Paper Admission Decisions under Imperfect Classification: An Application in Criminal Justice

Incarceration diversion programs aim to rehabilitate justice-impacted individuals and reduce recidivism, but limited capacity necessitates careful admission decisions that often rely on predictions of the risk of re-offending during the program. This paper examines how prediction errors impact decision quality in high stakes settings where online exploration is infeasible. We develop a framework that combines queueing models with uncertainty quantification to evaluate decision correctness and potential interventions. Theoretically, we show that a priority score policy solves both the ground truth and estimated admission control problems, and it remains optimal under a likelihood ratio ordering condition despite the presence of prediction errors. By decomposing decision uncertainty into priority score and decision boundary components, we further quantify when decisions are and are not reliable. Practically, using data from Adult Redeploy Illinois, our simulation based case study evaluates two interventions: collecting more data, and human-in-the-loop decision-making. While both interventions improve average decision quality, challenges such as non-monotonic cost reductions underscore the need for carefully designed interventions that balance algorithmic automation with targeted human oversight in high-stakes decision-making.

Get the paper