↓ 跳过正文

Algorithmic Decision-Making

Routinized Deference: Predictive Screening and Bureaucratic Capacity in Child Protective Services

JPART — 2026-10-16 Original Routinized Deference: Predictive Screening and Bureaucratic Capacity in Child Protective Services Jiaen Wu Abstract Abstract Predictive algorithms are increasingly used to support frontline decision-making in public administration, but evidence remains limited on how they affect bureaucratic capacity. This study examines county-level adoption of predictive screening tools in U.S. Child Protective Services using NCANDS data (2010–2023) and a stacked difference-in-differences design. Leveraging staggered adoption, I find no change in the total number of investigations opened and a sizable decline (about 15%) in the share of investigations that are substantiated. These results are inconsistent with operational capacity gains and more consistent with misclassification that erodes analytical capacity. These findings are consistent with a mechanism of routinized deference: under workload and accountability pressures, algorithmic recommendations can become procedurally defensible defaults. Overall, this study argues that predictive screening is not capacity-enhancing by default, as routinized deference can undermine the professional judgment central to bureaucratic work.

When Do Citizens Resist the Use of AI Algorithms in Public Policy? Theory and Evidence

JOP 88/4 pp. 1301-1317 2026-10-01 Original When Do Citizens Resist the Use of AI Algorithms in Public Policy? Theory and Evidence Shir Raviv Abstract In recent years, there has been a significant rise in the use of algorithmic decision-making systems (ADS) to assist or replace human decision-making in a wide range of policy areas as policing, criminal sentencing, and social welfare assistance. How do citizens view the incorporation of this technology in guiding high-stakes decisions? I introduce a new theory to explain the conditions under which citizens view ADS as legitimate, fair, and accurate, and test it using data from original experiments embedded in a national U.S. survey. I show that across a wide range of policy domains, citizens exhibit aversion to the use of ADS in decisions that are seen as designed to sanction rather than to assist, and when they are required to make inferences about individuals rather than collectives. Evidence from a second experiment suggests that the employment of ADS in such contexts can significantly undermine the legitimacy of the policy actions they inform. Together, the study offers a framework to identify where AI-based tools will be deemed appropriate and where they might trigger backlash, highlighting the importance of accounting for citizens’ values in AI development and implementation to maintain legitimacy and democratic accountability.