A man was arrested for entering a New York subway station through an emergency gate without paying the fare. When police patted him down, they found he was illegally carrying a loaded gun. When another man was arrested for fare evasion at a different stop, police found he was carrying 38 decks of heroin and a loaded gun that had been reported as stolen. Yet another was arrested for not paying the subway fare in another part of the city, and it turned out he was carrying a gun, ammunition, and crack cocaine. New York police argue that enforcing a seemingly minor offense (fare evasion) helps them catch people like these who are up to no good—perhaps preventing violent crime.
And yet, how to handle such offenses has been hotly debated for decades. Some worry that allowing arrests will result in violent confrontations or jail time for people already struggling to make ends meet. For what, a $2.90 subway ticket? While some of those arrested for fare evasion will have a track record of crime, for many others this would be their first arrest. Is a criminal charge really our best option?
In 2023, the board of directors of BART—Bay Area Regional Transit, the San Francisco subway agency—voted to oppose a bill moving through the California legislature that would decriminalize fare evasion. “The public is speaking very loud to us right now—and they have been—about the lack of enforcement of rules in our system,” said BART board member Debora Allen. Local residents were worried about safety on the subway system, with stories about violent crime at the top of everyone’s minds. “I can’t help but say we could help prevent some of the bad behavior in our system by getting tougher on fare evasion.” Daly City resident Howard Bernstein agreed that such a move would only embolden offenders: “The more we decriminalize criminal behavior, the more criminal behavior we’re going to experience.”
A few years earlier, in 2017, Manhattan district attorney Cy Vance Jr. made headlines when he announced he would no longer prosecute fare evasion cases. “Prosecuting for turnstile jumping is counter to this city’s efforts to be a sanctuary,” Anthony Posada of Legal Aid NYC told reporters at the time. The move was generally motivated by a recognition that fare evasion was a crime often committed due to poverty. Clearly, pressing criminal charges in such situations would not address the problem—that the person had no money—and risked making the problem worse. But in 2023, the debate raged on. The New York Times journalist Ana Ley reported that the Metropolitan Transportation Authority (MTA) had lost $690 million to fare evasion the previous year. Turnstile hoppers were apparently undeterred by tickets written by the police hovering nearby, and seeing some people ride without paying made others feel like suckers for buying a ticket. A year later, Ley followed up with an equally dire statistic: 48 percent of bus riders in the city did not pay the required fare. (The number for subway riders was 14 percent.) “If the transit system does not work and nobody plays by the rules, it feels lawless. It is lawless,” said Janno Lieber, chief executive of the MTA. “This is really tearing at the social compact of New York.”
Current debates about how to handle low-level nonviolent crimes are reminiscent of those from thirty years ago. In 1993, Rudy Giuliani was elected mayor of New York City. He came to office promising to make the city safer, and quickly adopted a strategy that had previously been discussed only in academic circles: “broken windows” policing. The idea underlying this approach was that disorder begets disorder. By addressing low-level offending—trespassing, vandalism, prostitution, urinating in public—police could deter more serious offenses. People would know that police were paying attention and that the community was well cared for, and so they would behave better. Fix the broken windows, and arrest the guy who broke them, and you’ll prevent violent crime as well.
Crime fell dramatically in New York City during the 1990s, and Giuliani claimed victory. In his farewell address in 2001, he credited this strategy—cracking down on low-level offenders as a way to prevent crime from escalating—as the key to his success. “The broken windows theory replaced the idea that we were too busy to pay attention to street-level prostitution, too busy to pay attention to panhandling, too busy to pay attention to graffiti. Well, you can’t be too busy to pay attention to those things, because those are the things that underlie the problems of crime that you have in your society.”
Since then, it’s become less clear how helpful that strategy really was. Crime fell dramatically throughout the US in the 1990s and early 2000s, not just in New York City. And yet this theory is still appealing to voters, and still motivates the policies of many police departments and prosecutors’ offices. Fast-forward to the present, when many cities are struggling with persistent and increasingly brazen public drug use, homeless encampments, vandalism, petty theft, and, yes, turnstile hopping. More serious crime, including organized retail theft rings, carjackings, and even homicide, also became serious problems in many cities—a shock after decades of declining crime rates. In 2022, the San Francisco resident and political commentator Richie Greenberg described what he saw as a turn toward lawlessness, to The Atlantic’s Annie Lowery. “People are sick and tired of the whole atmosphere of the city. It’s not fun to live here anymore,” he lamented.
The past several years have prompted calls to return to a Giuliani-style zero-tolerance approach to disorder, with stiff penalties for even the most minor offenses. Today, when I speak with policy stakeholders across the United States, I’m frequently asked if a broken windows approach could be the solution to their problems. Could a tougher approach to minor offenses be the key to reducing more serious crime?
Many voters have decided it’s at least worth trying—again—and not just in the subway. In 2023, police in San Francisco began cracking down on public drug use, arresting more than seventeen hundred people between May and December. “You’ll never hear me say that arresting folks will solve addiction, but these are still crimes,” Police Chief Bill Scott told The San Francisco Standard. San Francisco sheriff Paul Miyamoto saw punishment as a compassionate incentive for people to get the help they need: “Justice-involved persons with substance use disorder sometimes need the threat of jail time to compel them to remain in programs that successfully address the root causes of addiction,” he said, explaining the city’s efforts to dismantle open-air drug markets. In Philadelphia’s Kensington neighborhood, police crackdowns targeted public drug use as well as other minor offenses and quality-of-life issues—they shooed away people who were loitering, towed unregistered cars, and cleaned vacant lots. “The neighbors really didn’t complain,” Deputy Commissioner Pedro Rosario responded when asked about community pushback. “They were happy. They were thumbs-upping me a lot.”
There is at least a kernel of wisdom in this approach. We know that people respond to incentives, and swift and certain punishment for bad behavior deters future offending. That is important, and to the extent that broken windows means making consequences more likely, it could indeed reduce crime and put people on a better path. But increasing the probability of punishment is different from making the punishment harsher. Many proponents of the broken windows approach don’t simply want to arrest people for their bad behavior, they want to throw the book at them. (Because of this, broken windows, as a philosophy, extends beyond what police do and into the courtroom, where prosecutors and judges decide the consequence for an offense.) It’s not clear that harsher punishment would be productive. On top of that, it’s possible that prosecuting and punishing low-level offenders has other, detrimental effects that cancel out any benefits we get from deterrence—especially if it’s someone’s first brush with the law.
For instance, a criminal record makes it more difficult to find a job, and it means you might lose the job you had before your conviction. This creates economic hardship that can make criminal behavior more likely, as a way to make ends meet. A criminal record also makes it more difficult to find housing, as most landlords run background checks just like employers do. Without a safe place to live, you might find yourself in more dangerous situations, with less to lose, and more vulnerable to future charges for offenses such as trespassing when you have nowhere else to go. To the extent that your previous offenses were the result of untreated mental illness or substance use, the stress of criminal charges and any punishment could make those problems worse.
How much do all these factors matter in practice? Would reducing the consequences for low-level offenses be helpful? Those who say no—like proponents of broken windows policing—think the threat of harsh punishment has a big deterrent effect. But others say yes. Reform-minded prosecutors, elected in many cities over the past decade, have promised to go easier on minor offenses so that they can focus more attention on violent crimes. In 2019, Wesley Bell, the prosecuting attorney in St. Louis County, Missouri, told NBC News that diverting nonviolent offenders from jail to rehabilitative programming “not only brings our crime rates down, but most importantly, it helps people and families.” Still others believe that the challenges faced by those who commit low-level crimes—poverty, limited education, untreated mental illness—are so large that only a massive reform of our social safety net will keep people from cycling back through the criminal justice system. Who’s right?
* * *
Prosecution in Suffolk County
With such diverging opinions on what to do with people arrested for low-level offenses, figuring out the best path forward requires turning to data. I teamed up with the economist Amanda Agan, now at Cornell University, and political scientist Anna Harvey, from New York University, to study this issue. We wanted to know what effect the decision to prosecute someone for a nonviolent misdemeanor—minor offenses like trespassing, shoplifting, and minor drug possession—would have on a defendant’s future criminal justice involvement. Would cracking down on that low-level offense reduce their likelihood of reoffending (as broken windows proponents expect) or increase it (as reform prosecutors argue)?
We managed to get data from the District Attorney’s Office in Suffolk County, Massachusetts, where Boston is located. The DA there at the time, Rachael Rollins, also wanted to know the answer to this question, and so enthusiastically handed over her office’s data, no strings attached. This is a researcher’s dream. Many policymakers are hesitant to share data when they can’t control the results of the study, and can’t block unfavorable results from being published. But such restrictions are a nonstarter for researchers like us. Luckily, then-DA Rollins was on board with our scientific approach and wanted to follow the evidence. She understood that this was the best way to figure out how to improve public safety—and perhaps end this long-standing debate.
In Suffolk County, once police make an arrest or issue a summons, and then determine that probable cause exists for the charge, the case goes to an arraignment hearing. In that hearing, an assistant district attorney (ADA) representing the government decides whether to pursue the charges or dismiss the case. They are essentially deciding whether they think the case is a good use of prosecutors’ time. This is the decision we were interested in. What if more cases were dismissed up front? Would that lead to more recidivism, or less?
What if more cases were dismissed up front? Would that lead to more recidivism, or less?
Simply comparing people who were prosecuted with those who were not wouldn’t answer this question, because prosecutors intentionally choose whom to prosecute. If we found that those who were prosecuted were more likely to reoffend in the future, we wouldn’t know if this was the effect of the prosecution decision, or because prosecutors only move forward with cases against higher-risk defendants. Prosecution might be correlated with recidivism, but that doesn’t necessarily mean prosecution causes recidivism.
So once we had the data, it was time to hunt for natural experiments that would allow us to distinguish correlation from causation. The ideal experiment in this context would randomly assign some defendants to be prosecuted and others not. We could then attribute any differences in future behavior across these two groups to the effect of being prosecuted, without worrying there are other underlying differences between them that explain their differences in behavior.
Of course, no one would agree to prosecute cases at random (nor should they). But it turns out that the way nonviolent misdemeanor cases are assigned to ADAs mimics this ideal experiment.
What determines which ADA handles each case? Handling arraignments is the “grunt work” of the prosecutors’ office. (The more interesting work comes later in the case proceedings.) So everyone takes a turn, especially junior ADAs who haven’t specialized yet.
ADAs are assigned to the arraignment courtroom in an ad hoc way that changes week to week, depending on their other meetings and case schedules. This Monday, Tom might be assigned to handle arraignments, but next Monday, Anne might be assigned to that task. This assignment schedule is unrelated to the types of cases expected on that day—this is the key. Because of this, we don’t need to worry that ADAs are selected to handle particular cases on account of their expertise or preferences—at least for the nonviolent misdemeanor cases we are interested in. (They might pull someone with more expertise in for more serious offenses.)
It’s important to understand the huge volume of these cases that go through the courts in any given week—misdemeanors make up 70 to 80 percent of all cases. For these very minor offenses, ADAs have just a few moments to decide whether to proceed with a case or drop it. The goal is to keep the cases moving; this is the only way the courts don’t become completely overwhelmed by minor charges and grind to a halt.
Because everything moves so quickly, and because ADAs’ schedules are so unpredictable, it is not possible for defendants to game the system to get a particular ADA. When their case is at the top of the pile on the ADA’s desk, it’s their turn. They get what they get.
This all means that which ADA handles a particular case is effectively random—there is no correlation between case characteristics and the characteristics or relative harshness of the ADA.
Human discretion as a natural experiment
So we have randomization of ADAs, but this doesn’t help if all ADAs behave the same way. What we also need from this natural experiment is randomness in the decision to prosecute.
This is where we could rely on human nature, and a fact that we see in every domain where humans make decisions: People have different preferences, and so they will use any discretion they have in different ways. And prosecutors have a lot of discretion. In this context, this means that two different prosecutors considering identical cases might make completely different decisions. One might drop the case immediately, while the other might choose to move the case forward with the goal of conviction and punishment.
This probably sounds extremely unfair. Shouldn’t identical cases get the same outcome regardless of who the prosecutor is? That is certainly the ideal, but in contexts like this, there is no right answer about what should happen in a case. We count on human decision-makers to use their best judgment. This leads to differences in outcomes that we’d rather not have.
In the U.S., we tend to swing back and forth between limiting the discretion of criminal justice actors like prosecutors and judges and giving them more discretion. We hear about big differences in outcomes across similar cases—for instance, Black defendants receiving harsher sentences than similar white defendants—and we demand restrictions on discretion. This is part of the reason for policies like sentencing guidelines and mandatory minimum sentences. These tie the hands of prosecutors and judges, at least on some dimensions.
Having your case dismissed rather than pursuing prosecution—reduced the likelihood of showing up in court again with new charges by 53 percent, and it reduced the number of future charges by 60 percent.
But then we hear about a case that, based on these standardized rules, resulted in an outcome that seems totally unfair given some extenuating circumstances, and we demand that decision-makers have more discretion to deviate from those rules when it is warranted. We want them to use their judgment to provide the best outcome. And then when they do, we wind up with different outcomes across similar cases, and we swing back toward wanting less discretion.
What we want is for prosecutors and judges to use their discretion only for good—to reach the decision we think is most appropriate. But the problem is that different people disagree about what is appropriate. Allowing people to use their best judgment has trade-offs, and we have to take the bad with the good.
One silver lining to this complicated dilemma is that human discretion creates great settings for research. In Suffolk County, random assignment of cases across ADAs meant that we effectively had random assignment of cases to different treatments—the ideal experiment we’d hoped for. Some defendants get lucky and their case is handled by a lenient ADA; because of this, they are more likely to have their case dismissed outright. Other defendants are unlucky and their case is handled by a harsh ADA; their case is more likely to move forward to the next stage. Through the luck of the draw—which ADA happened to be in that courtroom that day—we have identical cases that are treated in different ways.
What happened to those lucky defendants whose cases were dropped because they happened to be in the right courtroom at the right time with a lenient ADA? Proponents of broken-windows-style punishment as a deterrent would predict that those defendants would be emboldened. Facing little consequence for their actions the first time, they’d realize the costs of bad behavior were low and commit even more crime in the future.
But that’s not what the data showed. It turns out that leniency at this early stage—having your case dismissed rather than pursuing prosecution—reduced the likelihood of showing up in court again with new charges by 53 percent, and it reduced the number of future charges by 60 percent. The effects were larger for first-time defendants—those with no prior arrest or conviction on their record.
* * *
The power of leniency
David Eil is an assistant public defender in Mecklenburg County, North Carolina, where Charlotte is located. He has seen firsthand the damage that a first conviction can do. And—unlike most lawyers—he used to be an economist. So he has a keen eye for natural experiments. This makes him a great person to compare notes with about how the criminal justice system works in practice.
“I had two misdemeanor clients who were similarly situated,” he told me recently. “Both were facing the same charge, but their cases had different outcomes due to random chance.”
Both clients were charged with misdemeanor assault on a government official—a charge that sounds really bad, but David assured me that such a charge at the misdemeanor, rather than felony, level is almost always pretty minor. (“I literally had a client get charged with assault of a government official for not saying, ‘Excuse me,’ when moving past a cop through a doorway,” he recalled.) The first client, Tiffany, arrived at the courthouse for her hearing to find that the officer accusing her of assault had not shown up. His partner, who had witnessed the incident, was there, but the prosecutor told him he could leave, then asked the judge for a continuance—a delay to a new date—so that they could get the first officer to the court. The judge denied that request. This led the prosecutor to scramble to get the second officer back to court, to testify as a witness. He succeeded—“a miracle for the prosecutor,” David noted—and Tiffany was convicted. This was devastating for her. As a result of that conviction—her first—she was not able to obtain an employment certification she’d been working toward. The training program she had invested time and money into was suddenly worthless. This first criminal record changed her trajectory for the worse, even though (because the charge was so minor) no meaningful punishment was handed down by the court.
David’s second client, Sam, faced the exact same charge. When he showed up to court, the accusing officer was not there. Just as in Tiffany’s case, the officer had failed to appear. (It turns out that this is a pretty common occurrence, which isn’t great for the smooth functioning of our justice system.) This time there wasn’t another officer who could testify as a witness. The judge again denied the prosecutor’s request for a continuance, and this time the case was dismissed. Sam was relieved—he worked as a security guard, but had been suspended from his job because of this pending charge. (A criminal record—even a misdemeanor like this one—is typically disqualifying for a position focused on public safety.) He had a limited amount of time before his employer would have needed to replace him. Even if the judge had granted the continuance and the case had been dismissed a month later, it would have been too late; he would have lost his job. Because the dominoes fell as they did that day, Sam kept his clean record and got to return to a job he liked, continuing his life as it had been before.
Just like the similar defendants in our study, who were lucky or unlucky depending on which prosecutor they faced during their arraignment hearing, David’s clients faced different consequences as a result of luck rather than anything about them or their cases. It is easy to see how such luck can play out case after case, day after day, putting similar people—like Tiffany and Sam—onto radically different paths. The results of our study support David’s observation that a first misdemeanor record can do a great deal of harm—at least for nonviolent defendants. (The assault charges that Tiffany and Sam faced would not be in this category, of course, but future studies may show similar impacts for violent misdemeanor charges like theirs.) With worse employment options, additional criminal behavior is more likely.
I asked David how aware his clients are of the impact of a first criminal record. “It’s usually me who brings it up,” he told me. “It’s more common that people first feel the impact of a pending charge, and then learn—probably accurately—that if the case is resolved quickly, even if it’s not in their favor, the problem goes away.” For example: “A class 2 misdemeanor for carrying a concealed weapon—a very common misdemeanor in North Carolina that is viewed as relatively minor—probably wouldn’t bar you from driving for Uber, but a pending criminal case will.” So there is a strong incentive to quickly take a plea deal so that you can keep your current job. Even so, there might be longer-term consequences from that conviction, depending on where you live or what types of jobs you might want in the future. So David tries to persuade his clients that waiting for a dismissal might be worth some temporary pain.
“Sometimes I’m in a position where I’m trying to talk somebody out of [taking a plea bargain immediately], reminding them that once a conviction is there [on your record] it will be there until it can be expunged, which is a long time from now and you might not qualify when that time comes. It might be worth trying to get to the next court date and beating this case, even if it’s going to cause you some more short-term difficulty. . . . I’m often the one who is trying to describe for them the problems of having a conviction, and they’re the one saying, ‘Look, I just want to get this over with.’”
More lessons from Suffolk County
Our results from Suffolk County tell us that prosecutors should avoid putting first-time defendants in this situation. They should err more toward leniency at the arraignment hearing and focus instead on trying fewer, more serious cases. This would give us more results like Sam’s, and fewer like Tiffany’s.
As an economist, I’m used to most policies involving trade-offs. And when my colleagues and I started working on the Suffolk County study, I fully expected to find some costs to leniency. Surely we would see some increase in criminal behavior, if only in minor offenses like trespassing and drug possession. The question in my mind was whether those costs outweighed the benefits, like how much time it saved everyone (including defendants) when a case was dropped. But when we followed the data, we found only benefits. Criminal behavior didn’t increase, it fell. And it fell by a lot. The other benefits to the court and defendants were icing on that cake.
As researchers, we are skeptical by nature. We wondered: What if these defendants didn’t offend again because they realized they’d gotten lucky? What would happen if there were an actual policy change that pushed all ADAs to become more lenient? Those defendants might change their behavior, and members of the community might hear about the change and decide obeying the law wasn’t worth it. Would we see crime rates go up then?
Luckily, we had such a policy change to consider. DA Rachael Rollins had implemented a “presumption of nonprosecution” for a list of fifteen nonviolent misdemeanor offenses when she took office. This meant that she instructed her ADAs to dismiss such cases unless they had a good reason not to. (A good reason might be that that person had already cycled through the court several times, and leniency was clearly not working.) This pushed all ADAs to be more lenient, particularly toward first-time defendants—exactly the group we’d found benefited the most from such decisions.
When we looked at the effects of this policy change, we found similar benefits as before: For defendants, we found that dismissing their cases led to fewer future charges, not more, just as we’d found was the case before the policy change. (This rules out the “I got lucky” effect.) And when we looked at local crime rates, we found no increases. Some types of reported crime may even have fallen.
Declining to prosecute low-level offenses isn’t unheard of. All prosecutors consider which cases are worth pursuing and which are not, and decline to prosecute many of them, for a variety of reasons. Under Rollins’s more conservative predecessor Dan Conley, the DA’s office chose not to prosecute 34 to 38 percent of nonviolent misdemeanor cases. Under Rollins’s policy, this rate increased by 5 to 8 percentage points—so her office became more lenient, but it wasn’t a radical shift. The change was on the margin, but it was big enough to make a difference for quite a few people.
“We have to give credit—DA Conley was doing this,” Rollins told reporters, poking fun at those who had criticized her approach as too liberal. “He just wasn’t as vocal about it as I was, and we’ve increased it a bit as well.”
The data vindicated DA Rollins, who had faced extreme pressure from local police officials to be tougher on misdemeanor defendants. Reporters asked what would have happened if the study’s results had come out the other way. “We would be adapting right now because at the end of the day, it’s not about policies, it’s about what are we doing to keep the people of Suffolk County safe,” she told The Boston Globe at the time—an answer that any researcher would love. “What I hope this does is say we are really serious about data-based, and evidence-based, solutions. This data shows the policies we proposed are working.”
Economists like to say there’s no such thing as a free lunch—there are always trade-offs. But we’d found a free lunch! Erring toward leniency, particularly for first-time defendants, made everyone better off.
This was for nonviolent misdemeanor cases, the most minor type of offense. There are many such cases, so this could make a big dent in the number of charges going through the courts, but what might leniency look like for people charged with more serious crimes?
* * *
Leniency in felony cases
Nonviolent felony cases are more serious offenses, like burglary and motor vehicle theft. These types of cases are much less likely to be dismissed outright. But in many places, prosecutors have the option to wait and see if a defendant is a public safety threat before convicting them.
In Texas, this option is called “deferred adjudication.” When prosecutors choose this option, the defendant begins a probationary period. If they successfully complete that probation with no new offenses, their initial charges will be dropped completely and they avoid that conviction. On the other hand, if they do get into additional trouble, their conviction goes into effect, along with some punishment (usually community supervision).
The country star Zach Bryan made headlines in 2023 when he was arrested in Oklahoma on obstruction charges. “It was ridiculous, it was immature, and I just pray everyone knows I don’t think I’m above the law,” Bryan said the next day. “I was just being disrespectful, and I shouldn’t have been. It was my mistake.” He received a deferred prosecution—that state’s equivalent of a deferred adjudication—and completed the terms of his probationary period six months later. A spokesperson for the Craig County District Attorney’s Office explained that Bryan “admitted responsibility and followed all the rules and conditions of probation. [Deferred prosecution agreements] are commonly used in cases where the person has no criminal record. It is an opportunity to take responsibility for their actions, follow probation rules, and avoid having a criminal conviction on their records.”
Does this second chance to avoid a felony conviction lead to more future offending, or less? Again, we face the same potential trade-offs: Reducing the consequences for committing a crime might embolden the defendant, leading to more crime in the future. On the other hand, avoiding a conviction could help them keep their job and housing, allowing them to course-correct on their own.
The net effect is an empirical question, and only real-world data can tell us what happens in practice.
The University of Michigan economist Michael Mueller-Smith and Simon Fraser University economist Kevin Schnepel were able to study this question in Harris County, Texas, where Houston is located. They linked a variety of administrative datasets that allowed them to see not only criminal justice involvement but also employment and earnings.
Linking such datasets is surprisingly difficult in the United States. These datasets are maintained by separate government agencies at county and state levels, and linking them requires complex negotiations and lengthy data use agreements. Many agencies simply say no when researchers ask to use their data, and even more say no when researchers ask to link their data with data from other agencies. This makes it difficult to understand how our criminal justice system affects other aspects of people’s lives—like whether they have a job.
But Mueller-Smith and Schnepel pulled these negotiations off. As a result, they had amazing data on felony defendants in a major American city (one that I now call home).
As before, data wasn’t enough. These researchers couldn’t simply compare defendants who received a deferred adjudication with those who were prosecuted and convicted as usual, because prosecutors carefully choose who gets this second chance. Deferred adjudications might be correlated with lower recidivism, but that could be because prosecutors give this option only to lower-risk defendants.
Mueller-Smith and Schnepel needed a natural experiment—something that sorted similar defendants into “deferred adjudication” and “regular conviction” groups, as if at random. Luck was on their side: They found not just one natural experiment but two.
Policy change one
The first natural experiment was a policy change that had unintended consequences. In 1993, the Texas legislature enacted a reform that imposed a new probationary requirement for low-level offenders. The policy sounded like it was in line with diversionary goals—that is, helping first-time offenders avoid being pulled into the criminal justice system—but in practice it made diversion less appealing to prosecutors. If they granted a defendant a deferred adjudication, and that defendant did not comply with the terms of the probationary period, they could not simply revert to the original conviction and sentence. The new policy said they’d have to put them on probation again before the sentence could go into effect. This gave the first probationary period no teeth.
Prosecutors warned the legislature that this policy could backfire, but to no avail. The policy went into effect on September 1, 1994. Deferred adjudications immediately dropped, by 24 percentage points for first-time defendants.
This meant that identical defendants, charged with the same nonviolent felony offense but who committed their crimes just before and just after September 1, 1994, faced different consequences. The person who offended just before the policy change was dramatically more likely to receive a deferred adjudication than the person who offended just after the policy change.
This created the first natural experiment. The date of the policy change—September 1, 1994—sorted defendants into treatment and control groups, as if at random, based on the date of their offense. Nothing else changed at that date. The only difference between these defendants was whether they got a second chance to avoid a felony conviction.
It turns out this second chance was very helpful. First-timers who got lucky and received a deferred adjudication committed fewer crimes going forward. They were 31 percentage points less likely to be convicted of any new crime over the next ten years—a 44 percent reduction compared with the control group.
First-timers who got lucky and received a deferred adjudication were 31 percentage points less likely to be convicted of any new crime over the next ten years—a 44 percent reduction compared with the control group.
This second chance also increased labor market participation. Employment rates increased by 18 percentage points (49 percent relative to the control group), and total earnings over the following 10 years grew by more than $85,000 (93 percent relative to what the control group earned). A large share of those who received this second chance took full advantage of it.
Policy change two
In the early 2000s, policymakers in Harris County, Texas, were increasingly worried about overcrowding in the local jail. By 2005, there were nearly two thousand inmates sleeping on mattresses on the floor—very bad conditions that were clearly unsustainable.
Government officials sought to raise money to expand jail capacity. In November 2007, they put an initiative on the ballot in the county election to fund construction of a new jail facility. Particularly in conservative states like Texas, such initiatives are usually successful.
This one was not. To everyone’s surprise, the ballot initiative was narrowly defeated, with 50.6 percent of voters voting against it—even as they overwhelmingly approved additional statewide funding for increasing prison capacity. This defeat shocked local policymakers and criminal justice practitioners. They suddenly realized they’d need to solve their overcrowding crisis some other way.
One result was a sudden increase in diversion—either deferred adjudications or outright case dismissal—for nonviolent felony defendants. First-time offenders who committed a crime just after the failed ballot initiative got lucky—they were dramatically more likely to get a second chance. Overnight, the probability of diversion increased by 18 percentage points.
Again, this set up a beautiful natural experiment. Mueller-Smith and Schnepel could compare defendants sentenced on either side of the election on November 6, 2007. The only difference between those sentenced before and after this date was that those sentenced after were much more likely to avoid a conviction. This difference wasn’t because of underlying differences between these defendants or their cases; it was because of the failed ballot initiative. This gave the researchers confidence that any future differences in recidivism or employment would be due to the diversion decision and not to something else about those defendants.
Just as in 1994, there were big benefits to greater leniency. As the likelihood of diversion suddenly increased, the likelihood of new, future convictions fell, by 26 percentage points (46 percent). This is a dramatic change. Nearly half of the first-time offenders who would have committed another crime in the future if they’d been prosecuted and convicted as usual cleaned up their acts and avoided future crime when their cases were dropped or they received a deferred adjudication. It is really difficult to find interventions that reduce recidivism this much. This second chance to avoid a first felony conviction had a much bigger impact than most rehabilitation programs do.
* * *
Putting it all together
When I first saw these results, they seemed too good to be true. We can cut recidivism by half simply by not convicting first-time defendants? This feels like magic, particularly in a context where many highly praised reentry programs struggle to reduce recidivism at all.
But these studies are extremely compelling, with natural experiments that closely mimic ideal experiments with control groups. And the findings are remarkably consistent—a second chance for first-time defendants cuts future crime by half, in all these contexts. As more research comes out, I become more and more convinced that a criminal record is a terrible bludgeon that we should use much more sparingly than we do now. It is very difficult to undo the damage of a criminal record, once it has been imposed.
I become more and more convinced that a criminal record is a terrible bludgeon that we should use much more sparingly than we do now.
Best of all, these dramatic results don’t require dramatic reforms. Small shifts, making leniency the default rather than the exception, are enough. Cutting recidivism in half for first-time offenders will quickly reduce reported crime and court caseloads, allowing greater attention on those who do offend again.
Leniency doesn’t mean no consequences
When we first released our research from Suffolk County, some responded that we should completely decriminalize minor offenses like disorderly conduct and shoplifting. But that’s not what this research showed. Dropping someone’s charges at their arraignment hearing doesn’t mean there were no consequences for their actions.
That person had likely been arrested and booked in jail, and had to show up in court for that initial hearing. This might mean taking time off work, and it certainly meant worrying about what might happen during the hearing. All this isn’t nothing—it is an inconvenience at best and a costly and stressful event at worst.
In the Harris County study, “lucky” defendants also had to successfully complete a probationary period, during which they had to demonstrate that they could refrain from future criminal behavior. That involved following additional rules to earn their second chance.
What this research shows is these earlier steps in the case—arrest and an initial hearing for nonviolent misdemeanors, and a probationary period for nonviolent felonies—are often punishment enough. Adding a criminal record on top is what has big, detrimental effects, at least for first-time offenders. Helping someone avoid that first conviction gives them a second chance. It’s as if they’re standing at a fork in the road, considering what to do next. One direction leads toward more criminal behavior and criminal justice involvement, and the other leads toward a productive, law-abiding life. It turns out that many first-time defendants will choose the better path if we simply get out of their way.
Excerpted from The Science of Second Chances by Jennifer Doleac, published by Henry Holt and Co. Copyright © 2026 by Jennifer Doleac. All rights reserved. Reprinted with permission.
