Recidivism and measuring success after prison
In the United States, recidivism rates are the primary measure to evaluate the success of correctional and re-entry programs. Recidivism estimates can be controversial though, particularly given limitations of currently available data. A new report from the National Academies of Sciences, Engineering, and Medicine (NASEM) highlights some of these challenges. A more controversial part of the report argues that the effectiveness of correctional and re-entry programs can be better understood by looking at things like education and employment outcomes rather than focusing on recidivism specifically. Unfortunately, focusing solely on the latter does not tell the whole story, and does not accurately reflect whether a particular program is successful or not.
The National Institute of Justice (NIJ) defines recidivism as “a person’s relapse into criminal behavior, often after the person receives sanctions or undergoes intervention for a previous crime.” Current measures of recidivism are inadequate, though. Administrative data are limited in that they only describe specific legal system actions (e.g., arrests, convictions, incarceration). These data don’t capture crimes that go undetected, nor do they capture unsolved crimes. Additionally, rates might be skewed by inconsistencies of the criminal justice system, such as variations in police presence and discretion in terms of who to arrest. National recidivism rates are calculated using administrative data, and the wide variation in estimates reflects some of these inconsistencies, with three-year estimates ranging from 30% to 60% and five-year estimates ranging from 38% to 71%. Administrative data does have advantages though, such as allowing for larger samples and more advanced statistical techniques.
In addition to administrative data, self-report data are sometimes used to supplement administrative records. However, self-report data can be costly to obtain and also suffer from problems related to precision and recall. For example, respondents’ reports may be influenced by characteristics of the interviewer or length of the survey instrument. Other times, descriptive terms aren’t universally understood by all respondents, such as when a person inaccurately characterizes a burglary as a robbery. Finding participants that will consent to an interview is also challenging, especially when people in this type of sample are typically highly mobile. Those who do consent to interviews still might under-report criminal activity or report it as less serious than it was.
Administrative data are the records maintained by criminal legal agencies (e.g., law enforcement, jails, courts, etc.) to document the activities of their agents and personnel. This is how criminal history information (AKA, a “rap sheet”) is recorded. A completed criminal history record typically contains information about arrest charges, dispositions, sentences, and custody dates. These records provide researchers with the most comprehensive and accessible source of data on recidivism as measured by criminal legal agencies, but there are a number of reliability concerns related to things like missing or purged records and state-level variation regarding the details that are reported to repositories. Thus, the completeness of the criminal history records varies among states. In 2018, 49 states reported having final disposition data for 68 percent of all arrests in state databases.
As stated above, one major problem with administrative data is that it doesn’t reflect the large portion of crime that goes unreported. Victimization surveys indicate that this represents a large portion of criminal behavior. In 2020, results from the National Crime Victimization Survey (NCVS) showed that victims reported only 40% of violent victimizations and 33% of property victimizations to the police.
Administrative data on recidivism also will not capture information on unsolved crimes. If a crime is unsolved, the perpetrator is not held accountable and there is no record of them committing the crime. Unfortunately, unsolved crimes account for a large percentage of overall crime. Federal statistics indicate that, in 2019, only 45.5% of violent crimes and 17% of property crimes were cleared. Generally the more serious the crime, the more likely it is to be cleared. In 2019, 61% of murder offenses were cleared, in comparison with 52% of aggravated assaults, 33% of rapes, and 30.5% of robberies. These numbers put into perspective how many violent and nonviolent perpetrators remain unidentified.
Recidivism is also limited in that it is a binary measure, typically indicating whether one has been re-arrested, re-convicted, or re-incarcerated at a certain point in time. This measure doesn’t account for contextual details that might predict recidivism from person to person. For example, people who have been incarcerated once are less likely to reoffend when compared with people who have been incarcerated multiple times.
Another persistent problem that there is no shared definition of recidivism. One re-entry program might measure recidivism in terms of re-arrests and technical violations, while another may measure only re-incarceration. A 2017 study compared recidivism outcomes among a sample of felony offenders sentenced to prison versus sentenced to probation, while controlling for crime severity. They found that most instances of recidivism stemmed from parole violations rather than new felony convictions, showing how technical violations can inflate recidivism rates.
Alternatively, one state might track recidivism outcomes for five years, while another may track it for three years. Consequentially, it is difficult to reliably compare recidivism rates across programs or across jurisdictions.
Efforts to Improve Administrative Data
Several efforts to improve administrative data focus on establishing common definitions of data elements. When states use common outcomes but measure them differently, comparisons between states are unreliable. Efforts to develop national standards for criminal offenses as part of law enforcement statistics are ongoing through the Uniform Crime Reporting (UCR) Program’s National Incident-Based Reporting System (NIBRS), which provides incident-level data on crime incidents and multiple attributes of arrests, victimization, and individuals involved in the criminal legal system. Unfortunately, the reliability of NIBRS data across participating jurisdictions has not yet been fully assessed.
Efforts to develop standards for state courts include those promoted by the National Center for State Courts through its National Open Court Data Standards, which supports the creation, sharing, and integration of court data by developing rules by which data are described and recorded. Under this effort, states may still define events differently, but the differences would at least be documented. The National Open Court Data Standards has not been implemented yet, though.
Within states, the state criminal history record repositories integrate arrests with prosecution and adjudication outcomes, but wide variation exists among states in the completeness of records. The repositories have a program for enhancing data quality, known as the State Repository Records and Reporting Quality Assurance Program, which offers voluntary standards for information maintenance and reporting requirements. Unfortunately the data quality issues associated with this program have not been assessed at this time, though efforts are currently underway.
National and State-Level Recidivism Estimates
Bureau of Justice Statistics (BJS) studies are the primary source of national statistics on recidivism. Using data from state and federal criminal history records, reports provide statistics on large samples and include several different measures of recidivism, including re-arrest for a new crime (in-state or out-of-state), re-conviction, and re-incarceration. While there is no way to calculate national recidivism rates due to the limited number of states providing data, the BJS reports are the closest option. BJS has standardized data collection for 34 states, representing 79% of all individuals released from state prisons.
One of BJS’ recent recidivism studies included 92,000 people released from prison in 34 states, all of whom had served a sentence of one year or more. In the sample, over 33% were arrested for a new crime within one year of their release, 60% were re-arrested within three years, and 71% within five years. Importantly though, these results are not nationally representative (though they are generalizable to the 34 states in which data were collected).
Studies conducted by the U.S. Sentencing Commission (USSC) are another official source of recidivism estimates for people released from federal prison. These studies track re-arrest rates for up to eight years using data collected by the USSC along with criminal history records. One of their more recent studies examining 32,000 federal offenders found that 18% of people were re-arrested within one year of their release, and 49% were re-arrested within eight years. Violent offenses comprised about 31% of re-arrests. Criminal history was strongly associated with recidivism, with offenders in the most serious criminal history category having the highest re-arrest rates at 76%.
The Office of Probation and Pretrial Services (OPPS) of the U.S. Courts has provided some estimates on national-level recidivism estimates, but not in recent years. A 2013 study by the OPPS followed 245,000 offenders on federal community supervision, where they found that 30% of offenders were re-arrested within three years and 38% of offenders were re-arrested within five years. They found several risk factors associated with recidivsm, including criminal history, drug abuse, mental health issues, and unemployment. On the positive side, they found that certain factors decreased recidivism risk, such as having a strong social support system, marketable skills, and motivation to change. The study also found that several community-level factors affected recidivism rates, such as population size and average household income. For example, offenders who returned to impoverished neighborhoods had higher recidivism rates.
State Departments of Corrections (DOCs) also use multiple measures of statewide recidivism. For example, the Pennsylvania DOC reports on re-arrest and re-incarceration rates and breaks these down by variables such as offense type, demographics, and geographic location of releases. But state estimates are generally too imprecise to draw meaningful conclusions.
When reviewing these types of studies, pay attention to what types of crimes are excluded or included in the analysis (e.g., technical violations, traffic offenses). Furthermore, know that every definition will underestimate the “true” recidivism rate because rates are based on official data and do not account for crime that goes undetected.
Academic Literature on Recidivism
Many academic studies on recidivism rely on official records, but they vary in the specific measures used, the level of severity that constitutes recidivism, and the time periods over which recidivism is measured. For example, a 2021 study of recidivism following California’s Public Safety Realignment Initiative examiend only felony re-arrests, thereby limiting their analysis to more serious offenses. Another example is when re-incarceration is the primary outcome, but results are disaggregated by type of return, i.e., for a parole violation vs. for a new crime. Disaggregation by class or severity can help researchers better understand variation in recidivism measures. A 2017 study examining both re-conviction and re-imprisonment found reductions in re-incarceration rates, but not in re-conviction rates. By disaggregating data, the authors were able to attribute the difference to technical violations rather than new crimes.
A major challenge to assessing the causal impact of incarceration on recidivism is selection bias. A primary example of this is that judges are typically more likely to hand out lengthier sentences to people who are already higher-risk. Older studies have attempted to account for selection bias with the use of matching and regression techniques to control for measures such as age, sex, offense type, and prior records. However, these types of studies are still susceptible to unmeasured covariates affecting selection bias and recidivism. More recent studies use “instrumental variable regressions” to exploit naturally occurring variability in the use of incarceration, often in the form of random assignment of cases to judges. The random assignment helps ensure that both unmeasured and measured case characteristics (e.g., criminal history, offense seriousness) are the same across judges. Judges with identical caseloads but differential use of incarceration can be compared to see if recidivism differences are caused by differential uses of incarceration. Other times, studies will exploit naturally occurring variation across sentence severity as determined by pre-existing sentencing grids.
A 2022 research review that examined judge instrumental variable studies found both null impacts on recidivism as well as recidivism reduction effects. Though, the correctional programming provided in prisons is so inconsistent that it is hard to know which types of correctional experiences are linked to reduced offending. The research review mentions that recidivism reductions were more likely among those who received rehabilitative correctional programming, but it is unclear exactly what these rehabilitative programs were, whether they applied to all offenders or just a subset of offenders, or whether the programming was implemented in a prison or community-based setting.
This brings us to another problem that is rampant in incarceration research: defining the right population and sample. Some individuals will be more susceptible to change, and this can also vary based on differential circumstances. If a study focuses on people who are on probation and receiving drug treatment, versus an in-prison substance abuse program, the differences in recidivism can be large. This is because offenders sentenced to probation versus prison differ substantially from each other, and respectively, their treatment programs are also not comparable. When it comes to recidivism, it can be extraordinarily difficult to know what types of correctional programming are effective in reducing recidivism, and for whom.
There are various risk factors that can reliabily predict an post-release criminal behavior, but differences in the type of risk factors might explain why some people can be rehabilitated while others cannot. Namely, the difference between static and dynamic risk factors is worth considering. “Static” risk factors are risk factors that are either not subject to change or are not amenable to intervention (e.g., age, criminal history) whereas “dynamic” risk factors are modifiable and might be amenable to intervention (e.g., substance abuse problems, serious mental illness). Because dynamic risk factors are more amenable to change, it probably makes sense to focus rehabilitative efforts on people with more dynamic, rather than static, risk factors.
Correlates of Recidivism
Criminal offending of formerly incarcerated individuals tends to correlate with neighborhood context, particularly concentrated disadvantage. Structural disadvantages weaken interpersonal ties among residents and thus weaken the capacity of neighborhoods to act collectively and to regulate behavior. As a result, repeat offending is partly a consequence of a neighborhood setting that lacks the capacity to exert informal social control over unwanted behavior. Concentration of a large number of formerly incarcerated individuals in one area can also contribute to erosion in neighborhood structure and culture, both through the housing market, collective legal cynicism, and legal estrangement.
Conversely, living in a neighborhood with ample resources may mitigate negative outcomes. In one 2020 study, community cohesion (i.e., one’s networks and social ties to the community) was protective against returning to prison. However, this impact was dependent on neighborhood-level resources. This might explain why people released from prison fail when they return to resource-depleted communities. Another study found that former prisoners who received more support from parole officers were less likely to recidivate than those who received less support from parole officers.
Housing instability and homelessness can also contribute to an increased likelihood of recidivism, but it is unclear whether a causal relationship exists, as housing issues are often related to other criminogenic needs, such as severe mental illness or substance abuse histories. Mental illness and substance abuse problems are very common among ex-prisoners and can certainly increase recidivism risk. Returning individuals often identify drug use as the primary cause of many of their past and current problems including family, relationship, employment, legal, or financial problems.
Another major risk factor of recidivism is gang involvement. A 2021 study found that current gang members had the highest risk of recidivism in terms of re-arrests, re-convictions, and re-incarcerations. Former gang members were more likely to get re-arrested than non-gang members, but there were no differences in rates of re-conviction or re-incarceation.
Employment is another large concern for many formerly incarcerated people. Many believe that employment can reduce recidivism among ex-offenders, but the evidence is mixed regarding causality. More nuanced research shows that it might not be employment itself that prevents recidivism, but rather, the type of employment and amount of earnings. For example, a 2022 study found that formerly incarcerated people with higher paying jobs were less likely to recidivate than those with lower-paying jobs. In that study, workers in the top quartile of wage earnings were half as likely to be reincarcerated, while the lowest-paid individuals returned to prison as often as their unemployed counterparts. A 2017 study found that exposure to construction and manufacturing opportunities (that generally pay more) was associated with significant reductions in recidivism, whereas exposure to lower-wage opportunities had no influence on recidivism. Importantly, studies with longer follow-up periods have found more modest reductions or no effect on recidivism.
Strong family ties and social bonds have been widely documented as an important factor in successful re-entry. Those with family supports and emotional supports have a lower risk of post-release offending. Even visitations by family members in the months leading up to release from prison are correlated with a lower probability of post-release criminal behavior. Such family support is also associated with higher rates of employment, lower rates of substance use, and fewer mental and emotional problems. In theory, these factors all coupled together may lead to decreased recidivism. Instrumental support provided by family (e.g., housing, employment, transportation) also might lower the risk of recidivism, though this finding has been met with contention in some research. Rebuilding family relationships and being around people not involved in criminal behavior also, not surprisingly, can reduce recidivism. In any case, it is unclear what the causal mechanism is between social support and cessation of criminal behavior. This is something worth exploring further in future research.
For people without prosocial bonds, re-entry will be much more difficult. Hardships experienced during the reentry process, such as trouble with familial and peer relationships, can lead to negative emotional states and maladaptive coping. With this in mind, it is not surprising that family conflict is correlated with higher risk of post-release criminal behavior and substance use. Many individuals returning home from prison will return to families and communities facing similar circumstances that resulted in their incarceration, which can be a barrier to forming prosocial bonds. Once again, the importance of being around people not involved in criminal behavior cannot be stressed enough in terms of preventing recidivism.
Correctional Programming and Recidivism
When it comes to knowing “what works” in terms of correctional programming, the NASEM report makes it sound like there are several programs with beneficial effects. The authors claim that inmates would be better prepared for release if they received more support in their re-entry process, such as assistance finding housing, employment, or drug treatment. In theory, if individuals can make constructive use of their time in prison, they will be better able to return to society and desist from crime after release. Unfortunately though, the current evidence on desistance-focused correctional intervention options is still very mixed and correlational, with little causal evidence regarding the effects on recidivism. Correctional programs can be divided into those that focus on internal change (e.g. therapy, drug treatment) and external change (e.g., employment, education).
Various evaluations have found that employment-focused interventions (e.g., transitional jobs, job training) may improve post-release employment outcomes, but such programming only minimially affects longer-term employment and has little impact on recidivism. Regarding prison labor, the evidence supporting recidivism reduction effects is minimal. Although some older research has reported that prison employment reduced recidivism, other studies have not found significant effects. As stated above, a growing body of evidence suggests that not all jobs will have the same impact on recidivism; rather, finding high-quality employment (not just any employment) is a more important determinant.
Educational opportunities may help individuals in myriad ways, such as supporting personal growth, development of new interests, increased mutual support, and positive socialization. But whether educational programs reduce recidivism varies based on the individual. For example, prison education seems to be more effect for people with large education deficits. Effects are also greater when participants actually complete the program, though, this presents a selection bias issue in terms of estimating causal effects. In other words, the type of people participating (and completing) these types of programs are likely different from those who do not, so it is hard to know whether successes are related to the programs themselves or if there is something about those individuals (e.g. greater motivation to change) that makes them more successful. For example, students may initially enroll in an educational program for extrinsic reasons, but those who actually graduate from the class are intrinsically motivated and realize that they are enthusiastic about learning.
Many purport that psychological interventions such as cognitive behavioral therapy can reduce recidivism. For example, a 2021 research review of 29 randomized controlled trials (RCTs) on rehabilitation programs found that psychological interventions were associated with reduced recidivism. However, this effect disappeared when small studies (<50 participants in the intervention group) were excluded. Cognitive behavioral therapy has also been regarded as an effective approach in reducing recidivism, but this tends to be more effective for youth, while the impacts on adult recidivism remain unclear. Relatedly, many people believe that substance abuse treatment greately reduces recidivism, but this doesn’t appear to always be the case. A 2018 meta-analysis on in-prison substance abuse treatment found that medication-assisted therapies reduced substance use, but treatment didn’t lead to significant impacts in recidivism.
Among the research on correctional programming, there is a lack of specificity regarding the causal mechanisms that might lead to reduced recidivism. Many people presume that correctional programming will influence intermediate outcomes, such as employment, stable housing, and substance use, which will then reduce recidivism, but these findings have not been validated. Taking this into consideration, it is premature to assume that outcomes like employment, stable housing, and reduced substance use are indicative of “success” after release, when the research is unclear as to whether these factors reduce recidivism.
The current report by NASEM makes very good points, particularly about the limits of recidivism measures. However, the conversation about “other kinds of success” merits some critique. The committee views post-release success through the lens of flourishing and well-being. As part of this approach, they say that we need to look to things like employment, housing, family relationships, and health outcomes as measures of success rather than looking at recidivism as the only possible indicator of success. But given the research discussed above, this is likely not capturing the whole story. While education and employment outcomes may be useful in their own right, they don’t necessarily support the cessation of criminal activity.
In sum, individuals are likely to vary as to what factors and behaviors are more (or less) important for their overall reintegration, and this is probably why the research has been so mixed.
Future research should use more rigorous methods to examine the impacts of correctional programming, as most existing research on this is correlational and relatively weak. It is tempting to focus on these non-criminal-justice outcomes as ways to measure progress or marginal success (e.g., changes in substance abuse, communication skills, etc.) because they are easier to affect, while recidivism rates are more long-term. Unfortunately though, failing to look at recidivism rates doesn’t tell the whole story of “success” and this presupposition could lead to poor policymaking.
The evaluation of success is also complicated by the lack of shared definitions. As a result, it is difficult to reliably compare recidivism rates (or other post-release outcomes) across different programs or different jurisdictions. Uniform national standards for measuring success among individuals released from prison would support program evaluations and improve the utility of administrative and other data across multiple policy domains.
Further, the research on recidivism is hindered by restrictive data-sharing practices across criminal legal institutions. Most prominently, there is no ability to link data across agencies and policy domains. Linking data across multiple recordkeeping systems would facilitate the development of national standardized ways to measure success among formerly incarcerated people. This would enhance the ability to compare outcomes across jurisdictions.
More immediately, the academic community could develop a website of core success measures, instruments, and validation studies from multiple administrative domains that is accessible to researchers, practitioners, and policymakers. Such a toolkit could be developed by a combination of partnerships between private foundations and government agencies.
Broad generalizations about “the recidivism rate” need to be avoided. Rather, recidivism rates should be connected to their study populations and to the purpose of each inquiry. Because there are many recidivism events that can be measured, the general term “recidivism” needs to be explicitly defined (e.g., re-arrest, re-conviction, etc.). Similarly, cross-jurisdictional comparisons of recidivism rates are subject to misinterpretation.
Better data can play a key role in informing policy development and ensuring more effective programming for those in or recently released from prison. Given the rehabilitative function of prisons and the extensive network of reentry supervision and programming, improved measurement can also enable correctional and reentry leaders to better identify program and policy impacts, document successes, and refine best practices.
Authors contend that post-releases success is multi-faceted and shouldn’t just be based on recidivism. The other outcomes they mention that might be indicative of “success” are employment, housing, health, family and community attachment, and personal well-being. Less is known though about how these factors are associated with recidivism though. Thus, relying on these types of measures as indicators of success is less than ideal. From a policy standpoint, recidivism outcomes are still very important for determining whether correctional programs and policies are successful.
Still, there are a few factors that seem to decrease recidivism that should be looked into more. Firstly, high-quality, full-time employment with decent wages might reduce risk. Secondly, social support and prosocial connections seem to decrease recidivism risk as well. Psychological interventions seem to help for some individuals, but this is more impactful for juveniles rather than adults.