One bad apple may not spoil the bunch, but it can disease the rest, slowly turning all of them into bad apples. A new study from Northwestern University supports that idea when it comes to police misconduct. According to the study, police misconduct is a group phenomenon—one that can be tracked through machine learning and network analysis tools.
In a first-of-its-kind study, researchers used insights from three known cases of police corruption—including the ongoing case of ex-Chicago Police Sergeant Ronald Watts—to create a statistical model to identify possible crews of officers engaging in misconduct and criminal behavior.
“This paper shows we can identify possible crews of bad cops using historical examples, like the Ronald Watts case, as a point of calibration,” says study author Andrew Papachristos, director of the Northwestern Neighborhood and Network Initiative. “The Watts case is shaping up to be one of the largest police corruption scandals in U.S. history, and our paper shows what we’re learning here can possibly help us find other groups of criminal-oriented cops.”
Watts and his team ran an extortion racket at a Chicago housing project for more than a decade. Prosecutors have said Watts and the officers under his command repeatedly planted evidence and fabricated charges in order to further their own gun and drug trade. Already, 200 convictions tied to Watts have been overturned, although some organizations project the final number to be closer to 500.
While certainly an unwanted situation, Papachristos and his team are taking advantage of the data coming out of the Watts trial.
Scouring decades of reports and news investigations into the harms perpetrated by known corrupt Chicago police officers, Papachristos and team designed parameters for a network cluster analysis of the data. The machine learning model was then used to analyze publicly available complaint and arrest data on Chicago police officers from 1971 to 2018.
“We know that not all discernible clusters of police officers who have been repeatedly co-accused of misconduct can be automatically identified as bad actors or criminal networks within the police department,” said study co-author Rajiv Sinclai. “But by learning about the specific patterns and particular network characteristics that are exemplified by the known crews, we were able to narrow down the discernible clusters to just those crews that are most similar to these three well-coordinated groups of known bad actors.”
The study, published in PLoS ONE, detected approximately 160 potential “crews” of officers algorithmically linked by formal/informal work assignments and co-allegations. That number comprises less than 4 percent of all Chicago police officers, but accounts for approximately 25% of all use of force complaints, city payouts for civil and criminal litigations, and police-involved shootings.
The study results also show that the 160 detected crews contributed to racial disparities in arrests and civilian complaints, generating nearly 18% of all complaints filed by Black Chicagoans and 14% of complaints filed by Hispanic Chicagoans.
The research team then reviewed each crew by analyzing investigative documents, gathering available reporting and interviewing people with direct experience with crew members to more fully assess the validity and significance of each identified crew. Following the investigative feedback and validation process, the “hot lead” crew clusters will be added to the model as additional exemplars to refine the selection of network characteristics to help identify future crew clusters.
“The study has yielded a tool of immediate utility to police departments and oversight agencies,” said James Kalve, founding executive director of the Invisible Institute, which houses the data on Chicago police used in the study. “It is a critical component of an early warning system that enables supervisors to identify groups of officers that have characteristics resembling those of crews of officers known to be criminal. It is important to be clear: Such patterns do not in themselves constitute proof of criminality. They are, rather, prompts for supervisors to investigate.”
Click here to learn more about the project.