Commentary

Racial Profiling — Not Always Black and White

By Greg Ridgeway

Greg Ridgeway is a statistician at RAND.

How prevalent is racial profiling? Anecdotal evidence abounds that racial profiling occurs, but to make sound policy, we need more than just anecdotes — we need hard data. One fortunate consequence of racial profiling being such a high-profile issue of late is that we now have lots of hard data; hundreds of law enforcement departments are either voluntarily collecting such data or being compelled by the U.S. Justice Department to do so.

With the data collection has come a spate of studies — using those data — that purport to show that racial profiling is either a problem (statewide studies in Texas and Massachusetts) or not a problem (a local study in Sacramento). So do such studies actually tell us whether racial profiling is occurring? Unfortunately, the answer is not black and white.

Determining whether racial profiling occurs in the decision to stop a motorist seems simple enough: Compare the collected data on the racial distribution of traffic stops against some “benchmark” of the racial distribution that we would expect if police are not racially profiling; any difference between the two distributions could be evidence of racial profiling. In Oakland, California, for example, 56 percent of those stopped are black, while, according to the census (one of the more commonly used benchmarks), only 35 percent of the population is black.

However, all benchmarks in urban settings have fatal flaws. Using the census has been widely discredited, because driving patterns and behavior may differ by race. For example, many blacks live in high-crime areas to which police devote more manpower, and, thus, blacks may be stopped more than whites simply because blacks are more exposed to the police.

Our work analyzing data from Oakland shows that it’s possible to get around the problem of benchmarks by bypassing them altogether. We looked directly at whether an officer’s ability to identify a driver’s race in advance influences whom the officer stops. Because an officer’s ability to identify a driver’s race in advance degrades as the day moves from daylight to darkness, we compared the distribution of the driver’s race in stops made in daylight to those made after dark. In particular, we compared stops occurring near the boundary of daylight and darkness — or dusk — a brief but crucial interval during the transition from day to night and during which the driving population cannot quickly change. This innovative approach produced credible results that showed little evidence of racial profiling in the decision to stop motorists.

Of course, what happens after the stop can also reveal racial profiling. In Oakland, we did see some signs of racial bias, for example, in the greater likelihood of blacks being pat-searched for weapons than whites. But had we not built comparison groups to control for factors other than racial bias — factors such as region, time of day, and age of driver — our results would have overstated the problem of racial profiling significantly and, more likely than not, led to inaccurate headlines in the local newspapers.

Racial profiling will continue to be a flash-point issue, driven on all sides by contentious debate from multiple stakeholders. While we are on the right track by collecting data — a trend that will increase if the U.S. Congress passes the End of Racial Profiling Act mandating data collection for all law enforcement agencies receiving federal funds — we must do more to ensure that credible approaches are used to analyze that data. If we fail in that charge, we run the risk of inflaming stakeholders and public opinion and, even more important, of leading policymakers to make ill-informed decisions to implement the wrong remedies.

PHOTO CREDIT: DIANE BALDWIN