← Knowledge universe

V — Probability & Statistics

Base rates, Bayes, and the shape of uncertainty

Conditional probability, prior and posterior, p-values, effect sizes, confidence intervals, regression to the mean.

Epistemic status · Well Established

The base rate does most of the work

A 99%-accurate test for a condition affecting one in ten thousand people still yields mostly false positives. The evidence updates the prior; it does not replace it.

What a p-value is not

It is not the probability the hypothesis is false. It is the probability of data this extreme if nothing were going on. Effect size and interval say more than significance.

Regression to the mean

Extreme measurements drift back toward average on their own. This alone manufactures apparent cures, coaching effects, and curses.

Practice

Before reading a study's result, estimate the prior probability the claim is true.

TEST A BELIEF

Next

Taking the irrational mind seriously without surrendering to it