Criminal Careers: Onset, Persistence, and Desistance
Author: Ukrainian Psychological Hub · Published: October 11, 2026 · Editorial Policy
A criminal career is the sequence of a person's offending behavior across time. In criminology, the term describes an analytical history of behavior, including when offenses begin, how often they occur, how their seriousness or type changes, and when offending declines or stops. It does not imply that offending is a profession, a fixed identity, or an inevitable lifelong pattern. The foundational National Research Council framework was developed precisely because aggregate crime rates conceal substantial differences between individuals and across life stages.
The central question is developmental: why does one individual commit an offense once, another offend repeatedly for several years, and another show a longer, changing pattern? Research distinguishes onset, participation, frequency, persistence, duration, specialization, escalation, and desistance. These dimensions have different denominators, measurement problems, and explanatory demands. This article explains the entire criminal-career framework and the limits of what an observed trajectory can tell us about any particular person.
Ukrainian Psychological Hub · Psychology for the AI Era · Reviewed October 11, 2026.
What does “criminal career” mean in criminology?
A criminal career is a time-ordered sequence of criminal acts attributed to an individual, whether identified through self-report, police contacts, arrests, court dispositions, convictions, or another specified indicator. Researchers should state the precise outcome. An arrest career describes arrests, a conviction career describes convictions, and a self-reported offending career describes disclosed acts; those records can portray substantially different histories. The term “career” is a technical description of repeated opportunities for observing acts, not a formal legal category or clinical diagnosis.
The National Research Council's Criminal Careers and “Career Criminals” distinguishes participation—the extent to which people are involved at all—from dimensions observed among active offenders. Its framework allows researchers to ask whether changes in the population crime rate come from changes in the number of people participating, the number of acts per participant, offense seriousness, or career duration. Without these distinctions, even a genuine decline in recorded crime can be explained incorrectly.
Core dimensions of a criminal career
Onset refers to the first occurrence of a specified offending outcome. Its observed age depends on what counts as an offense and when observation begins. Participation or prevalence is the proportion of a population who offend during a defined interval. Incidence describes the rate of offenses in a population; frequency or individual offending rate counts acts by an active offender within a defined period. Persistence refers to continued offending across successive periods, while duration is the span between operationally defined starting and ending points. Desistance concerns the processes through which offending decreases and ends. Offense mix, seriousness, specialization, and versatility describe the types and patterns of acts within that sequence.
Imagine two people each with four officially recorded offenses. One has four offenses in a single year during adolescence and no later recorded activity; the other has four offenses spread across fifteen years. Their observed totals are identical, but their age of onset, frequency, duration, and timing are different. A career analysis preserves that distinction rather than compressing both histories into a single offense count.
The stages of a criminal career: onset, continuation, and exit
Onset: when offending first occurs
Age of onset is the age at the first observed offense under a specified definition. Early onset is frequently associated with longer or more serious recorded offending histories at the group level, but association cannot establish an individual destiny. Observation begins at different ages, self-reports may be inaccurate, and official records systematically miss undetected offending. A first arrest at twenty-three is evidence of a first recorded arrest, not necessarily a first criminal act. Conversely, childhood behavior problems and school disciplinary events should never be relabeled as criminal convictions.
The relationship between childhood circumstances and later offending is cumulative and context dependent. A systematic review of fourteen meta-analyses identified family-related dimensions as prominent longitudinal predictors of general offending, while persistence was associated with a broader constellation involving peers, education or employment problems, family difficulties, certain mental health problems, and substance use. These are group-level statistical relationships. They cannot identify who will commit a particular offense or prove that any one factor caused it.
Persistence: why some observed sequences last longer
Persistence means that offending continues beyond its initial occurrence. It can involve frequent offending, sporadic episodes separated by long gaps, or changes in offense type. Researchers sometimes describe a small high-rate subgroup or longer-lasting trajectory, but the size and shape of those subgroups depend on the cohort, outcome, follow-up time, and statistical approach. Persistence is a property of an observed behavioral history, not a personality diagnosis.
Social exposure, opportunity, peer networks, cumulative disadvantage, family instability, contact with institutions, access to work, and changing responsibilities may all contribute to continuity. Some processes can reinforce one another: repeated justice-system involvement may make employment harder; limited lawful opportunities may increase contact with offending environments. These accounts are hypotheses that require suitable designs to disentangle causal pathways from selection effects. The same events can affect people differently.
Decline, interruption, and desistance
A decline in offending intensity is not necessarily complete cessation. A person may move from frequent to occasional offending, desist from one offense type while committing another, or experience a temporary gap during incarceration. Laub and Sampson distinguish termination—the point at which offending stops—from desistance as the broader process by which cessation develops and is maintained. That distinction matters because a single offense-free interval, especially a short one, does not by itself establish a stable endpoint.
Desistance can involve changes in routine activities, attachments, employment opportunities, relationships, identity, incentives, and aging. It is also shaped by broader institutions: housing, labor markets, discrimination, supervision requirements, and criminal-record barriers. Research should avoid suggesting that a marriage, a job, or a personal decision mechanically produces desistance in every case. Their apparent benefits can reflect who enters or sustains those circumstances, as well as effects of the circumstances themselves.
The age–crime curve: why criminal activity often peaks in adolescence
One of the most replicated aggregate observations in criminology is that many measures of offending rise during adolescence or late adolescence and decline through adulthood. Moffitt's developmental taxonomy was partly developed to explain how an age–crime curve can coexist with marked differences in continuity between individuals. The curve describes a distribution of behavior across ages; it is not a timeline guaranteed for everyone, nor does it establish the age at which any specific person will stop offending.
The shape of the curve varies by offense type, population, historical period, gender, data source, and the distinction between prevalence and frequency. A downturn in the total number of recorded offenses might occur because fewer people are participating, because each active participant commits fewer acts, or both. Offense-specific patterns may differ substantially. Juvenile justice definitions, policing practices, reporting rates, and exposure to opportunities also change across ages.
Age has several competing explanatory roles. Developmental changes may affect impulse regulation and social goals; increasing responsibilities may alter social networks and routine activities; individuals may withdraw from risky settings; and accumulated sanctions or barriers may constrain opportunities. No single mechanism explains every observed decline. Aging is a robust correlate of reduced offending in many cohorts, yet the policy relevance emerges from understanding how individual and social change interact.
Major developmental theories of criminal careers
Adolescence-limited and life-course-persistent patterns
Terrie Moffitt's influential 1993 account proposed two broad developmental patterns of antisocial behavior. In the adolescence-limited pathway, behavior is concentrated around adolescence and often declines as adult roles become accessible. In the life-course-persistent pathway, problems arise earlier and remain more continuous across stages, potentially through interactions between individual vulnerabilities and adverse environments. The theory organizes longitudinal hypotheses; it does not supply a diagnostic test for classifying a child as a future offender.
The distinction was historically influential, but later studies revealed greater diversity: low-rate intermittent offending, adolescent-onset persistent patterns, early-onset trajectories that desist, and later-onset activity. Moffitt's 2018 research review synthesizes long-term follow-up and the theory's continuing relevance. Its categories should be treated as useful theoretical contrasts, with observable overlap and classification uncertainty, rather than natural kinds that every individual must occupy.
Age-graded informal social control and turning points
Life-course perspectives emphasize the timing and quality of social bonds, transitions, institutional experiences, and turning points. The central claim is that pathways remain open to change even after extensive adolescent offending. Sampson and Laub's study of delinquent boys followed toward age seventy reported substantial decline in crime with age across modeled offender groups and found childhood characteristics insufficient to forecast individual long-term trajectories with precision. Their historical cohort is informative, but its male-only composition, historical context, and selective sampling limit universal generalization.
Turning points require careful interpretation. Employment may structure time and increase conventional opportunities, yet stable employment is partly selected by educational history, health, local labor demand, and discrimination. Supportive relationships may facilitate change, while troubled relationships may introduce additional stress. A person's own commitments and agency can change the meaning of the same social transition. Longitudinal association alone rarely separates these mechanisms completely.
Cumulative disadvantage, social learning, and opportunity
Cumulative disadvantage theories examine how earlier experiences alter later opportunities: disrupted schooling, stigmatizing records, exclusion from employment, unstable housing, or deteriorating social bonds can make pathways harder to reverse. Social learning perspectives consider reinforcement, peer behavior, and changing norms. Opportunity-oriented accounts emphasize access to situations in which offenses can occur. These perspectives can complement one another, and each should be evaluated against particular offenses and concrete empirical measures.
For example, a person might offend less after moving away from an offending peer group because opportunities change, motivations change, supervision increases, or several changes occur together. An apparent causal story inferred from a single biography remains incomplete until alternative explanations and selection are examined. Criminal-career research is strongest when it compares trajectories across people while preserving information about the contexts in which those paths unfold.
What longitudinal studies actually show
Long-term follow-up and heterogeneity
Longitudinal cohorts repeatedly observe individuals through multiple ages. They can describe within-person change more accurately than cross-sectional comparisons, and they allow researchers to examine the order in which exposures and outcomes occur. Piquero, Farrington, and Blumstein's criminal-career research, grounded in the Cambridge Study in Delinquent Development, illustrates the value of following one cohort over decades while analyzing participation, frequency, duration, and predictors. Yet a cohort of South London boys born in an earlier social context cannot automatically represent contemporary US women, rural communities, or populations in other legal systems.
Cohorts can establish temporal order and estimate average associations; causal claims require additional assumptions and designs. Attrition, mortality, migration, incarceration, missing records, changing offense laws, and changing enforcement intensity can distort a measured career. A sample recruited among incarcerated people selects a different population from a general-population birth cohort. Comparisons should begin with the sampling frame, outcome definition, follow-up period, and whether time without opportunity to offend in the community was handled.
What a 2026 systematic review adds about women
A 2026 systematic review by Coles and colleagues synthesized twenty-six studies in seven countries, covering approximately 92,000 participants across studies, while acknowledging possible overlapping cohorts. It identified multiple pathways among females, including adolescence-limited, early-onset desistant, intermittent, late-onset, and persistent patterns. The finding challenges any simple claim that persistent criminal careers occur only among males. At the same time, all included studies came from high-income countries, the included methodologies varied, and the reported percentage in a modeled group depends strongly on denominator and study design.
Female offending has historically been studied less extensively, and conclusions based on predominantly male samples deserve scrutiny. At the same time, subgroup averages cannot define the experience of all women or men. Sex and gender categories in historical datasets may have been recorded differently from contemporary research practices, and apparent disparities can reflect exposure, reporting, prosecution, and measurement differences alongside behavioral variation.
Trajectories are statistical summaries, not immutable identities
A widely used technique, group-based trajectory modeling, fits a small set of average trajectories to approximate heterogeneity across individuals. The apparent high-rate, low-rate, declining, or intermittent groups are model-derived summaries, not proof that nature contains exactly that number of offender types. Nagin's methodological paper specifically warns against treating group membership, the number of groups, or each individual's course as fixed facts. Nagin and Odgers likewise discuss the role and limitations of these models.
A group labeled “persistent” means that the model estimated relatively sustained offending within the observed window. It does not imply the same level continues indefinitely, that every member follows the group average, or that an individual can be assigned with certainty. Classification errors, differences in follow-up length, sparse observation, measurement of zero offending, and alternative model specifications matter. A pattern that seems stable through age twenty-five may change substantially when followed through age fifty.
How are criminal careers measured?
Self-report surveys
Self-report measures ask people about offenses and related behaviors, including events never formally detected. They can capture earlier and lower-level offending, but responses are subject to recall error, interpretation differences, willingness to disclose, and survey attrition. A question about “ever doing” something has a different time scale from a count of acts in the past year. Researchers should specify offense definitions, recall period, confidentiality conditions, and whether respondents understand the behaviors in the same way.
Police contacts, arrests, convictions, and incarceration
Administrative records can provide dates and legal outcomes over long intervals, sometimes with less participant burden. They nevertheless measure the decisions and reach of justice institutions alongside behavior. Police priorities, reporting by victims, prosecution policy, plea practices, diversion, record linkage, and expungement all affect what appears. A conviction is a legal outcome, not a standardized psychological measure of criminal propensity. An arrest represents a recorded enforcement event and is not a finding of guilt.
The US Bureau of Justice Statistics recidivism series offers a particularly clear illustration. Its analyses may report rearrest, reconviction, or return to prison, and these outcomes are derived from different available state datasets. They should not be used interchangeably. A five-year estimate differs from a ten-year estimate even for an otherwise comparable population, while release cohorts from different years face different legal and social conditions.
The observation window and time at risk
A study observing ages twelve through twenty-one cannot know what happens at thirty-five. A person who dies, migrates beyond record coverage, or spends years incarcerated no longer has the same observed opportunity to offend in the community. Time at risk needs to be defined explicitly. Statistical analysis may account for censoring and exposure time, but its quality depends on assumptions about missingness and observation. Apparent cessation can reflect a true behavioral change, limited detection, or loss of follow-up.
Specialization, escalation, and offense switching
An observed career can shift among property crime, violence, drug offenses, public-order offenses, and other legally defined categories. Some individuals show greater concentration in one category and others greater versatility. Apparent escalation might mean an increased number of acts, a change in measured seriousness, or a shift in legal classification; those are distinct claims. A crime's legal seriousness is determined by statute and jurisdiction, whereas psychological harm, opportunity, and behavioral mechanisms require separate assessment.
What predicts onset and persistence—and what does not?
Longitudinal predictors are variables measured before later outcomes that are statistically associated with those outcomes. A risk factor raises average probability under the study's conditions; a protective factor is associated with lower probability or modifies the effects of risk. Neither can automatically be treated as a sufficient cause. Predictors of general onset may differ from predictors of recurrent offending, violent offending, post-release rearrest, or successful reintegration.
The 2022 systematic review of meta-analyses by Basto-Pereira and Farrington supports a multilevel view. Family-related developmental conditions, peer exposure, school and employment problems, and substance use emerged among the relevant predictors depending on the outcome. The review synthesized studies with differing designs and operational definitions, so there is no defensible single risk-factor ranking applicable to every population. Even well-replicated group associations can have limited value for predicting an individual's trajectory.
Protective conditions may include consistent supportive relationships, educational engagement, meaningful lawful opportunities, stable housing, access to healthcare, and reduced exposure to contexts associated with offending. Evidence is strongest when a protective construct is operationalized and its effect tested for a specified population. Protective-factor language should never imply that a person is safe, incapable of offending, or legally responsible for conditions beyond their control.
Psychological traits and diagnoses
Impulsivity, behavioral regulation, sensation seeking, hostility, and other measured constructs may contribute to explanatory models, but psychometric scores vary in reliability and validity across instruments and groups. Clinical diagnoses have their own criteria and purposes. A diagnosis neither constitutes an offense nor supplies proof of future criminal behavior, legal responsibility, or a particular criminal career. Childhood difficulties, neurodevelopmental variation, psychosis, and substance-related conditions should be discussed with careful attention to treatment access, adversity, specific behavior, and the fact that most people with a diagnosis are not defined by criminality.
A developmental theory may describe interacting vulnerabilities and environments without endorsing a deterministic biological story. Criminal acts have specific legal elements, while psychological evidence concerns behavior under particular conditions. In an individual legal proceeding, the presence of a risk factor is neither evidence that the alleged act occurred nor a substitute for legally admissible proof.
Can a criminal career be predicted for one person?
Prediction and explanation are different tasks. An observed association can help explain differences between cohorts without providing an accurate individualized forecast. A risk model must be evaluated for its defined population, time horizon, outcome, discrimination, calibration, missing data, and transportability. Discrimination asks how well it separates people with different outcomes; calibration asks whether predicted probabilities match observed frequencies. A model can discriminate better than chance yet be poorly calibrated in a new jurisdiction.
False positives occur when someone is forecast to reoffend but does not meet the specified outcome during follow-up; false negatives occur when someone is forecast not to reoffend yet does. Both rates depend on thresholds and outcome prevalence, as do positive predictive values. Administrative-record errors, unequal policing, and differences in community support complicate interpretation. A criminal-career trajectory fit retrospectively is especially unsuitable as an unvalidated individualized forecasting instrument.
Forensic opinion must state its referral question, methods, limitations, and relationship to the jurisdiction's legal standard. A probability of rearrest from a reference sample cannot establish guilt, criminal intent, competence, insanity, dangerousness under a particular statute, or whether an identifiable person will commit an offense. Judicial conclusions belong to the responsible legal decision-makers; scientific inference should inform rather than replace legally required reasoning.
Recidivism versus persistence versus desistance
Recidivism usually concerns a new specified outcome after an index event such as release from custody, completion of a sentence, or an earlier conviction. It is a study-design concept, and jurisdictions and agencies operationalize it differently. Persistence concerns continuity of offending across a broader sequence, potentially beginning long before correctional contact. Desistance concerns reduction and cessation across time. A person can show a long history of intermittent offending without registering a post-release recidivism event in a given follow-up window, or a recorded rearrest can occur without a new conviction.
In the BJS study of state prisoners released in 2012, 71% were rearrested within five years across thirty-four participating states. This is a descriptive statistic for a selected release cohort, outcome, and window—not the probability that any US resident, any previously convicted person, or every released prisoner will commit another crime. The study also uses different state subsets for reconviction and return-to-prison measures. Its published numbers therefore require their denominators and definitions.
Because the main question here is the structure of a career, detailed recidivism instruments, calibration studies, post-release intervention effects, and policy comparisons belong in a dedicated recidivism article. The full mechanisms and empirical debates about why people stop offending belong in a dedicated desistance article. Their separation protects the specificity of each search intent.
Practical implications for prevention, courts, and corrections
Prevention across developmental periods
A developmental perspective supports matching prevention to a person's age, circumstances, and actual needs. Earlier support can address educational exclusion, unstable living arrangements, family stress, victimization, substance-related difficulties, or exposure to harmful peer networks without labeling children as future criminals. Interventions should be evaluated with outcomes meaningful to participants and communities, including wellbeing, school engagement, safety, and later offending where ethically and methodologically appropriate.
Individualized assessment in correctional practice
Professionals working with justice-involved populations may examine a documented history, current circumstances, strengths, support networks, education, work, housing, and barriers to participation. Developmental history provides context, but services and legal decisions require contemporaneous evidence and an appropriately validated assessment procedure. Review points should account for change rather than freezing a person in the worst period of their record. Intervention programs need independent outcome evaluation; successful participation alone is not evidence that future offending has been eliminated.
Sentencing, selective incapacitation, and legal fairness
Criminal-career evidence has historically informed debates over incapacitation and selective sentencing. Its ethical challenge is the gap between group-level patterns and uncertain individual futures. Assigning a long sentence mainly because a person resembles a high-rate group risks punishing statistical resemblance instead of legally proven conduct. Sentencing rules, risk admissibility, due-process requirements, and protections against discrimination differ across US states and countries. Psychologists must identify the jurisdiction-specific legal question rather than converting population-level criminology into a universal legal rule.
Reentry and opportunities for change
Reentry planning can consider continuity of care, housing, employment barriers, family relationships, transportation, community supports, and the effects of supervision. A person with a lengthy recorded history may still experience later declines or cessation: long-term evidence cautions against treating childhood risk profiles as lifelong verdicts. Programs should measure relevant outcomes over an appropriate horizon, distinguish technical supervision violations from new offenses, and attend to unanticipated harms as well as intended benefits.
What criminal-career research cannot establish
A sequence of arrests is not a complete sequence of crimes. A statistical cluster is not a fixed human type. An early onset association is not proof of lifelong offending. A clinical diagnosis is not a crime predictor by itself. An apparently predictive score is not evidence that a specific allegation is true. A recorded lull is not necessarily durable desistance. A documented historical correlation between marriage and offending does not prove marriage caused a particular person's change.
Researchers should also separate psychological constructs from legal findings. The questions “What behaviors occurred?”, “What official outcomes are recorded?”, “Which factors were associated with later behaviors?”, “What causal mechanisms are plausible?”, and “What does the applicable law permit a court to infer?” require different evidence. Blurring them can create unjustified certainty in research translation, professional reports, media narratives, and public policy.
Frequently asked questions
Is a criminal career the same as being a career criminal?
No. “Criminal career” is a research framework for events over time, including very short sequences. “Career criminal” is a popular, administrative, or occasionally jurisdiction-specific label, usually implying repeated offending; its meaning depends on who uses it. Neither expression is a clinical diagnosis.
What is the difference between onset and age at first arrest?
Onset refers to a first occurrence under the study's offense definition. Age at first arrest is the first officially recorded arrest. Undetected or unreported acts, differences in enforcement, and record gaps mean those ages may differ.
Does early offending mean someone will keep offending as an adult?
No. Early onset is associated with elevated persistence in many samples, but individual paths are heterogeneous. Long-term cohort research demonstrates later decline even among people with extensive youthful histories. Group probabilities cannot establish a person's future.
Do most criminal careers end?
Many observed offending trajectories decline as people age. Whether a career has truly ended depends on how offending is defined, the age and duration of follow-up, and whether gaps reflect behavior or observation. Criminal-career scholarship studies this uncertainty rather than assuming a universally visible endpoint.
Are there distinct offender types?
Theoretical typologies and statistical trajectory groups can summarize recurring patterns, but they are imperfect simplifications. Their boundaries and numbers vary with data and modeling. They should not be treated as definitive person-level classifications.
What makes criminal-career evidence legally relevant?
It can provide context for research-informed policy and properly scoped expert testimony, such as describing the limits of risk inference or patterns observed in comparable populations. Its relevance and admissibility in a particular case depend on the question and governing law. Neither a trajectory nor a psychometric score can replace proof of the charged offense.
Can a person desist after many years of offending?
Yes. Longitudinal studies document substantial within-person change over time, including among individuals with long earlier histories. How and why that happens varies and may involve opportunities, commitments, relationships, context, and aging. The timing and stability of cessation must be evaluated, rather than inferred from labels.
Is recidivism a measure of criminal careers?
It is one possible segment of a recorded criminal career. A recidivism study starts from an index event and records a specified subsequent outcome; a full career study may begin before that event and cover participation, frequency, seriousness, continuity, and exit.
Conclusion
Criminal-career research turns an offense count into a longitudinal question. It asks when involvement begins, how participation and frequency change, which pathways persist, which decline, and how context shapes those transitions. Its main scientific achievement is showing heterogeneity and the possibility of change while documenting regularities such as the age–crime curve. Its responsible application preserves the difference between behavior and records, association and causation, population patterns and individual forecasts, and psychological evidence and legal conclusions.
For broader context, see Why Do People Commit Crimes? What Psychology Actually Shows.
Related Articles
References
Basto-Pereira, M., & Farrington, D. P. (2022). Developmental predictors of offending and persistence in crime: A systematic review of meta-analyses. Aggression and Violent Behavior, 65, 101761. https://doi.org/10.1016/j.avb.2022.101761
Coles, C., Wickersham, A., Lankester, M., Jolliffe, D., Blackwood, N., & Dickson, H. (2026). Female offending trajectories: A systematic review. Journal of Developmental and Life-Course Criminology, 12, Article 8. https://doi.org/10.1007/s40865-026-00295-y
Durose, M. R., & Antenangeli, L. (2021). Recidivism of prisoners released in 34 states in 2012: A 5-year follow-up period (2012–2017). Bureau of Justice Statistics. https://bjs.ojp.gov/library/publications/recidivism-prisoners-released-34-states-2012-5-year-follow-period-2012-2017
Laub, J. H., & Sampson, R. J. (2001). Understanding desistance from crime. Crime and Justice, 28, 1–69. https://doi.org/10.1086/652208
Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100(4), 674–701. https://doi.org/10.1037/0033-295X.100.4.674
Moffitt, T. E. (2018). Male antisocial behaviour in adolescence and beyond. Nature Human Behaviour, 2, 177–186. https://doi.org/10.1038/s41562-018-0309-4
Nagin, D. S. (2005). Developmental trajectory groups: Fact or a useful statistical fiction? Criminology, 43(4), 873–904. https://doi.org/10.1111/j.1745-9125.2005.00026.x
Nagin, D. S., & Odgers, C. L. (2010). Group-based trajectory modeling (nearly) two decades later. Journal of Quantitative Criminology, 26, 445–453. https://doi.org/10.1007/s10940-010-9113-7
National Research Council. (1986). Criminal careers and “career criminals,” Volume I (A. Blumstein, J. Cohen, J. A. Roth, & C. A. Visher, Eds.). National Academies Press. https://doi.org/10.17226/922
Piquero, A. R., Farrington, D. P., & Blumstein, A. (2007). Key issues in criminal career research: New analyses of the Cambridge Study in Delinquent Development. Cambridge University Press. https://doi.org/10.1017/CBO9780511499494
Sampson, R. J., & Laub, J. H. (2003). Life-course desisters? Trajectories of crime among delinquent boys followed to age 70. Criminology, 41(3), 555–592. https://doi.org/10.1111/j.1745-9125.2003.tb00997.x
U.S. Bureau of Justice Statistics. (n.d.). Recidivism patterns explorer. https://bjs.ojp.gov/recidivism-patterns-explorer