A new paper introduces the "Likelihood Ratio Wall," a statistical barrier limiting the accuracy of pretrial risk assessment tools for rare violent offenses. The research demonstrates that even with highly discriminative tools, achieving a 50% positive predictive value for rare events like violent re-offense is statistically improbable. The paper also highlights a "Surveillance Ceiling" effect, where over-policing can structurally lower achievable precision for certain groups, suggesting that current data regimes may not support high-confidence detention decisions for rare violent outcomes. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Highlights statistical limitations in AI-driven risk assessment, potentially impacting fairness and accuracy in legal applications.
RANK_REASON Academic paper introducing a new statistical concept and its implications for risk assessment tools.