Question

Why control FDR instead of family-wise error rate?

HChcole· 4 days ago

Coming from genomics where we run tens of thousands of tests: FWER control (Bonferroni) is so conservative it kills power. FDR feels like the right trade-off, but how should I think about choosing the threshold? Is q < 0.05 just a convention inherited from p < 0.05, or is there a principled way to set it?

Statistics

1 Reply

Accepted answer

WEweizh4 days ago

The threshold is a tolerance for the proportion of your discoveries that are false — pick it based on the cost of chasing a false lead in your setting. In screening (cheap follow-up) you might accept q=0.1; for expensive confirmation, tighten it. It's a decision, not a law.