This study examines the emergence of cohort bias from the use of predictive risk assessment instruments (RAIs).
Social science research and policy increasingly rely on predictive risk assessment instruments (RAIs), including those using machine-learning algorithms. This paper shows that the relationships between risk factors and future arrest are unstable over time when measured across sequential birth cohorts. As a result, prediction models reliant on risk factors are prone to systematic and substantial error. Such cohort bias, arising from the dynamics of social change, requires algorithmic updating and accounting for social factors affecting entire cohorts. Cohort bias can generate inequality in criminal justice contacts distinct from racial bias and has implications not only for the tailoring of RAIs but also for efforts aiming to provide preventative interventions to high-risk groups targeted based on individual-level risk factors alone. RAIs are widely used to aid high-stakes decision-making in criminal justice settings and other areas such as health care and child welfare. Because societies are themselves changing and not just individuals, this assumption may be violated in many behavioral settings, generating what the authors call cohort bias. Analyzing criminal histories in a cohort-sequential longitudinal study of children, the authors demonstrate that regardless of model type or predictor sets, a tool trained to predict the likelihood of arrest between the ages of 17 and 24 y on older birth cohorts systematically overpredicts the likelihood of arrest for younger birth cohorts over the period 1995 to 2020. Cohort bias is found for both relative and absolute risks, and it persists for all racial groups and within groups at highest risk for arrest. The results imply that cohort bias is an underappreciated mechanism generating inequality in contacts with the criminal legal system that is distinct from racial bias. Cohort bias is a challenge not only for predictive instruments with respect to crime and justice, but also for RAIs more broadly. (Published Abstract Provided)
Downloads
Similar Publications
- A Multi-Stream Fusion Approach with One-Class Learning for Audio-Visual Deepfake Detection
- The Experience of Social Distancing During the COVID-19 Pandemic: A Qualitative Lens on Variability in Compliance
- Examining the Black Box: A Formative and Evaluability Assessment of Cross-Sectoral Approaches for Intimate Partner and Sexual Violence