Why Most Discovered True Associations Are Inflated

John P. A. Ioannidis

Argues that even a genuinely true effect, once it clears a significance threshold in an underpowered or heavily tested literature, will on average have its estimated size inflated by the same selection process that produces false positives.