Can exploratory algorithms reduce discrimination in hiring?

How to cite

Baumann, Julia; Kübler, Dorothea. 2025. "Can exploratory algorithms reduce discrimination in hiring?." AEA RCT Registry. AEARCTR-0016496. https://doi.org/10.1257/rct.16496-2.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Julia Baumann* , Dorothea Kübler

* Primary investigator

Registry ID
AEARCTR-0016496
DOI
https://doi.org/10.1257/rct.16496-2.0
First registered
Last updated
Status
In development
Trial dates
2025-08-26 – 2026-08-26
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/16496

Abstract

This study examines managers' willingness to follow different hiring algorithms and their impact on reducing discrimination against minority groups. Focusing on the exploration-exploitation trade-off, we compare explorative algorithms, which encourage hiring from less-known applicant groups, to exploitative ones favoring familiar candidate profiles. We investigate which algorithms managers prefer, their effects on hiring gender minorities, and how these tendencies relate to internalized gender stereotypes. Our findings aim to inform the design of fairer algorithmic hiring practices.

Experimental design

We conduct a hiring experiment to study how managers respond to different algorithmic recommendations in recruitment decisions. Managers choose between male and female applicants performing a sports knowledge quiz. While men typically perform better on sports trivia quizzes than women (Bordalo et al, 2019, AER), our sports quiz is designed so that men do not perform better on average than women. If managers hire women despite the initial stereotype they may have against them in this task, they can learn about this true distribution and should thus not discriminate against women. Participants are assigned to one of three treatments: no algorithm (control), an explorative algorithm encouraging hiring from underexplored groups, or an exploitative algorithm favoring groups with higher observed productivity. Managers receive feedback on hired applicants and state beliefs about group performance over multiple rounds. We measure which algorithm managers follow more willingly and how this affects gender-biased hiring discrimination, also considering the influence of managers’ gender stereotypes.

Design details

Randomization method
Randomization done in office by a computer
Randomization unit
Individual
Clusters
For a pilot study, we collect 120 observations. Based on the effect size in this pilot, we will calculate the required sample size to reach at least 80% power in the main data collection.
Observations
120 in pilot study
Arms
40 for each of the three treatment arms in pilot study
Primary outcomes
- Number of hires from male versus female applicant pool - Willingness to follow algorithm’s recommendations (number of times the hiring choice was the same as what the algorithm recommended)
Secondary outcomes
- Beliefs about ability of men and women in quiz - Weight of early negative/positive experiences on subsequent hiring - Managers’ bonus earnings

Cited by

No citation data available for this registration yet.

Keywords