TLDR
When researchers test many ideas at once, some may seem true by chance. This method limits the share of false alarms among the findings reported as important, while usually detecting more real effects.
Summary
This content was automatically synthesized by Credo's AI models directly from the original source text.
AI summaries can make mistakes — double-check important details against the original source.
1 Study Aim
Benjamini and Hochberg aim to develop a better way to handle multiple significance tests. They compare controlling the false discovery rate (FDR), the expected share of rejected claims that are false, with controlling the familywise error rate (FWER), the chance of making any false rejection. The authors argue that FDR control can provide greater statistical power when some tested claims are genuinely true. The study seeks a practical testing rule that limits false findings without discarding too many real discoveries.
2 Study Design
The authors formulate the FDR criterion and compare it with FWER control. They introduce a sequential Bonferroni-type procedure, which evaluates ordered test results using progressively adjusted thresholds. The paper proves that this procedure controls the FDR when test statistics are independent. A simulation study assesses its statistical power, and examples demonstrate how researchers can apply the procedure and judge whether its criterion fits their goals. The researchers develop the method mathematically, test it in simulations, and illustrate its use with examples.
3 Findings
The study shows that FDR and FWER are equivalent when all tested hypotheses are true. When some hypotheses are false, FDR is smaller, creating potential for higher power. The authors prove that their sequential procedure controls FDR for independent test statistics. Simulations show a substantial power gain compared with stricter familywise-error control. Examples illustrate practical use and the importance of choosing an error criterion suited to the research question. Researchers can use this approach when finding more real effects matters and a limited share of false findings is acceptable.