Algorithmic Risk and Discretion: Evaluating Competing Management Tools in State Surveillance Efforts

• In Event: 41039 - Crime, Law and Deviance Section Roundtables
  In Refereed Roundtable (60 minute): 2: Technology and the Criminal Justice System

Abstract

Abstract: Recent work finds that bureaucrats utilize decision-making tools, including risk assessment scores and other contextual information, to different ends. I utilize 15 years of administrative data from a U.S. state parole agency to determine how these tools are related to harsher or more lenient sanctions after an individual violates a technical condition. I will also evaluate if different endowments across risk score levels explain much of the differences in likelihood of sanctioning departures. Using a preliminary sample of n = 25,611 sets of violations and subsequent sanctions, I use multinomial models to regress departures from a baseline sanction unto risk scores and other private information. Results show that risk scores are not significantly related to more lenient sentencing than that warranted by violation severity, but that a high risk score compared to a low risk score is significantly associated with increased odds of a harsher sentence. These results suggest that risk assessments might be a tool used to increase the relative punitivity of a sanction rather than limit it.

Author

• M. Evangeline Lopoo, Northwestern University