Behavioral science & gun violence

Abstract

Gun violence, like many self-destructive behaviors (suicide, drug overdose, etc.), has proven hard to change. Conventional wisdom assumes these behaviors stem from people making deliberate, rational benefit–cost calculations, which has led to a policy focus on incentives (sticks and carrots). But most shootings are not premeditated or motivated by economic considerations; they’re in-the-moment arguments that spiral and end in tragedy because someone has a gun. In such settings, behavior seems more automatic than deliberate. Behavioral science gives us a way to understand and change that behavior. A growing body of research shows that this seemingly intractable problem is remarkably responsive to the right policies, including changes to the choice environment and even efforts to “change the chooser.”

Gun violence is the number one problem facing most cities in the United States, a public health crisis that disproportionately affects low-income minority communities and drives people and businesses out of our cities.^1,2 Every murder reduces a city’s population by a total of 70 people. ¹ This is a problem unlike in any other wealthy nation in the world; almost all of the differences in murder rates between the United States and other rich countries are due to murders committed with guns (see Figure 1).

Figure 1.

Homicide rates per capita, by country & weapon

If most of America’s “homicide exceptionalism” is due to murders with guns, and given that compared to other countries the United States has more lax gun laws and far more guns per capita (around 400 million guns for a country of 330 million people), is simply changing gun regulation the solution? The best available data suggest that reducing the number of guns in the United States would reduce the number of murders, perhaps substantially. ² But those 400 million guns are not going anywhere anytime soon, and the prospects for major new federal gun laws are unlikely.

If we are going to make progress on this problem, it will have to come from recognizing that gun violence equals guns plus violence. If the guns part of this equation is fixed for the foreseeable future, progress will require changing people’s willingness to use these readily available weapons to hurt or kill one another. But on that front, we need to do much better than we have been. The U.S. murder rate today is almost exactly the same as it was in 1900, echoing similarly grim trends for other self-destructive behaviors like suicide and drug overdose.

This lack of progress is, as I argue in my new book Unforgiving Places: The Unexpected Origins of American Gun Violence (2025), largely due to how we have misunderstood the problem. Surveys show that most Americans think of gun violence as the result of rational benefit–cost calculation. This has led to a focus on disincentivizing gun violence either through bigger sticks (prison) or bigger carrots (jobs or social programs to lure people away from crime).

But this conventional wisdom misunderstands what gun violence is. Most shootings are not the result of premeditated, deliberate, rational behavior. They’re arguments that go sideways and end in tragedy because someone has a gun. Many people in the immediate aftermath recognize they’ve made a mistake that they would not repeat if they had the chance to do it over again.

Behavioral science helps us understand these mistakes and which public policies can prevent them, which generally focus on either changing the choice environment or “changing the chooser.” For a problem many Americans have given up ever solving, behavioral science gives us new hope.

Conventional Wisdom & Its Limitations

It was at 69th and Calumet on the South Side of Chicago, at 10 p.m. on Halloween Eve, 1996, when Brian Willis, age 18, was arguing with Alexander Clair, 23, about a used two-door beige Ford LTD parked in front of Little Hobo’s restaurant. Clair had sold Willis the car a few days earlier and complained that Willis hadn’t paid for it yet. Willis was angry that Clair had reportedly entered the car earlier that night and tried to take it back.

Regarding payment, Willis told Clair, “I’m not going to give you shit.” Regarding adherence to the transaction’s terms, Clair replied, “If I catch you in the car—if I see the car or I catch you in the car—I’m going to burn the car up.”

The two argued in the street for another 10 minutes. Willis eventually broke off and ran across 69th, past the car and behind the building at 352 East 69th Street. Clair followed. Meanwhile, 25-year-old Jewel Washington, Clair’s girlfriend, was trailing behind when she heard two loud gunshots. These would turn out to be the gunshots that killed Clair, fired from a short-handled 12-gauge pump-action shotgun, with one shotgun blast to Clair’s stomach and one to his head.

Washington turned and tried to run back toward Calumet Avenue. Willis, still holding the shotgun, yelled, “Where do you think you’re going, bitch?” and again, “Where the fuck do you think you’re going, bitch?” A witness would later report hearing a high-pitched voice outside her window, pleading, “Please don’t, please don’t.” Three more shotgun blasts, and Washington was dead along with Clair.

Brian Willis was later convicted of two counts of first-degree murder and sentenced to life in prison. In effect, three lives, not just two, were lost that night.

For most Americans, the tragedy at 69th and Calumet in Chicago would be explained in one of two ways. The first is that shootings like this stem from characterologically bad people. Whether born bad or raised badly, the perpetrators of gun violence in this view have no moral compass or fear of the justice system. The implication is that the only way to prevent gun violence is to disincentivize it through the threat of harsher punishments.

A second perspective is that gun violence stems from root causes—that is, from a set of social conditions that fuel gun violence, leading economically desperate people to do whatever it takes to make ends meet. In this view, the only way to prevent gun violence is to disincentivize it by improving the alternatives to crime and by doing things like ending poverty.

The survey data confirm that most Americans believe crime and violence are due to some version of one of these two conventional wisdoms. To see the limits of these ideas, let’s return to 69th and Calumet. While it is possible that Willis was the kind of “super predator” that some commentators have warned Americans against, he not only got his undergraduate degree but also attained a master’s degree after being sentenced to life in prison.

Neither does it seem that economic desperation drove the shootings. The 1996 Blue Book value for the car at the center of the disagreement was $3,500. The statistical chances Willis would get caught for murder were about 50%. What person, no matter how economically desperate, would be willing to flip this coin: heads, a $3,500 used car; tails, life in prison?

In our search to understand the motives here, conventional wisdom leaves us largely lost.

A Behavioral Science Perspective

Interestingly, the conventional wisdoms of the left and right have one point of implicit agreement: the assumption that shooters are acting rationally. Both sides believe that before anyone pulls a trigger, they’ve carefully and deliberately thought through the pros and cons. The main point of disagreement is whether the right solution is bigger sticks versus carrots.

But these conventional wisdoms misunderstand what gun violence in the United States actually is. Most shootings are not premeditated or motivated by economic considerations. They are instead garden-variety arguments that escalate and end in tragedy because someone has a gun. In a typical year in Chicago, for example, around eight of 10 shootings will be not crimes of profit so much as crimes of passion, recognizing that one of the most powerful of human passions is rage.

Who gets into arguments? You. Me. Everyone. And if you’ve ever been in a heated argument, you can attest that the last thing you are in that moment is a perfectly rational benefit–cost calculator, or what my University of Chicago colleague Richard Thaler calls “homo economicus.”

My family’s dog, Aiko, is a 70-pound mix of hound dog and German shepherd. She’s built like a moose but acts like a lap cat. Her favorite place in the house is wherever one of her humans is. Anyone who sits on the couch will find Aiko right next to them, either asleep with her snoot resting in their lap or lying on her back looking at them with her big brown eyes, showing her belly in a “you know what to do” sort of way. This is to say, Aiko is a lover, not a fighter.

I take Aiko out for a walk every morning around the University of Chicago campus. On Wednesdays, I do this while having a weekly check-in call with the leadership team of my research center. One Wednesday I had leashed up Aiko, put in my earbuds, and patched into my weekly work call. I was walking down the sidewalk when a neighbor’s dog came barreling down their driveway, barking, snarling, and baring its teeth, then attacked Aiko.

What to do?

If I had acted rationally, using the sort of slow, deliberate, effortful thinking that behavioral scientists sometimes call “slow thinking” or System 2, ³ I would have realized that I could have just put myself between the other dog and mine, or picked Aiko up, or, worst case, resigned myself to yet one more vet visit. Sadly, my System 2 deliberate self was nowhere to be found at that moment.

The key insight from the dual-systems model of cognition is the recognition that our minds do not only engage in slow, deliberative System 2 thinking. As this other dog was charging toward mine, my other channel of thought—the effortless, automatic responses that happen below the level of consciousness, or System 1—led me away from a rational, conciliatory response. Some of my System 1 responses that are typically so helpful for me in routine, low-stakes situations got overgeneralized and led me into trouble.

For example, the tendency for System 1 to engage in egocentric construal—to personalize and think everything’s about me—is helpful in lots of day-to-day situations. When my wife or daughter (or both) come into the kitchen in the morning with a furious look on their faces, there really is a reasonable chance it’s due to some stupid thing I did. But when the neighbor’s dog was charging down the driveway—most likely an accident, a dog slipping its leash, etc.—egocentric construal led me to feel like “Why is he letting his dog do this to me?”

Similarly, throughout the long arc of humankind, it’s been helpful for System 1 to be really fast in its responses—out on the savannah no good thing comes to those who vacillate. If you see everyone else sprinting away from the creek, it’s more helpful for System 1 to choose from a limited set of response options—run or don’t run—rather than an overly long list of responses. This tendency to choose from a limited set of options also shapes how we assess any situation. So when the neighbor’s dog is charging down the driveway, and System 1 has to choose between “totally fine” and “end of the world,” it’s clearly not the former, so it must be the latter.

I catastrophized. I made a negative event feel even more negative than it really was, like nothing in the world could be worse than letting my dog get attacked. (A transcript of how I articulated these feelings to the owner of the other dog would include a long list of four-, seven- and 12-letter words that a distinguished behavioral science journal is not in the regular habit of printing.) If my dog being attacked feels like the end of the world, by definition any other course of action is less bad than that. So catastrophizing meant I didn’t bother to consider other candidate actions or the consequences of the course of action I wound up taking.

From my neighbor’s perspective, he assumed I must know what he knew—that this was just an accident, his dog had slipped the leash (a version of what behavioral scientists call the “curse of knowledge”)—so screaming the most horrible profanities imaginable at him right in front of his home, where his wife and young children were having breakfast, made me the unreasonable one in this interaction. We both believed we were the good guy here.

Here is where everyone involved got lucky: Hyde Park is chock full of what journalist Jane Jacobs, in her wonderful book The Death and Life of Great American Cities, famously called “eyes on the street”—people around the neighborhood who enforce a shared set of social norms. ⁴ Being near the University of Chicago, Round 2 of our disagreement occurred in an area heavily patrolled by campus security guards. As we were screaming at one another, one of the guards drove by, stopped, and asked, “Everything okay?”

I jumped at the lifeline. I yelled to the security guard, “This guy jumped off his bike and got in my face!” The security guard stopped his car and picked up his radio to call the campus police. That led the neighbor to hop on his bike and ride off—giving me the double bird as he went.

Why did I lose my mind at an off-leash dog? Why did Brian Willis shoot Alexander Clair and Jewel Washington? Odds are they were all the same thing: a normally helpful System 1 automatic response overgeneralized into a difficult situation.

Once System 1 has made a mistake, all of our cognitions and behaviors that sit downstream from there can go sideways as well. If it never occurs to us to revisit the upstream System 1 mistake that is leading us astray—as it may well not, since System 1 operates below the level of consciousness and is invisible to us—we can spend countless hours thinking about what to do and still not make a good decision. That means System 1 can lead to mistakes that play out over an extended period, not just mistakes that look clearly rushed in the moment. Princeton psychologist Anuj Shah calls this not thinking fast but “thinking past.”

Maybe the clearest way to describe the behavioral science perspective on gun violence comes from a conversation I had a few years ago with a staff supervisor at the Cook County Juvenile Temporary Detention Center. The supervisor noted that in his view a modest subset of the kids in the facility were genuinely dangerous: “If you let them out, they’ll go on to hurt other people.” But he said he always tells the other 80%, “If I could give you back just 10 minutes of your lives, none of you would be here.”

Changing the Choice Environment

One way to give someone 10 minutes of their life back is by changing the choice environment. In the case of gun violence, this can take the form of having someone else step in and save you from yourself when you make a mistake in a heated moment, to interrupt the argument before it escalates into tragedy. Think of it as a sort of life-saving (and labor-intensive) nudge.

Note that conventional wisdom suggests that any sort of violence interruption like this should be a waste of time. Since (in this view) violence stems from big societal incentives, and those incentives aren’t changed by interruption, violence interrupted is merely violence delayed. The behavioral science perspective makes a different prediction. If arguments escalate because people are behaving automatically in a difficult moment, the motivation for violence in such cases should be relatively fleeting. So violence interrupted can be violence prevented.

In my Hyde Park confrontation with my neighbor, I got lucky that the University of Chicago hires so many security guards and university police officers that someone happened to be around to step in and de-escalate things. This sort of violence interruption—prevention—is part of what police regularly do—one reason, I believe, for why research shows that when cities hire more police, not only do serious crimes decline but arrests for serious crimes also decline. ⁵

Unfortunately, most urban neighborhoods in the United States don’t have a University of Chicago nearby to provide Hyde Park-levels of resources. So what to do instead? One of the most important (and low-cost) things cities can do to get more violence prevention from what they already spend on police and other first responders is capitalize on the fact that violence risk is highly concentrated by time and place and use data to focus resources there.

Why is violence risk so predictably concentrated? Consider the liquor store not so far from the University of Chicago that closes at midnight, while most other South Side liquor stores close by 10 p.m. From 10 p.m. to midnight, everyone from all over the South Side, with the grudges and beefs they’ve been carrying around against those from other neighborhoods, all converge at the same spot. The result is, predictably, lots of arguments and lots of shootings.

No human being, not even a trained police intelligence analyst, will be able to detect that needle in a haystack: The location is a gun violence hotspot, but, according to the best available statistical algorithm, only for that two-hour window each day.

This suggests a low-cost (almost no-cost) and highly effective policy intervention for cities: Use algorithms, rather than human intuition, to predict the high-risk places and times to concentrate first responders. A large-scale randomized controlled trial carried out by the Los Angeles Police Department several years ago showed that this strategy can fully double the amount of crime prevention cities can get per existing officer. ⁶ While some people worry that using data to target police resources may exacerbate bias in the criminal justice system, research has shown that targeting using algorithms rather than humans does not substantially change the racial composition of who gets arrested. ⁷ Given the potential for fewer crimes at low cost, what cash-strapped mayor (or city resident) wouldn’t want that?

The main challenge for cities has been getting cops and police commanders to use these data tools, what behavioral scientists call algorithm aversion. ⁸ In my experience, few cities are making nearly as much use of data and algorithms as they could. Part of the problem is that humans give up too quickly on the algorithm if it makes a mistake, not recognizing the benchmark for the model is not perfection but rather the model’s alternative (humans). ⁹

Part of the problem is also human overconfidence in their own judgments. I asked a Chicago cop once what he thought of some risk prediction algorithm the city was using. He said “It’s so dumb, it predicts these people to be high risk when I know they’re really not. Plus, I mean . . . I never saw a computer program put cuffs on anyone.”

The holy grail for behavioral science is to figure out how to get humans and algorithms to work together to do better than either can alone. Statistical models can predict on average more accurately than humans can.^10,11 But humans often have information that is not in any dataset and so have their own source of comparative advantage. Solving this “man plus machine” problem, to help people learn their comparative advantage relative to the algorithm (and vice versa), would be of enormous value for solving all sorts of policy problems.^12,13

Another way to nudge people away from gun violence at low cost is to realize violence interruption doesn’t need to be done just by cops; any community resident can serve as an eye upon the street and interrupt trouble. This is a key insight because there are way more civilians than police. If there were some way to harness the capacity for everyone to help step in and de-escalate conflict, what criminologists call informal social control, that could be a game changer.

Indeed, a growing body of randomized experiments and natural experiments shows that almost anything that gets more people out in public can prevent violence. For example, in one study, cleaning up vacant lots or fixing up abandoned buildings reduced shootings in the vicinity by 10%.^14,15 In another, changing the presence or absence of retail stores changed foot traffic, which in turn changed crime rates by up to 20%. ¹⁶ Improving street lighting can reduce serious crime by 45%.^17,18

The implication is that urban planning can play a surprisingly important role in public safety. There is a big debate underway nationwide about the “abundance agenda,” to reduce government regulations in ways that make it easier to change the built environment. ¹⁹ Why not roll out any new abundance agenda first in the most underserved urban neighborhoods where gun violence is disproportionately concentrated?

Behavioral scientists also have an important role to play here. Sociologists have found big differences across neighborhoods in residents’ willingness to prevent trouble by intervening, or collective efficacy. ²⁰ But no one knows how to create more collective efficacy. That seems like a natural extension of some of the bystander effect work in psychology from decades ago that, if solved, could help save many, many lives.

Change the Chooser

When I helped start the University of Chicago Crime Lab in 2008, the conventional wisdom in behavioral science was that we’re limited to changing the choice environment because changing the chooser was just too hard. System 2 slow thinking is just too effortful to be constantly deployed, so System 1 fast-thinking mistakes are just an unavoidable fact of life. ³

Lucky for me, I didn’t know any behavioral science at the time. So we stumbled into a series of R&D partnerships with local government and nonprofits to test different ways of changing the chooser, which I later realized is reflected in the growing interest in psychology on boosts. ²¹ The data, perhaps surprisingly, suggest this can indeed be very helpful.

Our first Crime Lab project was the study of a program called Becoming a Man (BAM), which was submitted by the nonprofit organization Youth Guidance. BAM, like a lot of the submissions we received, included many CBT elements that come out of clinical psychology. Aaron Beck of the University of Pennsylvania, one of CBT’s founders, wrote a wonderful book called Prisoners of Hate, in which he argued that most violence was committed by someone who, in their own mind, was actually the victim—the good guy. ²² Clinical psychologists started looking into the distorted thinking patterns possibly leading to this conclusion and how much of that happens below the level of consciousness.

Part of what BAM does is help people recognize when it’s worth trying to slow down their cognition. Even if it is too exhausting to use deliberate thinking all the time, if someone is in a situation that looks like it might be leading to conflict, BAM helps people realize the value of “stop, look, and listen.” Other programs like this encourage people to, for example, count to 10 backward using every other number, which for most of us requires deliberate System 2 thinking.

But BAM and other programs like this don’t just get people to avoid thinking fast; they also push people to revisit specific System 1 assumptions that might lead to trouble that the person might not have even realized they made (that is, thinking past). Without this prompt, people might spend a lot of time engaging in deliberate thinking downstream of some mistaken upstream System 1-generated feeling or assumption and never catch the upstream mistake they’ve made.

The core idea here is illustrated by a study in which Stanford University undergraduates were asked to answer a hypothetical question: “You drive up to San Francisco with friends in order to celebrate the end of the [academic] quarter. The plans include dinner and then some entertainment afterward. How much money will you personally spend on the dinner?” ²³ Subjects come up with a dollar amount answer. Some students were randomly assigned to think about what assumptions they’d made about the scenario that the question hadn’t explicitly spelled out, which specific friends would go with them, what those friends liked to do and would be willing to spend themselves, and so on. This reflection exercise got the students to realize what implicit assumptions they’d made and how they might be wrong. They then update their forecasts and realize they should be less certain about what would actually happen, that they’d jumped to conclusions and hadn’t considered all the possibilities. (Other examples of such prompting in the literature include the two-column reflection exercise that helps people recognize what was thought but not said during some interaction. ²⁴ )

How does BAM induce this sort of reflection? The first BAM exercise divides teens into pairs. One teen is given a ball; the other has 30 seconds to get it. Almost all of them rely on force to try to complete the assignment by prying the other person’s hand open or wrestling or even pummeling the other person. During the following debrief, a BAM counselor asks why no one asked for the ball. Most youth respond by saying their partner would have thought they were a punk (or something worse—you can imagine). The counselor then asks the partner what they would have done if asked. The usual answer: “I would have given it; it’s just a stupid ball.” This exercise puts people into scenarios where they can see how their System 1 had made an implicit assumption about the situation they were in. They get feedback in a low-stakes situation that they mindread the other person incorrectly, with the hope that the next time they’re in a high stakes situation they slow down and prompt themselves to think about whether there are any key upstream assumptions they’ve missed.

Conventional wisdom says BAM shouldn’t prevent violence. Yet it does. The first experiment we carried out, with a few thousand middle- and high-school-age male students in Chicago Public Schools, showed that the program increased high school graduation rates by 20% and led to a drop in violent crime arrests of nearly 50%. ²⁵ A few years later, we did another experiment and again saw nearly 50% reductions in violent crime arrests. We also saw declines in weapons offenses (like illegal gun carrying) by 33%.

Importantly, the evidence for this type of change-the-chooser intervention doesn’t come just from BAM. For example, Chicago Public Schools partnered with Children’s Home & Aid (now Brightpoint) and Youth Advocate Programs on Choose to Change, which combines BAM-like programming with a mentor to serve teens who are only lightly connected to school and are at higher risk than those in BAM. Through the first six postprogram months, violent crime arrests declined by about 50%; through 36 months there were still signs of an effect. ²⁶ Heartland Alliance’s READI Chicago program (READI stands for Rapid Employment and Development Initiative) focuses on adult men at extremely high risk for gun violence involvement and combines a subsidized job with BAM-like programming that the participants nicknamed CAD (for control-alt-delete). The job is unlikely to be the active ingredient here since the evidence shows that jobs programs for adults do not prevent violence.^27,28 So program impacts are more likely to come from the CAD component. While the data are a bit noisier than we’d like, the estimated impact on shooting arrests taken at face value implies a 64% reduction. ²⁹

Moreover, it’s not just “name brand” behavioral science programs that have shown results; we’ve seen encouraging results from all sorts of other interventions that help promote reflection. Restorative justice gets people to write letters of apology that include some reflection on the event that got someone into trouble. Nearly a dozen different experiments suggest that restorative justice reduces crime involvement.^{30 –32} A study of reflective journaling with inmates in an Ashville, North Carolina, jail found a 25% decline in recidivism. ³³ A Midwestern juvenile detention center combined reflective journaling, mentoring (which often prompts reflection), and reading and talking about the “great books” and found large reductions (on the order of 60% or more) in serious crimes. ³⁴

These findings suggest a strategy for policymakers worried about gun violence but without lots of money to solve the problem: diffuse BAM-like programs as much as they can in every adult and juvenile detention facility they have. This can be done at low cost since, as we found out in Chicago, it’s possible to train existing detention staff to effectively deliver the program themselves, thus avoiding the cost of hiring outside specialists to do it. ²⁵ This was remarkably cheap and yet still reduced recidivism by about 20%. Interventions like these can have real impact at scale given the outsized share of people involved in gun violence who pass through these detention facilities; data from Chicago, for example, show that around 90% of shooters and 80% of shooting victims have a prior arrest record.

Behavioral scientists can help make these policies as effective as possible and help us better understand how and why these programs work, how to make them work better, and how to scale them. Right now, these programs somewhat try to “cover the waterfront” in addressing a wide range of cognitive errors, with the hope that they capture the biases and heuristics and other errors that are most important in contributing to gun violence. But we don’t really know the landscape of the cognitive distortions that most frequently lead to shootings. Nor, for that matter, do we know the most effective ways to help people recognize and avoid these violence-inducing mistakes and generalize the stylized program lessons to different real-world settings.

A particular challenge is scale. For example, we have seen that BAM can have amazing results, but as the program has scaled over time in Chicago there is a real question of whether impacts can be sustained. ³⁵ Could new generative AI technologies help scale programs like this by helping participants practice and reinforce the skills BAM is teaching in ways that are both low cost and replicate perfectly (since software scales perfectly)? ³⁶ This possibility is suggested by 2025 research by Brynjolffson et al., who studied a generative AI tool’s effects on the productivity of frontline customer service workers. The tool gives the employees suggestions about how to respond to customers. Not only does the tool improve worker productivity, the recommendations also appear to help workers learn how to better navigate these situations: At some point during the study period, the AI technology failed and the recommendations stopped, yet the gains in worker productivity persisted.

A Reason for Optimism?

For the past 100 years, public policy has assumed that gun violence, our most serious, socially costly crime, is the result of deliberate, rational benefit–cost calculation. The result has been that our policies have focused largely on how best to disincentivize violence, either bigger sticks (more prison) or more enticing carrots (jobs or social programs to lure people away from crime). Those efforts have been expensive, because changing incentives by enough to really change behavior is often expensive, and they haven’t been particularly effective. The murder rate today is about the same as it was in 1900, which has led many Americans to conclude that the problem of gun violence might be just too big to fix.

Behavioral science gives us a very different way to see the problem: that even for such a serious crime, one with such life-changing consequences, most may be driven by fast, automatic System 1 thinking. Once we can see the problem more clearly, it seems to be much more solvable than we’ve long thought. A growing body of evidence shows that behaviorally informed policies that change the choice environment and even change the chooser can have large effects in reducing shootings. And many of these are, unlike policies to change incentives, quite cheap to implement.

In other words, behavioral science finally gives us the thing that has been in shortest supply for the problem of American gun violence: a reason for optimism.