Monitoring prisoners preparing for release: Who ‘fails’ in open prison conditions?

Abstract

Open prisons play a vital role in offender rehabilitation and resettlement but absconds, temporary release failures (TRFs) and re-offences have damaging implications for the legitimacy of these institutions. Identifying and mitigating the risk for such ‘failures’ is crucial. The present study examined predictors of failure in a sample of 316 adult male prisoners in two open prisons in England and Wales. Almost one-third (n = 100) of the sample failed in open conditions, the greatest proportion (n = 83, 26.3%) instigated by the prison to maintain security and good order (security recall). Yet, only seven re-offended in the year following custodial release. Absconds, custodial re-offences, and TRFs were rare events. Regression analysis identified five factors predicting security recall. Current behaviour, rather than static/historical risk factors, more reliably predicted such failures. Behavioural monitoring and systemic policy re-evaluation are proposed as way of mitigating failures in open prisons.

Keywords

Abscond behaviour monitoring open prisons recidivism risk management temporary release failure

Minimum security or ‘open’ prisons, and in particular, the temporary absence/release schemes utilised in such regimes, play an important and unique role in the rehabilitation, preparedness, and resettlement of ‘high’ risk and long-term prisoners, as they transition back to the community. Gradual re-exposure to the community may convey benefits for prisoners and the wider community, foremostly moderating the risk of recidivism (Cheliotis, 2008; Hillier and Mews, 2018; Mastrobuoni and Terlizzese, 2022; Ministry of Justice, 2015, 2019). Temporary release activities supporting successful resettlement include finding stable accommodation, securing work or training, building family ties, and reintegration into local communities (Ministry of Justice, 2013a, 2013b). Indeed, temporary release – the process by which prisoners are permitted leave from prison for the purpose of engaging in activities to support their resettlement – provides opportunities to test prisoners in conditions akin to those faced on leaving prison. Temporary release failures (TRFs) are typically rare. Hillier and Mews (2018) report that there were 7000 individuals released under the Release on Temporary Licence (RoTL) scheme from prisons in England and Wales in 2016 (on average, 47 incidences of RoTL per individual) with less than 0.1% resulting in release failure, that is, breach of licence, failure to return to custody, or alleged offending; a failure rate of just 75 per 100,000 RoTLs. Yet release failures, irrespective of victim harm, can have damaging repercussions for the legitimacy of an institution (cf. Dawar and Davis, 2014; HM Inspectorate of Prisons, 2014). While escalating risk can be mitigated in custodial settings by transferring the prisoner back to secure conditions (i.e., security recall), this is costly and potentially iatrogenic. Cautionary tales resound from Alexander (2006) and Beijersbergen et al. (2016) who attribute adverse outcomes such as abscond and re-offending to procedurally unjust rule enforcement. To maintain a rehabilitative environment that prepares for community reintegration, identifying those individuals genuinely at heightened risk for failure is a necessary first step. Below we review factors known to be linked with adverse outcomes such as abscond and re-offending.

Predicting risk of ‘failure’

There is a substantial body of evidence underpinning the prediction of recidivism by offenders released into the community (see meta-analyses by Eisenberg et al., 2019; Gendreau et al., 1996; Katsiyannis et al., 2018) which serves as a potential starting point for predicting failure amongst those residing in open prisons, given consistency observed between custodial and community re-offence behaviours (McDougall et al., 2013). These meta-analyses broadly support Bonta and Andrews’ (2017) general personality and cognitive social learning (GPCSL) theory of a ‘central eight’ risk factors for recidivism. These risk factors are as follows: a history of criminal behaviour (early, persistent and varied criminal activity), antisocial personality pattern (impulsive, callous and aggressive disposition), procriminal attitudes (attitudes condoning antisocial behaviour), antisocial companions, family/marital (lack of prosocial support and conflictual intimate relationships), school/work (poor engagement, performance and relationships within), leisure (low engagement and satisfaction in prosocial leisure pursuits) and substance abuse. A recent meta-analysis by Goodley et al. (2022) focused specifically on adults sentenced to and released from custody. Their review identified 17, largely static/historical, predictors of recidivism, the majority of which could be accounted for within the four domains of antisocial potential (‘big four’: Andrews et al., 2006) such as number of previous contacts with the Criminal Justice System, holding procriminal attitudes, and institutional misconduct. These risk factors showed a consistent albeit small relationship with recidivism across jurisdictions and heterogeneous samples.

Substantially less attention has been given to identifying risk factors for the outcomes of absconding and failure on temporary release. Absconding from forensic hospital facilities has received most attention. The following are consistently reported as predictors of absconding in low secure psychiatric units: a history of absconding attempts; substance misuse; higher risk ratings on the Historical Clinical Risk-20 (HCR-20); and evidence of non-compliance (Cullen et al., 2015; Martin et al., 2018; Wilkie et al., 2014). Absconders in these samples were variously motivated. The predominant motivation was instrumental in nature (goal-directed) and often resulted following substance misuse and/or an adverse event (e.g., conflict on the unit; family crises). Boredom and frustration were also widely represented, as were absconds following periods of mental illness. Absconding impulsively/opportunistically or incidentally, for example, losing track of time on release was uncommon in these samples (Martin et al., 2018; Mezey et al., 2015; Simpson et al., 2015; Wilkie et al., 2014). In a prospective study of incidents of absconding, Cullen and colleagues (2015) found that in-hospital behaviour, namely verbal aggression and substance misuse, discriminated those who absconded from those who did not, with greatest reliability.

Presently there are just three published studies with custodial samples from which data on absconding or TRF can be delineated, each of which – except the Emirali et al.’s study – conflate absconding with escape risk (Emirali et al., 2020; McSweeney et al., 2011; Mews, 2014). McSweeney et al. (2011) explored predictors for absconding in a sample of 2312 prisoners at risk of absconding. It is unclear why those identified as a potential risk were included in the sample given there were 362 absconding events during the study period. Nonetheless, they identified the following correlates predictive of increased risk: having served two or more custodial sentences; having outstanding needs in each of the ten criminogenic areas assessed in the offender assessment system (OASys); serving a sentence for robbery; and having an offending career longer than average (i.e., spanning 7 years). Other potential predictors were factors typically associated with re-offending such as being male, having served previous custodial sentences, and, presenting with a high likelihood of reconviction (measured using the OASys assessment). Emirali et al. (2020) compared a group of absconders against a control group and identified one predictor of absconding: number of previous offences. However, this study was based on a small sample of prisoners (N = 61). Mews (2014) analysed predictors of absconding in a sample of 23,701 prisoners, 347 (1.5%) of which were recorded as having absconded or escaped. This larger study identified two predictors for absconding: a previous absconding incident, and previous TRF. Only the latter remained statistically significant after controlling for several other historical and demographic variables. However, the study was limited to exploring the predictive ability of a small number of static variables, which, given Cullen et al.'s (2015) analysis showing the relative value of in-hospital behaviour, are perhaps sub-optimal for discriminating absconding events.

A handful of unpublished qualitative studies conducted in English prisons have also been conducted into the motivations underpinning the decision to abscond, revealing potentially promising dynamic risk variables (Chant, 2015; Papworth, 2015; Picksley, 2016). These studies identified several ‘push’ (avoidance-focused) and ‘pull’ (approach-focused) factors. Push factors included prisoners feeling unsafe (e.g., threats, intimidation), often linked to drug debts; facing barriers to progression (e.g., denial of temporary release); struggling to adapt to the less structured regime of open prison; avoiding further punishment for misdemeanours (e.g., failing a drugs test); and experiencing poor relationships with staff. Pull factors included absconding to convey contraband (e.g., substances) into the prison; and responding to family pressures. These findings, however, remain speculative.

The current study

Following a series of serious further offences by prisoners granted temporary release from open prisons in England and Wales in 2013, the then Justice Secretary commissioned an internal review into the use of the RoTL scheme. The review concluded that a uniform approach had developed to managing all prisoners, with access to RoTL having become an expectation and not, crucially, a privilege predicated on comprehensive risk evaluation (HM Inspectorate of Prisons, 2014). This led to new guidelines being issued (NOMS, 2015a) alongside an enhanced behaviour monitoring (EBM) system (NOMS, 2015b) designed to ensure that ongoing risks (i.e., absconding, harm, re-offending) posed by prisoners in open conditions are appropriately identified and managed. Behavioural monitoring has demonstrated efficacy as an individualised and temporally responsive means of risk identification (Clark et al., 1993; McDougall et al., 2013; Pearson & McDougall, 2017) as an alternative or complement to structured professional judgment (SPJ) tools which can be unwieldy to administer (Green et al., 2010; Ho et al., 2018).

EBM consists of two elements: (i) a psychologist-led case file review (CFR) – detailing the risks posed by the prisoner in open conditions; and (ii) a behavioural monitoring intervention – a wraparound package of support consisting of collaborative goal setting, behavioural feedback, and coaching of positive alternative behaviours – reserved for those deemed at heightened risk for ‘failing’. However, there is little guidance beyond ‘current risk management concerns’ regarding which prisoners should be targeted for the additional monitoring and support of EBM.

Thus, the primary aim of the present study was to isolate a set of factors predictive of increased risk of failure in open conditions. Given the limited research we consider this crucial; open prison placement and temporary release schemes prior to release are expected to impact positively on recidivism rates (Cheliotis, 2008; Hillier and Mews, 2018; Mastrobuoni and Terlizzese, 2022; Ministry of Justice, 2015, 2019) but their existence is frequently threatened by release failures (cf. Dawar and Davis, 2014). The secondary aim of this study was to identify the extent to which failure in open conditions equates to later community re-offending and so consider whether EBM should be targeted at those who infringe custodial rules. Ultimately, by modelling those at ‘heightened’ risk of failure, we aimed not only to ensure allocation to the EBM scheme in England and Wales is data-driven but, more widely, to develop a set of guiding principles to help policymakers nationally and internationally utilise open prison placements safely, given their resettlement benefits.

Method

Study sample and design

The current study adopted a retrospective cohort design. Prisoners selected were resident in two open prisons within England and Wales holding male prisoners aged 18 and above, and subject to psychologist-led CFRs as part of the EBM scheme. Eligible or ‘restricted RoTL’ prisoners were those who met one or more of the following criteria as per Prison Service Instruction 13/2015 (NOMS, 2015a): serving an indeterminate sentence; subject to multi-agency public protection arrangements (MAPPA); and/or, assessed as high or very high risk of harm on the OASys risk assessment tool (Home Office, 2006). The whole sample had progressed through higher security classifications of prison to open conditions.

Prisoners on whom CFRs had been completed between 2 June and 22 July 2014 were selected for the study. The time period reflects the initial phase of EBM implementation in open prisons across England and Wales – the psychologist-led CFR. This time period was selected to avoid any confounding effect that the second phase of EBM – a behavioural monitoring intervention – might have on failure rates and thus on identifying predictors of failure. Basic demographic details pertaining to those eligible for the study were recorded on electronic spreadsheets from which it was possible to trace their custodial records.

The eligible population was comprised of 322 prisoners. Six were removed from the sample: three because it was not possible to trace their records,¹ and three due to their being removed for procedural reasons (i.e., they no longer met the eligibility criteria). In total, the final sample included 316 participants. The average age of the sample was 40 years old (SD = 13.1) and the majority of participants were serving sentences for violent offences which is representative of the open prison population of England and Wales, as calculated using the prison population tool (Ministry of Justice, 2021). Our sample however, contained a higher proportion of men convicted of sexual offences and smaller proportion of drug offenders. There is limited data about the proportion of men with substance misuse or mental health problems but our sample was consistent with prevalence rates reported in recent studies (Office for Health and Improvement Disparities, 2022; Tyler et al., 2019). Table 1 shows the sample characteristics.

Table 1.

Sample characteristics.

Variable	Measure
Age, M_years (SD, Range)^a	40.1 (13.1, 20–81)
Previous convictions, M_number (SD, Range)^b	7.2 (7.9, 0–40)
Previous incarcerations, M_number (SD, Range)^c	2.3 (3.6, 0–21)
Time at risk (open conditions), M_days (SD, Range)^d	548.3 (350.4, 55–2180)
Index offence (%)
Burglary	5.1
Drug	2.5
Fraud	1.6
Robbery	14.2
Sexual	31.0
Violence	38.6
Other	7.0
Sentence type (%)
Determinate	39.2
Indeterminate for Public Protection	32.0
Life	28.8
Substance misuse (%)
Yes	54.4
No	45.6
Mental illness (%)
Yes	26.6
No	73.4

Mdn = 39

Mdn = 4

Mdn = 1

Mdn = 511

Data and measures

Predictors

Risk predictors were identified from the research literature into absconding, release failure and re-offending (reviewed above). The list was augmented by consulting with a group of applied psychologists with expertise in working in open conditions. Data pertaining to these predictors were accessed from a variety of data sources including: the EBM CFR record, and national operational databases for the management of offenders such as the National Offender Management Information System (NOMIS), OASys, and the Public Protection Unit Database. Twenty-six risk factors potentially relevant to the above outcomes were identified. These included socio-demographic factors such as age at CFR, and previous employment prior to custody, and, criminal history variables including number of previous convictions, number of previous incarcerations, index offence type, range of previous convictions, sentence type, and, as identified by McSweeney et al. (2011), whether the offending career spanned a period of 7 years or more. Prior parole revocation, prior absconding, and prior TRF were also recorded. Medical history variables included history of mental illness, personality disorder, and substance misuse disorders. Behavioural risk factors included the number of behavioural incentives and warnings recorded in the prisoner's electronic wing behaviour record in the year prior to transferring to open conditions, and in the 6 months prior to release or failure event. These written behavioural evaluations are used to determine the privilege entitlement for each prisoner. Total number of adjudications across the prison sentence, and in the 6 months to release or the failure event, were also recorded. Risk assessment variables included the Offender Group Reconviction Scale (OGRS3) (Ministry of Justice, 2009), OASys violence predictor (OVP) (Howard and Dixon, 2012) score, and a summary ‘highest’ risk score; the highest risk category attributed to that individual on any static risk assessment tool.²

Outcome measure

There were four types of outcome measure: absconding, custodial re-offence, TRF, and security recall. Collectively they are referred to as ‘failure’ here. Absconding was defined as unlawfully gaining liberty for 15 minutes or more without overcoming a physical security restraint, and TRF was defined as a failure to adhere to any condition included on the individual's temporary release licence. These are the official definitions used for recording absconding and TRFs in His Majesty's Prison and Probation Service (HMPPS) (Justice Data Lab, 2022). Custodial re-offence was defined as reconviction for any alleged offence, occurring either within or outside of the prison boundaries, whilst resident in open conditions. Security recalls were recalls to secure conditions, initiated by a Governor, typically following a breach of the prison rules, deemed indicative of an intolerable increase in the individual's risk. Participants were tracked until release or failure in open conditions – whichever occurred first – and this was used to classify the outcome event. This follow-up period represents time period 1.

Time period 2 commenced from the date of release from custody. Participants were tracked for one year or until a community re-offence – whichever occurred first. The purpose of the community follow-up was to identify those who re-offended in the 1 year following release from open conditions, to understand the community re-offending outcomes of those who succeeded and failed in open conditions. Six of the individuals successful in graduating from open prison (time period 1) re-offended in the community (time period 2). Given their offending shortly following release, it seemed contraindicative to categorise these individuals as ‘successful’ open prison graduates; arguably they had masked behaviours indicative of ongoing risk which quickly manifested in the community, and which might otherwise have been addressed in custody, had the concerns been identified. Nonetheless, they were technically successful in navigating open conditions so they were removed from the data analysis as a precaution.

The data were primarily located using NOMIS and were cross-referenced against other records to validate the failure or custodial re-offence. In a small number of cases the custodial failure could not be validated; the decision to fail was clearly overturned within a 28-day period and the individual returned to open conditions. Such cases were not classified as a failure and the follow-up period continued until the release or valid failure.

Data pertaining to custodial failure and community recidivism were collated in December 2020. All participants had graduated or ‘failed’ in open conditions at the end of the follow-up period and were included in the analyses. The average time spent in open conditions was 548 days.

Analytical plan

The data analysis was split into two parts. Part one focused on the security recall group; the largest ‘failure’ group (n = 83, 26.3%, see section ‘Results’). The other failure groups (absconding, TRF, custodial re-offence) amounted only to an additional 5% of the sample. Given the substantive differences between the failure types, pooling the groups into a general failure outcome variable was inappropriate. We saw merit in preventatively identifying the security recall group in open conditions for future intervention, given the long-term community outcomes for those recalled to secure conditions in this sample did not result in community re-offending (n = 7, 9.7%, see section ‘Results’).

Logistic regression analyses were employed to examine which variables predicted security recall. Harrell and colleagues (1996) suggest that to have predictive discrimination that validates on a new sample, no more than m/10 predictor variables should be examined in the multiple regression model where m is the number of participants in the less frequent outcome category, in this case, security recall (n = 83). As such, it was necessary to reduce from 26, to a maximum of eight, the number of predictor variables included in the final model. This was achieved by following the purposeful model building strategy in Hosmer et al. (2013), the stages of which are outlined in Table 2. We also included an additional step of removing variables with collinearity of [r] > 0.8; retaining for bivariate analyses only the collinear variable which showed the strongest relationship with security recall. Field (2005) advocates r > 0.8 as an appropriate indicator for when collinearity distorts model estimation and prediction.

Table 2.

Selection of variables in final regression model.

Step 1: Chi square and Kruskal–Wallis tests
Variables excluded:	X²		p
Age at Case File Review (CFR)	49.89		.596
Step 2: Collinearity tests
Variables retained:	R				Variables excluded:
Number of previous convictions	0.863				Number of previous incarcerations
	0.786				Number of different offence types
Step 3: Backward logistic regression
Variables retained:	B	S.E.	Wald	p	Exp(B)	Variables excluded:	Score	p	−2ll change
Substance misuse in 6 months to outcome	3.07	.572	28.70	<.001	21.46	Positive behaviour in 12 months prior to transfer to open	0.02	.963	0.02
Adjudications in 6 months to outcome	2.46	.488	25.51	<.001	11.74	Parole violations	0.20	.889	0.20
Diagnosis of personality disorder				.001		Highest risk			0.43
Negative diagnosis	−0.39	.527	0.55	.459	0.68	Low	0.42	.811
Subthreshold traits	1.87	.510	13.50	<.001	6.51	Medium	0.18	.894
Positive diagnosis	0.50	.594	0.71	.400	1.65	High	0.15	.696
Property index offence	0.94	.500	3.53	.060	2.56	Number of previous convictions	0.08	.775	0.08
Behavioural warnings in 12 months prior to transfer to open	0.36	.082	19.10	<.001	1.43	Positive behaviour in 6 months to releasePrior TRF	0.070.08	.785.778	0.080.08
Total adjudications across sentence	0.06	.030	4.24	.039	1.06	Sentence type			0.47
OVP risk assessment score	0.04	.015	8.20	.004	1.04	Determinate	0.47	.791
Treatment completion	−0.82	.462	3.10	.078	0.44	IPP	0.00	.982
Treatment completion	−0.82	.462	3.10	.078	0.44	Life	0.24	.627
						Employed prior to custody	0.14	.713	0.14
						Behavioural warnings in 6 months to release	0.16	.687	0.16
						OGRS score	0.18	.668	0.18
						Adverse event	0.40	.526	0.41
						Previous abscond	0.66	.415	0.66
						History of substance misuse	0.94	.331	0.94
						OASys needs			12.29
						One	0.22	.638
						Two	0.21	.648
						Three	4.03	.045
						Four	0.16	.692
						Five	0.00	.947
						Six	0.07	.793
						Seven	2.47	.116
						Eight	0.00	.959
						Nine	1.31	.253
						Ten	3.96	.047
						History of mental illness	1.16	.282	1.14
						Offending career of 7 years or more	1.19	.275	1.20
Constant	−3.59	.680	27.82	<.001	0.28
Step 4: Removal of non-statistically significant variables from those retained at Step 3
Variables retained:					Variables excluded:
Substance misuse in 6 months to outcome					Property index offence
Adjudications in 6 months to outcome					Treatment completion
Diagnosis of personality disorder
Behavioural warnings in 12 months to open conditions
Total adjudications
OVP risk assessment score

Six independent variables were included in the final model. These were: Substance misuse in the 6 months prior to release/failure; adjudications in the 6 months to release/failure; diagnosis (subthreshold traits) of personality disorder; behavioural warnings in the 12 months prior to transfer to open conditions; total adjudications across sentence; and OVP risk assessment score. The magnitude of the effect sizes was determined using Cohen's (1988) rules and converting these into the corresponding effect sizes outlined in Salgado (2018). Thus, Cohen's d of .2 (small), .5 (medium) and .8 (large) are represented by ORs of 1.434, 2.488 and 4.258, respectively.

Part two of the data analysis concerned the other failure groups (absconding, TRF, custody failure, and community failure). Given the small number of failures related to each category, the data were analysed using conceptual content analysis (Weber, 1990) to assess the extent to which these groups were an extension of, or indeed diverged from, the security recall group. Each failure was summarised into a case description to code the content analysis; the most ‘typical’ examples of which are presented in the supplemental materials (online version).

Results

Descriptive statistics

The majority of the sample graduated successfully from open prison conditions (n = 216; 68.4%). One hundred (31.6%) individuals ‘failed’. Three (0.9%) of the total sample (N = 316) re-offended whilst resident in open conditions, four (1.3%) absconded and 10 (3.2%) breached the terms of their licence whilst on temporary release. The remaining and most significant proportion of failures in open conditions were ‘security recalls’ (n = 83; 26.3%), recalls initiated by the prison. The failure event in open conditions typically occurred after 357 days on average (range: 55–1735) with absconding typically occurring earlier (M_days = 272) and TRFs occurring later (M_days = 443).

Of those released from open conditions (n = 216), six (2.8%) re-offended within 1 year of being released and 18 were recalled (8.3%). Eighty-three of the 100 men who failed in open conditions were released from closed prison during the follow-up period. Nine re-offended (10.8%) and 16 were recalled (19.3%). Seventy-two of these 83 men were security recalls. Seven of this group re-offended (9.7%) in the 1 year following release.

Model of security recall

A logistic regression analysis was used to assess which of the six independent variables accounted for unique variance in predicting recall to prison for security reasons. The final model is presented in Table 3. Three variables produced large effect sizes. Substance misuse in the 6 months to recidivism/failure was the best predictor, increasing the odds of recall by a factor of 17.46 (95% CI: 5.99, 50.89). Each additional adjudication in the 6 months to recidivism/failure increased the odds of recall by a factor of 10.63 (95% CI: 4.21, 26.85). Personality disorder also predicted recall – albeit only those formally assessed as having problematic personality traits subthreshold for diagnosis were more likely to be recalled (OR = 5.20, 95% CI = 2.01, 13.46). Likewise, each additional behavioural warning in the 12 months to transfer to open conditions (OR = 1.42, 95% CI = 1.21, 1.67) and each unit increase in the OVP score (OR = 1.05, 95% CI = 1.03, 1.08) increased the odds of recall; both small effect sizes. Total unadjusted number of adjudications across sentence was not a significant predictor and was dropped from the final model. Overall, the model accounted for 57% of the variance and correctly classified 85% of the sample (No recall = 93.8%; Recall = 62.7%).

Table 3.

Logistic regression model for prediction of security recalls.

Variable	β	S.E.	Wald	p	Exp(β)	CI
Substance misuse in 6 months to recidivism/failure	2.86	0.55	27.45	<.001	17.46	[5.99, 50.89]
Adjudications in 6 months to recidivism/failure	2.36	0.47	25.04	<.001	10.63	[4.21, 26.85]
Diagnosis of personality disorder			14.38	.002
Negative diagnosis	−0.47	0.52	0.81	.369	0.63	[0.23, 1.67]
Subthreshold traits	1.65	0.47	11.54	.001	5.20	[2.01, 13.46]
Positive diagnosis	0.39	0.60	0.43	.510	1.48	[0.46, 4.78]
Behavioural warnings in 12 months to transfer to open	0.35	0.82	18.45	<.001	1.42	[1.21, 1.67]
OASys violence predictor (OVP)	0.05	0.14	14.03	<.001	1.05	[1.03, 1.08]
Total adjudications across sentence	0.05	0.03	3.10	.078	1.05	[0.99, 1.16]
Constant	−4.16	0.55	56.46	<.001	0.16

Cox and Snell = 0.40; Nagelkerke R² = 0.57; −2 log likelihood = 200.046; Correctly classified = 85.0%.

Table 4 provides a breakdown of the mean scores for the security recall group versus those who were successful in open conditions and on release according to the five predictive independent variables included in the final logistic regression model.

Table 4.

Security recall versus non-failures according to the five independent variables.

Variable	Total (N = 293)	No recall (N = 210)	Recall (N = 83)	Statistic/significance
Adjudications in 6 months to recidivism/failure
M (SD)	0.2 (0.4)	0.1 (0.2)	0.4 (0.6)	Χ₍₁₎² = 44.00
Median	0	0	0	p < .001
Behavioural warnings in 12 months to transfer to open
M (SD)	1.4 (2.0)	1.0 (1.7)	2.5 (2.4)	Χ₍₁₎² = 51.38
Median	1	0	2	p < .001
Diagnosis of personality disorder
Not assessed	168	126	42	Χ₍₁₎² = 11.80p = .01
Negative diagnosis	58	47	11
Subthreshold traits	45	25	20
Positive diagnosis	24	14	10
OASys Violence Predictor (OVP)
M (SD)	23.1 (13.5)	20.7 (12.4)	29.1 (14.4)	Χ₍₁₎² = 56.76
Median	21	19	27	p = .13
Substance misuse in 6 months to recidivism/failure
No	254	206	48	Χ₍₁₎² = 76.39
Yes	41	6	35	p =< .001

The mean scores and standard deviations were used to produce cut-off scores to distinguish the security recall group from the non-recall group, on each variable. Where there was overlap in the standard deviation of the means, 99% confidence intervals were applied to select the cut-off score. The mean number of items observed for the security recall group was calculated and compared against the other failure groups (abscond, reoffending, TRF) to identify the extent to which these groups could be considered an extension of the security recall group. These are outlined in Table 5.

Table 5.

Security recall cut-off scores.

Variable	Non-recall	Security recall
Substance misuse in last 6 months	No	Yes
Number of adjudications in 6 months to outcome	0	1
Personality disorder diagnosis	Not assessed/negative diagnosis	Subthreshold traits
Behavioural warnings in 12 months to open conditions	0–1	2+
OVP score	0–23	24–100

Only one case in the security recall group met the criteria for every variable. On average, they met the criteria for two of the five variables (M = 2.23; SD = 0.99).

Case descriptions

Below we briefly summarise case descriptions of the ‘rare event’ groups. Their similarities/differences were compared to the security recall group to assess the extent to which these groups were an extension of, or indeed diverged from, the security recall group. Fuller descriptions of cases ‘typical’ to each category are available in the supplemental material.

Absconders

Four individuals absconded from open conditions. On average they met the criteria for fewer than one of the five ‘security recall’ variables (M = 0.75, SD = 0.96). Three individuals were serving indeterminate sentences and three had been recalled to custody at least once during the sentence following release. Despite this, three of the four individuals were assessed as a low risk for re-offending and in the three cases where details of the absconding were investigated, all three reported relational problems with peers in the prison acting as a catalyst to the absconding.

Temporary release failures (TRFs)

Ten individuals breached their licence whilst in the community on release on temporary licence (RoTL). On average, they endorsed one of the variables predictive of security recall (M = 1.40; SD = 1.07). All except one were serving an indeterminate sentence. All individuals had accessed structured treatment to address their offending behaviour, and the majority (seven) of individuals were assessed as a low risk of re-offending. Nonetheless, all attracted behavioural warnings in the 6 months prior to the TRF. In most cases, the TRFs related to institutional offence behaviour such as misuse of substances, or engagement in sexually inappropriate behaviour/relationships.

Custody re-offenders

Three individuals were reconvicted whilst resident in open conditions. On average they met the criteria for less than two of the variables associated with the security recall group (M = 1.67; SD = 1.15). All were serving indeterminate sentences. Drawing patterns from the data is difficult given two of the three individuals were convicted for bringing unauthorised mobile phones into a prison. This is an indictable offence in England and Wales but arguably might be better framed as a breach of licence. The third prisoner was convicted of sexual assault. In all three cases, the level of harm caused was reduced compared to their respective index offences. These prisoners were also younger (age < 30) than the security recall group and claimed to have been extorted or influenced by peers.

Community re-offenders

Six individuals, released from open prisons, were reconvicted within 1 year of release to the community. All displayed similar behaviours to those recorded in their respective index offences and one-half were reconvicted of offences of equivalent seriousness (i.e., level of harm). On average, these individuals endorsed one or two variables predictive of security recall (M = 1.50; SD = 1.05). All six had offence histories spanning 7 years or more, were serving indeterminate sentences and had a history of substance misuse, reflecting a history of antisocial behaviour. The age range (31–45 years) was relatively narrow. All were considered to have completed treatment commensurate with their re-offending risk yet all but one had attracted a behavioural warning of some form in the 6 months prior to release to the community.

Discussion

The primary aim of the present study was to isolate a set of factors predictive of increased risk of failure in open prisons. Scant previous research has examined risk factors for custodial release failure yet such understanding is crucial, not only for the effective management of risk but also for maintaining the legitimacy of open prisons and temporary release schemes, given their positive impact on recidivism rates (Cheliotis, 2008; Hillier and Mews, 2018; Mastrobuoni and Terlizzese, 2022; Ministry of Justice, 2015, 2019).

Sufficient data were available to statistically isolate factors predictive of ‘security recall’, and three factors produced large effect sizes. Subthreshold traits of personality disorder (large effect), each additional unit increase in OVP score, and each additional behavioural warning in the 12 months prior to transfer to open (both small effects) are each measurable prior to receiving prisoners in open conditions and so could be used to pre-emptively assign cases to EBM. Substance misuse and each additional unit increase in adjudications in the 6 months prior to recall produced large effect sizes, highlighting the importance – in predicting adverse outcomes – of continuously monitoring behaviour (Clark et al., 1993; McDougall et al., 2013). On average, these behavioural indicators of risk typically emerged at 1 year (357 days) from transfer to open conditions and could be used to reactively assign prisoners to EBM.

The factors predictive of security recall are shown consistently to predict recidivism and are accounted for by Bonta and Andrews’ (2017) GPSCL theory. The number of institutional infractions (measured here by adjudications in the 6 months to failure and behavioural warnings in the 12 months to transfer to open conditions) is likely a marker of antisocial persistence (Andrews et al., 2006; Bonta and Andrews, 2017) given behavioural schemas are relatively stable across time and context (Mischel and Shoda, 1995) and act frequency (i.e., repeated acts of antisocial behaviour) provides a reliable index of antisocial persistence (Buss and Craik, 1983). Notably, behavioural warnings in the 6 months prior to recall did not feature in the final model. Open prisons operate with a smaller staffing model than closed prisons, so it is plausible that antisocial behaviours go undetected more frequently, and it is only those observed behaviours of sufficient severity to warrant adjudication, which correlate with recall. It is also possible that a timeframe of 6 months was insufficient to tap into antisocial persistence.

Antisocial personality is one of the ‘big four’ risk factors for recidivism and in this study, personality disorder diagnosis predicted recall. Still, it was those with subthreshold traits of personality disorder who were more likely to be recalled. There are several possible hypotheses to explain the lack of relationship between diagnosed personality disorder and security recall. First, the variable captured any form of personality diagnosis, not just those with antisocial orientation and thus included diagnoses which do not correlate strongly with re-offending (cf. Roberts and Coid, 2010) diluting the relationship. Yet, Burt et al. (2016) found that even in highly psycho-pathic offenders, re-offending rates can be moderated by factors such as age, the presence or absence of particular traits, and level of community support. Given subthreshold traits of personality disorder did predict security recall, another perhaps more plausible hypothesis is that those with diagnoses of personality disorder were screened into and received more intensive rehabilitative intervention (cf. Skett et al., 2017). That is, these individuals were either rehabilitated, that is, ‘walking the walk’, or had learned at least, how to ‘talk the talk’. Substance misuse also significantly predicted recall and represents one of the ‘moderate four’ risk factors for criminal recidivism (Andrews et al., 2006; Bonta and Andrews, 2017).

The final factor – OVP score – is potentially anomalous within the GPSCL theory. Whilst the OVP has predictive validity for violent offending (Howard and Dixon, 2012; 2013), meta-analyses consistently find that a history of violent offending is less predictive of recidivism compared to other offence types, particularly burglary (Goodley et al., 2022). Furthermore, violent offenders were no more likely to be recalled in the current sample. However, the OVP does tap into stable dynamic risk factors consistent with the GPCSL theory (e.g., attitudes, employment), perhaps accounting for unique variance in the model.

There were notable omissions from the final model predicting security recall. Neither prior absconding nor TRF correlated with security recall despite predicting absconding elsewhere (Mews, 2014). However, the numbers of men with such histories were too low to draw firm conclusions in this study. Likewise, there was no identifiable link between security recall and the ‘push’ and ‘pull’ factors described in qualitative studies of open prison failure (Berman-Roberts, 2015; Chant, 2015; Flowers, 2014; Papworth, 2015; Picksley, 2016). Atkinson and Mann (2012) describe a number of procedural and individual level reasons why important risk-related information such as these ‘push’ and ‘pull’ factors goes unrecorded by staff which might be further exacerbated in open prisons given their lower staffing profile compared to closed prisons. Finally, factors which consistently predict re-offending such as age, criminal history (i.e., previous convictions, incarcerations), and some measures of antisocial persistence (e.g., offending history length, offence versatility) did not correlate with security recall. Since Eno-Louden and Skeem (2013) have previously identified the prevalence of bias in the recall decisions of community probation officers, assessing the basis for decision-making amongst open prison managers is worthy of exploration.

Does failure in open conditions equate to later community re-offending?

A secondary aim of this study was to explore whether failure in open conditions equates to later community re-offending and to consider whether wraparound support packages in open prisons, like EBM, should always be targeted at those who infringe custodial rules. The present study found that whilst open prison graduates re-offended at the lowest rates (2.8%), there was, nonetheless, a mismatch between the number of men being returned to closed prisons for security reasons (n = 72) and their community re-offending rates (n = 7). Violations inevitably create methodological and policy-level challenges (see Ostermann et al., 2020 for a discussion). Returning a prisoner to closed conditions prevents the re-offending outcome from being determined; all or none of these individuals might have been on a re-offending pathway. Relatedly, returning the individual to closed conditions represents an intervention that might change the trajectory of offending.

Whilst determining the effectiveness of security recall as an intervention is beyond the scope of this paper, some conclusions can be drawn from the current data and wider literature. First, those who utilise resettlement opportunities afforded by temporary release schemes in open prisons, are less likely to re-offend (e.g., Cheliotis, 2008). They re-offend at lower rates compared to closed custodial samples (Ministry of Justice, 2019) and in the present study, open prison failures. Second, the current data suggest that the specificity of decision-making in open conditions is effective. Arguably, only 17 adverse events – three custodial re-offences, four incidents of absconding, and ten release failures – were missed. Yet, to achieve that sensitivity, there is a trade-off with specificity. Over 90% of those returned to closed conditions for security reasons did not re-offend in 1 year of release. Given the best interventions reduce recidivism by 40% (Lipsey and Wilson, 1998) and meta-analyses indicate that discipline-based approaches delivered without rehabilitative support are unlikely to reduce re-offending (Barnett and Howard, 2018), we argue that prisoners are too readily being recalled. The research indicates that factors such as positive punishment (Schaefer, 2016), procedural injustice (Beijersbergen et al., 2016), stigma, and lack of hope (LeBel et al., 2008) all increase re-offending. Therefore, overzealous recourse to recalling prisoners to closed prison conditions is more likely to impede than accelerate desistance processes and alternatives to recall are likely to be less costly both psychologically and financially, as discussed below.

EBM and rare event outcomes

Statistically, there was little difference between the profiles of the security recall and other failure groups. Whilst those falling into the other failure groups endorsed fewer variables predictive of security recall on average, this difference was not significant indicating that custodial rule infringement (i.e., adjudications, behavioural warnings, substance misuse) may be relevant to preventing other adverse outcomes. Nonetheless, it was tentatively possible to isolate patterns of behaviour in the qualitative data, specific to each failure group, providing direction for further research. For instance, three of the four absconders were serving indeterminate sentences, three had previously been recalled to custody during the current sentence, and three reported relational problems in the prison (e.g., bullying). These patterns mirror both the loss of hope observed in those serving IPP sentences – particularly those recalled to custody (Beedon, 2020; Harris et al., 2020; Merola, 2015) and the ‘push’ factors reported in custodial absconder samples (e.g., Picksley, 2016). Serving an indeterminate sentence was also a consistent feature amongst those who failed on temporary release and re-offended whilst in custody perhaps also reflecting dangerousness and antisocial persistence amongst IPP prisoners. It is also noteworthy that those who breached the terms of their temporary release had done so by engaging in offence-related behaviour, attracting behavioural warnings in the 6 months prior to the breach. Pearson and McDougall (2017) discuss the important contribution to risk evaluations that can be made by aggregating ‘lower level’ prison behaviours such as insults, threats, and bullying.

Given the limitations of the data and the relative frequency with which some of these variables are observed in the population (e.g., IPP sentences), it would be neither practicable nor cost-effective to target wraparound support packages, such as EBM, at all prisoners with the characteristics observed in the absconding, TRF, and custodial re-offending groups. Nonetheless, the data highlight the value of monitoring and attending to behaviours displayed by prisoners in open conditions, particularly those serving IPP sentences and those recalled to custody. Likewise, adverse outcomes such as absconding and re-offending whilst resident in custody in England and Wales, are rare events; hence, ensuring temporary release is purposeful and based upon a current assessment of risk, has likely impacted rates of failure (cf. HM Inspectorate of Prisons, 2014; Simpson et al., 2015). Consequently, there is a degree of futility in trying to predict serious recidivistic events in open conditions based on individual-level factors only, without producing excessive false positive errors and either recalling, or providing costly support services to, a sizeable proportion of individuals who would not otherwise have failed in open conditions. Beyond effective risk management, rates of abscond and re-offending in open prisons can likely be lowered by addressing the ‘push-pull’ factors previously described (e.g., Picksley, 2016); good relational security (good-staff prisoner relationships) (Mezey et al., 2015) and developing rehabilitative cultures (Auty and Liebling, 2020).

Practical implications

The present findings could be practically applied in a number of ways. First, the criteria used to predict security recall could be used to distinguish between those at greater risk of failure in open conditions compare to those at lower risk. Given the model was relatively more effective at identifying non-recalls (93.8%) than recalls (62.7%), the criteria could be most usefully applied to identify those who are unlikely to require additional support in open conditions, with any resource efficiencies being diverted into interventions for those exhibiting indicators of risk-related behavioural deterioration. Indeed, adverse outcomes typically occurred after 357 days on average. As such, the decision to trigger wrapround support packages to those at greater risk of recall could reasonably be deferred until there are behavioural indications related to one of the stronger proximal predictors of recall (i.e., substance misuse; behavioural warnings, particularly adjudications). This study also provided tentative evidence that behavioural deterioration in particular groups (e.g., indeterminate sentence prisoners, recalled prisoners) is potentially indicative of raised risk of more serious adverse outcomes (i.e., absconding, reoffending), with wraparound support packages providing the scaffolding needed to aid behavioural stabilisation. However, this was based on a qualitative analysis of a small sample of serious recidivistic failures and further research is required to establish the veracity of these conclusions.

Arguably this paper lifts the lid on more systemic problems in open prisons in England and Wales, the learning from which is applicable to open prisons holding both male and female prisoners internationally. In our sample, almost one-third of prisoners transferred to open conditions ‘failed’ and of those prisoners recalled to closed conditions, 90% did not re-offend in the first year of release. There was overreliance on a costly system of dealing with risk-related behaviour perhaps reflecting a decision-making bias which has been observed amongst related staff groups (Eno-Louden and Skeem, 2013). Further research is required and where evidence exists, training to help staff correctly identify behaviours indicative of an increase in risk is indicated (cf. Clark et al., 1993; McDougall et al., 2013).

Yet, the high number of recalls likely reflects the serious political implications that adverse events can have on the legitimacy of an institution (cf. Dawar and Davis, 2014; HM Inspectorate of Prisons, 2014). Given this cohort of prisoners will principally be released to the community, and graduated release is associated with reductions in recidivism (Cheliotis, 2008; Hillier and Mews, 2018; Mastrobuoni and Terlizzese, 2022; Ministry of Justice, 2015, 2019), the scientific community has a key role in challenging existing narratives about open prison failure, and educating both government and the wider public, on the relative benefits of open prisons. This includes acknowledging that risk exposure is necessary, to a degree, although open prisons can garner public safety assurance by reducing excessive risk exposure. Arguably, this is best mitigated through multi-level systemic changes relevant to all open prisons, thereby reducing the burden on, and bias inherent within, relying on open prison staff to ‘catch’ the residual risk. Indeed, a review of the recategorisation process in England and Wales is warranted. Currently, open prisons in England and Wales have little control over who is recategorised and, where such recategorisations are imposed and inappropriate, this may translate to risk aversion. Open prisons should have decision-making powers in recategorisation assessments whilst other bodies involved (e.g., closed prisons, Parole Boards) should be made accountable for understanding the managed risks in open conditions and justifying their decision-making. Integrating open prison sites into closed prisons is another means of improving selection and preparation of prisoners. Alternatively, building secure accommodation on open prisons sites might also afford prison managers time to adequately investigate aberrations before initiating recalls. Developing rehabilitative prison cultures (Auty and Liebling, 2020) and investing in good relational security (good-staff prisoner relationships) (Mezey et al., 2015) also show potential as risk assurance measures.

Limitations

There are methodological limitations to this study. First, the sample was extracted from two open prisons in England and Wales during a narrow time period. Whilst these open prisons were selected at least partially on the basis of their representativeness of the wider open estate in England and Wales, there was evidence of both over- and under-representation of some groups (see study sample and design). Evidently the applicability of the identified risk factors needs to be demonstrated over both time and place, including in other European jurisdictions, to aid understanding of failure in open conditions and the decision-making underpinning recall to closed conditions. Second, the logistic regression model was primarily populated with individual-level static/historical risk factors. Whilst we included risk factors containing a dynamic component as per the emerging literature on push/pull factors for absconding (Chant, 2015; Papworth, 2015; Picksley, 2016; Roberts, 2016), their coding was likely insufficiently nuanced to understand the interaction between the person and the situation. Indeed, Mischel and Shoda (1995) propose that behaviour is person-situation specific such that situational cues trigger mental representations of beliefs, memories, images, etc., and subsequently activate associated behavioural strategies. The tolerance level for absconding, for instance, will likely differ between individuals faced with the same situational triggers, if indeed the behavioural repertoire exists at all for some of those individuals. Given the significance of current behaviour to the identification of failure outcomes evidenced in this paper, there is imperative in ensuring that our risk and information management tools adequately capture these behavioural manifestations of risk and interface with one another. Nomothetic risk assessment, particularly in open prison environments, is undermined by a range of practical (e.g., low base rates, sample attrition) as well as wider conceptual issues (Heffernan et al., 2019), and we argue that individual-idiosyncratic assessment has greater potential to respond to changing risk (cf. McDougall et al., 2013).

Conclusions

The current study contributes to understanding, predicting, and mitigating the risk of serious failure outcomes in open prisons. The evidence indicates that open prisons and in particular, temporary release schemes positively impact on recidivism (e.g., Cheliotis, 2008) and arguably, in the pursuit of successfully resettling high-risk prisoners back in their communities, should be an essential component of prison systems globally. Failures inevitably undermine the legitimacy of such institutions but we have shown that many failures occur following observable behavioural indicators of risk-elevation and thus behavioural monitoring interventions, such as EBM or equivalent initiatives, are likely to have utility and some precision in detecting whom would benefit from a package of support as well as informing post-release licence conditions (McDougall et al., 2013). Serious failure outcomes were few in the current study perhaps reflecting that the adoption of clear risk assessment protocols, particularly around temporary release, is sufficient to mitigate risk (HM Inspectorate of Prisons, 2014; Simpson et al., 2015). Indeed, predicting serious adverse outcomes (e.g., abscond, re-offending) based upon static factors was futile given the limited variability within the population (e.g., all high-risk offenders) and low base rates for such events, prompting our call for open prisons to focus upon relational security and rehabilitative cultures to identify and mitigate the potential ‘push-pull’ underpinning these events. Managing the risk cynically through transferring prisoners back to closed conditions is potentially iatrogenic (cf. Alexander, 2006) and at the very least, in our study, was a costly intervention that correlated poorly with community recidivism outcomes.

Supplemental Material

sj-docx-1-euc-10.1177_14773708231183570 - Supplemental material for Monitoring prisoners preparing for release: Who ‘fails’ in open prison conditions?

Supplemental material, sj-docx-1-euc-10.1177_14773708231183570 for Monitoring prisoners preparing for release: Who ‘fails’ in open prison conditions? by Gary Goodley and Dominic Pearson in European Journal of Criminology

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Gary Goodley

Supplemental material

Supplemental material for this article is available online.

Notes

References

Alexander

(2006) Patients’ feelings about ward nursing regimes and involvement in rule construction. Journal of Psychiatric and Mental Health Nursing 13(5): 543–553.

Andrews

Bonta

Wormith

(2006) The recent past and near future of risk and/or need assessment. Crime and Delinquency 52(1): 7–27.

Atkinson

Mann

(2012) Prison officers as observers of offence paralleling behaviours: An untapped resource? The Journal of Forensic Psychiatry and Psychology 23(2): 139–155.

Auty

Liebling

(2020) Exploring the relationship between prison social climate and reoffending. Justice Quarterly 37(2): 358–381.

Barnett

Howard

(2018) What doesn’t work to reduce reoffending? European Psychologist 23(2): 111–129.

Barnett

Wakeling

Howard

(2010) An examination of the predictive validity of the risk matrix 2000 in England and Wales. Sexual Abuse: A Journal of Research and Treatment 22(4): 443–470.

Beedon

(2020) Hope deferred, humanity diminished: An ethnographic enquiry into the improvement of pastoral care offered to those serving an indeterminate sentence of Imprisonment for Public Protection (IPP) [Unpublished doctorate thesis]. University of Birmingham.

Beijersbergen

Dirkzwager

Nieuwbeerta

(2016) Reoffending after release: Does procedural justice during imprisonment matter? Criminal Justice and Behaviour 43(1): 63–82.

Berman-Roberts

(2015) A qualitative investigation into the context, motivation and potential risk factors of prisoners who fail to return when Released on Temporary Licence (RoTL) in open conditions. Unpublished manuscript.

10.

Bonta

Andrews

(2017) The psychology of criminal conduct, 6th ed. New York: Routledge.

11.

Burt

Olver

Wong

SCP

(2016) Investigating characteristics of the nonrecidivating psychopathic offender. Criminal Justice and Behaviour 43(12): 1741–1760.

12.

Buss

Craik

(1983) The act frequency approach to personality. Psychological Review 90(2): 105–126.

13.

Chant

(2015) It’s time to go: exploring prison life of those who abscond. An interpretative phenomenological analysis. Unpublished manuscript.

14.

Cheliotis

(2008) Reconsidering the effectiveness of temporary release: A systematic review. Aggression and Violent Behaviour 13(3): 153–168.

15.

Clark

Fisher

McDougall

(1993) A new methodology for assessing the level of risk in incarcerated offenders. The British Journal of Criminology 33(3): 436–448.

16.

Cohen

(1988) Statistical power analysis for the behavioural sciences, 2nd ed. New Jersey: Erlbaum.

17.

Cullen

Jewell

Tully

, et al. (2015) A prospective cohort study of absconsion incidents in forensic psychiatric settings: Can we identify those at high-risk? PloS one 10(9): 1–16.

18.

Dawar

Davis

(2014) Scandal of our ‘holiday camp’ open prisons. Daily Express, 21 May. https://www.express.co.uk/news/uk/497892/Scandal-of-our-holiday-camp-open-prisons.

19.

Eisenberg

van Horn

Dekker

, et al. (2019) Static and dynamic predictors of general and violent criminal offence recidivism in the forensic outpatient population: A meta-analysis. Criminal Justice and Behaviour 46(5): 732–750.

20.

Emirali

O'Rourke

Friendship

(2020) Desistance from absconding: “You’ve got to have hope”. The Journal of Forensic Practice 22(4): 223–234.

21.

Eno Louden

Skeem

(2013) How do probation officers assess and manage recidivism and violence risk for probationers with mental disorder? An experimental investigation. Law and Human Behaviour 37(1): 22–34.

22.

Field

(2005) Discovering statistics using SPSS, 2nd edn. London: Sage Publications.

23.

Flowers

(2014) Understanding the experiences of prisoners who have been returned from open conditions to a closed prison: an interpretative phenomenological analysis. Unpublished manuscript.

24.

Gendreau

Little

Goggin

(1996) A meta-analysis of the predictors of adult offender recidivism: What works!. Criminology; An Interdisciplinary Journal 34(4): 575–608.

25.

Goodley

Pearson

Morris

(2022) Predictors of recidivism following release from custody: A meta-analysis. Psychology, Crime and Law 28(7): 703–729.

26.

Green

Carroll

Brett

(2010) Structured risk assessment in community forensic mental health practice. Australasian Psychiatry 18(6): 538–541.

27.

Harrell

Jr Lee

Mark

(1996) Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Statistics in Medicine 15(4): 361–387.

28.

Harris

Edgar

Webster

(2020) ‘I’m always walking on eggshells, and there’s no chance of me ever being free’: The mental health implications of imprisonment for public protection in the community and post-recall. Criminal Behaviour and Mental Health 30(6): 331–340.

29.

Heffernan

Wegerhoff

Ward

(2019) Dynamic risk factors: Conceptualization, measurement, and evidence. Aggression and Violent Behavior 48: 6–16.

30.

Hillier

Mews

(2018) The re-offending impact of increased release on temporary licence. London: Ministry of Justice.

31.

HM Inspectorate of Prisons (2014) Release on Temporary Licence (RoTL) failures. London: HMIP.

32.

Lesneski

Hsu

(2018) What we don’t know may hurt us: An examination of systematic bias in offender risk assessments. Deviant Behavior 39(12): 1566–1577.

33.

Home Office (2006) Offender assessment system OASys user manual. London: Home Office.

34.

Hosmer

Jr Lemeshow

Sturdivant

(2013) Applied logistic regression. New Jersey: John Wiley and Sons.

35.

Howard

Dixon

(2012) The construction and validation of the OASys violence predictor: Advancing violence risk assessment in the English and Welsh correctional services. Criminal Justice and Behaviour 39(3): 287–307.

36.

Howard

Dixon

(2013) Identifying change in the likelihood of violent recidivism: Causal dynamic risk factors in the OASys violence predictor. Law and Human Behaviour 37: 163–174.

37.

Justice Data Lab (2022, July 18) Public protection. London: Ministry of Justice. https://data.justice.gov.uk/prisons/public-protection

38.

Katsiyannis

Whitford

Zhang

, et al. (2018) Adult recidivism in the United States: A meta-analysis 1994–2015. Journal of Child and Family Studies 27(3): 686–696.

39.

LeBel

Burnett

Maruna

, et al. (2008) The chicken and egg of subjective and social factors in desistance from crime. European Journal of Criminology 5(2): 131–159.

40.

Lipsey

Wilson

(1998) Effective intervention for serious juvenile offenders: A synthesis of research. In: Loeber

Farrington

(eds) Serious and violent juvenile offenders: Risk factors and successful interventions. Thousand Oaks, CA: Sage, 313–341.

41.

Martin

McGeown

Whitehouse

, et al. (2018) Who’s going to leave? An examination of absconding events by forensic inpatients in a psychiatric hospital. The Journal of Forensic Psychiatry and Psychology 29(5): 810–823.

42.

Mastrobuoni

Terlizzese

(2022) Leave the door open? Prison conditions and recidivism. American Economic Journal: Applied Economics 14(4): 200–233.

43.

McDougall

Pearson

Willoughby

, et al. (2013) Evaluation of the ADViSOR project: Cross-situational behaviour monitoring of high-risk offenders in prison and the community. Legal and Criminological Psychology 18(2): 205–228.

44.

McSweeney

Turnbull

Bowling

(2011) Fight or flight? Absconders, fugitives, and European law enforcement responses. London: Institute for Criminal Policy Research.

45.

Merola

(2015) Young offenders’ experiences of an indeterminate sentence. The Journal of Forensic Practice 17(1): 55–68.

46.

Mews

(2014) Analysis on the impact of absconding incident history as a predictor of future absconding incidents. London: Ministry of Justice.

47.

Mezey

Durkin

Dodge

, et al. (2015) Never ever? Characteristics, outcomes, and motivations of patients who abscond or escape: A 5-year review of escapes and absconds from two medium and low secure forensic units. Criminal Behaviour and Mental Health 25(5): 440–450.

48.

Ministry of Justice (2009) OGRS 3: The revised offender group reconviction scale. London: Ministry of Justice.

49.

Ministry of Justice (2015) Justice data lab re-offending analysis: HMP Kirklevington Grange. London: Ministry of Justice. https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/627518/hmp-kirklevington-grange-report.pdf

50.

Ministry of Justice (2019) Prison-youth-accomm-probation-data-tool-jan19-mar19. London: Ministry of Justice. https://view.officeapps.live.com/op/view.aspx?src=https%3A%2F%2Fassets.publishing.service.gov.uk%2Fgovernment%2Fuploads%2Fsystem%2Fuploads%2Fattachment_data%2Ffile%2F989739%2Fprison-youth-accomm-probation-data-tool-jan19-mar19.xlsxandwdOrigin=BROWSELINK

51.

Ministry of Justice (2021) Offender management statistics quarterly. London: Ministry of Justice. https://www.gov.uk/government/statistics/offender-management-statistics-quarterly

52.

Ministry of Justice (2013a) Transforming rehabilitation: A summary of evidence on reducing reoffending. London: Ministry of Justice.

53.

Ministry of Justice (2013b) Analysis of the impact of employment on reoffending following release from custody, using Propensity Score Matching. London: Ministry of Justice.

54.

Mischel

Shoda

(1995) A cognitive-affective system theory of personality: Reconceptualizing situations, dispositions, dynamics, and invariance in personality structure. Psychological Review 102(2): 246–268.

55.

National Offender Management Service (2015a) Release on temporary licence: (PSI 13/2015; PI 10/2015). London: National Offender Management Service.

56.

National Offender Management Service (2015b) Enhanced behaviour monitoring: (PSI 24/2015; PI 16/2015). London: National Offender Management Service.

57.

Office for Health and Improvement Disparities (2022) Alcohol and drug treatment in secure settings 2020 to 2021: Report. London: Office for Health and Improvement Disparities. https://www.gov.uk/government/statistics/substance-misuse-treatment-in-secure-settings-2020-to-2021/alcohol-and-drug-treatment-in-secure-settings-2020-to-2021-report#mental-health

58.

Ostermann

Hyatt

DeWitt

(2020) The influence of technical violation revocations on parole efficacy: Employing competing risks survival analysis to address methodological challenges. Journal of Criminal Justice 43(3): 323–341.

59.

Papworth

(2015) An interpretative phenomenological analysis of prisoners’ experiences of absconding from an open prison. Unpublished manuscript.

60.

Pearson

McDougall

(2017) Serious offenders: Using evidence to predict and manage the risk. Behavioural Sciences & the Law 35(3): 239–252.

61.

Picksley

(2016) A qualitative investigation into the experiences of prisoners in an open prison who abscond or fail to return when released on temporary licence (ROTL). Unpublished manuscript.

62.

Roberts

Coid

(2010) Personality disorder and offending behaviour: Findings from the national survey of male prisoners in England and Wales. The Journal of Forensic Psychiatry and Psychology 21(2): 221–237.

63.

Salgado

(2018) Transforming the area under the normal curve (AUC) into Cohen’s d, Pearson’s r pb, odds-ratio, and natural log odds-ratio: Two conversion tables. European Journal of Psychology Applied to Legal Context 10(1): 35–47.

64.

Schaefer

(2016) On the reinforcing nature of crime and punishment: An exploration of inmates’ self-reported likelihood of reoffending. Journal of Offender Rehabilitation 55(3): 168–194.

65.

Simpson

Penney

Fernane

, et al. (2015) The impact of structured decision making on absconding by forensic psychiatric patients: Results from an AB design study. BMC Psychiatry 15(1): 103–115.

66.

Skett

Goode

Barton

(2017) A joint NHS and NOMS offender personality disorder pathway strategy: A perspective from 5 years of operation. Criminal Behaviour and Mental Health 27: 214–221.

67.

Tyler

Miles

Karadag

, et al. (2019) An updated picture of the mental health needs of male and female prisoners in the UK: Prevalence, comorbidity, and gender differences. Social Psychiatry and Psychiatric Epidemiology 54(9): 1143–1152.

68.

Weber

(1990) Basic content analysis. Beverly Hills, CA: Sage.

69.

Wilkie

Penney

Fernane

, et al. (2014) Characteristics and motivations of absconders from forensic mental health services: A case-control study. BMC Psychiatry 14(1): 91–104.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB