Risk Assessment in Criminal Justice

Risk assessment in criminal justice examines how validated instruments predict reoffending and guide sentencing, supervision, and reentry. The systematic assessment of an individual offender’s probability of future criminal behavior has become the most consequential application of social science research in American criminal justice. Risk assessment instruments inform pretrial detention decisions affecting millions of defendants annually, shape the intensity and conditions of probation and parole supervision for millions more, determine eligibility for correctional programming, guide parole release decisions, and increasingly influence sentencing itself. This article, part of the Clinical Criminology section of the broader Criminology resource, examines the development, methodology, and application of risk assessment in criminal justice, evaluates the evidence on instrument validity and limitations, and addresses the policy and ethical controversies that surround the use of predictive tools in decisions about individual liberty.

Introduction

The use of risk assessment in criminal justice reflects a fundamental shift in the conceptual foundations of offender management—from a classification paradigm based on offense severity and criminal history to one based on the empirically identified predictors of future behavior. This shift has been driven by three decades of meta-analytic research demonstrating that the factors most strongly associated with recidivism can be reliably measured, that structured instruments incorporating these factors outperform unstructured professional judgment in predicting outcomes, and that matching supervision intensity and programming to assessed risk and need levels produces larger reductions in recidivism than undifferentiated approaches (Andrews & Bonta, 2010).

The adoption of structured risk assessment has accelerated dramatically since 2000, driven by the convergence of the evidence base with fiscal pressures on state and local governments seeking alternatives to expensive incarceration. The National Institute of Corrections has promoted the adoption of validated risk assessment as a core component of evidence-based corrections, and professional organizations including the American Probation and Parole Association and the Association of State Correctional Administrators have endorsed risk assessment as a practice standard. Today, virtually every state correctional system and most large probation and parole agencies use some form of validated risk assessment instrument, making risk assessment one of the most widely implemented evidence-based practices in American criminal justice (Desmarais & Singh, 2013).

The Development of Risk Assessment

Generational Progression

The development of risk assessment instruments in criminal justice has followed a generational progression that parallels the broader history of clinical prediction in psychology and medicine. First-generation assessment relied entirely on unstructured clinical judgment—the practitioner’s professional impression formed through interview, file review, and personal experience. Research consistently demonstrated that unstructured judgment performed substantially worse than statistical approaches and was subject to systematic biases including overconfidence, inconsistency across practitioners, anchoring effects, and sensitivity to vivid but non-predictive information. The foundational critique by Paul Meehl (1954) documenting the superiority of actuarial over clinical prediction across multiple domains established the intellectual foundation for structured assessment, though decades passed before criminal justice agencies widely adopted the implications of this research (Grove et al., 2000).

Second-generation instruments addressed the limitations of unstructured judgment by identifying through statistical analysis the factors most strongly associated with recidivism in large samples and combining them into numerical scores. These instruments—exemplified by the Salient Factor Score used by the U.S. Parole Commission and early versions of the Statistical Information on Recidivism (SIR) scale in Canada—relied exclusively on static factors: prior criminal history, age at first arrest, offense type, number of prior incarcerations, and similar variables that could be reliably coded from official records. Static instruments provided more accurate and more consistent predictions than clinical judgment, but their exclusive reliance on historical factors meant they could not identify targets for intervention or measure change resulting from programming or maturation (Andrews, Bonta, & Wormith, 2006).

Third-generation instruments incorporated dynamic risk factors—criminogenic needs that are both associated with recidivism and amenable to change through intervention. The Level of Service Inventory-Revised (LSI-R) and its successor, the Level of Service/Case Management Inventory (LS/CMI), assess domains including criminal history alongside education and employment, family and marital relationships, leisure and recreation, companions, substance use, procriminal attitudes, and antisocial pattern. By incorporating dynamic factors, third-generation instruments serve dual purposes: they classify offenders by risk level for supervision intensity decisions and identify the specific criminogenic needs that should be targeted in case management and programming. This dual functionality transformed risk assessment from a purely classificatory exercise into a clinical tool that drives individualized intervention planning (Andrews & Bonta, 2010).

Fourth-Generation and Integrated Systems

Fourth-generation instruments extend the third-generation model by incorporating responsivity factors—individual characteristics that affect how an offender responds to specific intervention approaches—and by integrating risk classification, need identification, case planning, and outcome monitoring into integrated management systems. The LS/CMI, the Ohio Risk Assessment System (ORAS), and the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) all incorporate elements of fourth-generation design, though the extent of integration varies across instruments. The theoretical foundation for fourth-generation assessment is the Risk-Need-Responsivity (RNR) model, which specifies that effective intervention requires matching intensity to risk level (risk principle), targeting criminogenic needs (need principle), and adapting intervention delivery to the learning style, motivation, and cognitive abilities of the individual offender (responsivity principle) (Andrews et al., 2006).

The implementation of fourth-generation systems requires organizational infrastructure that connects assessment findings to programming decisions, supervision strategies, and case management activities. In agencies that have fully implemented fourth-generation models, the risk assessment drives the entire supervision trajectory: the initial assessment determines supervision level and identifies priority criminogenic needs; the case plan specifies the programming and supervision strategies that will address those needs; reassessment at regular intervals measures change in dynamic risk factors and triggers modifications to the case plan; and outcome data feed back into the system to improve the accuracy of risk classification and the effectiveness of programming. This integration of assessment, planning, delivery, and evaluation represents the most sophisticated application of evidence-based practice in criminal justice, though achieving full implementation fidelity remains challenging for most agencies (Desmarais & Singh, 2013).

The proliferation of risk assessment instruments has created a complex landscape in which agencies must choose among dozens of validated tools, each with different factor structures, administration requirements, and validation populations. The selection of an appropriate instrument requires attention to the population for which it was developed and validated (general offenders, sex offenders, juveniles, domestic violence offenders), the decision point at which it will be applied (pretrial, sentencing, classification, parole), the resources available for administration and training, and the evidence of validity in the specific population and jurisdiction where it will be used. Instruments validated on one population may not perform equivalently in another, and cross-validation in the target population is essential before operational deployment (Singh, Grann, & Fazel, 2011).

The Central Eight Risk and Need Factors

Static and Dynamic Predictors

The empirical foundation of contemporary risk assessment rests on meta-analytic research identifying the factors most strongly and consistently associated with recidivism across studies, populations, and measurement approaches. Andrews and Bonta’s synthesis of this literature identified the “Central Eight” risk and need factors that account for the largest share of variance in recidivism prediction: criminal history, antisocial cognition, antisocial associates, antisocial personality pattern, family and marital circumstances, school and work functioning, substance abuse, and prosocial recreational activities (Andrews & Bonta, 2010).

Criminal history—the strongest single predictor of recidivism—is a static factor that reflects the accumulated behavioral record of prior offending. Early onset of criminal behavior, frequency of prior offenses, diversity of offense types, and prior violations of conditional release are all independently associated with elevated recidivism risk. The predictive power of criminal history derives from its function as a proxy for the underlying criminal propensity that drives offending: individuals with extensive criminal histories have demonstrated through their behavior that they possess the combination of antisocial attitudes, peer associations, personality characteristics, and environmental circumstances that produce persistent offending (Gendreau, Little, & Goggin, 1996).

The dynamic factors—antisocial cognition, antisocial associates, antisocial personality pattern, family difficulties, school and work problems, substance abuse, and lack of prosocial leisure—are both predictive of recidivism and modifiable through intervention. Antisocial cognition—attitudes, values, and beliefs that support criminal behavior—is among the strongest dynamic predictors and the most directly targeted by cognitive-behavioral interventions that have demonstrated the largest treatment effects in correctional programming evaluations. Antisocial associates—the presence of criminal peers and the absence of prosocial relationships—exerts its influence through social learning mechanisms that reinforce criminal behavior and neutralize prosocial influences. Together, the Central Eight provide both a predictive framework for risk classification and a prescriptive framework for intervention targeting—the dual functionality that makes them the conceptual backbone of evidence-based correctional practice (Andrews & Bonta, 2010).

Protective Factors and Strengths

The incorporation of protective factors and strengths into risk assessment represents a significant development that addresses criticisms of the deficit-focused orientation of traditional risk instruments. Protective factors—individual characteristics and social circumstances that reduce the probability of offending independent of risk factors—include strong prosocial bonds, stable employment, educational achievement, positive family relationships, engagement in prosocial activities, and the cognitive and emotional competencies that enable effective self-regulation. The Structured Assessment of Violence Risk in Youth (SAVRY) incorporates six protective factors alongside its risk items, and the Good Lives Model developed by Tony Ward provides a theoretical framework for integrating strengths-based assessment into risk management (Ward & Maruna, 2007).

The empirical relationship between protective factors and desistance from crime has been documented in longitudinal research demonstrating that marriage, stable employment, military service, and residential relocation are associated with reduced offending among previously active offenders—the “turning points” that Sampson and Laub (1993) identified as mechanisms of desistance. Instruments that assess protective factors alongside risk factors provide a more complete picture of the individual’s circumstances and prospects, enabling case plans that build on existing strengths as well as addressing deficits. The practical implication is that effective offender management requires not only the reduction of risk factors through treatment but also the cultivation of the prosocial capital—relationships, skills, opportunities, and identity transformations—that enable legitimate lifestyle achievement (Maruna, 2001).

The balance between risk-focused and strengths-focused assessment reflects broader theoretical debates within clinical criminology about the nature of desistance and the mechanisms through which intervention produces change. The RNR model emphasizes the reduction of criminogenic needs as the primary mechanism; the Good Lives Model emphasizes the development of prosocial goods—meaningful work, intimate relationships, community participation, creativity, spirituality—as the primary mechanism, with risk reduction occurring as a byproduct of positive lifestyle development. In practice, the most effective interventions likely integrate both approaches, targeting criminogenic needs while building the prosocial competencies and opportunities that give offenders viable alternatives to criminal behavior (Ward & Maruna, 2007).

Application Across Decision Points

Pretrial Risk Assessment

Pretrial risk assessment—the use of validated instruments to inform decisions about whether defendants should be detained or released pending trial—has become one of the most consequential and most controversial applications of risk assessment in American criminal justice. The pretrial detention decision affects approximately 10 million jail bookings annually in the United States, and the population of pretrial detainees—individuals who have been charged but not convicted—constitutes approximately two-thirds of the local jail population. Research consistently demonstrates that pretrial detention produces negative outcomes beyond its immediate liberty deprivation: detained defendants are more likely to plead guilty, receive longer sentences, lose employment, and experience family disruption than similarly situated defendants who are released, creating cascading consequences that extend far beyond the pretrial period (Lowenkamp, VanNostrand, & Holsinger, 2013).

The Public Safety Assessment (PSA), developed by the Laura and John Arnold Foundation (now Arnold Ventures), is the most widely adopted pretrial risk instrument, implemented in more than 40 jurisdictions including entire state court systems. The PSA uses nine factors—drawn from criminal history and current charge information available at booking—to generate scores for failure to appear, new criminal arrest, and new violent criminal arrest. The instrument was developed and validated on a national sample of more than 1.5 million cases and is designed to be administered quickly from readily available data without requiring a defendant interview. Its implementation has been associated with reductions in pretrial detention rates in several jurisdictions, though the magnitude and equity of these effects vary with local implementation decisions (Desmarais & Singh, 2013).

The pretrial risk assessment movement has generated significant opposition from both civil liberties advocates and bail industry stakeholders. Civil liberties organizations have raised concerns about racial disparities in instrument scores, the use of group-based statistical predictions to make individual liberty decisions, and the potential for risk assessment to legitimize detention decisions that would otherwise face greater scrutiny. Bail industry stakeholders have opposed risk-based pretrial release as a threat to the commercial bail system that generates billions of dollars annually. The resulting political dynamics have produced a complex reform landscape in which some jurisdictions have embraced risk-based pretrial reform while others have retreated from or rejected it (Stevenson, 2018).

Table 1. Risk Assessment Applications Across Criminal Justice Decision Points

Decision Point	Primary Purpose	Common Instruments	Key Considerations
Pretrial	Detention vs. release; conditions of release	PSA, VPRAI, ORAS-PAT	Speed of administration; equity concerns; flight risk vs. public safety
Sentencing	Inform sentencing guidelines; alternative sanctions	LSI-R, COMPAS, ORAS	Judicial discretion; proportionality; use of risk in punishment decisions
Classification	Security level; housing; program eligibility	LSI-R, LS/CMI, COMPAS	Institutional safety; program matching; over-classification costs
Supervision	Supervision level; conditions; caseload allocation	LS/CMI, ORAS, STRONG	Risk-need matching; officer workload; contact standards
Parole	Release timing; conditions; supervision intensity	SIR, COMPAS, LS/CMI	Prediction accuracy at low base rates; victim concerns; public safety
Reentry	Service referrals; housing; employment support	ORAS-RT, LS/CMI, SVORI	Transition planning; community capacity; continuity of care

Correctional Classification and Supervision

Risk assessment instruments serve as the primary basis for correctional classification—the process of assigning inmates to security levels, housing units, and program tracks within correctional institutions. Objective classification systems, which replaced the subjective judgment of classification committees in most state prison systems during the 1980s and 1990s, use validated instruments to determine the minimum security level consistent with institutional safety and to identify programming needs that correctional institutions should address during the period of incarceration. Research demonstrates that objective classification reduces the proportion of inmates housed at security levels higher than their risk warrants—a phenomenon known as over-classification—saving correctional costs while improving outcomes by placing inmates in less restrictive environments where rehabilitative programming is more accessible and the criminogenic effects of high-security confinement are reduced (Austin, 2006).

In community supervision, risk assessment determines the intensity of supervision—the frequency of reporting, the extent of monitoring, and the conditions imposed—for individuals on probation and parole. The risk principle’s central insight—that intensive supervision should be reserved for high-risk offenders while low-risk offenders should receive minimal intervention—has been validated through research demonstrating that intensive supervision of low-risk offenders produces worse outcomes than standard supervision, apparently by disrupting the prosocial routines, employment, and relationships that constitute the low-risk offender’s protective factors. Agencies that have implemented risk-based supervision report improved outcomes and more efficient allocation of officer time and program resources, though achieving full implementation requires overcoming organizational resistance to reducing supervision intensity for any offender population (Lowenkamp & Latessa, 2004).

The integration of risk assessment with case management—using dynamic need scores to drive individualized case plans that target the specific criminogenic factors elevated for each offender—represents the highest level of evidence-based supervision practice. Agencies that have implemented this integration report that officers who use structured case plans based on validated assessment produce better outcomes than officers who rely on unstructured supervision approaches, provided that adequate training, organizational support, and quality assurance mechanisms are in place. The practical challenge is ensuring that the programming and services identified in case plans are actually available in the community—a matching problem that requires investment in treatment, employment, housing, and family support services that many jurisdictions have not made at sufficient scale (Andrews & Bonta, 2010).

Validity, Bias, and Limitations

Predictive Performance

The predictive validity of risk assessment instruments has been evaluated in hundreds of studies across diverse populations, instruments, and outcome measures. Meta-analyses consistently report moderate predictive accuracy, with area under the curve (AUC) values typically ranging from .65 to .75 for general recidivism prediction—meaning that instruments correctly rank a randomly selected recidivist as higher risk than a randomly selected non-recidivist approximately 65–75 percent of the time. This level of predictive accuracy exceeds that of unstructured clinical judgment and is comparable to the accuracy of diagnostic instruments in medicine, though it falls short of the certainty that individual liberty decisions might ideally require (Singh et al., 2011).

The performance of risk instruments varies across populations, outcome measures, and follow-up periods in ways that have practical implications for implementation. Instruments generally perform better at predicting general recidivism than violent recidivism, because violent offending has a lower base rate and is therefore harder to predict. Predictive accuracy tends to decline over longer follow-up periods, as the influence of static historical factors diminishes relative to the influence of dynamic factors that change over time. And instruments validated on one population may not perform equivalently in another—an instrument validated primarily on white male offenders may predict less accurately for women, racial minorities, or specific offense types—making cross-validation in the target population essential before operational deployment (Desmarais & Singh, 2013).

The false positive problem—the incorrect classification of individuals who would not reoffend as high risk—is an inherent limitation of prediction at the individual level that cannot be eliminated through instrument improvement. When the base rate of the outcome being predicted is low—as it is for serious violent recidivism—even instruments with good overall accuracy will produce a high ratio of false positives to true positives, meaning that many individuals classified as high risk will not in fact reoffend. The consequences of false positive classification—extended detention, more restrictive supervision, denial of parole—fall on individuals whose actual future behavior would not justify these restrictions, creating a systematic injustice that the aggregate accuracy of the instrument conceals (Hart, Michie, & Cooke, 2007).

Racial Disparities and Algorithmic Fairness

The racial equity implications of risk assessment have become the most contested dimension of the field, particularly following ProPublica’s 2016 analysis documenting that the COMPAS instrument generated racially disparate false positive rates in a Broward County, Florida dataset. The analysis found that Black defendants were approximately twice as likely as white defendants to be falsely classified as high risk—incorrectly predicted to reoffend when they did not—while white defendants were approximately twice as likely to be falsely classified as low risk. These findings generated intense debate about the fairness of algorithmic risk assessment and its compatibility with equal protection principles (Angwin et al., 2016).

The subsequent methodological debate revealed a fundamental mathematical constraint: when base rates of the predicted outcome differ across groups—as recidivism rates do across racial groups in virtually every American jurisdiction—it is impossible for an instrument to simultaneously achieve equal false positive rates, equal false negative rates, and equal predictive values across groups. This impossibility result means that any definition of fairness that requires equality on one metric will necessarily produce inequality on another, and that the choice among fairness criteria is a value judgment rather than a technical determination. The practical implication is that agencies implementing risk assessment must explicitly decide which form of fairness they prioritize and must acknowledge the trade-offs that their choice entails (Chouldechova, 2017).

The sources of racial disparity in risk scores include both the instrument’s factor structure and the criminal justice system’s differential treatment of racial groups. Criminal history variables—which carry substantial weight in most instruments—reflect not only the individual’s actual criminal behavior but also the enforcement practices, prosecutorial decisions, and judicial outcomes that have produced a documented record. If Black individuals are more likely to be arrested, charged, and convicted than white individuals who engage in similar conduct—as extensive research on racial disparities in policing, prosecution, and sentencing suggests—then criminal history variables will embed and reproduce these disparities in risk scores. Removing criminal history from instruments reduces racial disparity but also reduces predictive accuracy, creating a trade-off between equity and validity that has no purely technical resolution (Stevenson, 2018).

Policy Controversies and Future Directions

Risk Assessment in Sentencing

The use of risk assessment to inform sentencing decisions—rather than merely classification, supervision, and release decisions—represents the most controversial expansion of risk assessment in contemporary criminal justice. Several states have incorporated risk assessment into sentencing guidelines, either by permitting judges to consider risk scores as a factor in sentencing or by using risk classification to determine eligibility for alternative sanctions. Virginia’s sentencing guidelines, for example, use a risk assessment instrument to identify low-risk non-violent offenders who are recommended for alternative sanctions rather than incarceration, diverting thousands of individuals from prison annually (Monahan & Skeem, 2016).

Opponents of risk-based sentencing argue that it violates fundamental principles of criminal law by punishing individuals not for what they have done but for what they are predicted to do—a form of preventive detention that is inconsistent with the retributive justification for punishment. The use of demographic and socioeconomic variables in risk instruments—including factors such as employment, education, and neighborhood characteristics that correlate with race and class—raises equal protection concerns that have been litigated in several courts, with mixed results. The Wisconsin Supreme Court’s decision in State v. Loomis (2016) upheld the use of COMPAS in sentencing while imposing limitations on its use, including requirements that risk scores not be used as the determinative factor in sentencing and that defendants be informed of the instrument’s limitations (Starr, 2014).

The future of risk assessment in criminal justice will be shaped by the resolution of tensions between predictive accuracy and equity, between individual liberty and public safety, and between the efficiency of algorithmic decision-making and the legitimacy of human judgment. Emerging approaches include the development of race-neutral instruments that exclude variables correlated with race, the use of fairness constraints in instrument development that optimize predictive accuracy subject to equity requirements, and hybrid approaches that combine actuarial risk scores with structured professional judgment to retain the accuracy advantages of statistical prediction while preserving the individualized assessment that legal and ethical principles require. The field’s capacity to resolve these tensions will determine whether risk assessment fulfills its promise of making criminal justice decisions more accurate, more consistent, and more just—or whether it becomes a mechanism for embedding existing inequalities in algorithmic form (Monahan & Skeem, 2016).

Implementation Science and Organizational Change

The effectiveness of risk assessment in practice depends not only on instrument validity but also on the quality of implementation within the organizations that adopt them. Research on implementation science in criminal justice has identified several factors associated with successful implementation: strong leadership commitment, adequate staff training, fidelity monitoring and quality assurance, integration of assessment into case management workflows, availability of programming and services matched to identified needs, and the organizational culture necessary to sustain evidence-based practice over time. Agencies that treat risk assessment as a compliance exercise—administering instruments without using the results to drive supervision and programming decisions—achieve little or no improvement in outcomes relative to agencies that do not use assessment at all (Taxman & Belenko, 2012).

The training requirements for valid administration of risk assessment instruments are substantial and ongoing. Officers must be trained in the standardized administration protocol for the specific instrument being used, including interview techniques, scoring rules, and the interpretation of results. They must receive regular refresher training and calibration exercises to prevent drift from standardized procedures. And they must understand the conceptual foundations of risk-need-responsivity principles sufficiently to translate assessment findings into case management strategies. Agencies that invest in this training infrastructure achieve higher fidelity and better outcomes than those that provide only initial training without ongoing support (Andrews & Bonta, 2010).

The sustainability of evidence-based risk assessment requires institutional commitment that outlasts individual champions and survives changes in leadership, budget priorities, and political conditions. Agencies that have embedded risk assessment into their organizational identity—through policy codification, performance measurement, data-driven quality improvement, and professional development systems that reinforce evidence-based practice—are more likely to sustain high-fidelity implementation over time than those in which assessment remains a discrete program vulnerable to discontinuation when its institutional advocates depart or when competing priorities emerge. The comparative perspective provided by implementation research across jurisdictions reveals that organizational capacity is at least as important as instrument selection in determining whether risk assessment improves criminal justice outcomes (Taxman & Belenko, 2012).

Conclusion

Risk assessment in criminal justice represents the most consequential translation of social science research into operational practice in the field. Validated instruments incorporating the Central Eight risk and need factors provide more accurate and more consistent predictions of recidivism than unstructured professional judgment, and their integration with case management and programming through the RNR framework has produced the most rigorously evaluated body of effective practice in correctional supervision. The generational progression from unstructured judgment through static actuarial instruments to integrated fourth-generation systems reflects a steady accumulation of knowledge about the predictors of criminal behavior and the principles of effective intervention that has transformed the conceptual and operational foundations of offender management.

The limitations and controversies surrounding risk assessment are equally consequential. The moderate level of predictive accuracy that even the best instruments achieve means that substantial numbers of individuals are misclassified—detained, supervised, or denied opportunities based on predictions that prove incorrect. The racial disparities embedded in criminal history variables reproduce systemic inequalities in algorithmic form. The use of risk scores in sentencing raises fundamental questions about the purposes of punishment and the limits of preventive justice. And the gap between the potential of evidence-based assessment and the reality of implementation in under-resourced agencies means that many offenders are assessed by instruments whose results are not connected to the supervision and programming decisions that would give them practical significance.

The future of risk assessment lies in the development of instruments and implementation models that address these limitations—instruments that achieve greater predictive accuracy while minimizing racial disparities, implementation models that connect assessment to effective programming at scale, and governance frameworks that ensure risk assessment serves justice rather than merely efficiency. The evidence demonstrates that structured risk assessment, properly implemented, produces better outcomes for offenders and communities than the unstructured judgment it replaces. The challenge is to realize this potential while managing the ethical, legal, and political complexities that accompany the use of predictive tools in decisions about individual liberty and public safety.

References

Andrews, D. A., & Bonta, J. (2010). The psychology of criminal conduct (5th ed.). Anderson Publishing.
Andrews, D. A., Bonta, J., & Wormith, J. S. (2006). The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52(1), 7–27.
Angwin, J., Larson, J., Mattu, S., & Kirchner, L. (2016). Machine bias. ProPublica, May 23.
Austin, J. (2006). How much risk can we take? The misuse of risk assessment in corrections. Federal Probation, 70(2), 58–63.
Chouldechova, A. (2017). Fair prediction with disparate impact. Big Data, 5(2), 153–163.
Desmarais, S. L., & Singh, J. P. (2013). Risk assessment instruments validated and implemented in correctional settings in the United States. Council of State Governments Justice Center.
Gendreau, P., Little, T., & Goggin, C. (1996). A meta-analysis of the predictors of adult offender recidivism. Criminology, 34(4), 575–607.
Grove, W. M., Zald, D. H., Lebow, B. S., Snitz, B. E., & Nelson, C. (2000). Clinical versus mechanical prediction: A meta-analysis. Psychological Assessment, 12(1), 19–30.
Hanson, R. K., & Morton-Bourgon, K. E. (2009). The accuracy of recidivism risk assessments for sexual offenders. Psychological Assessment, 21(1), 1–21.
Hart, S. D., Michie, C., & Cooke, D. J. (2007). Precision of actuarial risk assessment instruments. British Journal of Psychiatry, 190(S49), s60–s65.
Latessa, E. J., & Lovins, B. (2010). The role of offender risk assessment. Victims & Offenders, 5(3), 209–219.
Lowenkamp, C. T., & Latessa, E. J. (2004). Understanding the risk principle. Topics in Community Corrections, 2004, 3–8.
Lowenkamp, C. T., VanNostrand, M., & Holsinger, A. (2013). The hidden costs of pretrial detention. Arnold Foundation.
Maruna, S. (2001). Making good: How ex-convicts reform and rebuild their lives. American Psychological Association.
Meehl, P. E. (1954). Clinical versus statistical prediction. University of Minnesota Press.
Monahan, J., & Skeem, J. L. (2016). Risk assessment in criminal sentencing. Annual Review of Clinical Psychology, 12, 489–513.
Sampson, R. J., & Laub, J. H. (1993). Crime in the making: Pathways and turning points through life. Harvard University Press.
Singh, J. P., Grann, M., & Fazel, S. (2011). A comparative study of violence risk assessment tools. Clinical Psychology Review, 31(3), 499–513.
Skeem, J. L., & Lowenkamp, C. T. (2016). Risk, race, and recidivism: Predictive bias and disparate impact. Criminology, 54(4), 680–712.
Starr, S. B. (2014). Evidence-based sentencing and the scientific rationalization of discrimination. Stanford Law Review, 66(4), 803–872.
Stevenson, M. T. (2018). Assessing risk assessment in action. Minnesota Law Review, 103, 303–384.
Taxman, F. S., & Belenko, S. (2012). Implementing evidence-based practices in community corrections and addiction treatment. Springer.
Ward, T., & Maruna, S. (2007). Rehabilitation: Beyond the risk paradigm. Routledge.
Bonta, J., & Andrews, D. A. (2017). The psychology of criminal conduct (6th ed.). Routledge.
Brennan, T., Dieterich, W., & Ehret, B. (2009). Evaluating the predictive validity of the COMPAS risk and needs assessment system. Criminal Justice and Behavior, 36(1), 21–40.
Campbell, M. A., French, S., & Gendreau, P. (2009). The prediction of violence in adult offenders. Criminal Justice and Behavior, 36(6), 567–590.
Dressel, J., & Farid, H. (2018). The accuracy, fairness, and limits of predicting recidivism. Science Advances, 4(1), eaao5580.
Hannah-Moffat, K. (2013). Actuarial sentencing: An unsettled proposition. Justice Quarterly, 30(2), 270–296.
Harcourt, B. E. (2007). Against prediction: Profiling, policing, and punishing in an actuarial age. University of Chicago Press.
Petersilia, J. (2003). When prisoners come home. Oxford University Press.

Related Articles