BookYol
tren

Concept

Incremental updating

Incremental updating is changing your estimate often and in small steps as evidence arrives, in proportion to how surprising the evidence is: from 65 to 68 percent, not from 65 to 90. It was the single behaviour that most distinguished Tetlock's superforecasters.

There are two common ways to handle new information and both are wrong. The stubborn person treats a first judgement as identity and dismisses everything after it. The jumpy person treats the latest headline as the whole story. Superforecasters did a third thing: they weighed each new fact against everything already known and moved a little. The logic is Bayesian in spirit; a friend's bad experience is one data point against a base rate built from twenty, and should move you slightly, while a pattern of three such reports should move you more. Small updates compound into accuracy; large swings cancel into noise.

A real-life example

In week three his new manager cancels two meetings and his probability that the move works drops from 65 to 55. In week six a project he asked for is approved and he moves back to 62. He never swings to 90 on a good week or 30 on a bad one.

How to use it

  1. 1When evidence arrives, ask how surprising it would be if you were right versus wrong, and move in proportion.
  2. 2Write the old number, the new number and the reason for the move.