This has to be one of the weirdest things I have ever seen. The authors of an often-cited study about racial bias in police shootings have asked for it to be retracted, while standing behind the data and statistical analysis. They’re retracting the paper only because they don’t like the way it’s being discussed in the media.
As I explained last year, the study has been controversial because of the roundabout way it approached the issue: It checked to see if, when police kill civilians, there is a correlation between the officers’ races and the suspects’. The idea is that if white cops are killing blacks out of bias, white cops should disproportionately be involved in killings of blacks.
As I wrote:
In the simplest analysis, the opposite is actually the case: Black suspects are disproportionately killed by black officers. But this isn’t as odd as it seems, because some parts of the country have higher black populations than others. In those places, both suspects and officers are more likely to be black.
The study addresses this by statistically controlling for the demographics of homicide victims in each county where a shooting happened — and the correlation between officer and suspect race disappears. In other words, someone shot by the police is more likely to be black if the area has a lot of black homicide, but not if the police officer involved happened to be white. This is inconsistent with the idea that white cops are killing black suspects out of anti-black racism.
Now, there are real limits to this approach, as I also laid out:
Black cops might disproportionately patrol black neighborhoods even within counties. If this is the case, black suspects and black officers might encounter each other more — above and beyond what’s accounted for in the statistical controls — which could cancel out the effects of white cops’ racism. To put it differently, black cops could shoot black suspects simply because they encounter a lot of black suspects, while white cops could shoot black suspects out of racism. And if these two phenomena occurred to roughly the same degree, the study would indicate that black and white cops are about equally likely to shoot black suspects.
But it seems the real issue is that the results are hard to describe. The study doesn’t calculate the chance that someone was shot — in the data the authors used, every suspect was shot. Instead, it calculates the probability that someone who was shot was of a given race, and how that probability changes depending on the cops’ races.
The authors previously made a correction to their own wording, changing “White officers are not more likely to shoot minority civilians than non-White officers” to “as the proportion of White officers in a fatal officer-involved shooting increased, a person fatally shot was not more likely to be of a racial minority.” And now they take issue with Heather Mac Donald’s description of the study in the Wall Street Journal: She wrote that “the more frequently officers encounter violent suspects from any given racial group, the greater the chance that a member of that group will be fatally shot by a police officer”; I imagine they’d prefer something like “when someone has been shot by the police, that person is more likely to be black if homicide victims in the surrounding county are heavily black, but not if the officer involved is white.” Even my own post quoted above toes the line by referring to how “likely to shoot black suspects” officers of different races are; despite the surrounding context that clarifies what I’m talking about, it might have been better to say “neither black nor white cops are disproportionately involved in shootings of blacks” or some such.
I’m all for using precise language and correcting the record when you fail to, but retracting the entire study is a bad decision. The lack of a correlation between officer and suspect race is noteworthy and deserves to be a part of the discussion.
For a different study reaching the opposite conclusion, see another post of mine here.