Large Replications Find 55% Success

A coordinated Nature study attempted to replicate 274 published social‑ and behavioral‑science findings (from 164 papers, 2009–2018) with high statistical power and preregistered protocols; only about 55% of effects replicated, and those that did averaged roughly half the original effect size. The study covered multiple fields (education, psychology, economics) and used original materials where possible, making it one of the most systematic tests of the replication problem to date. — If more than a third of published social‑science claims don’t hold up and effect sizes routinely shrink, policymakers, journalists, and institutions must change how they cite, fund, and build on single studies.