Algorithmic Bias and the Effects of Gender-Blind Wage Suggestions: A Field Experiment on a Job Platform

How to cite

Sun, Borui; Kelley, Erin; Lane, Gregory; Chiplunkar, Gaurav. 2025. "Algorithmic Bias and the Effects of Gender-Blind Wage Suggestions: A Field Experiment on a Job Platform." AEA RCT Registry. AEARCTR-0016347. https://doi.org/10.1257/rct.16347-1.0

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

Borui Sun* , Erin Kelley , Gregory Lane , Gaurav Chiplunkar

* Primary investigator

Registry ID
AEARCTR-0016347
DOI
https://doi.org/10.1257/rct.16347-1.0
First registered
Last updated
Status
Completed
Trial dates
2020-11-12 – 2022-09-06
Countries
India ()
Registry record
http://www.socialscienceregistry.org/trials/16347

Abstract

This study evaluates how employers and job seekers respond to wage suggestions generated by algorithms on a job posting platform. Motivated by concerns that algorithms trained on historical data may reflect social biases—such as gender-based wage disparities—we experimentally test the effects of removing demographic information from the algorithm’s input. Vacancy postings were randomly assigned to one of three groups: receiving wage suggestions from the platform’s status quo algorithm, receiving gender-blind wage suggestions, or receiving no wage suggestion at all. We examine how these design choices affect posted wages, employer behavior, and job-seeker engagement.

This trial was launched as a one-month pilot in partnership with a job platform but was discontinued before full rollout due to shifts in the partner company's priorities. As a result, the experimental data is limited to the pilot period, and no additional data can be collected.

Intervention

To study the effects of gender-blind algorithmic design, we partnered with a job platform that provides employers with suggested wage ranges when creating vacancy postings. These suggestions are generated by a model trained on historical job data with similar requirements, including the employer’s stated gender preference.

We restrict the sample to female-only postings. In the control arm, employers receive wage suggestions from the platform’s standard algorithm, which includes inputs such as education requirements, work experience, occupation, location, city, and gender. In the gender-blind arm, gender is removed as an input, and suggestions are based on male-only postings—effectively equalizing algorithmic wage recommendations across gender lines. In the no-suggestion arm, employers receive no wage recommendation.

Experimental design

The experiment was implemented on a job platform over a one-month pilot period. The sample was restricted to job postings that specified a female-only hiring preference. The unit of randomization was the individual job posting. Random assignment was implemented via a deterministic rule, and all interface elements were held constant across conditions to ensure a consistent user experience.

The partner company discontinued communication before the full rollout, and no additional data can be collected beyond the completed pilot.

Design details

Randomization method
Randomization was done in office by a computer
Randomization unit
Individual job posting
Clusters
Not applicable (treatment was assigned at the individual job-posting level)
Observations
14,199
Arms
14,199
Primary outcomes
Primary outcomes include: (1) the posted wage on the job listing; (2) the stated job requirements, including education, experience, and language proficiency; and (3) hiring outcomes, proxied by the number of applications received and the characteristics of applicants, such as education level, English proficiency, and prior work experience

Cited by

No citation data available for this registration yet.

Keywords