A post on X that went viral recently laid out a series of statistics about the percentage of serious crimes — murder, rape, robbery, assault, and so on — that are committed by people with prior arrests. All hovered between 60 percent and 79 percent. The post’s conclusion: “You can incarcerate your way out of crime. Facts.” Elon Musk, the platform’s owner, amplified the post to his hundreds of millions of followers and sharpened the point: “Either incarcerate or innocent people suffer.” To date, these two posts have nearly 50 million views each.
The claims in these posts are worth unpacking. First, Musk uses the correct metric: Reducing the suffering of innocent people is the proper goal of any criminal justice system, and public safety policy should be evaluated primarily by that standard. Musk is also correct in an important, albeit limited, sense: Failing to incapacitate genuinely dangerous people will lead to some level of crime and suffering that would have otherwise been avoided.
However, it’s a mistake to end the argument there. As economists like Thomas Sowell have noted, in public policy there are no solutions, only tradeoffs. And the relevant question in policy analysis is not “Does it work?” but “Compared to what?” Musk’s claim that incarceration works to control crime is the beginning of the analysis, not the end. The next question is whether it works better than the available alternatives.
Taking the original post literally leads to absurdity
The original X post argues that we can incarcerate our way out of crime because large majorities of those responsible for large fractions of serious crime have at least one prior arrest. The necessary implication is that the lever by which to achieve that goal is to incarcerate every arrestee for as long as it takes to reduce their threat level to zero.
If the U.S. adopted this strategy tomorrow, we’d take the 5 million people who will be arrested over the next year and put them all in prison for, say, 15-20 years. In year two, arrest totals would drop (perhaps substantially, since some fraction of offenders from year one would already be in prison in year two). But every year, new criminals begin their careers, some replacement occurs in drug markets, and a large number of low-rate offenders who escaped arrest in year one get caught in year two. The annual arrest numbers would never drop to zero, but the prison population would grow, year after year, because everyone would be held for decades to ensure they wouldn’t be released while they could still pose a threat to public safety. Within a few years, tens of millions of people would be behind bars; the huge majority would be low-rate and low-level offenders being warehoused even if it’s extremely unlikely that they would reoffend. Even a conservative estimate would put the cost of incarceration at that scale at hundreds of billions of dollars annually, dwarfing every other crime control investment combined.
So, taking the original post literally leads to obviously unserious policy implications. But what about a less literal interpretation? Perhaps the post’s author (or Musk) didn’t intend to imply that the U.S. should incarcerate every arrestee for decades. Instead, perhaps they just meant that since so many of those responsible for serious crimes are repeat offenders, we can incarcerate our way out of crime by targeting just those who are convicted of serious crimes, or who have racked up repeated felony convictions.
Selective incapacitation is insufficient
Criminologists call this strategy “selective incapacitation.” It is the logic behind many popular crime control strategies, most notably “three strikes” laws and other sentence enhancements based on an individual’s criminal history. These policies have obvious intuitive appeal. After all, one of the keystone findings of criminology is that roughly 5 percent of people who are ever arrested account for about 50 percent of all arrests. If we can incapacitate just that small group of prolific offenders, the public safety returns would be enormous and at relatively little cost.
Unfortunately, selective incapacitation is not the panacea some believe it is, for several reasons.
The first problem is epistemic. The familiar claim that a small fraction of offenders is responsible for an outsized proportion of crime is a retrospective observation, not a usable sorting rule ex ante. It tells us that, once criminal careers have fully unfolded, a small number of people will turn out to have generated a large share of criminal behavior. But selective incapacitation requires the ability to predict, for any given offender at the moment of sentencing, that offender’s future criminal career. The leap from general description after the fact to specific prediction before the fact is where selective incapacitation goes awry and collapses into speculation.
This doesn’t mean we have no idea, or that we can’t make reasonable guesses as to the likelihood of recidivism. Criminal history, for example, is an obvious potential proxy for future dangerousness. But it is an extraordinarily blunt one, imprecise and highly susceptible to misjudgment. Arrest records are not clean measures of dangerousness; they reflect policing patterns, surveillance intensity, local enforcement priorities, criminal incompetence, and bad luck.
In a landmark study of who actually gets caught committing crime, Jan and Marcia Chaiken found that prisons contained a significant number of what they called “low-rate losers” — people who commit relatively few crimes but are arrested nearly every time. Based on arrest records alone, the Chaikens found that it is nearly impossible to distinguish this cohort from genuinely prolific offenders. Two people with identical rap sheets can have wildly different actual offense rates. Even putting that aside, two offenders with nearly identical records at age 22 can follow different paths and reach different places by age 25. One may continue offending frequently, while the other finds steady work, enters treatment, forms a family, or simply stops committing crime. In fact, most people who commit crimes desist permanently without ever going to prison.
Of course, one might argue that repeat offender laws don’t use arrest records as proxies for dangerousness, but instead use convictions for serious felonies. Surely, the argument goes, a person with three felony convictions has shown an inability to live among the rest of us peacefully, or at least lawfully, and a long prison term is therefore justified on incapacitation grounds. But the epistemic fog covers this cohort, too. A person who accumulates three felony convictions over two years in his early twenties presents a very different risk profile from a person who accumulates three felony convictions over 15 years and is now in his forties. A sentencing scheme that assigns longer sentences on the basis of accumulated “strikes” does not take into account this critical distinction. It is designed to catch the people with the longest records, not the people who are currently the most dangerous.
Imagine an NFL general manager who decides to award the highest salaries to the players with the most career touchdowns. On its face, the logic seems sound; past performance is correlated with future performance. But in practice, the players with the most career touchdowns are more than likely the oldest players on the roster: veterans past their prime, more injury prone, and nearer to retirement. Paying them top dollar based on what they did five or 10 years ago, rather than what they are likely to do next season, would be a catastrophically bad investment and set the franchise back years. Three-strikes laws make exactly this mistake with prison beds instead of salary cap dollars.
The inability to rationally sort current offenders by future dangerousness is the unanswerable problem for selective incapacitation. When the target group is small, even a prediction rule that looks respectable in the aggregate will generate large numbers of false positives in practice. To capture most of the people who really will go on to account for a disproportionate share of crime, sentencing laws have to cast a much wider net in the hope that the high-rate targets are somewhere in the catch. But widening the net is exactly the problem selective incapacitation is intended to avoid. It means imposing long sentences not just on the rare true high-rate offenders— the 5 percent who are responsible for 50 percent of the crime — but on many others whose future offending would have been sporadic, modest, or quickly abandoned.
The problem is compounded by a further, less obvious error. Proponents of expanded incarceration routinely infer the future offense rates of the people they want to lock up from the offense rates of people already in prison. But research has shown that the free offender population has a mean offense rate at least an order of magnitude lower than that of current prisoners. Widening the net, in other words, does not just capture more people. It captures people who offend far less frequently than the ones already behind bars. That means the public safety return on each additional prison bed is smaller than the already modest estimates would suggest. And every bed occupied by an offender from two decades ago is unavailable for today’s dangerous criminal. As the late Mark Kleiman, a professor of public policy who wrote extensively about crime, observed, a prison bed used to house one offender for 20 years could also be used to house 20 offenders for one year. The net result of this inefficient allocation of scarce public safety dollars is more crime-related suffering, not less.
None of this should be read to imply that incapacitation is unnecessary or inherently inappropriate. Some offenders are clearly dangerous enough that removing them from the community prevents serious harm, and individual judges should be free to impose harsh sentences when appropriate. The difficulty is not recognizing that such cases exist, but identifying them in advance with enough precision to make broad sentencing enhancement policies effective.
What does all this add up to? The late William Spelman modeled all of this exhaustively in his book Criminal Incapacitation, the most thorough analysis of the subject to date. Spelman’s conclusion should give every advocate of expanded incapacitation pause. Even under the most favorable assumptions, Spelman found that no more than 22 percent of potential crimes can be prevented through incapacitation, and that even the most favorable selective policy would reduce crime by no more than 7 percent. Those are not trivial effects. But they fall far short of the claim that we can incarcerate our way “out of crime.” At its best, incapacitation is a partial solution. As Spelman put it: “Incapacitation cannot, by itself, solve the crime problem.”
The better option: Deterrence
If incapacitation is insufficient, where should those scarce public safety dollars be directed? The answer is to shift the focus from the back end of the criminal justice system to the front end. Rather than paying hundreds of billions of dollars to lock up every low-rate offender for 20 years, we should prioritize investments that deter crimes from happening in the first place.
The fundamental limitation of incapacitation as a primary strategy is that it is inherently reactive. It accepts that the crime has already occurred and the victim has already suffered, attempting only to prevent the next offense. Deterrence, by contrast, aims to prevent victimization before it occurs. A strategy that can only address crime after the fact is incomplete when tools exist to prevent a significant share of that crime in the first place.
Deterrence also offers an advantage in scale. Incapacitation is arithmetic: one prison bed neutralizes one offender for a given period. A credible deterrent signal, by contrast, alters the calculus of every prospective offender who encounters it simultaneously. And because it targets the decision to offend in the present rather than predicting who will offend in the future, deterrence sidesteps the epistemic problem that makes selective incapacitation so unreliable. We do not have to guess which 22-year-old is destined to become a career criminal. We have to make it credibly costly for anyone to commit a crime today.
Historically, policymakers have tried to generate deterrence by relying heavily on severity, operating under the assumption that the prospect of a draconian prison sentence will frighten potential offenders into compliance. But criminological research has consistently found that the severity of a punishment appears to have substantially weaker deterrent effects than the certainty of apprehension.
The reason for this failure lies in the reality of criminal decision-making. Most street crime is not the product of a careful, long-term cost-benefit analysis. It is heavily influenced by impulsivity, situational pressure, present bias, or substance use. To a prospective offender, a penalty scheduled to be served years in the future is heavily discounted, if it is considered at all. More importantly, severity only matters if the offender actually expects to be caught. If the perceived probability of apprehension is close to zero, multiplying that zero by a 10-, 20-, or 30-year sentence still results in a perceived risk of zero.
What actually changes offender behavior is the perceived certainty of apprehension. Unlike the distant, abstract threat of a long sentence, the immediate, palpable risk of being caught alters the environmental calculus in real time. When offenders believe that detection is highly likely, the specific length of the punishment becomes secondary; a major component of deterrence is the disruption, hassle, and immediate consequence of the arrest itself. Therefore, building an effective deterrence strategy requires shifting our focus away from the back-end severity of our sentencing laws and toward the front-end certainty of our enforcement mechanisms.
If deterrence is the superior strategy, and if certainty of apprehension is the central driver of deterrence, the central policy question becomes how to actually increase that certainty. Fortunately, there is a robust body of empirical evidence pointing toward specific, actionable levers that make the threat of getting caught far more credible to prospective offenders.
A deterrence agenda
The most fundamental lever is the sheer presence and visibility of law enforcement. Recent research demonstrates that expanding the size of a police force reduces violent crime, and especially homicide. Crucially, this reduction is not primarily achieved by making more arrests and feeding more people into the prison system. Instead, the presence of more officers acts as a general deterrent, changing the environmental calculus for potential offenders and preventing the crimes from occurring in the first place. In fact, credible evidence suggests American cities are under-policed.
But deterrence is a function of both quantity and quality; it requires not just more police, but better policing. Strategic deployments that focus resources on the small networks driving the majority of violence yield massive deterrence dividends. For example, Aaron Chalfin’s research on gang takedowns illustrates how targeted, precision enforcement against high-risk groups disrupts crime far beyond the individuals arrested. In New York City, gun violence in and around public housing fell by roughly a third in the year after a gang takedown, accounting for nearly a quarter of the overall decline in gun violence across the city’s public housing over eight years, and without any increase in stops or arrests for low-level offenses. When the justice system credibly demonstrates its capacity to dismantle organized violence, it sends a powerful deterrent signal to the broader community that the risks of engaging in similar behavior have dramatically increased.
Beyond physical police presence, technology plays a critical role in increasing the certainty of detection. Expanding DNA databases is one of the most cost-effective deterrents available. As research by Jennifer Doleac has shown, when an offender’s DNA is added to a state or federal database, their likelihood of reoffending drops. They know that if they leave biological evidence at a future crime scene, their capture is virtually guaranteed. DNA databases thus transform an abstract risk of arrest into a near certainty, and thereby suppress criminal activity.
Policies aimed at boosting overall crime clearance rates — for example, through better investigative resources and improved forensic capabilities — operate on the same principle. A high clearance rate is the ultimate deterrent signal; it tells prospective offenders that the system is highly effective at identifying those who break the law, raising the cost of crime in the process.
Finally, the principles of effective deterrence must extend to community supervision, where roughly 4 million Americans are currently on probation or parole. Traditional supervision is, in practice, an almost perfect inversion of what the deterrence evidence says works. Most violations go undetected, so the probability of any consequence for any given violation is close to zero. Violations that are detected accumulate for months before anything happens. And when the system finally does respond, it tends to use punishments that are out of proportion to the triggering violation. The result is a system that fails to change behavior in real time while filling prisons with people who could have been redirected months earlier at a fraction of the cost.
The effective alternative is the “swift, certain, and fair” model of community supervision. Under that model, the rules of supervision are clearly communicated in advance, and every detected violation is met immediately with a guaranteed but modest sanction — a few days in jail, for example. The certainty and celerity of the consequence, not its severity, drives compliance. This model has shown success in several iterations, most notably in Hawaii’s HOPE probation, and more recently in 24/7 Sobriety.
Together, these levers — adequate police staffing, focused enforcement, technological force multipliers, and swift accountability — form the architecture of a deterrence-first approach to crime control. They offer a blueprint for reducing victimization not by hopeless attempts to predict the future, but by increasing the cost of crime for today’s criminals.
None of this should be taken as a claim that deterrence is foolproof. Just as many offenders discount the distant threat of a long prison sentence, there is a smaller but real subset of offenders who will not be deterred even by a high probability of getting caught. Many crimes are the result of impulse, rage, or mental illness. The point is not to compare ideal deterrence to imperfect incapacitation. Both strategies have limitations. But when you compare them honestly, i.e., imperfect deterrence against imperfect incapacitation, each operating under real-world constraints, the best evidence suggests deterrence prevents more crime per dollar for the large majority of offenders whose behavior responds to incentives.
Costs and benefits of deterrence and incapacitation
Of course, one might offer an objection. Doesn’t incarceration generate general deterrence while also providing unique incapacitation benefits? After all, the threat of prison is the ultimate backstop of the justice system. Deterrence works in part because of the threat of incarceration, and common sense suggests that the prospect of spending years in a cell must prevent at least some crime.
It is true that a sufficiently draconian punishment can alter behavior at the margins. Consider, for example, the analysis of California’s “Three Strikes” law by economists Eric Helland and Alexander Tabarrok. By comparing the future offending rates of individuals convicted of two strike-eligible offenses against a carefully matched cohort convicted of one strike-eligible offense and one non-strike offense, they were able to isolate the specific deterrent effect of the third strike. They found that the looming threat of a mandatory life sentence did, in fact, reduce arrest rates among the two-strike cohort by about 17 percent to 20 percent. By that metric, the three strikes law worked.
Crucially, though, Helland and Tabarrok didn’t end their analysis there. Rather, they also found in the same study that California spent $4.6 billion enforcing Three Strikes, and prevented 31,000 crimes. Given the staggering cost of crime to victims, a reasonable case can be made that avoiding those crimes was worth it. But Helland and Tabarrok also found that the same amount spent on police hiring instead of expanded incarceration could have prevented approximately 1 million crimes. Again, the result of inefficient public safety policy is more crime-related suffering, not less.
Moreover, traditional walls-and-bars incarceration is not the only mechanism that can be used for incapacitation. Electronic monitoring can generate at least some of the effects of traditional imprisonment at a fraction of the fiscal and social cost. For example, Roman Rivera found that pretrial electronic monitoring in Cook County, Illinois, was an effective substitute for detention without a significant increase in serious crime. And when the federal Bureau of Prisons in President Trump’s first term transferred more than 13,000 lower-risk prisoners to home confinement, the re-offense rate was negligible, suggesting that prison provided no meaningful public safety benefit above what monitoring could deliver. None of this means electronic monitoring can replace prison for genuinely dangerous offenders. But every offender monitored electronically instead of imprisoned frees scarce prison space for someone who actually needs it.
Conclusion
Public safety budgets are zero-sum, and every dollar spent on one strategy means a dollar not spent on a different strategy. When we evaluate crime control policy through the lens of inherent tradeoffs, the math overwhelmingly favors the architecture of deterrence over incapacitation and certainty over severity. Every million dollars sunk into the diminishing returns of a decades-long prison sentence is a million dollars starved from police staffing, DNA database expansion, crime lab improvements, or homicide investigations. Crime and violence generate too much suffering, and the criminal justice system bears significant responsibility for that. If our goal is to maximize crime reduction —and by extension minimize the suffering of innocent people —we must systematically shift our marginal investments away from the expensive, back-end lever of incarceration and toward the front-end interventions that create credible threats of swift and certain consequences. A system built on the certainty of punishment will prevent more crime, at lower cost, than one built on the false hope that we can incarcerate our way to safety.
