🔋Statistical power

The chance that a test will detect an effect that is really there. One minus the probability of missing it, written 1 − β. It rises with sample size and with the size of the effect, and falls when the data are noisy.

By convention people aim for 80 percent, which still means one real effect in five goes unseen.

From 🎓 Lecture 10 - Power, sample size and stopping early. I had already used it without the word: 📌 Design the triage A-B test was the first test I sized myself, in the A-B test sample size sheet.

Origin: the concept is Jerzy Neyman and Egon Pearson's, from their work on testing in the 1930s. Power comes through Anglo-French from Latin posse, to be able.

In one line: an underpowered test that finds nothing has told me nothing.