Studying the Link Between Race and Police Killings

Since the Ferguson, Missouri protests of 2014, the issue of how race and police violence interact has consistently been a front-page news item in the United States. Recent weeks have seen the criminal conviction of three Minneapolis police officers who failed to stop the murder of George Floyd in May 2020; the acquittal of an ex-police officer connected to the Louisville, KY drug raid that ended with Breonna Taylor’s death two months earlier; and the decision to not file charges in the case of Amir Locke, a black man who was fatally shot by a Minneapolis SWAT team officer in February. The former decision has been applauded by those who see it as a step toward rolling back racist police abuses, while the latter two have been condemned as a perpetuation of the status quo. Meanwhile, two economists have just published the latest in a series of academic analyses of the issue, this one appearing in the American Economic Review under the title, ‘Does Race Matter for Police Use of Force? Evidence from 911 Calls.’

The new study, by Mark Hoekstra and CarlyWill Sloan, gets at a question that many of us ask when these tragedies happen: All other factors being equal, are black suspects more likely than white suspects to be harmed during encounters with the police (white officers, in particular)? Or is this impression created, at least in part, by the fact that killings of African-Americans by police officers now are likely to get media attention, while similar killings of white victims sometimes fly under the radar? The new study casts light on the issue by reporting findings from a city in which, as the researchers detail, white officers are indeed far more likely than their black colleagues to discharge their guns when sent on 911 calls in black neighborhoods—even when white and black officers work the same beats and shifts.

But the authors admit that they don’t know for certain why this is the case. Perhaps trigger-happy, racist white officer are killing people without justification in black neighborhoods. Or perhaps black officers are better at handling tense situations in such areas. Or perhaps suspects in these neighborhoods react differently to officers of different races. Or perhaps some combination of these factors (and various others, besides). Unfortunately, the city under study remains anonymous in the report, so there’s no way for third parties to investigate the individual cases further. Nor is there any way to determine how well the results would generalize to other places.

As I detail in a new report for the Manhattan Institute, “Fatal Police Shootings and Race: A Review of the Evidence and Suggestions for Future Research,” this field of research has witnessed immense strides in the past half-decade or so. But researchers still have not definitively answered the core question about officers’ mindset. And thus far, every existing fact-finding approach has come with significant limitations.

The Hoekstra-Sloan study is one of the methodologically strongest to date. And critics of the police will point to it as evidence that race matters, at least in some places and in some ways. But the academic literature has produced a wide range of findings. And proof of rampant, extreme forms of police racism hasn’t readily leapt from the data in any kind of unambiguous way, as many observers understandably expected would be the case. Instead, the debate has come down to complicated issues of analysis and interpretation.

One lesson that emerged from Ferguson and subsequent events is that the government has done a terrible of job of tracking police killings. As a stopgap measure, private actors have stepped in to help fill this knowledge void, including the Washington Post and a collaboration called Fatal Encounters, though each uses somewhat different criteria for inclusion in their databases.

By themselves, these raw numbers tell us a lot we didn’t know before. On-duty police officers shoot and kill about 1,000 people every year. And the total number and racial composition have been fairly steady since 2015, with black Americans accounting for a little more than a quarter of those shot by police—and about a third of the unarmed individuals killed (though it is worth stressing that someone without a weapon can still pose a serious threat to an officer, a bystander, or himself).

The question is how to turn such data into a useful analysis of police behavior. Blacks’ share of the general population is only about 13 percent. But this doesn’t tell us all we need to know, since crime (and crime victimization) isn’t evenly distributed across racial groups. And so even if this disparity is itself connected to underlying patterns of racism in American society in multiple ways, it still must be corrected for in any analysis that’s aimed at statistically isolating officers’ actual behavior. If crime isn’t evenly distributed across racial groups, then the deadly encounters associated with even the most fastidiously non-racist police corps will exhibit the same statistical pattern.

Some studies try to “benchmark” racial groups’ police-shooting rates to the crime rates associated with each group. However, racial disparities are different for different crimes, and there’s no consensus as to which crimes should be examined for purposes of this kind of analysis. Moreover, crime rates are difficult to measure accurately in and of themselves; and not everyone shot by the police is a criminal, or even a suspected criminal. Yes, the black share of known murderers (and cop-killers in particular) is higher than the black share of those shot and killed by police. But in light of all the other uncertainties at play, that fact alone is hardly enough to settle the debate.

Other studies focus on defined types of encounters, in which police officers either used lethal force or did not—for example, cases in which officers used a taser instead of a gun, or pointed a gun but didn’t fire it. The researchers who’ve produced these statistical models have tried to account for a wide range of important variables, such as the suspect’s weapon (if any), the reason the officer had been called to the scene, and so on; with the goal being to allow them to study what happened to white versus black suspects in reasonably similar situations.

The most famous study in this vein is Roland G. Fryer, Jr.’s 2019 paper, ‘An Empirical Analysis of Racial Differences in Police Use of Force.’ He found no anti-black bias in lethal force across numerous models (with the most rigorous analysis being limited to data from Houston). However, the information that informs such studies often comes, at least in part, from police departments themselves, which naturally raises all kinds of questions about bias. Moreover, no model can account for everything that’s important about every situation. This includes the question of whether racial bias presented itself as a significant factor before the decision to use lethal force was made.

Consider, as a thought experiment, data showing that a white police officer was less likely to shoot a black person once the officer had unholstered his or her gun. On the surface, this would look like evidence against racist intent. But this inference becomes much weaker—and perhaps even unsustainable—if one considers the possibility that officers are significantly more likely to unholster and aim their weapons at blacks to begin with. This field is full of puzzles such as this.

Still other studies aim to examine the races of the police officers themselves. And these have often found that white officers aren’t disproportionately involved in shootings of blacks. The most well-known of these studies was the subject of a bizarre 2020 academic fiasco in which, after a protracted pressure campaign, the authors retracted their work, even while maintaining that their statistical methods were “appropriate for investigating whether officer characteristics are related to the race of civilians fatally shot by police.” Going forward, this example suggests that we must not only be mindful of technical forms of statistical bias that may compromise any given data set or scholarly analysis; but also more general manifestations of political bias, which now may affect a scholar’s decision to report (or even investigate) certain facts in the first place.

The (not unreasonable) presumption in this area of inquiry is that black officers harbor less anti-black bias than their white colleagues. But this approach raises thorny issues as well. Information about the race of the involved officers isn’t always publicly available. Some locales have higher black populations than others, and these areas will tend to have more black suspects and more black officers. So studies of this type need to be run in a way that carefully accounts for local demographics. Complicating matters is the fact that even within some cities, black officers are disproportionately assigned to police black neighborhoods. (That’s why the Hoekstra-Sloan study was such a breakthrough: Their data are so detailed that they can account for the individual beats and shifts that police officers work, not just overall demographics.)

The question of geography presents other complications as well. A police officer who uses deadly force may be responding to fears (race-related or otherwise) that originate not only with the person being shot, but with the neighbourhood in which the encounter takes place. So there’s been work done on determining whether heavily minority neighborhoods have more police shootings, even after controlling for crime rates. Others have asked if locations in which white residents score poorly on controversial “implicit bias” tests also exhibit wider racial gaps in regard to police shootings. Reported measures of these correlations can be interesting; but as yet, it would be hard to use them to prove any kind of concrete thesis concerning officers’ motivations (even if all such studies pointed us in the same direction—which they don’t).

As I discuss in my new Manhattan Institute report, there are a number of useful ways in which this research literature can be developed. One would be for researchers to move past simplistic designs that merely “benchmark” police shootings to crime rates; I’d argue that there’s simply not much more to be learned on that front. Another would be to mine data from “natural experiments” that already offer pre-supplied control groups—as with the Hoekstra-Sloan study, whereby researchers were able to see how black and white officers responded to otherwise similar sets of 911 calls. Still another is for the government to finally collect these data in a comprehensive manner, several efforts at which are already underway; and to expand the information collected to include, for example, officer race and incidents that didn’t result in fatalities.

Given the widespread outrage that has followed in the wake of police killings of black victims, many Americans have understandably come to expect that clear proof of flagrant and widespread racism among officers would be provided by researchers. As explained above, however, we are still waiting for clear answers to emerge. Until then, our main consolation will be that Americans from all points on the political spectrum have at least been alerted to the need for better data, and for a deeper understanding of why police encounters too often end in tragedy.