The U.S. now incarcerates 25 percent fewer people than it did at its peak level of incarceration in 2009. Although this decline exceeds most projections from even 10 years ago, Stanford professor Keith Humphreys recently predicted in the Atlantic that even this smaller number would soon “fall off a cliff,” leaving us with a prison population that is 60 percent smaller than it used to be.
This decline has alarmed some people on the right, who believe we are now incarcerating too few people, rather than too many. All but the most extreme reformers accept that some number of people need to be incarcerated. The question then becomes, what is the optimal level for our society?
Ultimately, this question is not answerable scientifically. But there are some important trends in the current data that make me believe incarceration can (and will) continue to fall without any increase in crime. By the same token, any attempt to reverse the trend and incarcerate more people will not actually make us safer.
The biggest hint that the incarceration decline can safely continue is that it has been consistently paired with declining recidivism rates. This should have been expected in the early stages, assuming we began by intentionally releasing the least risky people. But, after that initial period, we might have expected that recidivism rates should increase, since only the riskiest people would be in prison. But that is not what has happened – instead, recidivism rates have continued to drop throughout the decline in incarceration.
Something else is happening, too. I believe the decline in prison populations is being driven by a seismic change in the crime rate, and subsequently the incarceration rate, of young people. In 2024, the arrest rate for 40-year-olds was higher than the arrest rate for teenagers — a fact that sounds silly to my criminologist ears, but is nonetheless true. Continuing to fight crime with the methods from 20 years ago is like air-dropping water on a forest where the fire has gone out. Policies intended to reduce crime must first reckon with the reality that among the most crime-prone population (i.e., young people), crime has already plummeted to rates previously thought impossible. Just as importantly, policy should be designed with humility about our ability to predict the behavior of individual humans.
The seductive siren of selective incapacitation
Going back 40 years, criminal-career researchers have repeatedly found that they can approximate the overall population of people who get arrested and convicted by modeling two populations – a large group of low-frequency, occasional offenders who have a good chance of staying out of trouble at any given moment, and a small group of high-frequency, chronic offenders with a low chance of keeping on the straight and narrow at any given moment. In fact, one of the keystone findings of criminology is that around 5 percent of the people who are ever arrested are responsible for about 50 percent of the arrests in an area. This reality creates a seductively attractive proposition – what if we could “selectively incapacitate” those chronic offenders relatively early in their criminal careers? This is the logic behind Three Strikes laws and other sentencing policies that are linked to prior history.
Unfortunately, we are not very good at sorting people with records into those who will offend more in the future and those who will not. This is not an unusual problem – it is hard to differentiate between members of highly selected groups. Noted criminologist Alfred Blumstein liked to point out that SAT scores were far better at predicting who was going to get into Harvard than who was going to be successful at Harvard.
Consider Covid-era policies that restricted the authorities’ ability to revoke parole – that is, to send parolees back to prison for violating their terms of release. If revocation policies were being used wisely to send back to prison the people who were most likely to commit new crimes, then limiting those powers during the pandemic should have led to more reoffending. This simply didn’t happen. The rate of new crimes committed among people released from prison did not rise when revocation rates fell.
This pattern also appears in pretrial detention. Pretrial reforms in New Jersey and New York were met with significant backlash, based in part on claims that they would result in more total crimes committed. Such claims are mathematically valid – setting loose 200 people with a probability of rearrest of 30 percent will generate more arrests than freeing 100 people who have the same probability of rearrest. But – and this is crucial – if the prior system was in fact detaining the highest-risk people (the “right” people), then those released under the new system should have a higher rate of rearrest than those released under the old. This is not what has happened; instead, the pretrial rearrest rate has remained constant. Our pretrial system is not detaining the highest-risk people.
Imagine a basketball team that decides they will double-team a member of the opposing team – but then chooses the target of their double-team at random. They can state, with a completely straight face, that they succeeded at preventing the targeted player from scoring. That straight face will eventually become a frown when they lose, repeatedly, because they often leave the best player unguarded.
Why are we so bad at identifying the riskiest people in the system? The main reason is that modern sentencing is not actually designed for it. Rather than seeking to surgically incapacitate the riskiest people, modern sentencing policy is driven by retribution. Within this framework, policymakers punish people based on the seriousness of the crimes they commit. Unfortunately, type of crime isn’t very predictive of recidivism risk after controlling for age and prior history. People who commit crimes do not specialize in certain crimes, nor do they systematically build towards more serious crimes. More frequent offenders are also more likely to have a violent offense for the same reason that they spend more time in prison – they offend more, period.
The way we talk about people in the criminal justice system reveals that we desperately wish that people did in fact specialize. People who commit rape are rapists, people who commit burglaries are burglars, etc. However, wishing won’t change the weather, and specialist labels, no matter how descriptive – drug dealer, violent offender – do not help predict what the person will do in the future. A retributionist sentencing system that assigns punishment primarily based on crime type will not incapacitate the most frequent offenders.
To be fair, modern sentencing systems do also focus on number of priors when assigning sentences. And, those with more priors do in fact have higher rates of recidivism even after serving long sentences. At least in principle, we could potentially get even more benefit with even longer sentences for those with more priors.
However, this strategy has an Achilles heel – the confounding problem of age. Five prior offenses at 25 is a good sign that a person is a frequent offender – five prior offenses at 45 is not. A system that assigns longer sentences to people with more priors will inevitably incarcerate older people more often than younger people – and these older people will have lower rates of offending either because they age out of crime or because they were simply a low-rate offender who distributed their occasional lawbreaking over a long time. In fact, the aging of our prison population is the key to the puzzle we began with: declining prison populations paired with declining crime rates.
The mystery of declining recidivism rates in a time of declining incarceration
Suppose we decide to reduce our prison population by only incarcerating the most crime-prone offenders. The smaller prison populations will be followed by higher recidivism rates. This makes sense – the prison population after the policy change contains a higher percentage of high-frequency offenders. Now consider the current situation in the U.S., where we have simultaneously reduced incarceration and seen a sizeable drop in recidivism rates among those exiting prison. This is very strange. What is happening?
The answer is both surprising and simple. The proportion of the prison population under the age of 30 plummeted between 2000 and 2020: from 44 percent to 24 percent. Younger people have higher recidivism rates than older people, so recidivism rates will drop as the age of the average releasee increases.
Why are we not incarcerating young people? Have we gone soft on crime? Nope – young people are simply not committing crimes at anywhere the rate they did before, a trend observed in both official and self-reported data from multiple countries. The crime rate among young people peaked during the crime boom of the 1990s and has been plummeting ever since.
This very fact, that the offending rates of different birth cohorts are influenced by crime booms, should send up warning flares about incarceration policies that focus on the number of prior offenses a defendant has on their record. Whole birth cohorts will acquire records at dramatically higher rates – and then have these records follow them into the future, even after the crime boom has ended. My colleague David Bjerk and I have shown that crime rates in a person’s state of residence when they are 12-15 years old help predict whether they are incarcerated at 30. An elevated rate of crime during the years when a person is at highest risk of crime involvement will generate a prior record that will then trigger longer sentences later in life. Essentially, a focus on prior records in combination with a crime boom that ensnares a large number of young adults will reverberate through the system for 20 to 30 years after the crime boom is over, even if the underlying arrest and conviction rate for this birth cohort in their later years is not any higher than for prior birth cohorts.
This helps explain the odd reality that American incarceration rates continued to increase for 20 years after the peak of the crime boom. This phenomenon was almost entirely driven by the incarceration of older adults. But as explained above, the older adults who are incarcerated only because of their priors will, in fact, have lower recidivism rates. That means these “hangover” incarceration rates that are generated based on priors accumulated years earlier will not have anywhere near the crime reduction benefit as earlier incarceration rates. At the same time, new policies that try to extend the incarceration periods of people who are currently cycling through spells of imprisonment will also not have large crime-reducing benefits. Those currently incarcerated have a median age of 39. Extending their sentences will incapacitate them well past their peak offending years.
Absent another crime boom, incarceration rates will continue to drop in the U.S. as members of Gen X, who came of age during the crime boom of the 1980s and 1990s, limp out of prison for the last time and are not replaced by members of Generation Z. Recidivism rates will continue to decline. Policies that try to target repeat offenders in the current environment will inevitably capture older offenders.
The key, in my view, is to stop thinking that all people are either innately good or bad. While it is true that crime is a concentrated phenomenon, both among people and places, it is also true that the actual levels of crime vary dramatically over time among similar people who live in similar places. This suggests that crime levels are highly dynamic, and the same person in a different time or place would have a different propensity to commit crime.
My friend and RAND colleague Roland Neil likes to pose this rhetorical question at dinner parties. Why do people always ask why crime has dropped among young people, but no one ever asks why young American men stopped killing Japanese and Germans after WWII ended?
The answer to the latter question is obvious – the war was over. Americans had not suddenly become less violent. Roland’s point is that the never-ending search to explain why young Americans are “less violent” than they were 30 years ago is pointless. The people are basically the same – but the circumstances that created the high violence rates of the 1980s no longer exist. The end was neither as clear nor as definitive as the signing of the Instrument of Surrender on May 7, 1945, but the unique circumstances leading to the problems did in fact end. Incarcerating people who participated in this uniquely violent situation because they participated in that situation, and not because of their current behavior, makes little sense.
The end of prison poker
The evidence is overwhelming that our current policies are not good at identifying and then incarcerating the high-risk people who will commit more crime in the future. Continuing to call down in poker even when presented with evidence that you are not good at poker is ultimately going to lead you to bust out. Continuing to invest in increased incarceration in the face of evidence that we do not regularly identify the most crime-prone will be far worse than going broke at the poker table – we will deprive people of their freedom, in all likelihood in a racially disparate way, with no real marginal benefits in terms of crime reductions. This is particularly true in a peaceful social context where our young people’s rate of offending is the lowest it has been in 50 years. Let’s focus on finding ways to maintain the changes that have created these more peaceful conditions in our most disadvantaged communities, rather than doubling down on the most expensive and bluntest instrument in our toolkit.
Shawn D. Bushway is a professor at the University at Albany, Rockefeller College of Public Affairs and Policy.
