Towards a National Database of Law Enforcement Accountability

• In Event: Intervention & Prevention Approaches to Minimize Police Violence

Abstract

There is increasing pressure for prosecutors to indict police in the wake of high-profile killings. However, there is as yet no centralized public database tracking how prosecutors respond to alleged fatal law enforcement misconduct at scale. I introduce an ongoing project to build a national database capturing how fatal use-of-force incidents and in-custody deaths are sanctioned, justified, or ignored by prosecutors. This database is a tool to answer pressing empirical questions about police accountability, such as: how many killings result in criminal charges against police? What racial, regional, economic, or legal factors impact this decision? To build this resource, I have scraped tens of thousands of court documents relating to fatal law enforcement interactions across eleven states. First, I sample jurisdictions as case studies to identify common missing data patterns and create qualitatively-coded documents. Next, I use these manually coded documents as a training set for supervised machine learning algorithms to generate classifiers to apply to novel documents from other jurisdictions. Finally, I deploy large-language AI models that parse messy documents without extensive preprocessing. To conclude, I discuss validation techniques and informed prompt engineering that may mitigate known racism in black-box AI training.

Author

• Kyla Bourne, University of California, Berkeley