Early Intervention Systems: Predicting Adverse Interactions Between Police and the Public
Jennifer Helsby, Samuel Carton, Kenneth Joseph, Ayesha Mahmud, Youngsoo Park, Joe Walsh, Lauren Haynes, Crystal Cody, & Estella Patterson
We have developed a prototype data-driven early intervention system that uses a diverse set of data sources to more accurately predict the officers who will have an adverse incident. Data sources include anonymized internal police department records on arrests, field interviews, citations, incidents, dispatches, trainings, traffic stops, and internal affairs data, in combination with publicly available data. We used machine learning to determine which aspects of the data best predict whether an officer has an adverse incident in the next year. We’ve built this prototype using data from the Charlotte-Mecklenburg Police Department. Validation is performed using historical data from the police department going back 10 years. The predictive approach is able to significantly improve accuracy compared with the existing early intervention system: preliminary results indicate a 55% reduction in false positives and a 15% increase in true positives. In addition to accuracy, the machine learning approach generates a continuous risk score that can be used to allocate based on resource levels.
Improving Use of Force Data Collection: One More Step Toward Constructive Commuity Dialogue with Criminal Justice Organizations
Jon M. Shane
In its current state, American policing cannot mount a productive counter argument for the vociferous minority view that U.S. policing is experiencing a “crisis” regarding use of force. Policing lacks the fundamental ingredient for such a conversation, which allows the minority view to hijack the conversation and which limits law enforcement’s ability to engage the community in a constructive dialog about the nature, extent and realities of police use of force. That ingredient is empirical data. There are no national standards that address use of force data collection; no national repository of incident-level use of force data to inform training, supervision and tactics; and no source for law enforcement to consult and make informed comparisons between jurisdictions on the nature and scope as legislators and appointed leaders debate new restrictions on the police. The current summary system maintained by the FBI entitled Justifiable Homicide is limited; some of the limitations have been discussed in policing research dating to at least 1979 and the empirical interest in measuring police use of force dates to at least 1963. These limitations keep the police, researchers and community members from addressing how a given incident is consistent with or divergent from the broader contextual patterns and trends across incidents and how agencies compare to each other.