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

Jessy Xinyi Han, Kristjan Greenewald, Devavrat Shah

Keywords: causal inference, algorithmic bias, recidivism, racial disparities

TL;DR: We present a multi-stage causal framework and an empirical test to study sources of racial bias in recidivism from a time-to-event perspective–revealing that fairness requires addressing deeper socioeconomic inequalities, not just fixing algorithms.

Abstract: Racial disparities in recidivism remain a persistent issue within the criminal justice system, increasingly exacerbated by the adoption of algorithmic risk assessment tools for decision making. Past works have primarily focused on understanding the bias induced by algorithmic tools, viewing recidivism as a binary outcome—i.e., reoffending or not. Limited attention has been given to the role of non-algorithmic factors (including socioeconomic ones) in driving the racial disparities in recidivism from a systemic perspective. Towards that end, this work presents a multi-stage causal framework to investigate the advent and extent of racial disparities by considering the time-to-recidivism rather than a simple binary outcome. The framework captures the interactions between races, the risk assessment algorithm, and contextual factors in general. This work introduces the notion of counterfactual racial disparity and offers a formal test using survival analysis that can be conducted with observational data to understand whether potential differences in recidivism rates among racial groups arise from algorithmic bias, contextual factors, or their interplay. In particular, it is formally established that if sufficient statistical evidence for differences in recidivism across racial groups is observed, it would support rejecting the null hypothesis that non-algorithmic factors (including socioeconomic ones) do not affect recidivism. An empirical study applying this framework to the COMPAS dataset reveals that short-term recidivism patterns do not exhibit racial disparities when controlling for risk scores. However, statistically significant disparities emerge with a longer follow-up period, particularly for low-risk groups. This suggests that factors beyond the algorithmic scores–possibly including structural disparities in housing, employment, and social support–may accumulate and exacerbate recidivism risks over time. Indeed, the use of survival analysis enables such nuanced analysis. This empirical analysis underscores the need for holistic policy interventions extending beyond algorithmic improvements to address the broader influences on recidivism trajectories.

Submission Number: 53