Michael Friedrich is a freelance writer living in Brooklyn who covers culture and social justice. His work has also appeared in The Nation and The New Republic.
Predictive tools can now track how a single shooting incident triggers a lethal cascade of gunshot violence—and predict who will be targeted next.
A young man who runs with the Black Disciples was shot in the foot in a dispute with rival Gangster Disciples on Chicago’s South Side, the New York Times reported late last year. Days later, two Gangster Disciples caught a bullet in a drive-by a block away. The day after that, the mother of a Black Disciple was shot in the foot, caught in the middle of more crossfire between the two gangs.
These were just a few of Chicago’s 4,368 shootings in 2016, but they’re telling: So much gun violence in American cities follows this tit-for-tat pattern of vendetta between people who know each other. What if there was a way to anticipate that and break the chain?
A new study says it’s possible to do just that. Researchers Ben Green and Thibaut Horel at Harvard and Andrew Papachristos at Yale used a social contagion model and tried to predict gunshot victimization in Chicago between 2006 and 2014.
Using police records of people arrested together for the same offense, they mapped a network of 138,163 subjects and looked at the spread of violence within it. Their model, based on the ones epidemiologists use to understand contagion, assumed that shootings were likely to spread between co-arrestees, who would have close social ties and engage in risky behavior together. When they ran probabilities on people linked to a shooting victim, what they found was staggering: 63 percent of the 11,123 total shootings in the network were part of a longer chain of gunshot victimization. The closer someone was to a victim, the greater the risk of being shot.
“Gunshot violence follows an epidemic-like process of social contagion that is transmitted through networks of people by social interactions,” concludes the study, published by the Journal of the American Medical Association last month.
This is important. Protecting the most vulnerable people starts with understanding who they are and how they’re connected.
In my work as a researcher for the National Network for Safe Communities, an organization based at John Jay College of Criminal Justice in New York that helps cities to face and avert community violence, I’ve become deeply familiar with this phenomenon. Our director, David Kennedy, developed Operation Ceasefire in Boston back in the 1990s. That basic approach has now helped many U.S. cities effectively reduce the spread of shootings by intervening directly in the group dynamics of gangs. We rely on data like this to identify patterns of shooting and design concrete solutions that police and communities can carry out together, with a focus on minimizing the use of enforcement and offering help to those at risk.
We already know that shooting victimization is highly concentrated in small social networks. Another recent study by Papachristos, for example, showed that 70 percent of Chicago’s nonfatal gun violence victims between 2006 and 2014 were found among a network of less than 6 percent of the city’s population—all people who had been arrested together at one time or another.
What this new study shows is a concentration within the concentration. Among the small network of victims, a single shooting incident can trip off a “cascade.” Intricate sequences of gun violence begin with a patient zero at the top and branch out in patterns of transmission that look like an awful Tree of Death. The figure below shows three such cascades. The largest in the study ended up touching as many as 469 shooting victims—all triggered by one earlier incident.
So how do we use this information on the ground? The study contains important guidance. “Violence prevention efforts that account for social contagion,” the authors propose, “have the potential to prevent more shootings than efforts that focus on only demographics.”
Many American cities are already using approaches that account for, and aim to disrupt, networks of victimization. Both police and communities have known about these violence cascades since long before there was sophisticated science to track them precisely: The conventional first question asked after any shooting is, “Did anyone get shot before this?”
Because these patterns have long been clear, there’s been action on them for just as long. Models like Cure Violence, first deployed in Chicago, use community-based “interruptors,” working largely on their own, to try to break the cycle of social contagion. Our own partnership-oriented approach includes “custom notifications,” a way for law enforcement, community members, and social service providers to talk face-to-face with high-risk people who are more likely to be caught up in retaliatory shooting. These notifications, used in cities such as New Haven, Connecticut, include a message from the community about how harmful the violence is, a community-based offer of help, and frank information from police about their legal risks.
In Tennessee, Chattanooga’s police and community leaders have formed a first-of-its-kind unit, the Chattanooga Police-Community Response to Victims of Violence, that visits every shooting victim in the city to offer protection and support with immediate needs. Sometimes this can be as definite as getting victims a hotel room out of town to protect them from further violence, talking their friends down from retaliating, or brokering peace with the gang of their assailant.
None of these approaches, as they stand, are good enough. But the growing understanding of network dynamics has allowed cities to shape new work that builds on what we know and sharpens it immensely. Cities are starting to set up routine “shooting reviews” with local police and state and federal law enforcement agencies to track likely new violence—almost in real time.
New Haven, for example, holds them daily, and Oakland weekly. In both cases, they’ve been vital to getting ahead of cycles of violence with both police responses and community action, like deploying street outreach workers. It’s no accident that both cities have seen historic declines in shootings in recent years—New Haven saw a 73 percent monthly average reduction, while Oakland had a 40 percent overall drop.
Primetime cop drama this is not. Instead, it’s workaday policing done anew, with thought and care and community involvement.
In recent years, a virtual cottage industry of criticism has taken aim at “predictive policing” methods that use computer modeling and algorithms. Are we trundling toward a Minority Report police state, where suspects are apprehended before committing their crimes? The roots of this criticism are, of course, well founded in anxieties over racial profiling and damage to civil liberties.
But the findings from this new study—if put to good use—would allow officials to do the exact opposite of profiling. This is about predicting who is most likely to be caught up in violence based on their position in the network, not their personal characteristics. And it emphasizes protecting high-risk people, not predicting shooters. While it’s true that today’s victim is often tomorrow’s shooter, it’s critically important that we shift our focus from offending to victimization.
One of the study’s most important contributions is to show that demographics alone won’t predict who gets shot. It’s not your age, race, sex, or even the neighborhood you live in. It’s who you hang out with and what you do together. “Treat gun homicide like a blood-borne pathogen, something transmitted from person to person through specific risky behaviors,” Papachristos has written elsewhere. “You don’t catch a bullet like you catch a cold.”
Yes, a young black man living on Chicago’s South Side faces higher-than-average risk. But if he’s a Black Disciple with friends, family, gang rivals, and co-arrestees who have been shot, that risk becomes unthinkably dire.
Our aim should be to reduce that risk. Officials can use network data to focus their efforts narrowly on protecting the most vulnerable people—instead of broadly across entire black communities, where trust in authorities is often damaged by over-policing. Community members can work with officials to stand against violence and offer pathways of support to the people most likely to be hurt.
In other words, using this data can help make policing fairer and communities safer.