Research Summary: Fairness Is More Than Algorithms: Racial Disparities in Time-to-Recidivism

This research highlights persistent racial disparities in recidivism, moving beyond traditional binary outcome analysis to examine 'time-to-recidivism.' It scrutinizes the role of both risk assessment algorithms and non-algorithmic contextual factors in these disparities. The study introduces a multi-stage causal framework and the concept of 'interventional racial parity' to understand the complex interactions driving these outcomes.

Why it matters

Addressing racial disparities in outcomes such as recidivism is critical for maintaining social equity and public trust in justice systems. This research contributes to a deeper understanding of the complex interplay between algorithmic predictions, human decision-making, and societal factors, which can inform the design of more equitable and effective risk assessment and intervention strategies.

Key insights

• Racial disparities in recidivism present an ongoing challenge.
• Previous research primarily focused on algorithmic prediction disparities and viewed recidivism as a binary outcome.
• Sociological and criminological studies have documented the influence of non-algorithmic factors on recidivism.
• It is unclear if current risk assessments fully account for racial disparities when acted upon by decision-makers.
• The research proposes a multi-stage causal framework for time-to-recidivism, capturing interactions between race, algorithms, and contextual factors.
• The study introduces the concept of interventional racial parity and a formal survival model within its framework.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2504.18629