By Jeff Horwitz

Facebook Inc. disproportionately shows certain types of job ads to men and women, researchers have found, calling into question the company's progress in rooting out bias in its algorithms.

The study led by University of Southern California researchers found that Facebook systems were more likely to present job ads to users if their gender identity reflected the concentration of that gender in a particular position or industry. In tests run late last year, ads to recruit delivery drivers for Domino's Pizza Inc. were disproportionately shown to men, while women were more likely to receive notices in recruiting shoppers for grocery-delivery service Instacart Inc.

The imbalance applied to postings for high-skill jobs as well, the study found. Facebook's algorithms were more likely to show a woman an ad for a technical job at Netflix Inc. -- which has a relatively high level of female employment for the tech industry -- than an ad for a job at Nvidia Corp., a graphics-chip maker with a higher proportion of male employees, based on data from federal employment reports.

The results suggest "a platform whose algorithm learns and perpetuates the existing difference in employee demographics," the paper says, noting that Facebook's algorithms appeared to produce skewed outcomes even if an employer intended to reach a demographically balanced audience.

Federal law prohibits discrimination based on gender, race, age and other traits in advertising for housing, employment and credit products. While the law's application to behavioral advertising remains contentious, the federal government has argued that ads must be distributed in ways that don't disadvantage protected classes of people's ability to see them.

The paper reflects Facebook's difficulties in understanding and managing the societal effects of its content-recommendation systems. Several large tech companies have teams working to study and find ways to eliminate bias in their algorithms.

"Our system takes into account many signals to try and serve people ads they will be most interested in, but we understand the concerns raised in the report," said Beth Gautier, a spokeswoman for Facebook. "We've taken meaningful steps to address issues of discrimination in ads and have teams working on ads' fairness today."

The USC researchers also ran near-identical tests of LinkedIn's delivery of job ads but found no evidence of gender skewing on the Microsoft Corp.-owned platform. The researchers didn't study other websites that show job listings and said running tests on more categories would improve confidence in the results.

Aleksandra Korolova, a former Google research scientist and one of the study's authors, said studying Facebook's systems was difficult without having access to company data but that she was surprised the company hadn't addressed the skewed distribution of job ads. "They've known about this for years, and it's an important question for society," she said.

LinkedIn Vice President of Engineering Ashvin Kannan said the USC paper's conclusions reflected LinkedIn's own findings, but the company remains concerned about bias in its systems.

Questions of discriminatory and illegal ad targeting have loomed large for Facebook for years. A 2016 report by ProPublica found that the company allowed employers to exclude older workers from seeing job postings and landlords to exclude ethnic minorities from ads.

The U.S. Department of Housing and Urban Development sued Facebook in 2019 over what it called biased ad targeting, alleging that the company was allowing landlords to exclude users with interests in topics such as Hispanic culture, mobility scooters and hijabs from apartment listings. Facebook settled the lawsuit and said it has removed such categories as targeting options and pledged to work with HUD to address the agency's other concerns.

It has also settled a lawsuit brought by the American Civil Liberties Union, agreeing to take steps including studying possible discriminatory outcomes produced by its algorithm.

While the research spotlights issues at Facebook, a lasting fix for the problem has so far eluded the industry, said Piotr Sapiezynski, a computer-science researcher at Northeastern University who collaborated with the USC team on past research into racial disparities in the delivery of job ads.

"Until we figure out how to do this right, the short-term solution is to turn off relevance matching for housing, credit and employment ads," he said.

Write to Jeff Horwitz at Jeff.Horwitz@wsj.com

Corrections and Amplifications

This article was corrected at 12:56 p.m. ET because it incorrectly said Piotr Sapiezynski was a computer-science professor. Mr. Sapiezynski is a computer-science researcher at Northeastern University.

(END) Dow Jones Newswires

04-09-21 0814ET