Researchers are raising red flags about a recent study on race and deadly encounters with police in the United States, highlighting the difficulties in measuring racial bias. The study claimed that white police were no more likely than their nonwhite colleagues to shoot minorities. But now, other researchers say the study was flawed and that it adds little to the debate over whether minorities have a greater chance of getting shot by police than white civilians.
“It’s just a completely indefensible conclusion to draw from the data that’s available,” says Dean Knox, a political scientist at Princeton University who published a critique of the study this month. To begin to justify such a claim, he says, researchers would need to know how often black and white civilians encounter police officers—something the authors of the original study did not consider in the paper.
Another criticism: The study did not investigate the possibility that all police—white and nonwhite—could be biased in shooting black men, says psychologist Phillip Atiba Goff at the Center for Policing Equity and John Jay College of Criminal Justice in New York City. “It’s not a serious framing of bias to think that white people have bias and other people don’t,” he says.
In the original study, published on 22 July in the Proceedings of the National Academy of Sciences (PNAS), researchers at Michigan State University and the University of Maryland compiled a list of more than 900 fatal U.S. police shootings in 2015 using crowdsourced databases from The Washington Post and The Guardian. Then, they asked police departments for information about the race of the officers responsible for the shootings. They found black police were more likely to kill black civilians than white civilians. However, the same held true for white and Hispanic officers: Each group of police was likelier to shoot civilians of their own race. That’s likely true, the researchers say, because police tend to be drawn from the communities they work in and are thus more likely to have deadly encounters with civilians of the same race. They concluded there were no antiblack or anti-Hispanic disparities across police shootings—which, critics say, should not be used to jump to conclusions of no racial bias.
The finding was picked up by major media outlets and rebounded across the internet, refracted through different political lenses. Without scrutiny, it seemed to undercut one of the central tenets of the Black Lives Matter movement: that unarmed black men have died too often at the hands of police.
Now, Knox and Princeton political scientist Jonathan Mummolo are pushing back against those conclusions with a critique that was published on the preprint server SSRN. They say the PNAS study is “uninformative” about racial bias because it assumes that black and white officers encounter black civilians in equal numbers. They illustrate their critique with a thought experiment in which a black officer encountered 90 black civilians and 10 white, whereas a white officer encountered the reverse. If both officers shot five black civilians and nine white civilians, the raw results seemed to validate the approach of the PNAS study: The white officer was indeed no more likely than the black officer to shoot a minority. But once the encounter rates were taken into account, it was clear that the white officer shot 50% of the black civilians they encountered and only 10% of the whites, revealing obvious racial bias, the authors write.
Knox and Mummolo submitted their critique to PNAS as a letter to the editor, but the journal declined to publish it. PNAS did not say why, citing the confidentiality of editorial decisions.
David Johnson, the lead author of the PNAS study and a socio-cognitive psychologist at the University of Maryland in College Park, says the criticisms are largely about how his team framed its research question. He argues that focusing on officer race in fatal shootings is valid, too, and has practical implications. For instance, he says his findings suggest that simply increasing diversity in police forces might not reduce racial disparities in fatal shootings, even if it leads to an increase in public trust. “It’s a smaller question, but it’s something we can actually answer with the data,” Johnson says. He and his co-author have published a formal response to the critique.
Knox and Mummolo say a better way to look for racial bias in police shootings is to compare incident rates to a benchmark, such as population or crime rates. For instance, a 2015 study of police shootings found that unarmed black men are 3.5 times more likely to be killed by police than unarmed white men, even after factoring in local crime rates.
An ideal benchmark, Knox says, would be the numbers of police-civilian encounters in similar circumstances. But these encounter rates are difficult to collect and nearly impossible to compare across jurisdictions. Researchers have begun to use footage from body cameras to examine racial bias in cities such as Oakland, California, but few police departments are willing to share their data. Knox, Mummolo, and their Princeton colleague William Lowe have proposed gathering video footage from highway speed cameras as a way to show what factors lead to driver stops in the first place. License plate numbers could be used to identify the race, age, and gender of the driver, and the footage itself could reveal information that might go unrecorded in administrative data—for example, whether a driver’s offensive bumper sticker could have angered the officer.
The question is not just about gathering more data, but the right data, Mummolo says. This year, the Federal Bureau of Investigation launched a “use-of-force” database in an attempt to collect records for nationwide statistics. It’s one of many new databases that researchers can use; most, however, still exclude information on incidents and encounters when police did not use force.
Documenting every use of force, Mummolo says, is important but insufficient. “What we’re trying to do is develop research designs that allow us to study how police come into contact with civilians in the first place.”
*Clarification, 16 August, 1:45 p.m.: This story has been updated to clarify the findings of the PNAS study team and comments from critics.