Part IV — Uncertainty

The number is indefensible. What is the machinery for saying how indefensible?

Two chapters, placed here rather than at the front of the book on purpose. The method is not useful until you have a model that has produced a number you cannot defend, and ch12 is where that happened.

ch13 · Monte Carlo gives you the whole technique, from first principles, in about as much code as fits on a page. It assumes you can read code and do arithmetic, and nothing at all about statistics. Four words of vocabulary arrive in it, each because a model has just raised a question that needs it.

ch14 · Correlation and convergence picks up the two things the first chapter assumed without establishing: that the inputs move independently, and that a hundred thousand samples was enough. Draw inputs independently when they in fact move together, and the model reports a narrower interval than the evidence supports. More samples do not narrow an interval at all. They settle where it sits.

Read both knowing what neither can do, which is ch20’s subject. Sampling quantifies the uncertainty a model can see, beautifully. It says nothing whatever about the uncertainty it cannot.