Marshall, you don’t need to be good at the question subjects, just as long as you don’t think you’re good when you’re not. Calibration tests aren’t about how many of the questions you can get right, they test if you’re over (or under) confident about your answers. They tend to use obscure questions which few people are likely to know for sure what the answers are.
Thanks Michael—I just don’t think calibrating on useless information is evidence of my rationality. I am 95% sure, that I don’t know all the time. Calibrating on whether I book my next dental appointment on time seems a better clue.
Interesting point. Does anyone know of any evidence about how well calibration test results match overconfidence in important real-life decisions? I’d expect it would give a good indication, but has anyone actually tested it?
There are a lot of tests that look plausibly useful but would be much more trustworthy if we could find a sufficiently good gold standard to validate against.
If we have enough tests that look plausibly useful, each somewhat independent-looking, we could see how well they correlate. Test items on which good performance predicts high scores on other test items would seem more credible as indicators of actual rationality.
We could include behavioral measures of success, such as those suggested by Marshall, in our list of test items. (Income, self-reported happiness, having stable, positive relationships, managing to exercise regularly or to keep resolutions, probabilistic predictions for the situation you’ll be in next year (future accomplishments, self-reported happiness at future times, etc.; coupled with actual reports next year on your situation) in our list of test items. If we can find pen-and-paper test items that correlate both with behavioral measures of success and with other plausibly rationality-related pen-and-paper test items, after controlling for IQ, I’ll say we’ve won.
Marshall, you don’t need to be good at the question subjects, just as long as you don’t think you’re good when you’re not. Calibration tests aren’t about how many of the questions you can get right, they test if you’re over (or under) confident about your answers. They tend to use obscure questions which few people are likely to know for sure what the answers are.
Thanks Michael—I just don’t think calibrating on useless information is evidence of my rationality. I am 95% sure, that I don’t know all the time. Calibrating on whether I book my next dental appointment on time seems a better clue.
Interesting point. Does anyone know of any evidence about how well calibration test results match overconfidence in important real-life decisions? I’d expect it would give a good indication, but has anyone actually tested it?
There are a lot of tests that look plausibly useful but would be much more trustworthy if we could find a sufficiently good gold standard to validate against.
If we have enough tests that look plausibly useful, each somewhat independent-looking, we could see how well they correlate. Test items on which good performance predicts high scores on other test items would seem more credible as indicators of actual rationality.
We could include behavioral measures of success, such as those suggested by Marshall, in our list of test items. (Income, self-reported happiness, having stable, positive relationships, managing to exercise regularly or to keep resolutions, probabilistic predictions for the situation you’ll be in next year (future accomplishments, self-reported happiness at future times, etc.; coupled with actual reports next year on your situation) in our list of test items. If we can find pen-and-paper test items that correlate both with behavioral measures of success and with other plausibly rationality-related pen-and-paper test items, after controlling for IQ, I’ll say we’ve won.