A much-touted study of Oakland police shows researchers’ determination to find racism, not cops’ bias.
Reprinted from City Journal.
The attempt to find systemic police bias has come to this: the difference between an officer saying “uh” and saying “that, that’s.” According to Stanford University researchers, police officers in Oakland, California, use one of those verbal tics more often with white drivers and the other more often with black drivers. If you can guess which tic conveys “respect” and which “disrespect,” you may have a career ahead of you in the exploding field of bias psychology.
In June, a team of nine Stanford psychologists, linguists, and computer scientists released a paper purporting to show that Oakland police treat black drivers less respectfully than white ones. The study, published in the Proceedings of the National Academy of Sciences, elicited a huzzah from the press. The Washington Post, the New York Times, and Science, among many other outlets, gave it prominent play. “Police officers are significantly less respectful and consistently ruder toward black motorists during routine traffic stops than they are toward white drivers,” gloated the New York Times.
Reading the coverage, one expected reports of cops cursing at black drivers, say, or peremptorily ordering them around, or using the N-word. Instead, the most “disrespectful” officer utterance that the researchers presented was: “Steve, can I see that driver’s license again? It, it’s showing suspended. Is that—that’s you?” The second most “disrespectful” was: “All right, my man. Do me a favor. Just keep your hands on the steering wheel real quick.”
The researchers themselves undoubtedly expected more dramatic results. Undaunted by the lackluster findings, they packaged them in the conventional bias narrative anyway, opening their study by invoking the “onslaught of incidents” involving officers’ use of force with black suspects that have “rocked” the nation. A cofounder of the Black Lives Matter movement helpfully commented in the San Francisco Chronicle that the study goes beyond individual racism to highlight a “systemic set of practices that has impacts on people’s lives.”
The study is worth examining in some detail as an example of the enormous scientific machinery being brought to bear on a problem of ever-diminishing scope, whether in police departments or in American society generally. The most cutting-edge research designs, computer algorithms, and statistical tools, such as Fisher’s exact tests, Cronbach’s alpha, and Kernel density estimates, are now deployed in the increasingly desperate hunt for crippling white racism, while a more pressing problem—inner-city dysfunction—gets minimal academic attention.
Lead researcher Jennifer Eberhardt, a Stanford psychology professor, specializes in implicit bias, the idea that nearly everyone approaches allegedly disfavored groups with unconscious prejudice. The Oakland Police Department has given Eberhardt virtually unlimited access to its policing data as part of a federal consent decree governing the department’s operations. Her first study of the department—on racial profiling in police stops—managed to run nearly 400 pages without ever disclosing black and white crime rates in Oakland. (Hint: they are vastly disparate.)
This latest study analyzed officer body-camera footage from 981 car stops that Oakland officers made during April 2014. Blacks were 682 of the drivers in those stops, whites 299. The resulting officer-driver conversations yielded 36,738 discrete officer utterances. In the first phase of the study, college students rated 414 of those officer utterances (1.1 percent of the total) for levels of respect. The students were shown what, if anything, the driver said immediately preceding each officer statement but were not shown any more of the earlier interaction between officer and driver. They were not told the race of the driver or officer or anything else about the stop. The students rated police utterances to white drivers as somewhat more respectful than those to black drivers, though the officers were equally “formal,” as the researchers defined it, with drivers of both races.
In the second phase of the study, the linguisticians tried to tease out which features of the 414 officer utterances had generated the student ratings. They came up with 22 categories of speech that seemed most determinative. On the positive scale were, inter alia, officer apologies, the use of surnames, the use of “um” and “uh” (known in linguistics as “filled pauses”), use of the word “just,” and what is referred to as “giving agency” (saying “you can,” “you may,” or “you could”). The eight negative categories included asking a question, “asking for agency” (phrases such as “do me a favor,” “allow me,” “may I,” “should I”), “disfluency” (a repeated word such as “that, that”), informal titles (“bro,” “my man”), first names, and, most disrespectful, the phrase “hands on the wheel.” If some of those distinctions seem arbitrary—“could I” is disrespectful, “you could” is respectful; “um” is respectful,” a word repetition is not—they are. More important, they are minute and innocuous. The 22 categories each received a score allegedly capturing their degree of respect or disrespect, with apologizing at the top of the respect scale and “hands on the wheel” at the bottom. There were no categories for swear words or even for unsoftened commands, presumably because officers never engaged in those forms of speech.
Finally, in phase three, the researchers turned their computers loose on all 36,738 officer utterances, using the 22-category rating system. They found that officers’ utterances toward white drivers scored somewhat higher in respect than utterances toward black drivers, even after controlling for whether the stop resulted in a search, citation, arrest, or warning. (The sample size for white arrests and searches was quite small, however: one arrest and two searches; black drivers were 15 times more likely to be arrested than whites.) Black officers scored the same as white officers in respect toward black and white drivers. White drivers were 57 percent more likely than black drivers to hear something from the top 10 percent of the respect categories, and black drivers were 61 percent more likely to hear something from the bottom 10 percent of the disrespect categories.
There is plenty to criticize in the study’s methodology and assumptions. Doing so, however, risks implying that the substantive claims are significant. They are not. Nevertheless, if it were the case that we should worry about whether an officer says “you can” (good) or “can I” (bad) to black drivers, the study leaves out critical components of officer-civilian interactions. The most disrespectful phrase in the disrespect scale is “hands on the wheel.” Black drivers are 29 percent more likely to hear those words than white drivers. Why might an officer ask a driver to put his hands on the wheel? Perhaps because the driver was not complying with an officer’s initial requests or was otherwise belligerent. Yet nothing about driver behavior is included in phase three’s regression analyses—not drivers’ words, demeanor, or actions.
Moreover, given crime rates in Oakland, a black driver is far more likely than a white driver to be on parole or probation, a fact that will show up when an officer runs his plates or his license. In 2013, blacks committed 83 percent of homicides, attempted homicides, robberies, assaults with firearms, and assaults with weapons other than firearms in Oakland, according to Oakland PD data shared with San Francisco Chronicle columnist Chris Johnson, even though blacks are only 28 percent of Oakland’s population. Whites were 1 percent of robbery suspects, 1 percent of firearm assault suspects, and an even lower percent of homicide suspects, even though they are about 34 percent of the city’s population. (The roadways draw on a population beyond Oakland, but Oakland’s crime disparities are repeated in neighboring towns.) Being on parole or probation could contribute to an officer’s hands-on-the-wheel request, but drivers’ criminal history is not included in the study’s models.
The authors claim to have controlled for the severity of any underlying offense that may have triggered the stop, but they do not show whether offense severity differed between blacks and whites. The proportion of male drivers in the black sample was higher than in the white sample, which will also skew the results toward a more crime-prone population. Males were 67 percent of all black drivers but only 59 percent of white drivers.
The study’s much-cited statistic that black drivers are about 60 percent more likely to hear a phrase from the bottom 10 percent of the disrespect scale is entirely accounted for by the “hands on the wheel” phrase, since there are only eight items on the disrespect list. The next two items on the disrespect list are first names and informal titles. Whites were 4 percent more likely to have a first name used with them, and blacks were 65 percent more likely to have an informal title used with them, by far the greatest discrepancy on the eight-item disrespect scale. An officer who uses “my man” or “bro” with a black driver in Oakland is likely trying to establish rapport through the use of street vernacular, hardly an invidious impulse; black officers were as likely to use such informal titles as white officers. The white drivers stopped were, on average, three years older than the black drivers. Though age had a greater effect on respect and formality than race in the regression models, the study did not test the connection between age and race. Given the socioeconomic profile of the Bay Area’s white population, class differences, too, could explain why officers are less likely to use “man” and “bro” with white drivers.
Whether a young black male in Oakland would feel affirmatively disrespected by “my man” is nowhere demonstrated. Eberhardt claimed in an e-mail exchange that black and white DMV patrons in a replication effort also rated utterances from the study’s phase one as “more respectful” toward white drivers, from which she concluded that “the use of urban vernacular by officers is not seen as more respectful by black citizens.” The question is, however: Are such street terms affirmatively experienced as disrespectful?
None of these methodological objections really matters, though, because the substantive results are so innocuous. Consider again the most disrespectful utterance provided by the researchers: “Steve, can I see that driver’s license again? It, it’s showing suspended. Is that—that’s you?” In no possible universe with any minimal connection to common sense should that utterance be deemed disrespectful. Why does it get that rating? A first name is used, which is the second most disrespectful item on the researchers’ disrespect scale. “Can I see” is “asking for agency,” the fifth most disrespectful thing an officer can say. Worse, “can I see” is part of a question, and questions are the eighth most disrespectful term on the list. If “can I see that driver’s license?” is now deemed racially disrespectful, it’s hard to see how police officers can do their jobs.
More demerits follow from “It, it’s showing.” The repeated “it” counts as a “disfluency,” fourth on the disrespect scale. The chance that a driver is even aware of such verbal tics is almost zero. The chance that he would distinguish a disfluency from a so-called filled pause (“um” or “uh”) and experience the one as disrespectful and the other as respectful is less than zero. The word “suspended” generates another strike because it is “negative.” Again, it is hard to see how officers can conduct traffic stops if such “negative words” are off-limits. The final sentence also racks up two demerits: “Is that—that’s you?” is a disfluency and a question. The question may have been asked to soften the fact that the driver is operating with a suspended license.
This is madness. In their franker moments, the researchers all but admit that their study makes a mountain of a molehill. “To be clear,” Dan Jurafsky, a linguistics and computer science professor told Science, “these were well-behaved officers.” The “differences are subtle,” Eberhardt said to Science. The language used with blacks was not “really disrespectful,” she added. No kidding. But the authors cannot resist pumping up their results to fit the conventional policing narrative. “We have found that police officers’ interactions with blacks tend to be more fraught,” they write at the end. They have found no such thing. Even if the professors had actually measured drivers’ reactions to the 36,738 officer utterances, rather than simply running those utterances through a computer algorithm, a de minimis difference on the respect scale is not tantamount to a finding of “fraughtness.” Nevertheless, Eberhardt repeated the “fraughtness” claim in numerous interviews. The study goes on to conclude that “we now have a method of quantifying these troubled interactions.” But the authors also did not measure whether the interactions were “troubled” from the driver’s perspective. Their method recalls campus-rape surveys that never ask alleged victims if they think they have been raped.
The authors titled their study “Language from police body camera footage shows racial disparities in officer respect.” A more accurate title would have been: “Language from police body camera footage shows that officers treat all drivers courteously but are more colloquial with young black drivers.”
In 2015, the last year for which full data are available, Oakland’s violent-crime rate was nearly four times the national average: 1,442 violent crimes per 100,000 residents, compared with 372 violent crimes per 100,000 residents nationwide. Oakland’s violent crime rate was 14 times higher than Palo Alto’s and twice as high as San Francisco’s. If police training starts insisting that officers refer to everyone as Mr. and Ms. and scrupulously avoid street appellations, there would be no loss. But it is the disparity in criminal offending and victimization that should concern race researchers, not whether police officers are more likely to repeat words or use “my man” with black drivers.