Only Guessing
You make two judgements on every decision: the answer, and how sure you are of it. This page measures the second one against the first, narrated by an AI that cannot tell them apart because it makes them in the same step.
Yesterday this lab took a decision apart into how fast evidence arrives, how much of it you demand, and the plumbing. Every one of those is a fact about the answer.
This is about the other thing you do on every trial and have never noticed doing. After the answer is out, something in you forms an opinion about whether it was right. That opinion has its own accuracy, and it is not the same number as the accuracy of the answer. Two people who score identically can differ completely in how well they know which of their own answers to trust, and one of them is far more dangerous to work with than the other.
The measurement is meta-d': the sensitivity an ideal observer would have needed in order to produce the confidence ratings you actually produced. Divide it by your real sensitivity and you get the fraction of the evidence in your eyes that reached the part of you that reports on itself. One means all of it. Most people are well under one.
- 👁️ A faint striped patch flashes for an eighth of a second, then gets wiped by a mask. It leaned left or right. Say which.
- 🤔 Then rate how sure you were, from pure guess to certain. Answer first, always: the rating cannot be allowed to change the answer it is rating.
- 🎚️ A staircase finds the contrast where you are right about seven times in ten and freezes it there. You are supposed to be wrong a lot. Confidence is only measurable where there is something to be uncertain about.
- 🚫 No feedback after the warm up. You will never learn how you are doing, which is the only way your ratings can stay about the trial instead of about the score.
About eight minutes, 104 measured trials after a staircase. Everything is computed in your browser. Nothing is recorded and nothing leaves this page.