Major Nature study exposes replication crisis in social sciences

A large-scale study has examined the reliability of research in social and behavioural sciences. Published in Nature in April 2026, the project re-tested 274 claims from 164 papers released between 2009 and 2018. The findings reveal significant challenges in replicating past results, with fewer than 60% of claims holding up under scrutiny. The team, led by researchers Tyner, Abatayo, and Daley, used rigorous methods to replicate earlier studies. Each attempt closely matched the original conditions and employed high statistical power, with a median value of 99.6%. Despite this, only 55.1% of the claims produced statistically significant results consistent with the first findings.

Success rates varied widely across disciplines, ranging from 42.5% to 63.1%. When assessed using 13 different statistical approaches, replication estimates stretched from 28.6% to 74.8%. The median effect size also dropped sharply: the original studies reported a Pearson's r of 0.25, while replications averaged just 0.10—an 82.4% decline in shared variance. At the paper level, nearly half of the studies failed to replicate even one core claim robustly. The research highlights systemic issues, including selective reporting and pressure to publish novel or positive results. These factors, the team argues, undermine the reliability of scientific findings. The initiative itself was designed as an open and collaborative effort. Protocols were preregistered, and datasets were made fully transparent. However, no specific funding bodies, universities, or research networks were publicly linked to the project.

The study calls for urgent changes to improve research reliability. Recommendations include better data sharing, stronger incentives for replication, and the use of advanced statistical frameworks. Without these steps, the authors warn, confidence in social and behavioural science findings could continue to weaken.