Replication

The crisis and what it taught.

What happened

Large replication projects in psychology, medicine and economics found that a substantial share of published findings did not hold up. The causes were mostly structural rather than fraudulent.

Small samples, flexible analysis and publication pressure explain most of it.
Pre-registration and larger samples measurably improve the situation.

How to read a study now

Ask about sample size, whether the analysis was decided in advance, and whether anyone has repeated it. A single striking result is a starting point, not a conclusion.