Learning to Suppress: Decision-Makers Learn to Control Irrelevant Information Based on Experience

Abstract

In selection processes, decision-makers sometimes rely on selection tests that are biased by irrelevant attributes such as gender, race, or appearance. If left uncorrected, this can lead to suboptimal decisions and discrimination. Decision-makers can mitigate this bias by treating the irrelevant attributes negatively as suppressor variables, a task they often struggle with when relying solely on descriptive information. In four pre-registered experiments (N = 470), we examined whether they could learn to do so through experience and feedback. Decision-makers predicted candidates’ future job performance based on a selection test and an irrelevant attribute (e.g., candidate’s height) that biased the selection test (e.g., interviewers favored tall over short candidates). Individual multiple regression equations showed that decision-makers learned to account for the irrelevant attribute’s effect by weighing it negatively, improving prediction accuracy. They performed well even without prior information about the bias and in a more complex environment. Results regarding the effect of attribute type (personal vs. situational) and whether successful learning improves description-based decisions are inconclusive. Our findings highlight people’s ability to learn to adjust for irrelevant information bias in selection processes, though further research is needed on the role of context. Better-calibrated predictions can reduce discrimination by improving candidate selection. The study further informs the relatively understudied learning of suppression by addressing limitations in past research and clarifying mixed findings.