Driving While Indian
Category: Best data-driven reporting (small and large newsrooms)
Country/area: United States
Organisation: InvestigateWest, Indian Country Today, Crosscut.com, Spokesman.com; also Tribal Tribune in Nespelem, Washington
Organisation size: Small
Publication date: 19/12/2019
Credit: Joy Borkholder
An exhaustive investigation of millions of traffic stops and searches of drivers by state patrol officers in the state of Washington (USA) exposes the phenomenon of “Driving While Indian”: Native Americans are searched 5 times more often than white motorists by the Washington State Patrol, often just outside Native reservations. Yet white drivers are far likelier to be found with drugs and other contraband. Black, Latino and Pacific Islanders also are searched at higher rates than white motorists. Legislators expressed outrage at the findings. The State Patrol now is taking steps to correct bias.
As a result of our two-year investigation, state lawmakers and tribal leaders expressed outrage at the findings; lawmakers said they will be looking at the issue during this year’s legislative session, while the Washington State Patrol said it’s moving forward to address the problem. The State Patrol says it has put some Seattle-area troopers through an anti-bias training course that is more comprehensive than what cadets receive in the training academy, and it is in talks with university researchers to look into the issue.
One of our main findings was that the State Patrol had discontinued a series of studies by Washington State University researchers examining stops and searches of people of color by Washington troopers. Those studies stopped in 2007 even though researchers said indicators of “implicit bias” required more research and further attention by the patrol. Within days of publication of our investigation, the patrol said it was “taking strides to jumpstart the research relationship with WSU.” A spokesman also said the agency was examining the anti-bias curriculum available through the Washington Criminal Justice Training Commission identified in InvestigateWest’s reporting; the agency had never before taken advantage of the training.
State Patrol chief John Batiste emailed all employees of the agency with a link to our investigation. “We must be clear we make mistakes from time to time … but when we do we must admit our mistakes, we must apologize, and we must commit to improvements,” Batiste wrote.
We used primarily RStudio, Excel and QGIS for processing, analyzing and visualizing the data – about 8 million police stops. Because of the size of the data set, we had to use statistical software, in this case RStudio, to customize the time range, create new variables for discretionary searches and police districts, and visualize the different cuts of data. We wanted to repeat and update the descriptive statistics that had last been done over 10 years ago by researchers at Washington State University. Ultimately, we used Excel for published visualizations, after experimenting with other programs like Flourish. We also used Excel as a database, starting with lists of stops exported from R. We then requested police reports from the state patrol, and processed Adobe narrative police reports (pdf files) into a file with driver demographics and stop information. We used this information to try to contact drivers. Finally, we used QGIS for analysis and visualization, ultimately publishing the heat map of Native American high-discretion searches, overlaid with counties and tribal lands.
What was the hardest part of this project?
As a small nonprofit, we did not have anyone ready to take on the data side of this reporting. Despite over ten years of applied research experience, Joy Borkholder had not learned some of the powerful open-source software used in this project, but was eager to do so. Over the course of working on this story, Borkholder attended hands-on workshops at NICAR 2018, including one with the team from the Stanford Open Policing Project. Stanford faculty and staff remained accessible for support and even script-checking throughout our reporting. Borkholder also took advantage of free online courses offered by the Knight Center for Journalism in the Americas, including R (using RStudio), QGIS, and Data Journalism and Visualization. This project is an example of the data journalism community supporting skill development, and thus, stories with impact.
Aside from the data aspects of the project: There are always difficulties covering under-represented communities that have histories of poor representation in the news media. Tribal leaders wouldn’t agree to meet with story co-author Jason Buch until we demonstrated our dedication to the story by assigning Buch to travel to Central Washington to visit their communities, pretty much speculating we could get interviews. Even then, tribal members remained guarded.
We had great difficulty finding drivers to talk to us: people who had been the subject of a high-discretion search, with contraband found, didn’t want to talk. We combed through hundreds of police reports and contacted drivers, and reported on the ground in Central Washington, and still struggled to find people willing to go on the record about their personal experience. Our resultant story has drawn reaction. “Omg! Someone is finally reporting on the WSP near the Coville Rez,” read one email we received after the story published. “THANK YOU!!!!”
What can others learn from this project?
You don’t have to be a data scientist or IT wizard to learn new programs that handle millions of records and visualize them. At first, Joy Borkholder learned and taught herself just enough RStudio to export slices of the data and figure out that there was a story there. She also did some basic work in QGIS, in both cases, starting with skills classes at the IRE/NICAR conference in 2018. The Knight Center for Journalism in the Americas offered in-depth R and QGIS classes as the reporting on this project continued. Some Knight Center classes remain publicly available as videos and tools for anyone to access, like this one: https://learn.r-journalism.com/en/. Additionally, the online community of R and QGIS users, like in Stack Overflow, is super helpful in figuring out glitches and scripts. Finally, programs like R and Python make transparency so simple; if you see a story with data crunching behind it, you might be able to click through to see the code, learn and copy.