Mark Wallheiser/Getty Images

Why It's So Hard To Measure Racial Bias In Police Shootings

What a headline-grabbing study misses when it comes to use of lethal force

What if the core of the Black Lives Matter movement is based on a lie? What if there is no racism involved in police killings? Harvard economist Roland Fryer Jr. has come out with a headline-grabbing study about police shootings: a new National Bureau of Economic Research working paper that arrived with a write-up in the New York Times' data blog The Upshot. According to The Upshot, the upshot (!) is that "surprising new evidence shows that there's racial bias in police use of force, but not in shootings." Fryer tells the Times that it's the most surprising result of his career — and if that were really what the evidence showed, it would be very important indeed. It would mean that the Black Lives Matter movement is, at least partially, animated by a myth.

But is that actually what the evidence shows? I'm not sold on that. Let's break it down.

Fryer uses four different data sets to examine racial bias in police decisions to use both nonlethal and lethal force. Using police-reported data from New York's controversial stop-and-frisk program, as well as national data from a survey of civilians, he convincingly demonstrates that police are more likely to use all types of nonlethal force — shoving, batons, pepper spray, etc. — against blacks and Hispanics than against whites. This is true even after Fryer accounts for a wide range of factors that could determine whether the use of force is called for. This finding matches the lived experience of people of color, conventional wisdom, and previous academic studies, so there's nothing too new or controversial here.

But Fryer is after more than just results on nonlethal force. He wants to find out if racism plays a role in whether police decide to use lethal force as well, and neither of those data sets include that information. While there is some data about police shootings, it tends to be relatively short on detail. (The woeful state of our data on police shootings is a huge issue that's the unspoken background to any statistical discussion of killings by police.) To find out whether police are disproportionately more likely to shoot at blacks, Fryer and a large team of research assistants painstakingly created their own detailed data set by examining hundreds of police incident reports supplied to them by the Houston Police Department. (By the way, this means that whatever his results say, they say it specifically about Houston and its police department, so we have to think carefully about how well they apply on the national level.)

It's important to emphasize here that the question Fryer is attempting to answer is not "are blacks more likely to be shot at by police than whites." It's not even "are blacks more likely to be shot at by police than whites, given that the police have already decided to stop them." He's trying to ask something even more precise: Given a situation in which the police officer is likely to be justified in using lethal force, is he more likely to choose lethal force if the subject is black? Fryer wants to know if race plays a factor in the decision to shoot, separate from whatever role race plays in the decision to initiate contact with the person in the first place by, for instance, deciding to make a traffic stop.

Some have already criticized the paper because of the way Fryer has formulated his query, saying that this narrow version is not useful: If the police are more likely to shoot black people out of racism, then it doesn't matter where the racism enters into their decision-making. But I disagree with these critiques. If it were true that racism only played a role in the cop's decision whether to interact with a civilian, that would be valuable information, not only for directing policy to reduce police killings of black people. It could mean, for instance, that getting rid of racism in police killings just means reducing the amount of discretion we give officers in their decisions to stop people. This would be much easier than rooting out a subconscious, deeply embedded fear of black people, which could be part of the task if the decision to shoot is itself racially motivated. The point is, Fryer is trying to answer a useful and important question.

But here's the tricky part: How do we observe a police officer deciding not to shoot someone? The data on officer-involved shootings, obviously, only includes situations in which the officer shoots. The way Fryer tackles this is to create a category of crimes for which lethal force is likely to be justifiable — attempted capital murder of a public safety officer, aggravated assault on a public safety officer, resisting arrest, evading arrest, and interfering in an arrest. If an officer arrests a civilian for one of these crimes without using lethal force, then Fryer counts it as an officer deciding not to shoot.

You can probably see one problem already. Whether someone fits into a category like "resisting arrest" or "assaulting a police officer" isn't a neutral question imposed by some objective third party. It's the police officer who makes the decision whether someone is resisting arrest at the time of the incident, and it's the police officer who decides whether to write down in their report afterward that someone was resisting arrest. If the police are racially biased against blacks when making either of these decisions, then the results will skew downward — they will show less discrimination than there actually is. Think of it this way: The probability of being shot at by the police is a fraction. In the top part of the fraction, the numerator, we've got the number of police shootings. In the bottom half, the denominator, we have the number of times the police could have shot. We can be reasonably sure that the numerator is right — it's relatively difficult to cover up the fact that a police officer fired their weapon. But if we overestimate the size of the denominator, then we'll underestimate the size of the whole fraction.

And obviously, there is some reason to believe that police are racially biased when making the decision in the moment as to whether an incident fits into a category where force is likely to be justifiable, and also when they decide how to write their reports. While systematic data about whether police lie about whether people are resisting arrest or threatening them with weapons is hard to come by, there's plenty of video evidence that this is the case. For instance, the police officer who killed Walter Scott said that Scott had grabbed at his gun, but the video evidence that eventually came out shows this to be false. Similarly, the officer who shot Samuel DuBose claimed that he'd been dragged several feet by DuBose's car before making the decision to shoot DuBose, but the video evidence, again, shows this to be false.

Fryer, to his credit, recognizes this problem and lists it as one of his caveats. But the problem is even deeper than this. Even if the police were totally objective about what constitutes, say, "resisting arrest," and totally honest about whether someone was resisting arrest when they make their report, we have to keep in mind that police behavior also affects whether a situation escalates to the point that there is a physical altercation. Police behavior can determine whether a situation escalates or de-escalates, and if the police are more likely to escalate with black civilians, Fryer's results will be skewed downward, showing less discrimination than there actually is. Again, we have evidence that this is already happening — in fact, it's in Fryer's own results. If the police are more likely to escalate to nonlethal force with black people, this could create more situations in which they eventually have to decide whether to use lethal force. And again, this would tend to pull his estimates of the amount of racial bias against black people downward.

So while the results of the study are definitely interesting, the evidence is too thin to support the conclusion that many will take away — that there's no racism in police shootings, and thus that the Black Lives Matter movement is misguided.

This is preliminary research. It's a "working paper" — an unfinished draft that hasn't gone through peer review. But it's a good contribution to the research on this topic, and the more well-done the research, the better. We need more studies of police shootings, and we also need better data.

What we don't need to do is baptize the status quo in the name of one paper.