When Franklin Zimring decided to study lethal violence by police for his recent book When Police Kill, he learned that official records could not be trusted. As he explains in the book, data collected by the FBI (or by Vital Statistics) accounts for no more than half of the shootings reliably counted by The Guardian or The Washington Post. This is an incredibly disturbing finding; if we want to have any hope of preventing such incidents, the first step is to know when, where, how, and to whom they occur. It is even more disturbing that efforts at counting these incidents stop with lethal violence, ignoring the many more instances of police use of force that have non-lethal (and yet serious) consequences.

A new study titled “Association of Legal Intervention Injuries with Race and Ethnicity Among Patients Treated in Emergency Departments in California” and published in the JAMA Network by Alyssa Mooney, Shannon McConville, my colleague Aaron Rappaport, and Renee Hsia attempts to bridge that gap by recurring to a neutral source of information: hospital-admission information. California maintains a central repository of reports on each and every outpatient ED visit and inpatient admission, and the study covers a nearly 11-year period (early 2005 to late 2015). This database is particularly helpful because the state requires all hospitals to collect patient race or ethnicity. The data also include special codes assigned by the hospitals to each injury, which shed light on how the injury occurred: the study covers injuries classified as stemming from “legal intervention,” i.e., injuries inflicted by law-enforcement agents in the course of suppressing disturbances, maintaining order, arresting or attempting to arrest offenders, or other legal action.

Over the study period, the authors found an overrepresentation of men aged 25 to 34 among those injured by police (31.8% of the injuries and 21.1% of the population). They also found overrepresentation of black individuals, which made up 18.5% of injuries but just 6.6% of the population. The rate at which black individuals were injured was high to begin with, but increased between 2005 and 2009 by 11% (compared to 85% among white individuals and 6% among Hispanic individuals). After 2009, rates declined for all populations (by 4% for blacks and 2% for whites); by 2015, risk among black individuals was 11% higher than white individuals.

In comparing 2005-2006 with 2014-2015, the authors found changes in injury severity. There was a substantial reduction in the proportion of incidents involving firearms, which was accompanied by an increasing proportion of incidents that required a hospital admission, rather than merely outpatient treatment. At the same time, the data showed an increase in the incidence of a behavioral-health diagnosis among those injured by law-enforcement agents (though this finding might merely indicate higher alertness on the part of hospital staff to the possibility of a mental-health issue in these scenarios).

Another intriguing finding was that, while arrest rates declined, there was a modest growth in injuries. The data showed an increase in legal-intervention injuries per arrest. The research team relied on this to try to explain the changes in racial disparity of the injured individuals. It may very well be that the decrease in arrests implied that arrests were used for more serious offenses. This suggests that the disparities in injury may stem from the fact that black individuals are more frequently exposed to legal intervention in general, and arrests in particular, and thus at higher risk of being injured in the course of such action.

In reporting these findings, the authors were well aware of the limitations of the dataset. Naturally, not all those injured by the police pursue hospital care. To me, what this suggests is that racial disparities might be even bigger than those the authors estimated, as I suspect that populations that are more mistrustful of the police—particularly populations of color and low income—might be less willing to engage further with the system by seeking medical care. Some work by Sarah Brayne suggests that people who are more vulnerable to criminal-justice intervention engage in “system avoidance.” which could include hospital visits and the paper trail they entail.

How much can we trust this database? The authors themselves express concerns about the consistency of hospital record keeping over time, as well as about changing practices that might affect the longitudinal validity of the data. But it is important to keep in mind that these alternative efforts to find out more about police-involved injuries stem from the inadequacy of the FBI database in identifying them at all. Not only is the database woefully incomplete, but also it does not cover non-lethal injuries, and in that respect this is a welcome addition.

All of this may change in the very near future. Recent legislation adopted by the state has led to the creation of URSUS, a database including reports of “serious injuries” from all of California’s law-enforcement agencies. The database is still new and includes very little information on the incidents themselves; it offers no room for narrative or details. But it does offer coverage of some of the information the authors sought to glean from hospital records. In particular, the proposition requires the agencies to report, for each incident involving use of force and serious injury:

(1) The gender, race, and age of each individual who was shot, injured, or killed.

(2) The date, time, and location of the incident.

(3) Whether the civilian was armed, and, if so, the type of weapon.

(4) The type of force used against the officer, the civilian, or both, including the types of weapons used.

(5) The number of officers involved in the incident.

(6) The number of civilians involved in the incident.

It might be interesting to follow the study up with another round of hospital records, ranging for a few years starting with 2018, to see whether law-enforcement agencies’ reports to URSUS are complete and reliable, or whether we need to trust newspapers and alternative sources to learn more about these disturbing incidents.