Reducing bias
Skill, not signals: what the research says about hiring bias
A summary of our field report on hiring bias and skills-first hiring, backed by landmark studies. Read the headline evidence, then get the full sourced report.
June 17, 2026 · 7 min read
Most hiring still starts with a name, a school, and a network. Decades of evidence say those proxies are poor guides to who can actually do the job, and reliable carriers of bias. We pulled the strongest public research into one place: our field report, Skill, not signals. This is the short version.
The name on the page changes the odds
In 2004, economists Marianne Bertrand and Sendhil Mullainathan sent nearly 5,000 fictitious résumés to real job ads in Boston and Chicago, randomly assigning each a white-sounding or a Black-sounding name. The résumés were otherwise identical. Those with white-sounding names drew 50% more callbacks, and the gap held across every occupation, industry, and employer size.
Hide the signal, and the bias drops
When American orchestras began auditioning behind a screen, economists Claudia Goldin and Cecilia Rouse found a woman’s chance of advancing past the preliminary round rose by about 50%. The orchestras did not ask juries to try harder to be impartial. They changed the structure of the evaluation so the only thing on display was the skill.
The degree is a weak, leaky proxy
Employers are loosening the four-year-degree filter: by 2024, 52% of US job postings listed no formal education requirement, up from 48% in 2019. But dropping the words changes little on its own. A 2024 Harvard Business School and Burning Glass study found announced skills-based policies reached fewer than 1 in 700 hires.
What actually predicts performance
Pooling 85 years of selection research, Frank Schmidt and John Hunter ranked how well each method predicts later job performance. Tests of the work sit at the top; the signals a résumé leads with sit at the bottom.
- Work sample test: r = .54
- Structured interview: r = .51
- Unstructured interview: r = .38
- Years of experience: r = .18
- Years of education: r = .10
What works
The evidence points to three moves, none of which ask people to be less biased. They change the process so bias has less to grip: anonymize until merit is established, measure the work rather than the proxy, and give everyone the same structured interview. This is exactly how Spoon Hire is built: recruiters browse an anonymized pool, every candidate takes the same AI voice interview, and verified skills tests put a like-for-like score on the work.
Frequently asked
Does the name on a résumé really affect callbacks?
Yes. In a landmark 2004 field experiment, Bertrand and Mullainathan sent nearly 5,000 otherwise-identical résumés to real job ads and randomly assigned white-sounding or Black-sounding names. The white-sounding names drew about 50% more callbacks, and the gap held across industries and employer sizes.
What actually predicts job performance better than a résumé?
Pooling decades of selection research, Schmidt and Hunter found that work-sample tests and structured interviews predict performance far better (around r = .51 to .54) than the things a résumé leads with: years of experience (.18) and years of education (.10).
Does dropping degree requirements fix it?
Not on its own. A 2024 Harvard Business School and Burning Glass study found that announced skills-based policies reached fewer than 1 in 700 hires, and about 45% of employers changed the words but not their behaviour, what the researchers call skills-based hiring 'in name only'.
Where can I read the full report?
The complete field report, 'Skill, not signals', is free. It walks through every study with charts and links to each original source. You can read it at spoonhire.com/resources/skill-not-signals.
Put it into practice with Spoon Hire.
Run fair, skills-first AI interviews and review anonymized, merit-ranked shortlists.