Recent public opinion surveys have revealed that the vast majority of Americans believe that use of racial profiling by the police is widespread.1 This is deeply disturbing for two reasons. First, it is disturbing because it undermines police legitimacy among the vast majority of our citizens. Second, it is disturbing because the vast majority of law enforcement officers I have known do not engage in bias-based policing. While racial profiling likely occurs among a small number of individual officers acting outside the bounds of their oath to uphold the Constitution, it is unlikely that racial profiling is systemic to law enforcement in the United States.
This begs the question, then, why do so many people perceive that racial profiling is widespread? We could blame individual members of the news media that seek to raise their ratings by stoking the flames of controversy, or certain protest organizations that seek to capitalize on distrust of the police. To be sure, these sources have contributed to the problem. Another factor that has also contributed to the problem, however, is the fundamentally flawed information that many law enforcement agencies have given the public through their biased-based policing data that was gathered and reported incorrectly.
Many law enforcement agencies gather data on the race and gender of the individuals their officers stop, search, and arrest. They report these data to the public in a biased-based policing report. Agencies produce these reports for a variety of reasons, such as statutory requirements, as part of their compliance with CALEA Standard 1.2.9.d, or simply out of a sincere desire to embrace transparency. While most law enforcement agencies, and individual officers, claim they do not racially profile, the vast majority of these reports show members of minority groups, especially African-American men, are disproportionately stopped, searched, and arrested. Why? One factor at work is the use of incorrect research methodologies and measures that are biased (often unintentionally) against officers from the start. One of the most damaging of these incorrect methodologies is the use of U.S. Census data as a benchmark comparison.
In order for any racial profiling data collection activity to be meaningful, the racial composition of police stops, searches, and arrests need to be compared to something. A benchmark is generally defined as a point of reference from which measurements may be made; something that serves as a standard by which others may be measured or judged; or a standardized problem or test that serves as a basis for evaluation or comparison. In the context of biased-based policing evaluations, a benchmark is the percentage of a racial or gender group that one would expect to be encountered if officers were not biased.
For example, imagine that 20% of the people speeding down a particular stretch of roadway were male and Hispanic. This makes 20% our benchmark for speeding stops of male Hispanics. We would expect that unbiased stops by police for speeding in this area would show that only about 20% of those stopped for speeding were male Hispanic drivers. However, where do we get these benchmarks? Unfortunately, most of the benchmarks used are fatally flawed. These flawed benchmarks consistently suggest officer bias, regardless of what officers are actually doing. The most common flawed benchmark is U.S. Census data.
The U.S. Census Bureau collects data on the social and demographic characteristics of the individuals who live within the U.S. This data is freely and easily accessible from the U.S. Census Bureau website and can be analyzed within different geographic regions, down to the zip code and census block levels. Many have used Census data as their benchmark for police activity because of its ease of access. The problem, however, is that the demographic characteristics of the people living at any one location have nothing to do with the driving population there, nor who is breaking the law in any specific area. We use our vehicles to travel to places away from our homes, as people generally do not work, shop, or recreate in their homes. Two studies illustrate this well.
The first study, conducted by sociologists Albert Meehan and Michael Ponder at Oakland University, examined the racial composition of drivers across one suburb in the Detroit area. According to the U.S. Census, the suburb they studied had a population that was 3% AfricanAmerican, but the city also contained a popular shopping district and a major auto factory. The researchers placed pairs of observers at major intersections across the three police beats in the city, and the observers recorded the races of 3,840 drivers who stopped at these intersections. Despite the city Census population of 3% African-American, in the police beat that bordered the city of Detroit, 49% of the drivers were African-American. The other two beats revealed 11% and 3% of the drivers were African-American.2
Think about that. What if the officers working these different beats stopped African-American drivers as the exact rates that African-Americans drove in these beats? Any study of this particular suburb using 3% African-American as its benchmark would falsely claim that officers working in two of the beats were racially profiling. When the stops from all three beats are combined as department-wide data, the whole department would incorrectly appear to be racially profiling because more than 3% of their stops were of African-American drivers, despite the fact AfricanAmericans actually made up far more than 3% of the drivers on the road.
Another example was a study, of which I was a part, that was conducted by a research team headed by criminologist Robin Engel at the University of Cincinnati. This study examined 315,705 traffic stops conducted by troopers of the Pennsylvania State Police. These stops occurred on interstate highways, U.S. highways, state routes, county roads, and village and city streets. An examination of these stops revealed that 96% of drivers stopped by the police were stopped outside of their home zip codes. Furthermore, 66% were stopped outside of their home county, and 27% were stopped outside of their home state. This study went on to conduct observations of the races of 66,741 drivers along various roadways in 27 counties of Pennsylvania. When compared to the Census statistics for each township where these observations were made, the Census statistics on race never matched the racial composition of the drivers that were observed.3 Census data is no reflection of who is driving in a given area.
Not only are Census statistics inaccurate measures of who is driving in any given area, Census data also fail to identify the racial and ethnic composition of who should actually be stopped by the police. Just because 49% of the drivers in a police beat are African-American, that does not mean 49% of the people stopped by officers should be African-American. If African-Americans were stopped just because they are driving in an area, without having done anything wrong, this would amount to stops for “driving while black.” Instead we need a measure of the racial composition of the drivers who are driving poorly by breaking traffic laws and driving unsafe vehicles (equipment violations). It is traffic law violators who should be at risk of traffic stops if no bias is present. The Census data in no way measures driving behavior.
So what should be used as a proper benchmark for these types of reports and studies? Hiring a group of researchers to go out and record the races and traffic violations of drivers across your jurisdiction is usually too time-consuming and expensive for most law enforcement agencies. A simple solution, however, is to collect race and ethnicity data on all traffic crashes in your jurisdiction and use this data as your driver benchmark. While no state currently collects race data on its state vehicle crash form, if your agency starts collecting race data in-house, your agency will eventually have a benchmark of bad drivers across the various beats of your jurisdiction.
Using traffic crash data as a traffic stop benchmark has a number of advantages. First, it identifies the drivers most likely to be stopped because crashes result from moving or equipment violations of the law. While there are some people who are blameless for their crash (such as the person waiting at a red light who is hit from behind), all crashes had at least one driver or equipment error at fault, and many had multiple drivers at fault. Second, officers investigating traffic accidents can verify the race and ethnicity of the driver when they complete their report, as opposed to a researcher trying to determine a driver’s race in a passing car. Third, as traffic crashes occur almost everywhere (even off of public roadways in parking lots and driveways) they are good samples of the bad driver or poorly maintained vehicle population throughout a district or beat. Research observers tend to focus just on certain thoroughfares. Finally, crash data come from the citizenry who report crashes to the police, so no suggestion can be made that there was bias by the police in gathering this data.4
If you agency is currently using Census data as your benchmark, it is imperative that you stop immediately and find a valid benchmark like the alternative discussed here. Using Census data is rigged against your officers as it almost always suggests disproportionate stops of minority group members, even when no officer bias occurred. If some outside individual or organization proposes to analyze your officers’ stops using Census data as their benchmark, oppose it vehemently, using the studies cited here to support your argument. If your state collects statewide data, as does Illinois, Missouri, and Texas, lobby your state lawmakers to stop using Census data as the benchmark comparison and begin to collect valid benchmark comparison data by modifying the state vehicle crash form to include race and ethnicity information.
The overwhelming majority of racial profiling studies done by academics, and biased-based policing self-examinations by police departments, have produced results that people of color, especially African-Americans, are disproportionately stopped by the police.5 It is likely, however, that the majority of these findings are in error as most relied on methodological errors that were guaranteed to show bias even when there was none. Using Census statistics as a benchmark, that in no way resemble the driving population or the traffic violator population, is just one of these many methodological errors.
The Dolan Consulting Group LLC now offers a training workshop that addresses these many errors, and offers suggestions on how to correct them. Biased-Based Policing Reports: Best Practices is a one-day course that teaches personnel from law enforcement agencies how these studies should be conducted and their reports written. It explains how to collect, analyze, and present your information in a manner that creates the least chance of misinterpretation or manipulation by the media, and presents the work of your agency in a fair manner. The information offered in this workshop is crucial to the creation of a data collection effort and report that is truly unbiased against the hard-working and principled officers who are policing their communities in a fair and impartial manner.
1 Weitzer, R., & Tuch, S. A. (2005). Racially-biased policing: determinants of citizen perceptions. Social Forces, 83(3), 1009-1030.
2 Weitzer, R., & Tuch, S. A. (2005). Racially-biased policing: determinants of citizen perceptions. Social Forces, 83(3), 1009-1030.
3 Engel, R. S., Calnon, J. M., Tillyer, R., Johnson, R. R., Liu, L., Wang, X. (2005). Project on Police-Citizen Contacts: Year 2 Final Report. Cincinnati, OH: University of Cincinnati.
4 Withrow, B. L., & Williams, H. (2015). Proposing a benchmark based on vehicle collision data in racial profiling research. Criminal Justice Review, 40(4), 449-469.
5 Withrow, B. L. (2006). Racial Profiling: From Rhetoric to Reason. Upper Saddle River, NJ: Pearson / Prentice Hall.