Recidivism: What Predicts Reoffending?
Author: Ukrainian Psychological Hub · Published: October 11, 2026 · Editorial Policy
Recidivism is the occurrence of further criminal-justice involvement after an earlier offense or justice-system event, measured over a defined period. In research, it is usually operationalized as a new arrest, conviction, or return to custody. The answer to “what predicts reoffending?” depends on which outcome is measured, whose histories are studied, the length of follow-up, and the social and legal conditions following release. Prior offending history and age are among the most consistent statistical predictors; changing circumstances such as substance use problems, exposure to offending peers, housing instability, and opportunities for employment can also matter. No factor, diagnosis, or score establishes that one person will commit another crime.
The most useful scientific distinction is between a group-level association and an individual prediction. Risk indicators can help institutions identify needs, plan support, and evaluate policies. Their accuracy is conditional, and their interpretation carries substantial consequences for liberty, supervision, and access to services. This article reviews the evidence on recidivism, explains why reported rates differ, and examines what risk assessment can and cannot establish.
What Is Recidivism in Criminal Justice?
The U.S. Bureau of Justice Statistics (BJS) defines recidivism through subsequent justice-system events and identifies three indispensable elements of a recidivism study: a starting event, a specified outcome, and a follow-up period. A study may begin at prison release, the start of probation, a court disposition, or another legally defined event. It must then specify what counts as a subsequent event and for how long people are observed. The BJS National Recidivism and Reentry Data Program sets out this framework.
Recidivism is a research and policy construct. It is not a psychiatric diagnosis, personality trait, or fixed property of an individual. In everyday language, “reoffending” usually means committing another criminal offense. In administrative research, however, the measured outcome is often arrest or reconviction, which can be an imperfect proxy for underlying offending behavior.
A person can be arrested and later acquitted or have charges dismissed. Another person may commit an offense that is not detected or recorded. A return to prison may follow a technical supervision violation rather than a new criminal conviction. These distinctions determine what a statistic actually tells us.
Rearrest, reconviction, and return to prison are different outcomes
Rearrest measures whether a person is arrested after the starting event. It is relatively easy to observe through administrative records, but arrest is not proof of guilt, and arrest probability is affected by policing, reporting, surveillance, and other circumstances.
Reconviction measures a subsequent conviction, ordinarily following a new arrest or charge. It has a different legal threshold and may be affected by charging decisions, plea negotiations, case delays, and jurisdictional rules. It does not measure every offense that occurred.
Return to prison or reincarceration measures a subsequent custody event. Definitions differ, and some datasets include returns for supervision violations. A return-to-prison rate therefore measures part of the operation of the correctional system as well as subsequent behavior. The BJS Recidivism Patterns Explorer explicitly separates arrest, conviction, and return-to-prison measures.
Researchers may also distinguish any subsequent offense from violent, sexual, drug-related, or property offenses; the first event from the number of repeated events; and short-term from long-term recidivism. A headline percentage is meaningful only when its denominator, outcome, period, and population are stated.
How Common Is Recidivism? What U.S. Statistics Actually Show
There is no single recidivism rate that describes everyone who has been arrested, convicted, incarcerated, placed on probation, or released from supervision. Populations have different histories and exposure to risk, and jurisdictions have different enforcement and record-keeping practices.
One important U.S. benchmark comes from the BJS study of people released from state prisons in 24 states in 2008. Approximately 66% were arrested within three years and 82% within ten years. Approximately 61% returned to prison within ten years for a new sentence or a parole or probation violation. The percentage arrested during each separate year declined from about 43% in the first year to about 22% in the tenth year. These results describe that particular release cohort and the specified administrative outcomes; they do not mean that 82% of all formerly incarcerated people commit a proven new crime. See Antenangeli and Durose's official BJS report.
The cumulative percentage of people with at least one arrest tends to rise as observation lengthens, because a longer follow-up provides more time for an event to occur. By contrast, the percentage experiencing an arrest in each individual year may fall. Those two patterns are mathematically compatible and should not be confused.
Comparisons between an older prison cohort and a newer probation cohort, or between different states, can be misleading even when both studies use the word “recidivism.” The BJS notes that differences in population composition and the completeness of criminal-history records complicate comparisons across its own historical studies. For the underlying sampling and measurement issues, consult the BJS Recidivism of State Prisoners program.
Why a recidivism rate can rise without a rise in offending
Administrative recidivism measures are jointly shaped by behavior and detection. More intensive supervision may detect more violations and produce more recorded arrests; improved interstate record matching may reveal events that older studies missed. A change in prosecutorial policy or the definition of a technical violation can alter reconviction or reincarceration statistics. The reverse is also possible: a lower recorded rate may reflect changes in enforcement or data availability rather than a corresponding decline in all offending.
A sound evaluation therefore examines the measurement process and, where possible, several outcomes instead of interpreting a single percentage as a direct measure of criminal behavior.
What Are the Strongest Predictors of Reoffending?
Research repeatedly identifies associations involving prior criminal history, age, aspects of social context, and certain changeable needs. But there is no universally valid ranking of predictors across every country, offense type, age group, gender, or follow-up interval. Predictive value is a property of a model tested in a specific population and setting.
A systematic review and meta-analysis by Yukhnenko, Blackwood, and Fazel examined 15 studies involving more than 246,000 people receiving community sentences. It reported associations for multiple criminal-history and psychosocial variables, while emphasizing variation in study quality, populations, and measurement. The findings establish useful population-level patterns rather than a causal explanation for any person's offending.
Criminal history and age
Past convictions, arrests, or repeated justice-system involvement frequently correlate with subsequent recorded offending. The relationship may reflect persistent exposure to risk, accumulated disadvantage, established behavior patterns, prior opportunities, or characteristics of the justice system itself. An official history is also affected by who is detected and prosecuted. Criminal-history measures are important because they are consistently associated with the target outcome, not because a previous conviction makes a future offense inevitable.
Age is another robust correlate. Across many criminal-justice populations, recorded offending is more common in late adolescence and early adulthood than at older ages, and the rate often declines as people age. Yet trajectories differ: some individuals stop earlier, some continue for longer periods, and some return to offending after a long interruption. Age-related averages must not be converted into an individualized timetable.
History and chronological age are often called static factors because their recorded values cannot be changed through an intervention. Static does not mean morally decisive, causally sufficient, or permanently determinative. Nor does a person's earlier history cancel the possibility of subsequent change.
Prior offending patterns, context, and opportunity
Patterns of repeated justice involvement can provide information beyond whether a person has ever been arrested. The number, timing, and nature of recorded events, age at first recorded event, and the duration of a recent offense-free period may each be relevant depending on the assessment purpose. However, apparent precision can exceed the quality of the data, and offense categories are themselves shaped by legal classifications.
Opportunity also matters. Housing, access to lawful income, substance availability, social networks, local violence, and the likelihood of detection are elements of a changing environment. Their association with recidivism should be considered within a broader developmental and institutional context. A statistical model that focuses on individual characteristics alone can miss conditions influencing both actual behavior and the chance of coming back into contact with law enforcement.
Dynamic Factors: What Can Change After Sentencing or Release?
Dynamic factors are variables that can change over time and may be useful for ongoing reassessment or case planning. The label does not establish that directly changing the variable will reduce crime. That causal question requires intervention research and careful designs that address confounding.
Substance use and substance-related problems
Problematic alcohol or drug use is associated with elevated recidivism in several criminal-justice populations. It may relate to impaired judgment in particular circumstances, economic pressures, exposure to risky settings, or interactions with an existing offending pattern. Substance use disorders also vary greatly in severity and consequences. Their presence alone does not establish criminal propensity, and an individual with a substance-related diagnosis should never be characterized as dangerous on that basis.
In the community-sentence meta-analysis by Yukhnenko and colleagues, substance misuse was among the psychosocial variables associated with recidivism. The magnitude of an association depends on definitions, statistical controls, the population, and the measured outcome. Support should be based on assessed needs and available clinical evidence, while the causal effect of any particular program on reoffending must be separately evaluated.
Peer relationships and social networks
Association with peers involved in offending is another frequently studied correlate. Networks influence opportunity, norms, perceived rewards, access to illicit markets, and responses to everyday conflict. But social selection also operates: people with overlapping histories and circumstances may become friends before a given outcome. The research therefore supports attention to social context without proving that one particular relationship caused a particular crime.
Stable and supportive relationships, prosocial community connections, and a sense of belonging may create alternative opportunities and reinforce change. Their significance depends on quality, safety, and individual circumstances; the mere existence of a partner or family contact cannot be treated as a universal protective factor.
Employment, income, and housing
Unemployment, unstable income, interrupted education, housing insecurity, and social exclusion are often associated with correctional outcomes. A stable job or home may help establish routines, reduce some immediate pressures, and support access to health care and treatment. Yet hiring discrimination, supervision requirements, local labor demand, disability, transport access, and family responsibilities influence whether those resources are available.
It is inaccurate to equate poverty with criminality or to infer wrongdoing from a person's unemployment. Socioeconomic conditions are relevant both as potential supports or constraints and as sources of unequal opportunity. High-quality program evaluation must ask whether improving a specific material condition actually changes a specified recidivism outcome.
Antisocial attitudes and thinking patterns
Some assessment frameworks include measures of attitudes favorable toward offending, perceived legitimacy of criminal opportunities, problem-solving, and self-regulation. Associations can be observed at the group level, but such constructs are often measured through interviews or self-report and may be sensitive to context, language, incentives, and evaluator judgment. A score must be interpreted within the instrument's validated purpose. It is not evidence that a person secretly intends to offend.
Mental health needs
Mental illness is neither a synonym for criminality nor a stand-alone explanation for reoffending. People experiencing depression, psychosis, anxiety, trauma-related symptoms, or other disorders have diverse behavioral histories and circumstances. Mental health needs may affect functioning, vulnerability, access to services, and supervision experiences, but relationships with criminal-justice outcomes are heterogeneous and can be confounded by substance use, homelessness, victimization, and socioeconomic disadvantage.
A finding that a mental health measure is statistically associated with recidivism in a particular study does not establish that the diagnosis itself caused an offense. For a closely related diagnosis-specific discussion, see Schizophrenia and Violence: What Research Really Shows. The scope of that article is clinical evidence; this article focuses on recidivism measurement and correctional prediction.
Static Risk Factors, Dynamic Needs, and Protective Factors
Risk factors are characteristics associated with a higher probability of a specified outcome in a studied population. Static factors, such as earlier convictions, generally cannot be changed. Dynamic factors, such as current living circumstances or substance-related problems, may change. Protective factors are characteristics or conditions associated with lower risk or with successful adjustment; they may include supportive relationships, accessible care, stable accommodation, skills, and realistic opportunities.
The distinction matters because prediction and assistance have different objectives. A historical variable can contribute to risk classification without identifying a practical intervention target. A current need may be important for a person's well-being even if it adds little to the accuracy of a recidivism model. A protective factor may help describe a person's present resources without guaranteeing any specific future behavior.
Time is central. A person's opportunities, health, relationships, legal restrictions, and aspirations can evolve substantially after the initial assessment. Reassessment can therefore be justified when circumstances materially change. Yet a revised score should only be interpreted as evidence of changed estimated risk when the instrument has been validated for repeated use and the outcome in question.
Recidivism and Desistance: Why Many People Stop Offending
Desistance refers to the process of ending or progressively reducing offending over time. It may involve changes in routine, identity, social ties, opportunities, and the legal environment. A period without recorded offending is an observation; the deeper process of desistance concerns how change unfolds and is sustained.
A person may have earlier convictions and later live for decades without another offense. Another person may experience an interruption followed by a relapse into offending. These trajectories show why a one-time assessment is an incomplete account of a life course. Longitudinal research, including the declining annual arrest percentages in the BJS ten-year follow-up, is consistent with the importance of change over time, although administrative rates by themselves do not reveal an individual's psychological reasons for desisting.
Desistance is a distinct subject that deserves its own research and conceptual treatment. Here it serves to clarify the limits of recidivism forecasting: statistical risk is conditional, future circumstances matter, and an earlier conviction does not identify a person's permanent identity.
How Do Criminal Recidivism Risk Assessment Tools Work?
Risk assessment instruments combine information about a person and their circumstances to categorize risk or estimate the probability of a specified future outcome. Some use fixed scoring rules; others employ structured professional judgment. They may be designed for pretrial decisions, sentencing, prison classification, probation or parole supervision, violence prevention, or offense-specific management. These purposes and outcomes are not interchangeable.
The U.S. federal Post Conviction Risk Assessment (PCRA) is an example used to support post-conviction supervision planning. Its existence does not make its scores universally applicable to all populations or legal decisions. A tool calibrated for general recidivism in a supervised adult population should not automatically be used to predict a violent offense, a sexual offense, or an outcome in adolescents.
Valid use requires clarity about the outcome, the original development sample, external validation, scoring fidelity, availability of accurate records, and local implementation. An assessment conducted by a trained professional remains subject to measurement error and the limits of the underlying evidence.
Discrimination: What does the AUC tell us?
Discrimination describes how well an instrument tends to rank people who later experience an outcome above those who do not. One common statistic is the area under the receiver operating characteristic curve, or AUC. An AUC of 0.50 is consistent with chance-level ranking; 1.00 represents perfect discrimination within the specified data. An AUC of 0.70 means that, when one person with the observed outcome and one without it are sampled from the evaluated dataset, the score assigned to the person with the outcome is higher about 70% of the time, with customary handling of ties. It does not mean a person assigned a score has a 70% chance of offending.
In a 2022 systematic review of sentencing risk tools, Fazel and colleagues identified 36 validation studies with 597,665 participants. In larger independent studies, AUCs ranged from 0.57 to 0.75. Performance varied, and smaller studies or studies with developer involvement tended to report more favorable results. The authors found inadequate reporting of calibration and important error measures. These findings support a restrained interpretation of risk classifications.
Calibration: Does a predicted percentage match observed outcomes?
Calibration addresses whether predictions correspond to observed rates. If an instrument assigns 100 comparable people a ten-year risk of 20%, a well-calibrated model would anticipate roughly 20 observed events in that group under comparable conditions. It does not identify which 20 people will have an event. A model can rank people moderately well while systematically overestimating or underestimating the absolute probability of the outcome.
Calibration can change when a tool is transported across jurisdictions, years, age groups, or institutional settings. Arrest and conviction rates, local treatment resources, supervision practices, and surveillance intensity all influence the observed outcome. A numeric risk estimate should be tied to the population, period, and event definition used to validate it.
False positives, false negatives, and base rates
A false positive occurs when a model classifies a person as likely to experience an outcome but that outcome is not observed during the specified period. A false negative occurs when a model classifies a person as unlikely to experience an outcome but the event is observed. Both matter. False positives can lead to unnecessary restrictions and stigma; false negatives can impair prevention and support planning.
The base rate is the proportion of the relevant population who experience the defined outcome. A test with the same sensitivity and specificity can have very different positive predictive values when the underlying event is common versus rare. A 2025 systematic review and meta-analysis of four established risk instruments underscored the importance of base rates and reported that positive predictive value becomes especially limited for low-frequency outcomes. Therefore, a person placed in a “high risk” category may still be more likely than not to avoid the specific event in the study's follow-up period.
The choice of decision threshold also changes the balance between false positives and false negatives. That threshold is partly a policy and legal judgment about consequences, not a universal psychological constant.
A worked example: why a high-risk label can mislead
Imagine a hypothetical group of 1,000 people in which 10% experience the specified outcome during a defined follow-up. Suppose a screening tool correctly flags 70% of those who experience the outcome, while incorrectly flagging 20% of those who do not. It would identify 70 true positives among 100 people with the outcome and 180 false positives among 900 without it. Of 250 people flagged, only 70 would have the outcome: a positive predictive value of 28%.
These are illustrative arithmetic assumptions, not results from a particular instrument. They show why sensitivity, specificity, base rate, calibration, and the consequences of classification must be considered together. Reporting only that a tool identifies elevated risk hides information decision-makers need.
Is an Actuarial Tool Better Than an Expert's Judgment?
Structured assessment can improve consistency and make assumptions visible. Unstructured opinion is harder to audit and may be more vulnerable to intuition, irrelevant information, and inconsistent weighting. Yet structure does not guarantee validity, unbiased implementation, or an acceptable error rate.
A systematic meta-analysis comparing risk assessment tools with unstructured judgments found an average predictive advantage for structured tools in the examined studies. The result does not establish that any available tool is sufficiently accurate for every legal use. The independent-validation and calibration limitations identified by Fazel and colleagues remain relevant.
The appropriate question is whether a particular instrument, administered correctly and independently validated for a relevant population and purpose, adds useful information beyond available alternatives. “Evidence-based” should refer to demonstrated performance and sound application rather than the mere presence of a questionnaire or algorithm.
Cultural Fairness, Race, Disability, and Jurisdiction
Risk estimates may reflect patterns in recorded arrests and convictions that are influenced by policing and institutional practices. If a model predicts arrest, it may partly predict who is more likely to be observed or contacted by law enforcement. Variables such as neighborhood, prior arrests, employment, or family circumstances can be connected to unequal structural conditions. These are substantive questions for model validity and justice, even when a protected characteristic is omitted from the equation.
A 2024 systematic review by Lawson, Narkewicz, and Vincent examined research on the disparate effects of adult and juvenile risk assessment instruments in legal decision-making. The findings reinforce the need to examine how a model is deployed and whether it produces different consequences across racial and ethnic groups. Evaluating statistical discrimination alone cannot settle whether a system is equitable.
Fairness also involves language proficiency, cultural interpretation, neurodevelopmental differences, disability accommodations, missing records, access to counsel, and institutional incentives. Researchers should examine local calibration, classification rates, false-positive and false-negative rates, and legal consequences across relevant groups. A tool developed in one national justice system cannot be presumed valid in another merely because its scoring instructions can be translated.
Does Recidivism Risk Differ by Offense Type?
Yes, but outcome definitions and populations need to remain distinct. General reconviction, violent reoffending, sexual recidivism, intimate partner violence, and technical supervision violations are different endpoints. Each has a different base rate and different measurement challenges. Some instruments were developed specifically for one category and should not be interpreted as universal violence or criminality tests.
For example, Hanson and Morton-Bourgon's meta-analysis examined predictors of sexual recidivism and associated distinct risk domains. Findings about sexual recidivism cannot be mechanically transferred to property offending or to all people convicted of an offense involving sexual behavior. The same principle applies to predicting interpersonal violence versus any new arrest.
Young people are also a distinct population. Adolescents experience rapid developmental and environmental changes, different legal rules, and different opportunities for intervention. An adult instrument should not be applied to a young person unless validated for that use. Factors such as school engagement, family conditions, neighborhood context, and access to services require developmental interpretation. For older adults, changes in health, disability, opportunity, and living circumstances can also alter the relevance of earlier data.
Do Prison, Probation, and Rehabilitation Reduce Recidivism?
Whether an intervention reduces reoffending is a causal effectiveness question. It is different from identifying characteristics associated with future arrest or conviction. An intervention may improve health, housing, relationships, or employment while having uncertain effects on recorded recidivism. Conversely, an apparent reduction in rearrest may reflect different surveillance practices rather than a comparable change in behavior.
The Risk–Need–Responsivity model proposes aligning intervention intensity with assessed risk, targeting potentially changeable needs related to offending, and adapting services to a person's ability to engage. It has shaped correctional programs and supervision practice. However, a 2024 umbrella review of 26 meta-analyses by Fazel and colleagues found mixed and often low-quality evidence for its component principles and urged stronger independent research. A widely adopted framework is not a substitute for high-quality causal evaluation.
Interventions may include substance use treatment, cognitive-behavioral programs, educational opportunities, employment support, housing assistance, psychiatric care, and coordinated reentry services. The question is which intervention benefits which population, through what mechanism, under what implementation conditions, and for which outcome. A program successful in one prison system may perform differently when staffing, participation, community resources, and legal requirements change.
Over-supervision is also a consideration. Requirements that are difficult to satisfy can increase administrative violations even when they have little demonstrated benefit for public safety. Proportionate practice balances documented risk, individual needs, rights, and access to constructive opportunities. The U.S. Courts explanation of PCRA-based supervision illustrates one institutional approach, not a universal rule for all jurisdictions.
Why good programs require good evaluations
Random assignment, when ethically and practically feasible, helps distinguish an intervention effect from the characteristics of people selected to receive it. Other rigorous designs may be appropriate, but selection bias, dropout, inconsistent program delivery, and different follow-up lengths remain serious problems. Researchers should preregister outcomes where possible, report harms and implementation failures, and distinguish new offenses from violations or administrative returns.
A credible claim that a program “reduces recidivism” identifies its comparison group, outcome, sample, study design, follow-up interval, and degree of uncertainty. It also considers whether benefits persist and whether they apply to populations excluded from the original study.
What Recidivism Predictions Mean in Court and Corrections
A risk assessment is an empirical input to a decision, not a legal verdict. It cannot establish guilt, prove a defendant's intentions, determine criminal responsibility for an earlier offense, or decide competency to stand trial. Criminal responsibility and competency are jurisdiction-specific legal questions requiring distinct criteria and evidence.
Risk classifications may affect sentencing, parole, probation conditions, housing, treatment intensity, or institutional placement. Because these choices influence liberty and opportunity, transparency and procedural safeguards matter. The relevant legal standard is determined by the jurisdiction and the decision being made. Research findings about statistical prediction do not replace law, and practices permitted in one U.S. state or country should not be treated as universal.
The American Psychology–Law Society's Specialty Guidelines for Forensic Psychology set professional expectations for competence, appropriate methods, attention to limits, and clear communication in legal contexts. The guidelines were extended through December 31, 2026. Where a forensic opinion uses a risk tool, the evaluator should explain the tool's purpose, relevant population, data quality, known error and bias, and the limits of inference.
Decision-makers should also recognize feedback loops: a higher classification can produce more intensive supervision, and more intensive supervision can change the probability of detecting a violation. Those changes may then enter future administrative records. Interpreting measured outcomes requires attention to how the assessment itself changed people's treatment by institutions.
Recidivism Research: Seven Frequent Misconceptions
“An arrest rate is the same as a crime rate.”
An arrest measures contact with law enforcement and can occur without a conviction. Undetected offending is missing from arrest data. Reports should call an arrest-based outcome what it is.
“A previous conviction means the person will offend again.”
Prior history is an important statistical correlate, but it does not determine a person's future. Age, time, circumstances, resources, and individual change also matter.
“A high-risk score proves dangerousness.”
A risk score summarizes information for a specified endpoint. It does not prove that an event will occur, and a general recidivism model may not predict violence at all.
“A diagnosis explains why a crime happened.”
A clinical diagnosis describes a recognized pattern of symptoms or functioning. It does not by itself establish motive, culpability, future offending, or a direct causal pathway to a criminal act.
“If a tool uses numbers, its decision is objective.”
Numerical rules can make scoring more consistent, but model design, outcome definitions, data quality, institutional practice, and selected thresholds still involve substantive choices.
“A declining recidivism rate always means rehabilitation worked.”
A lower measured rate can reflect several mechanisms, including changed populations, enforcement, follow-up, record quality, and available opportunities. A causal rehabilitation claim requires a design capable of testing the intervention.
“Risk factors and protective factors are permanent labels.”
Many circumstances change, and even historical risk indicators are statistical descriptors rather than moral identities. Reassessment should be responsive to relevant new information and supported by evidence.
How to Read a Recidivism Study Critically
First identify the study population: people released from prison, people on probation, individuals convicted of a particular offense, or another defined group. Next identify the starting event and the precise outcome. Then locate the follow-up period and determine whether the reported rate is cumulative or annual. Check whether the research counted first events, repeated events, technical violations, or only new criminal convictions.
For prediction research, examine whether the investigators externally validated the model, reported AUC alongside calibration, and disclosed false-positive and false-negative rates. Ask whether the test was developed in a different legal system, whether its base rate matches the setting where it is being used, and whether subgroup performance and fairness were evaluated.
For claims that a factor causes offending or that a program prevents it, look for designs that test causation rather than merely association. Finally, consider the ethical consequences: who may be restricted, who may receive services, whose history is represented in the data, and what appeal or review procedures exist.
These questions apply equally to traditional actuarial tools and newer machine-learning systems. Greater computational complexity does not automatically resolve problems of measurement, calibration, legal relevance, or fairness.
Frequently Asked Questions About Recidivism
What does recidivism mean in simple terms?
Recidivism generally means further offending or criminal-justice involvement after an earlier offense or legal event. In research, it usually means a specified subsequent arrest, conviction, or return to custody within a defined period. The exact measurement should always be stated.
What is the biggest predictor of recidivism?
Criminal history and age are among the most consistently replicated statistical predictors in many adult criminal-justice populations. However, their relative strength differs by population, offense type, and measurement period. A predictor's usefulness for a group should not be mistaken for a reliable forecast for one person.
What is the difference between recidivism and reoffending?
Reoffending refers directly to committing another offense. Recidivism is often operationalized through administrative outcomes such as rearrest, reconviction, or reincarceration. Those outcomes overlap with, but do not perfectly measure, actual offending.
What is the difference between recidivism and desistance?
Recidivism research records subsequent justice-system events; desistance research examines the process of reducing or stopping offending across the life course. Both are necessary to understand change, but they ask different questions.
Does everyone with a prior conviction have a high risk of reoffending?
No. People with prior convictions vary substantially in history, age, circumstances, needs, strengths, and later conduct. A prior conviction is information about a past legal event, not a verdict about the future.
Can a psychologist predict whether one person will commit another crime?
Psychologists and other trained professionals may provide carefully limited, evidence-based risk opinions for specified outcomes and populations. Their estimates are probabilistic and subject to error, especially when base rates are low or a tool is used outside its validated setting. No assessment can establish a particular individual's future conduct with certainty.
Does mental illness increase recidivism?
The answer depends on the diagnosis, co-occurring conditions, population, outcome, available support, and study design. Mental illness should not be used as a general proxy for criminality or dangerousness. Individual clinical and social needs deserve assessment on their own terms.
Are recidivism risk assessment scores fair?
Fairness has several meanings, including calibration, error rates, equitable consequences, and procedural safeguards. A tool may satisfy one statistical fairness criterion while raising concerns under another. Valid local evaluation and examination of institutional effects are essential.
Do jobs, housing, and therapy prevent recidivism?
They can be important aspects of reintegration and well-being. Whether a particular employment, housing, or treatment intervention causally reduces a defined recidivism outcome depends on the program and the evidence. Statistical associations alone cannot establish a preventive effect.
Does a lower rearrest rate prove a justice program worked?
Not by itself. A sound effectiveness evaluation compares appropriate groups and considers surveillance, data collection, baseline differences, and the intended outcome. Rearrest and reconviction should be reported separately when possible.
Is recidivism the same in every state or country?
No. Legal rules, supervision regimes, populations, event definitions, data infrastructure, and the availability of services differ across jurisdictions. Numerical comparisons require harmonized definitions and careful context.
Related Articles
References
American Psychology–Law Society. (2013; extension through 2026). Specialty Guidelines for Forensic Psychology. Official professional guidelines.
Antenangeli, L., & Durose, M. R. (2021). Recidivism of Prisoners Released in 24 States in 2008: A 10-Year Follow-Up Period (2008–2018). Bureau of Justice Statistics. Official report.
Bureau of Justice Statistics. (2021–present). National Recidivism and Reentry Data Program. Official definitions and methodology.
Bureau of Justice Statistics. (n.d.). Recidivism of State Prisoners. Research series and methods.
Bureau of Justice Statistics. (n.d.). Recidivism Patterns Explorer. Official recidivism measures and interactive data.
Fazel, S., Burghart, M., Fanshawe, T., Gil, S. D., Monahan, J., & Yu, R. (2022). The predictive performance of criminal risk assessment tools used at sentencing: Systematic review of validation studies. Journal of Criminal Justice, 81, 101902. https://doi.org/10.1016/j.jcrimjus.2022.101902.
Fazel, S., Hurton, C., Burghart, M., DeLisi, M., & Yu, R. (2024). An updated evidence synthesis on the Risk-Need-Responsivity (RNR) model: Umbrella review and commentary. Journal of Criminal Justice, 92, 102197. https://doi.org/10.1016/j.jcrimjus.2024.102197.
Hanson, R. K., & Morton-Bourgon, K. E. (2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73(6), 1154–1163. https://doi.org/10.1037/0022-006X.73.6.1154.
Lawson, S. G., Narkewicz, E. L., & Vincent, G. M. (2024). Disparate impact of risk assessment instruments: A systematic review. Law and Human Behavior, 48(5–6), 427–440. https://doi.org/10.1037/lhb0000582.
United States Courts. (n.d.). Post Conviction Risk Assessment. Official federal-courts resource.
Viljoen, J. L., Goossens, I., Monjazeb, S., Cochrane, D. M., Vargen, L. M., Jonnson, M. R., Blanchard, A. J. E., Li, S. M. Y., & Jackson, J. R. (2025). Are risk assessment tools more accurate than unstructured judgments in predicting violent, any, and sexual offending? A meta-analysis of direct comparison studies. Behavioral Sciences & the Law, 43(1), 75–113. https://doi.org/10.1002/bsl.2698.
Weber, M. A., Schnyder, N., Kirschstein, M. A., Graf, M., Endrass, J., & Rossegger, A. (2025). The key role of base rates: Systematic review and meta-analysis of the predictive value of four risk assessment instruments. Swiss Medical Weekly, 155, 3517. https://doi.org/10.57187/s.3517.
Yukhnenko, D., Blackwood, N., & Fazel, S. (2020). Risk factors for recidivism in individuals receiving community sentences: A systematic review and meta-analysis. CNS Spectrums, 25(2), 252–263. https://doi.org/10.1017/S1092852919001056.