Contact
University of Konstanz
Chair of Public Economics
Box 133
D-78457 Konstanz
Tel.: +49-7531-88-2565
Fax: +49-7531-88-4135
Time
Thursday, 14. November 2024
12:00 - 13:15
Location
G308
Organizer
Junior Professorship in Labor Economics
Speaker:
Patrick Arni (ZHAW & University of Bristol)
(Patrick Arni (ZHAW & University of Bristol) and John Körtner (University of Lausanne))
Abstract: The use of algorithms for practical policy making increasingly sparks interest und expectations. But little is known about how algorithmic predictions are factored into decisions by the deciding bureaucrats. We provide evidence for the case of algorithmic profiling in the context of unemployment insurance (UI). In a field experiment, UI caseworkers received access to algorithmic predictions of UI claimants’ unemployment duration in a randomly selected number of cases. We find that caseworkers reacted to (some) prediction signals by updating their beliefs and adjusting their behavior -- but not in intended ways. They increased their efforts for claimants with predicted good re-employment prospects, which increased unemployment exit rates for this group of easy-to-place job seekers. We interpret the caseworkers’ behavior to be driven by performance incentives, rather than by the initial intention of the intervention (to activate hard-to-place job seekers additionally). Based on a framework of the interplay between algorithms and caseworkers, we check several alternative explanations and potential behavioral mechanisms.
Time
Thursday, 14. November 2024
12:00 - 13:15
Location
G308
Organizer
Junior Professorship in Labor Economics
Speaker:
Patrick Arni (ZHAW & University of Bristol)
(Patrick Arni (ZHAW & University of Bristol) and John Körtner (University of Lausanne))
Abstract: The use of algorithms for practical policy making increasingly sparks interest und expectations. But little is known about how algorithmic predictions are factored into decisions by the deciding bureaucrats. We provide evidence for the case of algorithmic profiling in the context of unemployment insurance (UI). In a field experiment, UI caseworkers received access to algorithmic predictions of UI claimants’ unemployment duration in a randomly selected number of cases. We find that caseworkers reacted to (some) prediction signals by updating their beliefs and adjusting their behavior -- but not in intended ways. They increased their efforts for claimants with predicted good re-employment prospects, which increased unemployment exit rates for this group of easy-to-place job seekers. We interpret the caseworkers’ behavior to be driven by performance incentives, rather than by the initial intention of the intervention (to activate hard-to-place job seekers additionally). Based on a framework of the interplay between algorithms and caseworkers, we check several alternative explanations and potential behavioral mechanisms.