Major Nature study reveals crisis in social science reproducibility

A large-scale replication study published in Nature has cast fresh light on the reliability of research in the social and behavioural sciences. The project, led by Tyner, Abatayo, and Daley, re-examined 274 claims from 164 quantitative papers released between 2009 and 2018. Its findings reveal significant challenges in reproducing earlier results, with fewer than 60% of claims holding up under scrutiny. The team employed thirteen different statistical approaches to assess replication success. These methods produced widely varying estimates, from as low as 28.6% to as high as 74.8%. On average, however, only 55.1% of the original claims replicated successfully, showing statistically significant results in line with the first studies.

Effect sizes also shrank dramatically. The median Pearson's *r* in the original research stood at 0.25, but in replications, it dropped to 0.10—an 82.4% reduction in shared variance. The study further found that nearly half of the tested papers failed to replicate even a single core claim robustly. Replication rates differed across disciplines. Success rates ranged from 42.5% in some fields to 63.1% in others. The researchers pointed to systemic issues, including selective reporting and the pressure to publish novel or positive findings, as key factors behind the low replicability. Conducted with open and collaborative practices, the project featured preregistered protocols and fully transparent datasets. The authors have called for stronger incentives for replication, better data-sharing policies, and the adoption of more rigorous statistical frameworks to improve research reliability.

The study, released in April 2026, underscores persistent concerns about the reproducibility of social and behavioural science research. With less than 56% of claims successfully replicated and effect sizes markedly smaller, the findings suggest a need for structural changes in how research is conducted, reviewed, and published.