(Stanford Open Policing Project) 

Analyzing data from over 130,000,000 traffic stops in 31 states, a multidisciplinary team of Stanford researchers and reporters released a report on June 20 statistically confirming discrimination in police interactions with minority drivers. Their data found that officers search African American and Hispanic drivers on the basis of less evidence than when searching Caucasian drivers.

Along with their report, the team published the entirety of their dataset online in a national database of traffic stop data collected from 2011 to 2015 as part of the Stanford Open Policing Project.

Cheryl Phillips, visiting professor in the professional journalism department and founding member of the Stanford Computational Journalism Lab, said that the project took more comprehensive steps to address and assess the issue than previous undertakings.  

“This is the first time people have collected all of this data in one repository,” she said. “It’s also the first time there’s been any effort to actually analyze the data from all of these states. We’re able to take it a step beyond and say it’s not just disparities, that actually there is evidence of discrimination.”

In their analysis, the researchers found that minority drivers were more likely to be searched, ticketed and arrested than white drivers. However, they note that these disparities, or unequal forms of treatment, may not necessarily be indicative of racism.

Sam Corbett-Davies, a Ph.D. candidate in computer science and researcher with the Stanford Open Policing Project, notes that one of the most difficult parts of the data analysis was assessing when discrimination was actually occurring.

“Not every difference that you see in the data is necessarily discrimination,” he said. “It’s actually a really hard problem to determine whether discrimination is occurring.”

To measure whether or not biases were actually in play, three different statistical tests for discrimination were used: the benchmark test, which solely evaluates search rates, the outcome test, which examines success rates of searches, and the threshold test, a new test developed by the Stanford team, which examines the minimum level of suspicion required to conduct a search and whether that bar varies for different race groups.

Corbett-Davies said that by utilizing these three tests, they were able to determine that the disparities experienced by minority drivers were indeed the result of discrimination.

“It’s unambiguous that black drivers are searched more than white drivers, but that might just be because they carry contraband more often, and the evidence suggests that they do,” Corbett-Davies said. “But even taking that into account, they’re still searched with less evidence [and] searched even more than they should be considering the amount of contraband that they carry.”

The study also looks at the legalization of marijuana and its impact on stop and search disparities.

“One of the most common reasons why a search is conducted is because [of] drug possession, and probably the most common drug is marijuana, so when Washington state and Colorado legalized marijuana, it removed one big class of reasons why you could be searched,” Corbett-Davies said.

The study found that though legalizing marijuana reduced searches overall, disparities between racial groups remained, and police officers still required less evidence to search minority drivers than their white counterparts.

Camelia Simoiu, a Ph.D. candidate in management science and engineering on the research team, said the aim of the project is to increase accessibility of the data.

“We want to lower the bar for researchers, policymakers, police departments across the country to see what sort of patterns or similarities exist,” she said. “It took us two years to reach out to every single state … [and] get this data. It was a humongous effort to get the data, and we wanted to make it easily available to the public.”

Tutorials on how to access and interpret the information and open source code on how the researchers analyzed the data have been released for this purpose, according to Simoiu. She and her colleagues hope that their database will enable other researchers to examine more policies and innovate new tests for discrimination.

“One of our biggest goals is to improve transparency in police interactions with the public,” she said. “We hope this data will enable whoever is interested, be it policymakers, journalists or police departments themselves, to dig deeper.”

The Stanford Open Policing Project’s journey began in the winter of 2015 with a data negotiation project issued by Phillips to her journalism students.

“I wanted my students to learn how to negotiate for public information, and I also felt like it would be a good story at some point,” she said. “And then I ended up with millions of records.”

Phillips then reached out to assistant professor of management science and engineering Sharad Goel and his students, and the project began.

“Local data is really often the most important data when it comes to looking at accountability journalism, but it’s also the hardest to get,” Phillips said. “This for me was an attempt at seeing [if] is it possible to go at projects where you can collect local data and then aggregate it up for bigger picture work and make it available for journalists … to do local stories.”

Besides aiding researchers, reporters and policymakers, the team hopes that their findings will ultimately help reform and improve police practices themselves.

“Police practices have come into the front and center of national issues recently,” Simoiu said. “Policing is very complex, it’s a very nuanced role and officers are putting their lives on the line everyday to protect the community, and at the same time, that’s not a reason to hold back and not ask these difficult questions: what can be improved and how can we do that? And can we build better statistical tests to look at some of these hard questions?”

Contact Ashley Hitchings at ashitchings ‘at’ gmail.com