Risk Assessment as Policy in Immigration Detention Decisions
研究了美国移民和海关执法局2017年改变风险评估工具后，移除释放建议导致实际释放率从约10%降至约5%，官员仍高度遵循工具建议，且主管更可能推翻官员的释放决定。
A large literature examines the effects of algorithmic risk assessments on judges’ bail decisions in criminal cases. This article examines these effects in the immigration detention context. In 2017, US Immigration and Customs Enforcement changed its risk-assessment tool. Before the change, the tool could recommend detention, release, or referral to a supervisor; afterward, it did not recommend release—ever. Taking advantage of the suddenness of this change, I show that the removal of the release recommendation reduced actual release decisions by about half, from around 10 percent to around 5 percent of all decisions. Officers continued to follow the tool’s detention recommendations at only a slightly lower rate after the change, and when officers did deviate from the tool’s recommendation to order release, supervisors became more likely to overrule their decisions.