What Best Predicts Mass Shooting Risk? New Model Points To Local Gun Access

Berkeley Researchers Built a Mass Shooting Simulator. Here’s What It Found

A new computer model from researchers at UC Berkeley, published in the journal PNAS Nexus, finds that the density of gun dealers, pawnshops, and firearm manufacturers in an area predicts mass shooting risk better than population density does. Federal firearms license density, as the researchers call it, emerged as the strongest clue in their simulation.

Population density told a different story. More residents meant more potential perpetrators in direct proportion, but crowding itself only mattered up to a point. Once a place reached a certain density, adding more people barely moved the needle on risk. Gun dealer density kept driving risk higher the more dealers were nearby, with no such leveling off.

To reach that conclusion, the team treated a mass shooting as three factors lining up: a motivated person emerging, that person getting a gun, and potential victims being in the wrong place at the wrong time. Researchers ran that scenario tens of thousands of times across a grid resembling real American communities, using firearms license records, census data, and four decades of shooting reports, then checked which version matched real events.

Gun Dealer Density Predicts Mass Shooting Risk Best

Researchers built their tool as an agent-based model, a digital sandbox where individual “agents” move around and follow simple rules. Each agent stood in for a potential shooter who might emerge in a given area, then wander across a 30-by-30-mile simulated grid, running into places to get a gun and groups of people along the way.

Every simulated area drew on real information, from census population and land records to the locations of federally licensed gun dealers tracked by the ATF. Actual shooting events came from the Mother Jones mass shooting database, spanning 1982 through 2024, requiring at least three deaths and excluding cases tied to terrorism, gang violence, or domestic disputes.

Because the dataset is small, with well under 200 qualifying incidents across four decades, the team ran tens of thousands of simulated outcomes per community to find which underlying settings, like how often a “motivated” person appears or how far someone travels for a gun, best explained the shootings that occurred.

Researchers tested a few competing versions of this setup. One assumed a shooter grabs any available weapon and picks a random target nearby. Another added a one-in-ten chance of encountering an especially deadly weapon rather than a basic one. A third added a chance that the shooter deliberately seeks out the group that would result in the most deaths, rather than choosing at random.

Firearm Access Limited Most Simulated Mass Shootings

Checking these versions against real shootings, researchers found that factoring in access to high-lethality weapons made a real difference in how well the model matched actual events. That version, which mixed basic and highly lethal weapons but kept target selection random, performed best overall, beating both a simpler population-only baseline and a version that ignored weapon type entirely.

Researchers also tested a version assuming shooters actively hunt for the most casualties, what they called the “max-select” model, and it performed a bit worse. The estimated rate of that kind of calculated, casualty-maximizing behavior was very low, lining up with earlier research suggesting most mass shooters target locations or people they already have some connection to.

Digging into the numbers, the team estimated that the rate at which a “motivated” potential shooter emerges in the population is extremely low, somewhere around one in a million to one in ten million people per year. It’s a sobering way of quantifying something criminologists have long suspected: while anger, grief, or crisis are common, the leap to planning a targeted mass shooting is rare.

Getting a weapon, meanwhile, showed up as a major limiting factor in the simulation. Across American regions, the model estimated that only 10 to 40 percent of motivated individuals would successfully get their hands on any firearm, and fewer than 10 percent would obtain a high-lethality weapon. In other words, within the simulation, firearm access frequently became the bottleneck between a motivated agent and a mass shooting.

When researchers ran their model across roughly 5,000 American communities, predicted risk varied widely, from about 0.1 to 4 expected fatalities per million residents per decade. The riskiest spots on the map lined up closely with places that have already experienced a mass shooting, a sign the model is picking up on something real.

Lower Gun Access Cut Mass Shooting Risk in the Model

This research will not settle America’s gun debate; the authors acknowledge their model rests on simplifying assumptions. It treats behavior as constant across time and geography, and assumes a shooter’s path is basically a random wander rather than something shaped by personal history. Current firearm dealer data also stood in for historical gun access, which may not perfectly reflect earlier decades.

Still, building a working simulation rather than relying only on after-the-fact statistics lets researchers test hypothetical cause-and-effect scenarios that would be unethical or impossible to try in the real world. The model consistently showed that dialing down firearm access, or shrinking the radius in which someone could realistically obtain a weapon, reduced simulated shooting events across every version tested. That is not a policy prescription on its own, but it points at the same lever again and again: how easy it is to get a gun in a community appears to matter more than how densely populated it is.

---

Disclaimer: This article is based on a peer-reviewed study and reflects the findings of a computer simulation calibrated to historical data, not a real-world experiment. The model’s estimates are statistical predictions, not proof of what causes any individual shooting, and come with the uncertainties the researchers themselves describe.

---