The Digester
science

Half of social-science studies fail replication in seven-year SCORE project

A seven-year, multi-institution project found that only about half of tested social-science studies could be replicated, with poor data sharing and methodological gaps a leading cause.

Apr 1st 2026 · United States

Insights

  • The SCORE project reviewed 3,900 social-science papers from 62 journals over seven years.
  • When repeating full experiments, 49% of 164 focused studies were replicated with statistical significance.
  • Of 600 papers tested for computational reproducibility, only 145 provided enough information and 53% of those reproduced exactly.
  • In robustness tests, about 75% of 100 papers held up under alternative reasonable analyses and 2% reversed conclusions.
  • Major obstacles included missing data and incomplete methodological detail, alongside legitimate variation from new analyses.
  • The project was funded by DARPA and coordinated partly by the Center for Open Science to support automated confidence scoring of findings.
  • Separate analyses of 2022–23 papers show higher reproducibility, suggesting disclosure norms may be improving.

Sources