article
Why Most Published Research Findings Are False
Ioannidis models the chance that a 'significant' finding is true as a function of prior odds, statistical power, bias and how many teams chase the same question, then shows that under realistic values most claimed findings in many fields should be false. Published in 2005, it anticipated the replication crisis and is still the clearest account of why it happened. The arithmetic is simple enough to do yourself.
