Ten Years of Study and the Protesters are Right

What traffic stops tell us about racial bias in policing

Samuel Workman


Baumgartner, Epp, and Shoub 2018, Figure 3.3.

Written by Frank R. Baumgartner, Leah Christiani, Derek A. Epp, Kelsey Shoub, and Kevin Roach

In the wake of George Floyd’s murder, the nation and world have turned their attention to large-scale demonstrations emphasizing policing inequities and calling for reforms ranging from banning unnecessary chokeholds to the abolition of the police. The Black Lives Matter movement, like others before it, argues the police target black and brown people in their enforcement of the law.

We have studied policing in the context of traffic stops for years, and our studies have consistently corroborated what activists charge: racial disparities in policing are widespread, even when controlling for possible benefits to public safety. Our research has focused on the issue of identity-based differences in the likelihood of various outcomes following a routine traffic stop, mainly the odds of a search.

An officer’s decision to search is a strong signal that the officer views the person with suspicion. Since traffic stop encounters are both widespread and afford very little information about the driver beyond a brief conversation, they provide a lot of insight into what prompts an officer’s suspicion, and how accurate officers are in using their training to identify those involved in criminal activities. Without trying to suggest that we have all the answers, or even that we have studied all the questions, here are a few things we have learned about traffic stops.

With millions of publicly available records of police traffic stops, we consistently find that identity-based differences in experiences with traffic stops are vast. Police target black and Latino men for searches at two to three times the rate of similarly situated white men. These disparities are not constrained to any geographic region but are ubiquitous across the country. Some places have it worse. Municipalities relying on fines, fees, and forfeitures to fund their government, along with areas not instituting mandatory written consent forms for consent searches, tend to have higher levels of racial disparities in policing. The disparities cannot be explained away by accounting for things like poverty, crime rate, police department policy, or a few “bad apple” police officers. They persist in the face of any alternative explanation.

Consider these graphical illustrations. Figure 1 compares the percentage of blacks and whites among those stopped by police compared to their share of the population in the local municipality for the entire state of North Carolina. Black drivers, on the left, are consistently a higher share of the drivers stopped than their population numbers would predict. Whites, on the right, are generally under-represented.

Figure 1. Black and White Shares of Drivers Stopped and Local Population Shares.

Of course, we have to be cautious in comparing population numbers with traffic-stop statistics; we are on firmer ground when we analyze what happens after a stop. There, we have information on every person stopped and can ask what percentage see an adverse outcome, such as a search. Figure 2 shows that, on average, across over 600 jurisdictions where we were able to find the data, a comparison of the rate of search of black drivers to that of white drivers.

Figure 2. Black-White Search Rate Ratios

Source: Baumgartner, Christiani, et al. 2017, Figure 5.

Figure 2 shows a vertical line at 1.0: the point at which the black and white search rates are equal. Seven observations fall below that line, and 635 fall above it; the average value is 2.51. The numbers cannot be clearer. They cannot be “explained away” by factors like the time of day that black and white drivers are on the roads, why people get stopped, or others. We tested for that by conducting logistic regressions using every bit of data available in four different states’ data collection systems. Figure 3 compares the simple search-rate-ratio to the more sophisticated logistic regression odds-ratio based on our data. The two measures of disparate outcomes, one simple and one controlling for potential confounding factors, correlate at more than 0.95.

Figure 3. Simple Rate Ratios compared to Logistic Odds-Ratios

Source: Baumgartner, Epp et al. 2017.

The disparities we examine are consistent with the idea that police officers operate with highly stereotypical perceptions of suspicion — whether these are rooted in individual beliefs or institutional training. The “criminal suspect” looks a certain way; he drives a certain kind of car; he is more active at night; he lives in a particular part of town, etc. The police read a book by its cover, and the outward signs of suspicion are highly predictable: Young, Black, and Latino men are at the heart of this group.

Every time we find a dataset with another indicator for suspicion, such as poverty (e.g., age of car, neighborhood), or otherness (e.g., out-of-state plates), we find that those characteristics increase the stereotypical targeting over and above this baseline. We can also identify the visible cues that make officers extremely unlikely to view an individual with suspicion: Asian-American or white females over a certain age, driving relatively newer cars, in the daytime or (best) during the morning rush hour. The predicted probability that an officer will search a driver in Figure 4 displays these differences.

Figure 4. Predicted probabilities of search, by racial group, over vehicle age

Source: Christiani 2020, Figure 2

Leah Christiani’s analysis of millions of traffic stops showed the interaction of multiple identities. If we think of the age of one’s car as a possible surrogate for income levels (with wealthier people driving newer cars on average), Figure 4 shows that, for men, race-based differences in search rates are significant, even for those with new vehicles. As the cars get older, these racial differences increase dramatically. Latino, Black, and Native American men with older cars see search rates over 10 percent, and Latina and Black women do as well. In sum, identity matters, a lot.

We also found that people in neighborhoods that are stereotypically associated with “higher crime” were more likely to be searched, particularly fruitlessly (i.e., without the search being followed by an arrest). We identified eight possible “adverse characteristics” that a driver or a police interaction might have, based on race, age, gender, neighborhood, the purpose of the traffic stop, and the officer conducting the stop. The odds of a fruitless search ranged in order, from zero to almost 30 percent, as Figure 5 shows.

Figure 5. Odds of Fruitless Search

Source: Baumgartner et al. 2020, Figure 5c.

This suspicion is rooted in a charge to seek out and prevent crime, known as “proactive policing,” and disseminated along with the war on crime. But, we find that there is little public safety benefit to using traffic stops as part of this war on crime; the numbers just don’t add up. Stopping cars on the highway to find a few drivers who might be engaged in criminal activity such as transporting drugs or weapons would require much more information than appears to be typical. If an individual has an out of state plate, has a warrant for an unpaid fee from missing a previous court date, or has an expired tag, they are much more likely to elicit police scrutiny. But they most likely are not delivering a payload of drugs.

Here’s an example of just how wasteful these needle-in-the-haystack strategies can be. Across all traffic stops in North Carolina from 2002 through 2016, about 3 percent of the traffic stops led to a search. Police found contraband in about a third of these searches, or just less than 1 percent of the stops. The typical outcome from these contraband “hits,” however, was not to arrest the driver. Just under half were arrested among those found with contraband. The reason? The “contraband” is either not itself cause for arrest (e.g., it was “drug contraband” — an empty sandwich bag) or the amount was so small that it was residue from some previous incident, not a sign of current drug-related activity. Figure 6 shows how tiny these contraband “hits” typically are. It shows the number of ounces of contraband for all those contraband hits that amount to ounces only.

Figure 6. The Small Size of Contraband Seizures

Source: Baumgartner, Epp, and Shoub, 2018, Figure 5.1c.

Racial disparities in traffic stop outcomes are ubiquitous. Often, these police-initiated searches that target young, black, and Latino men at the highest levels do not even result in the uncovering of contraband. Yet, they can lead to alienation, as political trust understandably declines. Ten years of research shows that racial disparities are pervasive in policing, often without cause and without benefit to public safety.

*Figure source captions contain links to all published research.



Samuel Workman

Professor, Data & Statistical Consultant, West Virginian, Author of The Dynamics of Bureaucracy in the U.S. Government https://amzn.to/3ilKSuh