Algorithmic Monocultures in Hiring

5 hours ago

Over 90% of U.S. employers use hiring algorithms for screening job applicants, with many relying on the same vendors.
The study analyzed 3.4 million job applicants and 4 million applications across 156 employers, finding evidence of racial disparities and homogeneous outcomes due to algorithmic monoculture.
Black applicants are most likely to face adverse impact, with 30% applying to at least one position showing such impact against them.
Asian applicants experience the largest shortfall, with 29,000 additional applications recommended if they were selected at rates equal to the most selected group per position.
Systemic rejection occurs when applicants are rejected everywhere; 10% of applicants submitting 4 applications face this, exceeding expected baseline rates.
Excess homogeneity in outcomes is distinctive of centralized algorithmic assessment, as confirmed by comparisons with independent decision baselines.
Applicants need to submit 25 applications to ensure at least one recommendation with high probability under realistic behavior, compared to 10 for independent decisions.
Current regulations, such as Title VII and local laws like NYC's Local Law 144, often fail to address position-level adverse impacts and systemic issues.
Recommendations include measuring adverse impact per position, strengthening market surveillance, monitoring algorithmic monoculture, and expanding researcher access to hiring platform data.

Hasty Briefsbeta