New model could help police departments reduce excessive force incidents

After an incident where a police officer uses excessive force, investigators typically review whether there were any warning signs, such as a history of using serious force, says criminology professor Greg Ridgeway. Many police departments in the United States use commercial or in-house systems to flag warnings signs and identify officers whose use of force—ranging from restraining to potentially deadly force—exceeds a set threshold, such as two incidents in a given period.

However, Ridgeway notes that these methods don’t account for key differences in time and place. For example, two incidents might be reasonable for an officer working nights in the nightlife district but worrying for an officer working mornings in the business district, yet both would be flagged.

To address these limitations, Ridgeway developed a statistical model that estimates an officer’s likelihood of using a higher level of force than peer officers in the same circumstance. He did this by comparing only data from use-of-force incidents with multiple officers on scene and using a mathematical equation that cancels out time, location, and other situational factors. His findings are published in the Journal of the American Statistical Association.

“Importantly, flagging an officer as a statistical outlier does not prove that they are problematic since it could also signal that their peers are too passive or too risk averse or the officer is designated to be the officer to go hands-on first as part of a predetermined plan,” the paper states. Instead, the model “provides a principled approach for identifying elevated risk that should prompt a department to audit the officer’s force incidents, interview force subjects, and prioritize review of the officer’s body-worn camera videos.”

Ridgeway says once officers are identified, departments have a variety of remedies, such as training, basic admonition, reassignment to lower-risk positions, or punitive actions, such as suspension or termination.

He gives the example of an interaction involving three officers where one officer doesn’t take any action, another restrains a civilian, and a third uses pepper spray—respectively known as level 0, 1, and 2 force. (Level 3 force is deadly or potentially deadly). It’s possible everything in this encounter unfolded appropriately for the circumstance, says Ridgeway, “but if it’s always the same person going for the higher level or always the same person just being a witness, you start thinking, ‘Maybe there’s a real pattern here, and these are the officers that need closer scrutiny.’”

Collaborating with the Seattle Police Department, Ridgeway tested the model on their use-of-force data from April 2014 to March 2021. Ridgeway says he used this data because of the rigor and oversight SPD applies to their data collection—meaning that use of force gets documented appropriately. The individual identities of officers were anonymized when Ridgeway received the data.

By using a rule that flags officers likely to rank in the highest 5%, Ridgeway identified nine officers at the highest risk of escalating force more than their peers. For example, one officer used a more serious form of force than other officers in eight of the nine use-of-force incidents for which they were on scene, and in seven of those situations, they were the only officer to use force.

His analysis also identified 13 officers with a high probability of being in the bottom 5% of force escalation.

“There are some benign explanations, such as these officers having supervisory roles on site or being particularly skillful at de-escalating,” Ridgeway writes. “These officers may be well-positioned to model de-escalation strategies for others. However, police managers also need to review these officers to ensure they will act when the situation demands it, as the tragedies in Charlottesville, Parkland, and Uvalde underscore the consequences of inaction.”

In the future, Ridgeway hopes to work with other police departments to analyze their data, allowing him to compare findings across departments.

Greg Ridgeway is the Rebecca W. Bushnell Professor of Criminology, professor of statistics and data science, and faculty co-director of the Data Driven Discovery Initiative in the School of Arts & Sciences at the University of Pennsylvania.