Index > Course > 2021-03-02: Everyone is Overconfident, About Everything, All the Time
Heuristics:
How to account for subjective probability?
Non-experts show overconfidence. Calibration attempts to account for this.
How to perform a probability assessment
How to get reliable, consistent, accurate data from experts?
The Delphi Method:
Availability bias: we only really consider things that we know about.
Anchor and adjust bias: we know this number, and adjust to account for problems that we know about.
Am I gonna have to turn on mathjax lol
Probability Density function: What is the probability of X happening, given parameter Y? This is the classic bell curve.
Cumulative Density Function: The probability of X happening, given that parameter Y is at or less than some value. It’s the integral of the PDF.
This is going to take tikz in the pdfs. I barely know tikz.
When google fails.
It’s time to find an adult who knows what they’re doing. Or lots of them. But we can’t just ask them what they think:
A series of questions with numerical answers. No looking things up. Provide your best guess and your 90% confidence interval.
Question | Correct Intervals | Percentage |
---|---|---|
1 | 3 | 30% |
2 | 6 | 60% |
3 | 4 | 40% |
4 | 5 | 50% |
5 | 2 | 20% |
6 | 5 | 50% |
If we were good at making confidence intervals, we should capture the correct value 90% of the time. But at best, we saw 60%. Our ranges were too small, we were too confident in our answers.
I captured the right answer in 4 of 6 questions, or 66% of the time.
We were overconfident in the familiar questions, and still overconfident with unfamiliar questions.
We say that infrequent things, never happen. And common things, always happen.
“Everyone is overconfident; in Everything; All the time.” - Prof.
Except weather forecasters, apparently. They predict things, with percentages, every day. And they get quick feedback.
Good job, weatherman.
Index > Course > 2021-03-02: Everyone is Overconfident, About Everything, All the Time