Replication crisis
A large share of published findings fail to reproduce when independently repeated.
What it means
The replication crisis is the recognition that a substantial proportion of published results — prominently in psychology and the broader behavioral and biomedical sciences — do not hold up when independent researchers attempt to reproduce them with new data. Its causes are structural rather than mainly fraudulent: underpowered studies that chase noise, flexible analysis and undisclosed researcher 'degrees of freedom' (p-hacking), hypothesizing after results are known (HARKing), and a publication system that rewards novel, statistically significant findings while burying null results (publication bias). The combined effect is a literature inflated with false positives and overstated effect sizes that cannot be trusted at face value. The crisis prompted a wave of methodological reform, including pre-registration, registered reports, larger samples, open data and materials, and large-scale collaborative replication efforts. Several once-celebrated effects, such as ego depletion and some social-priming results, are now in serious doubt. It matters because it forces a more skeptical, cumulative stance toward individual studies and headline findings, and because the reforms it spurred are reshaping how credible behavioral science is produced and evaluated.
Examples
A large 2015 collaborative effort successfully reproduced fewer than half of 100 published psychology findings.
A lab runs a study five times, gets one significant result and publishes that one; the file drawer holding the four null runs never reaches the literature at all.
Corporate workshops built on willpower-as-a-fuel-tank kept selling long after the studies behind them were seriously challenged, because the headline travelled far further than the correction ever did.
First described in Open Science Collaboration (2015); Ioannidis (2005).