Get people to give 90% confidence intervals on 10 questions, and then at the end ask
“Ok, so who got all 10 within their intervals. 9? That’s what you should have got… ok, 8? Still no-one? Ok, how about 7?”
90% might not be the best number for demonstrating the idea of a confidence interval. It’s too close to 100%. There’s not much room to be underconfident. What about 50% confidence intervals?
Have you tried it? I have, and I can tell you most people I tried it on are over-confident when asked for 90% intervals.
Get people to give 90% confidence intervals on 10 questions, and then at the end ask
“Ok, so who got all 10 within their intervals. 9? That’s what you should have got… ok, 8? Still no-one? Ok, how about 7?”
90% might not be the best number for demonstrating the idea of a confidence interval. It’s too close to 100%. There’s not much room to be underconfident. What about 50% confidence intervals?
Have you tried it? I have, and I can tell you most people I tried it on are over-confident when asked for 90% intervals.