheld by a judge, we cannot fully isolate the impact of the criminal record because these individuals had been introduced to the judicial process; a judge makes the decision. A true diversion must therefore come at the pretrial stage, in police-led settings, not in conjunction with or following conviction, and even then, we must be able to estimate the effect of the criminal record alone. Therefore, among the available out of court disposals, the unconditional, simple caution is the only legal option fit for our purposes.

The Pathfinder Project

The Pathfinder scheme was commissioned by the Office of the Police and Crime Commissioner (OPCC) for Devon and Cornwall in 2017. The project was jointly funded by the OPCC and the Home Office, through the Police Transformation Fund. The primary goal of the project was to divert ‘low-risk’ adult offenders away from the criminal justice system through a programme of intensive case management provided by a team of caseworkers recruited specifically for this purpose (see more broadly on offender management schemes with caseworkers in Hadfield et al., 2021).

Pathfinder was founded on the same theoretical foundation as its predecessor projects, Turning Point and Checkpoint (see Neyroud, 2017 and Weir et al., 2019) , which took place in the West Midlands and Durham police jurisdictions in England. This theoretical foundation focuses on the deterrence argument, specifically, that the swiftness and certainty of an action are more important than its severity (Apel & Nagin, 2011; Sherman, 2011). Turning Point and Checkpoint emphasised the fear of punishment as a greater influence on future behaviour than the actual punishment. This is taken a step further in Pathfinder, where that punishment is not a court-based prosecution, but a caution—an out of court sanction. However, while significant, this is the only deviation from the principles of Turning Point. The term, ‘turning point’, refers to the point in an offender’s life at which they might pursue a direction away from crime (see Sampson & Laub, 2003; Andrews & Bonta, 2010). The central hypothesis here is that capitalising on this opportunity can lead to crime prevention: the offender chooses to turn away from criminal activity by dint of a change in their life circumstances. This premise is predicated on self-motivation (Gadd, 2006; Giordano et al., 2002; Sampson & Laub, 2003), which is also an intrinsic part of Pathfinder.

Methods

Trial design

We used a pretest–posttest control group design. Prospective randomised controlled trials are preferred for producing causal estimates when comparing the effect of two or more parallel treatment conditions (Shadish et al., 2002) . There are no true counterfactual conditions for testing the effect of the criminal record, or that of the Pathfinder scheme, because once the police have successfully investigated a case and secured admission of guilt, they are legally obliged to ‘do something’ with the apprehended offender. The bare minimum that can be done is to register the crime in the police database and then release the person without any further action. This is precisely what a caution is. Against this backdrop, we can estimate the effect of Pathfinder and the promise to expunge the criminal record at the end of a successfully completed scheme (on valid comparison groups, see discussion in Ariel, Sherman and Newton 2020).

Participants

We were interested in adult individuals who had committed low-level offences, as these cases were likely to result in a criminal record with no other punitive consequence by the criminal justice system (Cornelius, 2013). To be eligible for selection to the scheme, an offender had to satisfy the specific legal eligibility criteria set forth in the next section.

Whether a simple caution is suitable for an offender is an operational decision for the police, based on the circumstances of the case (see Allen, 2017). Generally, individuals could only be considered if they were over 18 years old, resided in and had offended in Devon or Cornwall, had admitted to having committed the offence and had committed an offence deemed appropriate under the simple caution legal guidelines (see UK Government Health and Safety Executive 14/05/2014).Footnote 1 Furthermore, eligible participants were individuals who had committed an indictable crime, such as common assault, shoplifting, drug possession, or any other crime category outside the excluded list of crime categories for which the police are not allowed to issue a simple caution. For example, domestic abuse, hate crime, possession of certain weapons or sexual offences would disqualify individuals from participating in the experiment (see further in Spyt et al., 2019 and criticism in Westmarland et al., 2018). However, we did not differentiate between first-time offenders and prolific criminals, and the legal basis for simple cautions does not remove seasoned offenders (although using our measurements we did not determine prior offending by conviction status—what we determine a prior offender may never have been brought to justice). Thus, the drivers of the decision to give an offender a simple caution are the specific circumstances of the case, not the offender and their background. Nevertheless, it is far more likely for this instrument to be used on first-time offenders than on habitual criminals (CPS 2018).Footnote 2

Interventions

Simple cautions

A simple caution is akin to a formal warning. It can only be given if not prosecuting the offender is in the public interest. Consequently, the case must meet the standard required by the evidential stage for prosecution (Smith, 2014). In practice, cautions are included in the England and Wales’s Enforcement Policy Statement (2018) as one of the possible responses to a criminal offence available to police officers at the rank of Inspector or above:

a statement by an Inspector[…] that the [offender] has committed an offence for which there is a realistic prospect of conviction. A simple caution may only be used where a prosecution could be properly brought.

When a police officer is satisfied that a simple caution is applicable in the case, they mark the prosecution papers as ‘not approved, subject to acceptance of a simple caution’ and set out their reasoning for the decision. Simple cautions are therefore cases in which a criminal record is made of the crime, but no further punitive sanction is afflicted on the offender (see implications in Ashworth & Zedner, 2008). However, unlike ‘no further action’ cases, which concern unproven, undetected or unsubstantiated suspicions, or are cases the criminal justice system has no interest in pursuing beyond the initial reporting of the crime, simple cautions are suitable for cases of a proven crime (HSE Guidance OG 00,018, 2018). Therefore, as cautions carry a formal criminal record, although unsanctioned (again, due to public interest considerations and overall legal policy in England and Wales to not prosecute or seek additional sentencing beyond the registration in a national offender database), simple cautions, like most other out of court disposals (see Neyroud, 2018a, 2018b), are not synonymous with dismissed investigations. Indeed, they are formally reported as ‘positive’ outcomes or ‘detections’ in official statistics.

To emphasise further, a simple caution is applicable only in cases in which the offender has admitted the offence and has agreed to accept the caution (a template of the admission letter can be found at www.hse.gov.uk/foi/internalops/og/og-00018-appendix-1b.pdf). If the suspect contests the crime, a caution cannot be applied, as a prima facie admission of guilt is a prerequisite for this legal instrument. A suspect is also entitled to refuse the caution, even if they admitted the crime, and they ‘should not be induced to accept the caution in any way and must not be pressed to make an instant decision on whether to accept a simple caution.’Footnote 3 Similarly, the implications of accepting a simple caution must be provided in writing to the offender, and they must be given reasonable time to consider and, if need be, take legal advice on the matter.

Should an individual who is cautioned reoffend, then having a simple caution is likely to affect the decision to prosecute them as a repeated offender, given that the caution is a proven crime and a criminal record, for all intents and purposes (see more broadly in Miller, 2016: 174–175). The caution can be referred to in court if there is a prosecution for any other offence, which has immediate implications for both culpability (e.g., habitualness, similar modus operandi) and punishment severity, as the sentencing guidelines, the law and legal practice more broadly view recurrent or frequent criminality as an aggravating factor (Rowe et al., 2022). However, unlike suspended sentences, a subsequent crime does not imply that the first crime (i.e. the one to which a simple caution was given) will then be prosecuted; these circumstances exist in ‘conditional cautions’, which are different from the simple cautions that we are exploring.

Procedurally, when a case is assigned a simple caution, the police follow legal guidance (see Sect. 14 of the HSE Guidance, 2022),Footnote 4 with the following protocol:

[A] obtain approval in writing from the Approval Officer and a more senior manager; [B] invite the [police officer] to attend a meeting and explain in writing how a simple caution would be administered and what it would mean; [C] meet with the [suspect]…; [D]… satisfy… that the recipient understands the nature of the simple caution and the consequences of acceptance, that s/he agrees to the procedure and that s/he admits the offence(s); [E] invite the recipient of the simple caution to sign two copies of the simple caution and retain one of them.

Pathfinder scheme

A case assigned to the Pathfinder arm is treated as described above. The first appointment with the offender is held at a police station, but subsequent sessions and interventions often take place in the community. Participants are required to work with their caseworker for 16 weeks, on a one-to-one basis. The caseworkers first conduct an initial ‘needs assessment’ with the offenders (Andrews & Bonta, 2010) and construct a tailored rehabilitation plan (the ‘contract’). The primary aim of the contract is to reduce subsequent criminal conduct, through a myriad of rehabilitative options. The contract includes conditions designed to support changes in behaviour and reduce the risk of reoffending. It is the duty of the caseworker to ascertain completion of the contract and, in cases of breach, to inform the police and/or the Crown Prosecution Office of the infringement.

The caseworkers also manage contact with victims in accordance with the Code of Practice for Victims of Crime, if necessary. The scheme therefore encourages the use of restorative justice conferencing, which has been found to be effective in multiple studies (see Angel et al., 2014; Bouffard et al., 2017; Sherman et al., 2015; Strang et al., 2013; Mills et al., 2019; Braddock, 2011; however, cf. Grace, 2022). When victim–offender restorative justice processes are deemed to be unsuitable in a case (e.g. drug or weapon possession), the caseworker works with the offender to find alternative suitable rehabilitation frameworks, including restitution, drug or alcohol therapy, and any other ‘pathways’ engagements intended to address the root causes of crime (see Williams & Ariel, 2013).

In this study, the team consisted of 13 caseworkers working in localised teams in Devon and Cornwall. As in Turning Point and Checkpoint, these caseworkers were not police officers, based on the premise that offenders would be more likely to engage with non-police personnel (see Neyroud & Slothower, 2013, 2015; Weir et al., 2019). Caseworkers conducted all communications with the subject once referred to the project.

Crucially, those who completed the 16-week scheme did not receive any form of criminal sanction and had their criminal record expunged. As noted by Donoghue (2014: 946–7), cautions ‘become immediately spent under the Rehabilitation of Offenders Act 1974’ once they are expunged, although ‘they can still be quoted on Standard and Enhanced Disclosure and Barring Service (DBS) checks.’ Moreover, they ‘may influence how the offender is dealt with, should they come to the notice of the police again, and may also be cited in court in any subsequent proceedings’ (ibid). However, it is still the case that, once an offender successfully completes the Pathfinder scheme, they no longer have a criminal record per se, with all the substantial adversity that comes along with such a label (see further discussion in Cunneen et al., 2018). On the other hand, the Pathfinder contract, which sets forth the conditions under which the expungement will be completed, informs the participant that should the contract be breached due to noncompliance, the caution will be reinstated (see UK Sentencing Council, 2022).

Outcomes

Crime data were collated by the Devon and Cornwall Police and relayed to the research team for statistical analysis. Records were linked to Pathfinder subjects by their unique reference number stored in the Devon and Cornwall’s Force Unified Intelligence System (UNIFI). We refer to an ‘outcome’ herein as any crime record with a ‘positive’ investigative outcome: a charge, caution, penalty notice or other form of out of court disposal for a subsequent crime.

Crime

Our first primary outcome was a count of the number of crimes suspected to be committed by the simple caution and Pathfinder arms of the trial. The suspected status is determined by the investigating officer following their investigations. Suspects may be removed if eliminated from enquiries, but if the investigation does not reach the evidential thresholds for charge or other outcome, the suspect will remain. In this sense, our measure of crimes is different to proven reoffending, which we deemed essential given that more than 90% of recorded crimes in England and Wales do not end in a charge or out of court disposal (ONS, 2022). Each offender in each group was assigned a cut-off date exactly 730 days prior to and after their randomisation to ensure comparable baseline and follow-up periods.

Crime harm

Crime data were then coded against the Cambridge Crime Harm Index (CCHI; Sherman et al., 2016). CCHI and versions of CCHI are a recent though established metric that has been used in multiple studies for tracking harm (Ariel et al., 2016; Bland & Ariel, 2015, 2020; Barnham et al., 2017; Dudfield et al., 2017; House & Neyroud, 2018; Linton & Ariel, 2020a, 2020b; Mitchell, 2019; Weinborn et al., 2017; Weir et al., 2019; however, see critique in Sarnecki, 2021). The index assigns a value to each classification of crime based on official guidelines for starting point sentences for first convictions with no aggravating factors. In practice, each crime type has an associated number equivalent to the minimum number of days in prison the offence attracts. Community sentences and fines have comparable values calculated on the minimum wage. After assigning these values, we calculated before and after totals of crime harm for each offender in the treatment and control groups as the second dependent variable to calculate the group means. Because crime harm values typically have a skewed distribution (homicide is above 5000 days, while many crimes are below 10 days), we used a Log10 transformation on these values.

Random allocation

Our experimental randomisation procedure commenced once an eligible offender was arrested in Devon and Cornwall. Following the individual’s apprehension, investigation and admission of guilt, an Inspector would then conclude that, prima facie, a simple caution was suitable in the case. The offender was then asked if they wished to participate in the experiment, which was duly explained to them. If they consented, the case was then randomly allocated to either the caution or Pathfinder arm of the trial. Randomisation was conducted using an online calculator (Ariel et al., 2012; Linton & Ariel, 2020a, 2020b), as shown in Fig. 1.

Fig. 1
figure 1

Distribution of cases by assigned group

Statistical methods

We used multiple methods to estimate the treatment effect. First, we applied descriptive statistics for measures of central tendency and dispersion. Second, we calculated post-assignment prevalence of recidivism and, separately, gain scores of the difference-in-difference for both frequency and crime harm measure (CHI), based on 2 years of pretest and posttest data. We compared the count gain score between the two groups using a Mann–Witney U test due to the skewed distribution of the gain scores (which we calculated using a Shapiro–Wilk test for normality). We calculated a t statistic for the differences in the Log10 of the crime harm gain scores(see Campbell and Stanley 1963: 23).

Dropping out is common in field settings where offenders are offered rehabilitative and treatments (see discussion in Mills et al. 2019: 83–6). Some individuals are expected ex ante to reject the offered treatment, despite the immediate ramifications in the form of a criminal record, and take the caution notice instead. Others failed to respond to caseworkers’ attempts to contact them, and others disengaged from the scheme after embarking on the initial sessions. In our experiment, those assigned to caution showed 100% compliance (i.e. received the simple caution), and none of these participants was subsequently allocated to the Pathfinder intervention. On the other hand, as shown in Fig. 1 by the distribution of participants, some did not engage with the Pathfinder caseworkers (n = 18). The majority of noncompliers (n = 15) opted out at the point of randomisation, in effect choosing a caution and a criminal record over the treatment. The remainder (n = 3) met with a caseworker but were unwilling to sign a treatment contract for the scheme.

Therefore, we first estimated the treatment effect under the intention-to-treat model, based on the participants’ randomisation position, not whether they were exposed to the treatment (Piantadosi, 2017). Next, we sought to correct for the dropout problem. We applied a solution offered by Angrist (2006): an instrumental variables approach (Angrist, Imbens and Rubin 1996), using a two-stage least squares regression model (2SLS) to estimate the local average treatment effect (LATE). The suitability of the 2SLS approach is decided based on two assumptions: first, that the prospective instrumental variable (compliance with random assignment) is correlated with random assignment and not with the error term; second, that there is no extraneous link between the instrumental and dependent variables, other than through random assignment (see Angrist, 2006). In the first stage of the 2SLS procedure, we regressed treatment delivered as the endogenous variable against random assignment as the exogenous covariate. In the second stage, we used fitted values from the first stage in place of random assignment. All 2SLS calculations were computed in SPSS.

Results

Baseline balance

In total, 341 offenders were randomly assigned simple caution or the Pathfinder scheme, with an even split of 169 in the Pathfinder condition and 172 in the caution condition. We collected data on gender, ethnicity, age and offences within the 2 years leading up to each offender’s referral to the trial (Table 1).

Table 1 Treatment (Pathfinder)—completers, non-starters and control (caution) groups: baseline characteristics, pretest and posttest crime means and crime harm scores

Treatment estimates of recidivism (intention-to-treat model)

Crime counts

A reduction in crime counts was observed in both groups. The prevalence of recidivism was 20.3% in the control group and 23.7% in the treatment group, respectively. In the 2 years prior to their referral to the trial, the control group was linked to 245 offences and the treatment group to 264. These totals decreased to 84 and 134, respectively, in the 2 years after random assignment. The marginal difference in the mean gain scores (0.17) favoured the control group, and the difference did not meet the conventional statistical significance threshold (u = 14,336; p = 0.808).

Harm

The gain scores of mean crime harm were also not statistically significantly different (t = 0.567, p = 0.532); however, the reduction in harm between the before and after periods was notably different between the groups. A 3.5% reduction in crime harm was reported in the control group, while the treatment group showed a 38.3% reduction.

Adjusted Treatment Estimates of Recidivism (LATE model)

Offenders who discontinued their Pathfinder treatment plan tended towards both higher numbers of repeat offences and more harm. Among this group, the prevalence of recidivism was 88.9%. For crime count, the four highest-count reoffenders did not complete the scheme and accounted for 35% of all reoffending cases. For crime harm, the pattern was much more acute, with the top four accounting for more than 80% of the group’s harm score post random assignment. Accordingly, the reduction in total crime among the treatment group who completed Pathfinder was 77% in count and 90% in harm.

The results of the 2SLS approach are shown in Table 2. In this analysis, both differences are statistically significant [(β =  − 0.175, p ≤ 0.01); (β =  − 0.125, p ≤ 0.05), respectively], with a steeper slope for harm and a complete change in direction for crime count. The overall effects remain modest, as the explained variance (R2) in the dependent variable remains unaccounted for by the instrumental variable. However, the adjusted model reveals significant effects of the treatment on both crime count and crime harm, which would otherwise have remained unobserved. Figure 2 summarises these differences.

Table 2 Pathfinder vs. caution conditions: 2-stage least squares crime counts and crime harm scores—analysis summary
Fig. 2
figure 2

Summary of outcomes of random assignment

Discussion

As far as we are aware, this study was the first to rigorously examine the impact of a programme that allows offenders to ‘work off’ their criminal record by participating in a rehabilitative scheme. By diverting cases outside the criminal justice system early on, that is, at the point of first contact with the police, we showed a non-negligible effect on reducing recidivism and subsequent harm. Within 2 years of being assigned to Pathfinder, offenders of low-harm crimes are better off participating in treatment than having a criminal record, providing they complete the treatment. These findings have theoretical, practical and research implications, which we now discuss.

Theoretical implications

The threat of a criminal record is a powerful deterrent

The first striking outcome of this experiment was the level of compliance with the alternative to a criminal record offered by the police: 136 out of 172 offenders (89%) have complied with the treatment scheme offered to them. One interpretation of this fact is that having a criminal record matters to people, and they will cooperate in order to expunge it. This is purely inferential, however, because we did not test the motivation of individual participants. The literature clearly shows that a criminal record is detrimental to individual wellbeing by limiting employment opportunities and impacting socioeconomic status. In addition, it can create difficulties in a wide range of social contexts. But as importantly, this high compliance rate is substantially higher than that for the usual participation in interventions: 72.9% (Olver et al., 2011; see also Cullen et al., 2011; Feder & Wilson, 2005; Romero-Martínez et al., 2019). We found that if offered the chance to ‘work off’ their criminal record while trying to manage the causes that led to their involvement in crime through treatment, most offenders complied suggesting that the police have at their disposal a powerful yet underutilised source of leverage that can be used to convince people to engage in some form of third-party rehabilitative program: one’s reputation.

The forced rehabilitation in exchange for a clean slate model

Whether offenders are better off taking up a rehabilitative programme remains a question, yet, as the available research, including the present experiment, suggests, pre-court diversion programmes tend to do well (Neyroud, 2018a, 2018b). The Turning Point project conducted in the West Midlands Police jurisdiction (Neyroud and Slothower, 2015) and the Checkpoint project run by Durham Constabulary in the North of England (Weir et al., 2019), for example, found that participating in out-of-court treatments has desirable impacts on recidivism; Turning Point reported a 36% reduction in harm, 45% lower costs and 45% higher victim satisfaction (Slothower, 2020). Similarly, we detected a 24% reduction in recorded harm relative to a caution-only group and a 29% reduction in frequency of recorded offences when adjusting for treatment compliance. Therefore, we concluded, like others, that the Pathfinder programme is a desirable intervention.

Crucially, however, these previous out of court disposal studies contrasted the tested interventions to another condition, which was often composed of a punitive component: sanctions, fines or prison sentences. These comparators include both a criminal record alongside a sanction, which, from a penological perspective, represent two (or more) responses to crime. To date, studies have tested the efficacy and cost-effectiveness of out of court disposals as an alternative to prosecution, post-conviction punishment or supervision (Barnes et al., 2012; Hyatt and Barnes 2017; D'Amico et al., 2016; Kleiman & Heussler, 2011; see also a review in Taxman, 2012). However, our focus was on a substitute at a more preliminary and possibly more crucial criminal justice system stage: the initial criminal branding. While the ‘no treatment’ conditions in previous experiments masked an interaction effect between two elements that cannot be disentangled, the criminal record (labelling associated with being found to have committed a crime) and a criminal sanction to which the out of court disposal was compared, we are able to separate the two and compare the Pathfinder intervention with one component only: the criminal record. Therefore, Pathfinder is different because its effectiveness is not compared to an alternative intervention or the usual criminal justice system route but to a nil sanction condition.

In practical terms, if we infer that having no criminal record is important enough for nearly 90% of people who committed a crime to be willing to attend a 16-week treatment, then we can conceptualise a new approach to crime policy: an out of court disposal offering a clean slate to offenders who go into rehabilitation. The removal of the label has a cost, but in this hypothesis, people are willing to pay the price to get their reputation back or avoid damage to their self-image or any other reasons. We cannot make conclusions about the motivations of individuals based on this study, but while these may be of scholarly interest for any replications of this study, from a practical standpoint the key fact is that completion of out of court rehabilitative programmes such as Pathfinder will reduce recidivism.

The proposed model, whereby the police give first-time offenders the option to expunge their criminal record in exchange for participating in treatment, is also palatable from a wider deterrence theory perspective. Studies have shown that sanctions are rarely guaranteed and therefore offenders are able to conceive scenarios in which their crimes would go undetected, unpunished or both (Nagin 2013). This lack of certainty is undesirable because it may create a preference towards committing crime over normative behaviour: the perceived odds of the punishment are underestimated (Gibbs, 1975). However, participants in a ‘rehabilitation or else’ programme are confronted with a guaranteed and immediate consequence for their actions. The result of noncompliance with Pathfinder is translatable to a certain outcome: labelling as an offender. While punitive outcomes may or may not materialise in some cases, not removing the criminal record is certain, should the offender opt out, and noncompliance becomes synonymous with a tangible cost (i.e. having a criminal record), without any further adjudication. Therefore, we hypothesise that when low-harm offenders are confronted with the binary option of taking the deal offered by the police, or being branded as a criminal, the vast majority of people are likely to take up the offer.

As crucially, while participation in Pathfinder may have been the result of pressure (hence the analogy with the Sword of Damocles), ‘forcing’ offenders to participate in this programme ultimately led to desirable consequences. This is an important finding because in practical terms, Pathfinder is not a new approach to rehabilitation; the concept of pathways and addressing the underlying issues that lead to involvement in crime has long been studied and shown to be effective (Sampson & Laub, 1995) and the treatment is freely available for people who want help. In the UK, a country that has relatively robust and well-functioning social services, voluntary treatment within the community is available and usually experiences a degree of success (e.g. www.princes-trust.org.uk; supportline.org.uk; prisonreformtrust.org.uk; www.gov.uk/guidance/offending-behavior-programs-and-interventions). This implies that by applying evidence-based rehabilitation programmes, such as Pathfinder and Turning Point, there is a non-negligible chance to reduce recidivism.

As importantly, we recognise that some people need reminders, nudges or threats to accept these services (Bersani & Doherty, 2018). We did not study a programme that offers treatment on an entirely voluntary basis. The threat of a criminal record can be an important ingredient. For those who are otherwise unlikely to seek help on their own, Pathfinder may provide such a tangible and certain consequence. However, importantly, the threat does not have to include a punitive element: the threat of the label, with all the informal and formal consequences it entails, was sufficient to drive eligible offenders into treatment, which, in turn, worked for many of them. A second chance for a clean slate in exchange for participating in a rehabilitation programme leads to less crime, without the need to apply additional sanctions. We therefore urge further research, with diverse populations, that would take advantage of the weight imposed by a criminal record on people’s willingness to desist.

This model is not for everyone

Nevertheless, we recognise that the generalisability of these results is limited by the population that was eligible to participate in Pathfinder: those who had committed common assaults, shoplifting, drug possession and other low-harm crimes. It may not be possible to caution criminals who are ineligible by law. Therefore, more research is required to understand the extent to which simple cautions and subsequent rehabilitation schemes are viable for other types of offenders.

In addition, if the operative threat is the criminal record, it seems unlikely that an offender with a rich history of crime would benefit from a forced rehabilitation in exchange for expungement of the index offence. We focused primarily on first-time offenders (two thirds of participants), but future research may wish to consider whether expungement of all criminal records in exchange for participating in rehabilitation would be beneficial, as such conjecture is outside the scope of the present paper.

Cost–benefit analysis

The process of charging, prosecuting and ultimately punishing offenders is an expensive business. Although we do not normally expect shoplifters or common assaulters to go to prison, adjudications are still costly, and avoiding them will have desirable economic consequences, with freed funds that can be used to facilitate rehabilitation instead of punitive outlets. From a law enforcement perspective, out of court disposals are the least expensive option, as they do not include prosecutorial expenses, and among them, the simple caution is the cheapest because it only includes the registration of the case in the police national crime database. This arrangement immediately follows the admission of guilt, and it is at this early moment that the criminal justice system, rather than a court of law and prosecution services, would ascribe the label of ‘offender.’ Therefore, institutional resources can be saved by applying this instrument more often.

Future research may wish to calculate the cost-to-benefit ratios of operating a programme such as Pathfinder, by looking at the total costing of this intervention (including the management of the offenders by caseworkers) relative to its benefits. In crude terms, we argue that the intervention is cost-efficient. The total economic cost of the intervention was £374,000, while the difference between the treatment and control conditions included 32 crimes. We estimate (again crudely, based on Heeks et al., 2018) that the cost of dealing with the control group’s post-randomisation offending approached £150,000, meaning the net spend per crime prevented was around £3600. This cost did not include officers’ time, which is rigid and nonreplicable (100% of crimes undergo a procedure of registering the criminal record initially), additional opportunity costs for the criminal justice system or the potential costs (and benefits) to each individual participant. Nor does this cost–benefit analysis include the expungement of the criminal record, which has short- and long-term tangible and intangible benefits. There might also be benefits to be gained from experimenting with different periods of case work enabling a greater capacity for the intervention and we might also consider expanding the analysis period up to 5 years.

Research implications

Field trials and clinical trials are equally important, but they do have one critical difference: attrition rates from the treatment group are far greater in real life settings, outside the sterile conditions of the laboratory (see review in Mills et al. 2019). In many clinical trials, attrition is assumed to be equally distributed in the experimental and control conditions because true double blinding can be achieved (e.g. both groups receive an identical looking pill). Thus, testing the policy of intervention can be akin to testing the efficacy of the intervention. However, in field experiments in criminology, attrition from the treatment group is not an exogeneous factor, and it interacts with the stimulus. Therefore, we think that an unbalanced dropout rate of offenders, with noncompliance in one treatment group but none from the control group, cannot simply be ignored.

Furthermore, attrition from a treatment programme should be both expected and considered clinically important. Although we should design studies that minimise attrition (see Ch. 5 in Ariel, Bland and Sutherland 2022), it is nevertheless unavoidable, especially in field settings with chaotic populations such as criminals. Subsequently, scholars are encouraged to conduct statistical analyses that adapt to these experimental settings, and in these conditions the instrumental variables approach becomes an informative complementary addition to the intention-to-treat model. The correction model does not assume there is no attrition, knowing very well that it exists; rather, it affects the outcomes of only one arm in the experiment by factoring in that outcomes may only be influenced by random assignment to treatment if the treatment is delivered.

Therefore, our findings serve as a further example of the importance of moving beyond standard intention-to-treat approaches to field trial analysis in policing and criminal justice. As Angrist (2006) set out, such frameworks underestimate the causal effect (in the absence of perfect compliance). While there is a ‘realpolitik’ aspect to considering results this way—they are, after all, a representation of the fact that not all offenders will participate in such schemes in practice—the true isolated effect of the treatment may be missed. This is not to say that the intention to treat effect is not important. As we have stated, not everyone will comply and it is essential to have measurement of the policy-level effects; this should not be up for debate. We propose, however, that complementary analytic strategies are beneficial. Underestimating the ‘average causal effect of treatment on the treated’ (Angrist, 2006: 28) also has the potential to bias theoretical understanding. The intention-to-treat model would challenge the broader literature on self-motivation being a critical component of desistance (Burnett & Maruna, 2006). We might also speculate that the intention to treat outcome is evidence that we have observed some form of crime prevention hormesis. Linning and Eck (2018) argue that such a phenomenon, normally observed in natural sciences, is possible in criminology. Hormesis is used to describe the phenomenon of the dose–response relationship, thereby inferring a ‘sweet spot’ for a dosage to achieve the optimal response. Too low a dosage or too high a dosage may have no impact or even backfire. Our results do not show the latter effect, which Latessa and Smith (2015) indicate is a possibility in circumstances in which interventions designed for high-risk offenders are applied to low-risk offenders; however, it is possible that 16 weeks of work with a caseworker is too high a dosage for low-harm offenders of the type participating in Pathfinder. Altering the dosage provided (to a lower amount, as in Project CARA in Hampshire, see Strang et al., 2017) may induce a greater reduction in reoffending or it may not. We do not think this is a strong argument in this circumstance, however, as the instrumental variables model has given us a more nuanced understanding. There are reductions to be had even from a low base rate. The repeat offending rate among the treatment group was just 3%, the mean number of crimes for which they were sanctioned was less than one per year and the level of harm was less than the equivalent of 20 days in prison. It is difficult to improve on that.

Additional limitations

This experiment focused on reoffending outcomes. As we have pointed out, Turning Point and Checkpoint both made estimates of cost reductions. Turning Point also considered the satisfaction of victims (Slothower, 2020). Both these elements are missing in our study but would benefit any replications. Cautions incur lower criminal justice costs than prosecutions (Neyroud & Slothower, 2015); therefore, we would expect to see lower marginal gains than those reported in the deferred prosecution projects. However, the extent of the potential reduction remains unclear and is of high importance to the future of tailored programmes such as Pathfinder that have no single funding framework (see Lundgaard, 2019 for a discussion on this subject).

Finally, all crime data are limited by the nature of crime reporting standards and the propensity of victims to present to law enforcement (see Ariel and Bland, 2019 for a further discussion). No alternative method of measuring the dependent variable was considered viable; however, it limits our overall understanding of the utility of Pathfinder. Other studies should invest in surveys with offenders to obtain additional information that is undoubtedly missing from our narrative.