EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3911. Accuracy, Awareness, and Acceptance: How Attention Drives People’s Use of Point versus Interval Forecasting Algorithms

Invited abstract in session WA-11: Scenarios and foresight practices: Behavioural issues I, stream Behavioural OR.

Wednesday, 8:30-10:00
Room: 12 (building: 116)

Authors (first author is the speaker)

1. Bas van Oudenhoven
School of Industrial Engineering, Eindhoven University of Technology
2. Rob Basten
School of Industrial Engineering, Eindhoven University of Technology

Abstract

Many operations rely on forecasts of future events, often determining the allocation of significant sums of money. Fortunately, algorithmic forecasts achieve high levels of accuracy, but many decision-makers reject algorithms upon observing they are not perfectly accurate. What determines whether decision-makers perceive an algorithm as inaccurate, and how are these perceptions related to their willingness to use algorithms when forecasting future events? We propose that the tendency to dismiss algorithmic advice is partially due to the common practice of presenting forecasts as single points rather than intervals. In Study 1, we find that people perceive equally imperfect forecasts as more accurate when these forecasts are presented as intervals rather than points. In Study 2, we examine how these differential accuracy perceptions affect people’s willingness to use algorithms. Here, we find that people do not necessarily favor interval over point forecasts when deciding whether to rely on the algorithm’s forecasts. Instead, such preference only occurs when people are explicitly prompted to evaluate the algorithm’s accuracy before deciding whether to commit to its forecasts. In sum, these findings underscore the complex processes underlying people’s perceptions of algorithmic errors and their subsequent reliance on algorithms. Moreover, they show how managers can apply interval forecasts to effectively increase human forecasters’ reliance on forecasting algorithms.

Keywords

Status: accepted

Back to the list of papers