The Chronicle Review

Police Violence, Out of Context

What a recent high-profile study about police shootings got wrong

Zach Gibson, Getty Images

Demonstrators gathered in Washington to protest the fatal police shootings of Alton Sterling and Philando Castile, July 7, 2016.
July 24, 2016

P hilando Castile, Alton Sterling, Janisha Fonville, Michael Brown, Eric Garner — these and hundreds of others have had their lives taken by the police, spurring the Black Lives Matter movement. While blacks are not the only victims of police violence, a wave of viral videos showing incidents in which blacks have been targeted by the police has left the impression that they are disproportionately victimized, and that police officers have an antiblack bias. But does the anecdotal evidence — the videos, the personal accounts — match up with empirical reality?

Roland G. Fryer Jr., a professor of economics at Harvard University, recently released a highly controversial report that suggests a complicated answer. According to his research, highlighted on the front page of The New York Times, police officers do tend to treat black and Latino citizens more roughly than their white counterparts. However — and this part of the study received the most attention — Fryer finds that when it comes to police shootings, officers are actually less likely to fire their weapons at blacks than at whites. That finding seemingly goes against the reams of anecdotal evidence that the police are willing to kill black citizens for a range of minor infractions. And it has come under significant scrutiny given both the timing of its release (the report appeared not long after Castile and Sterling were killed) and the general paucity of research on police-citizen encounters.

How did Fryer and his research team arrive at these conclusions?

They used four data sets — two publicly available (one from the New York Police Department, the other a national survey of civilians) and two unique data sets created for the study (one involving "officer-involved shootings" in 10 cities and counties, and the other involving police-civilian interactions in Houston).

“Does the anecdotal evidence of police violence against blacks -- the videos, the personal accounts -- match up with empirical reality?”
With the NYPD data set, which consists of more than five million individual stops from 2003 to 2013, Fryer’s analysis suggests that there are sizable and statistically significant differences between the way the police treat whites versus blacks and Latinos — with officers far less likely to use force (defined as anything from putting hands on a citizen to striking him with a baton) on whites than on blacks and Latinos. The national survey, which contains data on about 500,000 citizen encounters with the police from 1996 to 2011, generated similar results — whites’ encounters with the police were, on average, far less violent than were those of their black and Latino counterparts, no matter their income, the race of the officer involved, or the gender of the civilian. Those findings should come as no surprise to those who’ve argued for some time now that the police have an antiblack and anti-Latino bias.

However, the findings on the use of lethal force tell a different story. For officer-involved shootings, Fryer’s team constructed two data sets — one from the police records of three cities (Austin, Dallas, and Houston), Los Angeles County, and six Florida counties, and another constructed solely from police records in Houston, comprising all arrest categories in which an officer could have potentially used justifiable force (either a gun or a Taser). Fryer and his research assistants coded nearly 300 variables from 800 randomly selected cases and then collected 4,250 Taser-involved cases.

The results suggest that the police are not more but less likely to use deadly force against black and Latino citizens than against white ones. Given anecdotal evidence suggesting that officers are far more likely to mistakenly believe that a black citizen is carrying a weapon (see: Tamir Rice), Fryer’s results indicate that the police are more likely to discharge a weapon at a white citizen without having first been threatened than at either a black or Latino one. In light of what we’ve seen in video after video, these findings seem to say that our eyes have been lying to us.

F ryer’s study on the use of nonlethal force reflects the lived experience of blacks and Latinos. Almost every black male I know older than my children has had an encounter with the police. Almost every black male I know older than 40 has had more than one sketchy encounter with police.

The findings on the use of lethal force, however, reflect neither the anecdotal evidence nor, I believe, the empirical reality. And a deep dive into his study suggests that there are at least two dynamics that Fryer missed.

The first is sociopolitical context. We know that the rate of police shootings varies across municipalities and counties. Some counties have remarkably high rates, while others have low rates. Further, we know that how municipalities use police departments also varies. Some municipalities — Ferguson, Mo. comes to mind — increasingly turn to their police departments to generate revenue from citizens. (Until a recent settlement with the U.S. Justice Department, Ferguson received a significant amount of municipal revenue from police fines and legal fees.) Other authorities, like the New York Police Department, have deployed or continue to deploy "broken windows" tactics, aggressively policing minor infractions in an attempt to stave off larger ones. Counties also differ along other lines — some have high inequality and stark levels of racial segregation, others do not.

These contextual factors affect how the police interact with citizens. A police department that doubles as a revenue generator has an incentive to increase police interactions; a department that polices according to the "broken windows" principle has an incentive to hire officers interested and willing to interact with citizens in a certain way, and to train police officers in certain ways; and so on.

Such context is precisely what a 2015 paper by Cody T. Ross, an anthropology graduate student at the University of California at Davis, illuminates. Ross’s paper, which has gained attention in the wake of Fryer’s study, looks at the independent effect of a range of county-level indicators on the number of police shootings and finds that blacks are more likely to be killed than whites. Further, he finds that a variety of county-level indicators condition this rate. The higher the level of inequality within a county, the more killings. The higher the racial segregation, the more killings. The crime rate itself has no independent impact, suggesting that police shootings are a function not of the crime rate but rather of social control. Because Ross examines only aggregate shootings at the county level, we know nothing of the individual encounters that composed the data set. At the very least, however, his research suggests that context matters.

“We really have no sense of how many Philando Castiles there are.”
By contrast, Fryer’s theoretical model treats police-citizen encounters solely in individual terms. Did the stop occur at night? Did the citizen have a weapon? Was the officer black? How much money did the citizen make? It’s not that these individual-level factors don’t matter. Rather, it’s that these factors aren’t the only ones that matter — and, in fact, they may matter less than others. Police departments are not simply collections of individuals; they are institutions that shape individual behavior through rules and norms. And these institutions themselves are shaped by municipal, county, and state law. If Fryer had taken those factors into account, he would have focused not just on the individual encounter but also on the context surrounding that encounter. Did it occur in a municipality that employed "broken windows" strategies? Did it occur in a municipality that routinely tasks officers with producing revenue? If Fryer’s analysis of police shootings had looked at disparities across different police departments, he’d have had the ability to estimate the independent effect of context, something that Ross’s study gets at (albeit imperfectly).

The second major problem with Fryer’s study is that he doesn’t take into account differential rates of police stops. The police conduct investigatory stops of black citizens far more than they do of white ones, because of a combination of institutional incentives and individual biases. Every time any citizen is stopped by the police, there is a chance that stop will end up with an officer discharging his or her weapon. Stopping blacks disproportionately should almost automatically increase the odds that a police encounter with a black citizen will go wrong. It’s worth noting that Philando Castile was stopped because his "wide-set nose" matched the description of a robbery suspect, and that he was then killed as he was reaching for his gun permit. By ignoring differential rates of stops and focusing primarily on instances of high risk, in which the officer is likelier to use a gun anyway (traffic stops for more-mundane offenses, like a broken taillight, were not captured by the study), Fryer and his team did not account for the dynamics that arguably make blacks more likely to be shot by the police. (It’s important to note that Fryer says as much.)

Those problems, in addition to that of relying on police reports as the sole source of data on police shootings, render Fryer’s findings suspect. I’m not the only one to recognize the study’s problems. Which raises the question: Why report and publicize the research, which hasn’t been peer reviewed in the first place? Writing this three days after the three-year anniversary of the death of Tyrone West, and three days after a fourth officer was found not guilty of charges related to the death of Freddie Gray, I can’t help but wonder if claims like Fryer’s — claims that can have disproportionate influence on a fraught subject involving a marginalized population — should be subjected to a higher degree of academic rigor than The New York Times can provide.

We really have no sense of how many Philando Castiles there are. Let’s hope that the Times’s treatment of a deeply flawed, non-peer-reviewed study doesn’t impair other efforts to solve that mystery.

Lester Spence is an associate professor of political science and Africana studies at the Johns Hopkins University.