Using Algorithms to be Lenient: The Effects of Advice Valence and Algorithm Adjustment Decision Rights on Algorithmic Advice Use in Performance Evaluation

• In Event: Artificial Intelligence, Algorithms, and Creativity

Abstract

The use of algorithms to inform business judgments, including performance evaluation judgments, continues to grow. Despite the demonstrated potential of algorithms for improving judgment quality, recent research suggests that decision makers may be averse to algorithmic use due to various organizational and psychological factors. We conduct an experiment to examine whether and how managers’ willingness to use an algorithm advised performance rating is influenced by the valence of the advised rating and managers’ decision rights to adjust an algorithm. We find that managers are more willing to use an algorithm to evaluate subordinate performance when the advised rating indicates above-average performance of the subordinate than when the algorithm advises a below-average performance rating. This effect is due to managers’ motivation to maintain their relationship with the subordinate. Furthermore, we find that when an algorithm advises a below-average rating, allowing managers to adjust how the algorithm computes the rating leads to their greater willingness to use the algorithm, as compared to allowing managers to adjust the advised rating only after it has been generated. This effect is due to the improved perceived understanding of the algorithmic process when managers have discretion to adjust the algorithm’s computation. Our findings contribute to understanding the use of algorithms in making performance evaluation judgments, and inform organizations that are planning to implement algorithms to facilitate their performance evaluation process by suggesting a type of decision rights that can be granted to mitigate algorithm aversion.

Authors

• Mandy Man-sum Cheng, UNSW Sydney

• Kerry Anne Humphreys, UNSW

• Fangbin Lin, UNSW Sydney