11 May 2011 - significant result

Everybody knows that “statistically significant” means 95% guaranteed to be true, and everybody also knows that a lot fewer than 95% of published statistically significant results are true. And everybody knows why: It’s because you keep trying until you get a publishable result, and doing that violates the mathematical correctness requirements of the significance test. Knowledge is not power; inertia is power.

We need a new arbitrary publishability criterion that will tell us what’s interesting. I suggest numerology: Do the title and author work out to cool numbers?