Focused Deterrence Helped Baltimore Cut Homicides Without Saturation Policing

Baltimore’s focused deterrence strategy was designed around a simple empirical premise: a very small number of people and groups drive a large share of serious violence. Aaron Chalfin described that concentration as “the Pareto principle on steroids.” The strategy Baltimore adopted in 2022 did not try to flood high-crime neighborhoods with police or aggressively enforce low-level offenses. It tried to identify the gangs, groups, disputes, and individuals most closely connected to shootings, then direct both enforcement and services at them.

The citywide result, as Steven Davis framed it, was a 60% reduction in Baltimore’s homicide rate over the next three years, with shootings and carjackings also falling. But Max Kapustin and Chalfin did not claim that focused deterrence can be credited with the full citywide decline. Chalfin put the caveat plainly: “we will never be able to disambiguate” how much came from the focused-deterrence intervention rather than other forces, including reversion to the mean or changes in the prosecutor’s office.

The force of their claim is narrower and more empirical: several patterns point toward focused deterrence playing a material role, especially in gun violence. The sharpest early comparison is inside Baltimore. Focused deterrence launched in the Western District at the beginning of 2022 and was not extended to another district until 2023. During the first 18 months, Chalfin said, the Western District saw about a one-third reduction in homicides and shootings relative to comparable parts of Baltimore, and about a 40% reduction in carjackings.

The design of the evidence matters. The researchers compared the Western District with other Baltimore communities that had very similar pre-intervention trends, using what Chalfin called synthetic differences-in-differences. Starting in January 2022, when focused deterrence began, those trends diverged. Davis restated the logic in plainer terms: if a treated district experiences a large reduction in homicides and shootings while other similar Baltimore neighborhoods do not, the natural interpretation is that the treatment had a major positive effect, absent some unmeasured force unique to the Western District.

Kapustin added that the pattern was not a general crime decline. The reductions appeared in homicides, shootings, and carjackings — categories closely tied to gun violence — not in crimes further removed from that problem, such as property theft or home break-ins. That specificity, he said, is a signature one would expect if focused deterrence were working, rather than if something broad and coincidental were improving all crime conditions in the district.

The researchers also did not see evidence that focused deterrence worked by dramatically increasing overall enforcement. Kapustin said there was no detectable change in overall arrests, nor in more discretionary arrests such as drug arrests. What did change was the composition of enforcement: arrests for serious violent crimes increased substantially. To the researchers, that looked less like saturation policing and more like a redirection of police effort toward the violence the strategy was meant to address.

The cross-city evidence tells a similar but less clean story. Baltimore was not the only city to experience falling homicide in the years after 2022. Chalfin said many cities saw crime reductions over the last couple of years. But when Baltimore is compared with other U.S. cities that had similar homicide trends over the prior decade, its homicide reduction was about 25% larger than in similar cities. Again, the distinctive decline was concentrated in homicide. For other assaults and other forms of violence, Baltimore looked more like cities with similar trends.

That matters because some alternative explanations would predict broader effects. Chalfin gave prosecution as an example: if a change in prosecutorial practice were the main force, one might expect it to affect many violent crimes, not homicide alone. The observed pattern instead resembles the footprint of a strategy aimed at group-based gun violence.

The case for focused deterrence begins with the scale and concentration of homicide risk. Aaron Chalfin said that in 2021, before Baltimore’s decline began in earnest, about 21,000 Americans were killed in homicides. That number is large on its face, but he argued that the aggregate figure understates what homicide risk feels like in the neighborhoods and networks where it is concentrated.

For some young men, even after recent crime declines, Chalfin estimated the risk of dying by homicide over the next year may be around 3%. He compared that one-year single-cause mortality risk to the mortality risk faced by someone around 80 years old, and said it was greater than what U.S. soldiers faced at the peak of the Iraq War.

Steven Davis drew out the implication of the comparison. Soldiers who pass through combat zones are often understood to experience traumatic stress from the fear and violence around them. Young men in the highest-risk parts of some American cities live with that risk for years. It is not only the chance of being randomly shot, Davis said, but the constant awareness of danger — for oneself, a brother, a mother, or other family members.

Max Kapustin said roughly 100,000 Americans in a year may be homicide victims or people struck by gunfire who survive. If direct witnesses, victims of armed robbery, or assault victims are included, the affected population reaches into the hundreds of thousands. Because the violence is concentrated in a handful of communities, its consequences are not distributed evenly. For people at the center, he said, it governs day-to-day activities; for people on the periphery, it may mean knowing friends or family who have been victimized, suffering PTSD, or living in a community shaped by fear.

Chalfin also emphasized spillover costs beyond direct victimization. In violent neighborhoods, there is less investment in new businesses and less economic growth. People may tolerate long car or train commutes rather than live or work in places they see as unsafe. In an era when more professionals can work from home, he warned that “the entire project of cities” is at risk if violence is not controlled.

Baltimore fit that national pattern in an especially severe way. Davis noted that the city had among the highest homicide rates of large American cities for decades, and that conditions worsened after Freddie Gray died in police custody in 2015. Chalfin said homicide rose gently in a number of U.S. cities starting around 2015, but in Baltimore the rise was not gentle. Baltimore did not experience the same kind of 2020 spike that some other cities did — Chalfin contrasted it with Philadelphia — because violence was already very high and remained so.

Kapustin said that after 2015 Baltimore moved from being around the 95th percentile among large cities in homicide rates to trading off with St. Louis for the highest or second-highest position in a typical year. He also noted that Baltimore, and especially the Western District, has become culturally associated with gun violence, citing The Wire and other David Simon work. He said Mayor Brandon Scott has publicly lamented that association, but it has become hard to separate the city’s public image from its violence problem.

The Western District was the first site of the 2022 focused deterrence launch. Chalfin described it not only as Baltimore’s most violent police district, but as an extraordinarily poor place. He said the poverty rate is among the highest of any urban area in the United States, and that 36% of the housing is vacant. The vacancy is visible: boarded-up buildings, a physical condition he said is not seen in most U.S. cities. That made the district an unusually difficult place to live, to police, and to serve through social service providers.

Steven Davis pressed the distinction between focused deterrence and older models of proactive policing. The strategy is not to identify violent neighborhoods and put many more police on the ground. It is not to aggressively enforce small-scale infractions. Aaron Chalfin agreed: “It is not that.”

The difference is central. Chalfin acknowledged that concentrating police in high-crime areas can reduce crime, but described the payoff as modest. He also said the public-safety benefits of making many stops and low-level arrests are limited, because most people arrested for low-level crimes are not at high risk of shooting someone. Focused deterrence instead asks which gangs, groups, and disputes are actually driving violence.

That produces different treatment for different people. If someone is suspected of committing a shooting and there is evidence, Chalfin said the strategy is not to “give this guy a pass.” Law enforcement should try to make the arrest, and if others participated, prosecutors can bring conspiracy charges or use other available tools. The strategy includes proactive investigations, gang takedowns, and efforts to incarcerate those who have committed serious violence.

But focused deterrence also identifies people on the periphery of violent networks: people who may be committing crimes, may be at risk of committing gun violence, may be future victims, or may retaliate after a shooting. For them, the intervention tries to make law enforcement attention highly salient while offering a different path.

The core operational tool is direct notification. Teams go into the community and knock on doors. Chalfin described a typical team as including a police detective or officer, a community representative, and a social service provider. The law enforcement representative delivers the deterrence message: you are on our radar; we know what you and your friends are doing; we are watching you. The community member delivers what Chalfin called the community’s moral voice: gun violence is unacceptable and is destroying the fabric of the neighborhood. The service provider offers help with jobs, skills, and alternatives.

Davis argued that this is more than a carrot-and-stick model. It recognizes multiple human motivations: fear of incarceration, moral reasoning, aspiration, and opportunity. A person at risk of committing violence may also care about family, neighborhood, status, future prospects, and whether another life is realistically available. Chalfin called the strategy pragmatic and realistic. Someone who pulls a trigger, he said, is not necessarily “a monster.” The person may be caught in salient disputes and unable to see another way to settle them. The strategy tries both to deter that behavior and to make another path visible.

Max Kapustin explained why the directness of the message matters. Ordinary deterrence is often mediated by noisy signals: a person may hear about others encountering police, or see more officers in a neighborhood, and infer that enforcement attention has increased. Focused deterrence cuts through that noise. Kapustin described it as “literally having people show up on your doorstep and say you are a focus of this particular strategy.”

Baltimore’s implementation involved recurring identification of people and groups connected to recent shootings. Law enforcement examined incidents thought to have a nexus to group violence, then identified suspected trigger pullers, likely future victims, and potential retaliators. Those targeted for services were connected to jobs, education, life coaching, therapy, and related supports. Those targeted for arrest became the subject of individual arrests or group takedowns.

Kapustin said preparatory work before the 2022 launch allowed Baltimore to identify individuals involved in group-based violence before the program began. The city then waited a full year before expanding to another district. He called that restraint unusual. In local government, he said, when something appears to be working, there is often a strong incentive to push it out quickly to as many places as possible, risking dilution of impact. Baltimore deserved credit, in his view, for a more deliberate rollout.

The researchers tried not only to estimate whether focused deterrence reduced violence, but to understand how. Max Kapustin named the main mechanisms as deterrence, incapacitation, and services. Steven Davis separately pressed on messenger legitimacy — whether the presence of respected community members changed how the message was received.

The quantitative evidence is strongest for targeted enforcement and weaker for deterrence itself. The project had individual-level administrative data, which Kapustin said is unusual in the focused-deterrence literature. The researchers knew which individuals were identified by the strategy, whether they were selected for communication and services or for arrest, and what happened to them afterward in administrative data — including later arrests or victimizations.

For deterrence, however, the administrative data were noisy. The researchers looked at people who received doorstep notifications and services, but could not rule out that the message had little effect on later arrest outcomes. That does not mean deterrence did not operate; it means the administrative data do not cleanly reveal it.

The qualitative work gave a different kind of evidence. Kapustin credited their colleague Brian Wade, a qualitative researcher, with conducting focus groups and interviews with individuals who participated. Those interviews suggested that some recipients took the deterrence message seriously and changed their behavior. They reported avoiding certain people and locations that they believed would put them at risk of law enforcement contact or violent conflict.

Davis emphasized the value of that evidence. It is not the standard economist’s research style, he said, but in this context it helps get inside the heads of people whose behavior the strategy is trying to alter. He gave the example of a participant explaining that he no longer went to the corner grocery store because he might encounter people he was trying to avoid, people who could draw him into trouble.

Kapustin agreed. The qualitative interviews captured reasoning that arrest records alone could not. The participants were not merely becoming abstractly more compliant; some were reorganizing daily routines, avoiding places, and distancing themselves from particular peers.

The incapacitation evidence was clearer in administrative data. For individuals targeted for arrest, Kapustin said the researchers saw “an order of magnitude increase” in the likelihood of arrest in the first quarter after identification. Law enforcement did, in fact, apprehend many of the people selected for arrest.

But the data did not indicate long periods of detention. Kapustin reasoned that if people were held for extended periods, later arrests would fall simply because those people would be unable to be arrested again while confined. The researchers did not observe that pattern. That suggested relatively short stints of incapacitation.

Even short incapacitation may matter. Group violence often involves active and ongoing conflicts. Removing someone from the street for a short period can interrupt retaliation, prevent a victimization, or give a dispute time to cool. Davis also noted that arrest serves the deterrence channel by making the threat credible: if people hear that someone was arrested soon after police said they were watching, word gets around that enforcement is real. Kapustin agreed that following through on the enforcement threat is key to making deterrence operational.

The role of credible messengers is harder to quantify. Davis described it as “messenger legitimacy or credibility”: respected people from the neighborhood knocking on doors alongside police and service providers. Kapustin said the qualitative work suggests participants responded to the presence of those community moral voices. Given the general lack of trust in law enforcement, he said, the message might not have gotten “its foot in the door” without them. He could not assign a magnitude to their contribution, but saw them as plausibly important and complementary to the rest of the strategy.

Services also matter, but the researchers were cautious about how much of the Western District decline they can explain. Kapustin said the service component includes jobs, life coaching, and related assistance. In other work in Chicago, he and others have seen substantial reductions in violence among individuals offered programs such as cognitive behavioral therapy and jobs. Applying the best sense of those magnitudes to Baltimore, he said, services likely made a modest contribution. But services alone are unlikely to explain the size of the Western District’s decline.

Aaron Chalfin sharpened that point by distinguishing individual treatment effects from area-wide violence reductions. Many randomized controlled trials test promising services on relatively small groups — for example, 300 participants receive services and 300 do not — and find meaningful reductions in violence among those treated. But such programs are not necessarily intended to generate a district- or citywide decline. In Baltimore, only about 90 people received services, yet the observed reductions were at the level of an entire police district and, eventually, the city.

Davis reformulated the distinction: a strictly services-oriented strategy might plausibly reduce homicide substantially, but it would need to be delivered at very large scale to essentially all people at risk in the relevant neighborhood or city. Focused deterrence instead treats a small number of people directly and aims to affect many others indirectly through deterrence, group dynamics, and changed expectations. Chalfin agreed, adding that the strategy tries to create a broader equilibrium effect rather than only a partial equilibrium effect among treated individuals.

Focused deterrence is cheap enough to scale, according to Aaron Chalfin. The new financial outlays in Baltimore were roughly $3 million to $5 million, depending on how one counts them. Relative to the benefits of large reductions in homicide, he described that as a tiny amount of money. It does not require creating a new government agency. It works through existing institutions: police, prosecutors, courts, social service providers, and community organizations.

That does not make it easy. Steven Davis repeatedly emphasized that the strategy requires competence and coordination among institutions that do not always work well together: police, prosecutors, social workers, job and skills programs, and community leaders. In places where civic leaders are not on good terms with police, or where social service providers and law enforcement have clashing missions and little operational trust, implementation may be difficult.

Max Kapustin agreed that the strategy requires not only coordination but the pieces to coordinate. A city needs social service organizations capable of receiving participants and providing meaningful help. It needs agencies able to work together. Public safety, he said, is not a problem police can solve entirely on their own; it requires a whole-of-government or beyond-government approach.

Chalfin noted that Baltimore had tried focused deterrence before. The 2022 implementation was the city’s third attempt. Earlier efforts occurred in the 1990s and again just before Freddie Gray’s death in police custody, an event that helped derail that attempt. That history supported Davis’s concern: the idea may be sound, but execution is fragile.

Chalfin described focused deterrence less as a discrete intervention than as an “aspirational approach” to violence reduction. It is a way cities should organize themselves: directing law enforcement and social service effort toward the people most central to serious violence. But he also said there will be hiccups and that some jurisdictions will struggle.

The scalability question therefore has two sides. On one side, the economics and criminology are favorable. The strategy is inexpensive. It matches what researchers know about the concentration of crime. Cities now have more data than in earlier decades and better tools to identify networks, incidents, and individuals. Chalfin mentioned forensic investigation of cell phones and camera evidence as examples of investigative resources that can improve targeting.

On the other side, focused deterrence does not scale like infrastructure. Chalfin contrasted it with street lighting: if a city knows how to install one streetlight, it knows how to install a thousand. Focused deterrence depends on human capital. Police and prosecutors need to know how to work together, when to seek search warrants, what evidence is needed, and how to conduct longer-term investigations. Prosecutors need experience with complex cases. Social service providers need credibility and the ability to change lives. If good providers depend on visionary leaders, he said, those leaders are not in infinite supply.

Kapustin placed the challenge less in the number of inputs and more in sustaining fidelity over time. Smaller interventions such as tutoring or jobs programs often hit a scale constraint because skilled tutors or job coaches are scarce. Focused deterrence is already being studied at neighborhood and city scale. Its harder scaling problem is institutional durability: police, prosecutors, service agencies, and community partners all change leadership over time. The question is whether their working relationship can become a bedrock feature of how the city operates.

That question remains open. Kapustin noted that focused deterrence itself is not new; more than 50 U.S. cities, and likely other places abroad, have tried some version of it. The issue is not whether cities can attempt it, but whether they can implement it well and sustain it.

Davis closed on a cautiously optimistic note: the strategy can work and has worked in some settings, but it is not the easiest thing to implement. Chalfin said colleagues at the University of Pennsylvania Crime and Justice Lab helped Baltimore with some of the work and are active in other cities. He hoped for funding to support cross-city research that could produce better evidence on how well focused deterrence scales.