Elsevier

Journal of Criminal Justice

The predictive performance of criminal take chances assessment tools used at sentencing: Systematic review of validation studies

Abstract

Although risk cess tools have been widely used to inform sentencing decisions, there is incertitude about the extent and quality of evidence of their predictive performance when validated in new samples. Following PRISMA guidelines, we conducted a systematic review of validation studies of 11 commonly used adventure assessment tools for sentencing. We identified 36 studies with 597,665 participants, among which were 27 independent validation studies with 177,711 individuals. Overall, the predictive operation of the included run a risk assessment tools was mixed, and ranged from poor to moderate. Tool operation was typically overestimated in studies with smaller sample sizes or studies in which tool developers were co-authors. Most studies only reported area under the curve (AUC), which ranged from 0.57 to 0.75 in independent studies with more 500 participants. The bulk did not report cardinal performance measures, such as scale and rates of fake positives and negatives. In addition, about validation studies had a high risk of bias, partly due to inappropriate analytical approach used. We conclude that the research priority is for future investigations to address the key methodological shortcomings identified in this review, and policy makers should enable this research. More sufficiently powered contained validation studies are necessary.

Introduction

Chance cess tools are widely used to inform sentencing decisions for individuals convicted of crimes in many high-income countries (van Ginneken, 2019). These tools tin influence decisions on whether someone receives a prison or community-based judgement, sentence length, and associated restrictions, such equally electronic tagging (Monahan & Skeem, 2016). The choice between custodial and not-custodial outcomes is significant as prison is associated with many negative ramifications for accommodation, relationships, and employment (Harding, Wyse, Dobson, & Morenoff, 2014; Keene, Smoyer, & Blankenship, 2018; Western, Braga, Davis, & Sirois, 2015). Sentence length is determined past legislation and provides a range of judgement length options, which is typically based on certain individual factors, including the previous criminal history and take chances of hereafter serious offending. These tools can also assist with treatment decisions, especially for people with mental health issues, and atomic number 82 to referrals to diversion services. Risk cess tools need to be high quality because of the potential consequences for individuals in the criminal justice arrangement and public health and safe. For people in the criminal justice system, the utilize of inaccurate take chances assessment tools can lead to longer periods of time in custody. For public wellness and safety, it can lead to wasteful and poor allocation of resources if people remain in custody who could be released based on risk levels, and the consequences of repeat offending in people released without appropriate supervision and treatment. Furthermore, the wider ethical implications on public trust in criminal justice are relevant in that such tools should be transparently designed and reported, and not lead to amplification of systemic biases and also potentially mitigate against them.

Previous work reviewing these tools has combined different samples and settings, including forensic psychiatric hospitals, intra-institutional outcomes, and non-offending samples (Campbell, French, & Gendreau, 2009; Fazel, Singh, Doll, & Grann, 2012; Ramesh, Igoumenou, Vazquez Montes, & Fazel, 2018; Singh, Serper, Reinharth, & Fazel, 2011). In addition, it has combined development (likewise known every bit discovery) samples with validation ones, which does not reflect real-world performance every bit the development samples tend to overestimate predictive functioning (Pavlou et al., 2015). Furthermore, validation studies may accept been conducted by the tool developers, which tin can atomic number 82 to authorship bias (Singh, Grann, & Fazel, 2013), and separately examining such investigations conducted by independent groups needs consideration. Finally, previous reviews have used performance measures such as correlation coefficients and effect sizes that are not informative on their ain and not recommended in standard guidelines for systematic reviewing of prediction models (Moons et al., 2014). Rather, measures of discrimination (including truthful and faux positives and negatives), and calibration (how estimated and predicted risk scores compare) are necessary for any tool to be evaluated (Collins, Reitsma, Altman, & Moons, 2015).

To address these limitations, we have conducted a systematic review of validation studies of risk cess tools that are used to inform controlling in the criminal justice organization. Our primary outcomes were measures of predictive performance from independent validations, where tool developers are non co-authors. In so doing, we aim to provide a focused overview that will inform criminal justice and linked mental health services.

Department snippets

Protocol and registration

This systematic review was pre-registered under the Open up Science Framework (OSF). The protocol can be retrieved from: https://osf.io/59szj

Literature search

Equally a first step, we identified risk cess tools commonly used at the sentencing phase for criminal offences to estimate take chances of recidivism by searching PsycINFO, Medline, and EMBASE with the following keywords: (psychiatry OR forensic OR psychology) AND (sentencing OR sanctioning OR sanction OR violence) AND (backsliding OR re-criminal offence) AND risk assessment

Results

Overall, we identified 36 studies with 597,665 participants (PRISMA flowchart in Supplementary Fig. one), which were based in seven countries (Table ii). Of these, 27 were independent validation studies with 177,711 participants, reporting on 7 tools: COMPAS, HCR-20, LS/CMI, LSI-R, PCL-R, PCRA, and Static-99. No eligible validation studies were identified for IRAS and NVRA. The ORAS and OASys were but validated by the developers of these tools. The most common operation statistic reported was the

Discussion

In this systematic review of 36 studies of risk assessment tools used to inform decision-making in the criminal justice system that followed upwardly 597,665 participants for repeat offending, we identified functioning measures for nine tools. In the sixteen contained investigations with more than than 500 participants each, the nigh common reported outcome statistic was the AUC, which ranged from 0.57 to 0.75.

Overall, the extent and quality of evidence in back up of these tools is typically poor to

Conclusion

In this systematic review of external validation studies of 11 mutual run a risk assessment tools, most investigations solely reported the AUC equally an indication of model performance, just did not present other key measures including rates of false positives and negatives, and scale. Equally such, based on the current published evidence, the highest priority is for researchers to work towards addressing the primal methodological limitations identified in previous work. Jurisdictions that are because

Funding

SF is funded by a Wellcome Trust Senior Enquiry Fellowship (Grant no. 202836/Z/16/Z).

References (90)

  • et al.

    ROC curves for clinical prediction models role ane. ROC plots showed no added value above the AUC when evaluating the performance of clinical prediction models

    Journal of Clinical Epidemiology

    (2020)

  • E.W. Steyerberg

    Validation in prediction enquiry: The waste by information splitting

    Journal of Clinical Epidemiology

    (2018)

  • Thou. Rufibach

    Use of Brier score to appraise binary predictions

    Journal of Clinical Epidemiology

    (2010)

  • T. Ramesh et al.

    Use of risk assessment instruments to predict violence in forensic psychiatric hospitals: A systematic review and meta-assay

    European Psychiatry

    (2018)

  • K.O. Hajian-Tilaki et al.

    Comparison of three methods for estimating the standard error of the area under the curve in ROC analysis of quantitative data

    Academic Radiology

    (2002)

  • S. Fazel et al.

    Prediction of fierce reoffending on release from prison: Derivation and external validation of a scalable tool

    The Lancet Psychiatry

    (2016)

  • T. Douglas et al.

    Risk cess tools in criminal justice and forensic psychiatry: The need for ameliorate data

    European Psychiatry

    (2017)

  • K.-P. Dahle

    Strengths and limitations of actuarial prediction of criminal reoffence in a German prison sample: a comparative study of LSI-R, HCR-20 and PCL-R

    International Journal of Law and Psychiatry

    (2006)

  • Southward. Ægisdóttir et al.

    The meta-analysis of clinical judgment projection: Fifty-six years of accumulated research on clinical versus statistical prediction

    The Counseling Psychologist

    (2006)

  • A. Allan et al.

    Prediction of the take a chance of male sexual reoffending in Australia

    Australian Psychologist

    (2006)

  • D.A. Andrews et al.

    The Level of Service Inventory - Revised

    (1995)

  • D.A. Andrews et al.

    LS/CMI: The level of service/case management inventory

    (2004)

  • R. Barnoski et al.

    Washington'south offender accountability act: An analysis of the Department of Corrections' risk assessment

    (2003)

  • M.T. Boccaccini et al.

    Field validity of Static-99/R scores in a statewide sample of 34,687 convicted sexual offenders

    Psychological Cess

    (2017)

  • M. Borenstein et al.

    Introduction to meta-analysis

    (2011)

  • T. Brennan et al.

    Evaluating the predictive validity of the COMPAS risk and needs assessment system

    Criminal Justice and Behavior

    (2009)

  • Yard.A. Campbell et al.

    The prediction of violence in developed offenders: A meta-analytic comparison of instruments and methods of assessment

    Criminal Justice and Beliefs

    (2009)

  • J.C. Chambers et al.

    Outcome measures used in forensic mental health research: A structured review

    Criminal Behaviour and Mental Wellness

    (2009)

  • T.H. Cohen

    Predicting sexual practice offender backsliding: Using the Federal Post Conviction Risk Assessment instrument to assess the likelihood of recidivism among federal sexual practice offenders

    Periodical of Empirical Legal Studies

    (2018)

  • T.H. Cohen et al.

    How dangerous are they? An analysis of sex activity offenders under federal mail service-conviction supervision

    Federal Probation

    (2016)

  • Thousand.S. Collins et al.

    Transparent reporting of a multivariable prediction model for private prognosis or diagnosis (TRIPOD)

    Apportionment

    (2015)

  • Northward.R. Melt

    Use and misuse of the receiver operating characteristic bend in gamble prediction

    Circulation

    (2007)

  • M. DeLisi et al.

    Federal criminal careers: An empirical examination of the Mail service-Conviction Take a chance Assessment (PCRA)

    American Periodical of Criminal Justice

    (2018)

  • S.50. Desmarais et al.

    Performance of recidivism risk assessment instruments in U.S. correctional settings

    Psychological Services

    (2016)

  • 1000.Due south. Douglas et al.

    HCR-20V3: Assessing risk of violence—User guide

    (2013)

  • H.L. Dyck et al.

    Real-world employ of the risk–demand–responsivity model and the level of service/example direction inventory with community-supervised offenders

    Constabulary and Human Behavior

    (2018)

  • M. Egger et al.

    Bias in meta-analysis detected by a elementary, graphical exam

    BMJ

    (1997)

  • South. Etzler et al.

    Dynamic risk cess of sexual offenders: validity and dimensional construction of the Stable-2007

    Assessment

    (2020)

  • D. Farabee et al.

    COMPAS concluding written report

    (2010)

  • T.L. Fass et al.

    The LSI-R and the COMPAS: Validation data on two risk-needs tools

    Criminal Justice and Behavior

    (2008)

  • S. Fazel et al.

    Use of risk cess instruments to predict violence and antisocial behaviour in 73 samples involving 24 827 people: Systematic review and meta-analysis

    BMJ

    (2012)

  • S. Fazel et al.

    Selecting a take a chance assessment tool to utilise in practice: A 10-indicate guide

    Evidence-Based Mental Health

    (2018)

  • B.L. Garrett et al.

    Judicial reliance on risk assessment in sentencing drug and property offenders: A test of the treatment resources hypothesis

    Criminal Justice and Behavior

    (2019)

  • Due east.F.J.C. van Ginneken

    The utilize of risk assessment in sentencing

  • H. Gordon et al.

    An evaluation of the level of service/example management inventory in an Australian community corrections surroundings

    Psychiatry, Psychology and Law

    (2015)

  • B. Hanczar et al.

    Small-sample precision of ROC-related estimates

    Bioinformatics

    (2010)

  • J.A. Hanley et al.

    The meaning and use of the expanse under a receiver operating characteristic (ROC) curve

    Radiology

    (1982)

  • R.K. Hanson

    Assessing the scale of actuarial risk scales: A primer on the E/O alphabetize

    Criminal Justice and Behavior

    (2017)

  • R.K. Hanson et al.

    The field validity of Static-99/R sexual practice offender risk assessment tool in California

    Journal of Threat Cess and Management

    (2014)

  • D.J. Harding et al.

    Making ends meet after prison: Special symposium on qualitative and mixed-methods for policy analysis

    Journal of Policy Assay and Management

    (2014)

  • R.D. Hare

    Hare psychopathy checklist-revised: PCL-R

    (2003)

  • P.B. Harris et al.

    Field measures of psychopathy and sexual deviance every bit predictors of recidivism among sexual offenders

    Psychological Assessment

    (2017)

  • L.M. Helmus et al.

    Improving the predictive accuracy of Static-99 and Static-2002 with older sex offenders: Revised historic period weights

    Sexual Corruption

    (2012)

  • P.D. Howard

    The offender cess system: An evaluation of the 2d airplane pilot

    (2006)

  • P.D. Howard

    The prediction of reoffending by age, gender and ethnicity

  • Cited by (0)

    View full text