Dive Brief:
- A study of 4 million job applications screened by a hiring algorithm found evidence of “clear racial disparities” in applicant outcomes, with 26% of applications submitted by Black applicants and 15% of those submitted by Asian applicants being directed to positions that adversely impacted them, researchers at Stanford University said Thursday.
- Applications were screened through an algorithm created by vendor Pymetrics across 156 employers and 11 sectors. Researchers used one algorithm for all positions in order to study the effects of “algorithmic monoculture,” a term they said is reflective of the current state of the hiring sector, in which many employers screen applicants using the same few vendor-provided algorithms.
- The Pymetrics algorithm also yielded a systemic rejection rate — i.e., the rate at which candidates are rejected from multiple jobs — of 10% of all candidates who submitted at least four applications. This result “significantly exceeds the baseline rate expected under independent decisions,” the researchers said.
Dive Insight:
The tendency of employers to use the same hiring algorithm or any of a small group of algorithms that are designed in a similar way using similar data produces the kind of algorithmic monoculture outlined in the study, the researchers said.
“We’ve speculated in past work that if many firms relied on the same AI vendor to screen job applicants, that could prevent some applicants from getting any interviews,” Kathleen Creel, assistant professor at Northeastern University and co-author of the study, said in an article accompanying the publication. “But this study was the first time we were able to show this effect in real hiring data.”
Notably, the study found evidence that several of the jobs to which applicants applied did not conform to the “four-fifths” rule described in U.S. federal equal employment opportunity enforcement guidelines. Under this rule, federal EEO agencies consider the selection rate for a given job to have disparate impact against a particular demographic group if the rate is less than four-fifths of that of the group with the highest selection rate.
Among the positions measured in the study, 30% of Black applicants applied to at least one position that demonstrated adverse impact against them, as defined under the four-fifths rule.
However, Asian candidates experienced the largest “shortfall” of applicant groups in the study, also known as the difference between the number of actual candidates selected and the number that would be expected had Asian candidates been selected at the same rate as the most selected racial group for each position.
A key takeaway concerned the study’s methodology: whereas past research of algorithmic hiring discrimination analyzed aggregated selection data among all positions screened by a vendor’s algorithm, this study disaggregated and analyzed each position separately. This is notable as the vendor, Pymetrics, had previously published aggregated audits to show that its tools did not have measurable bias, said Sarah Bana, co-author and assistant professor at Chapman University.
“In that way, I was surprised because I thought that their algorithms would be an example of best practice,” Bana said. “When you read that something you’re buying has been audited, you tend to take that finding at face value — and that’s likely part of what is going on.”
Bana said employers that use hiring algorithms should find out who their algorithms are screening out for each applicable position. “This means you have to let, ideally, a random subset of applicants through that first stage and see how they fare,” she added. “This is probably worth doing regularly because your algorithm is probably not changing at the rate that your work is.
The study comes at a time where the vast majority of employers — more than 90%, according to a 2025 World Economic Forum estimate — use some form of automation to filter or rank job applicants. Concerns about discrimination by automated hiring tools has been spotlighted in an ongoing legal battle between vendor Workday and a group of job applicants who alleged the company’s tools discriminated against them.
HR departments can take steps to prevent algorithmic bias within their hiring processes, one management-side attorney wrote in a 2024 opinion piece for HR Dive. Conversely, a 2020 working paper by MIT researchers found that hiring algorithms could be designed in such a way as to improve both the diversity and quality of job applicants.