A recent arXiv preprint, Fairness Is More Than Algorithms: Racial Disparities in Time-to-Recidivism, shifts the focus of fairness research from the algorithm itself to the broader measurement of recidivism. Where many studies examine whether risk scores differ across racial groups in a binary yes/no outcome, this work looks at time-to-recidivism — how long until someone reoffends — and finds that disparities persist even when algorithmic outputs are scrutinized.
The paper argues that past work has primarily concentrated on disparities in the algorithm's predictions or decisions, which may miss important structural and contextual factors. By framing recidivism as a temporal process, the authors highlight that fairness is not simply a matter of calibrating a model, but also of understanding the data-generating environment and the real-world consequences of risk assessment.
Because this is a single preprint, there are no competing sources to compare against. Still, the central claim is clear: fairness in criminal justice risk assessment cannot be achieved by algorithmic fixes alone. The study suggests that researchers and practitioners need to look beyond model outputs to the underlying social and temporal dynamics that shape recidivism.