The Human Factor: Algorithms, Dissenters, and Detention in Immigration Enforcement

Abstract

ObjectiveThis article measures Immigration and Customs Enforcement (ICE) officers’ punitive biases in immigration enforcement, comparing rates of dissent with low‐ and high‐risk classification scores. It provides evidence that contextual political factors alter dissent with risk classification assessment (RCA) scores and describes how such dissent informs more punitive versions of ICE's RCA scoring rules.MethodsWe implement a statistical decomposition technique that compares the gap between low‐ and high‐risk dissent rates in risk assessments. Historical and qualitative evidence describes the effect of punitive biases in algorithmic editing.ResultsResults of this investigation show that larger gaps in dissent rates between low‐ and high‐risk classifications result in more punitive scoring updates in later versions of the RCA algorithm.ConclusionOur article provides evidence that risk assessment tools implemented to reduce biases in immigration enforcement were updated to accommodate the punitive bias of ICE officers and supervisors.