Last November, Robin described a study where subjects were less overconfident if asked to predict their performance on tasks they will actually be expected to complete. He ended by noting that “It is almost as if we at some level realize that our overconfidence is unrealistic.”
I think there’s a less perplexing answer: that at some level we realize that our performance is not 100% reliable, and we should shift down our estimate by an intuitive standard deviation of sorts. That way, we can under-perform in this specific case, and won’t have to deal with the group dynamics of someone else’s horrible disappointment because they were counting on you doing your part as well as you said you could.
Um, I don’t actually remember now– I thought that one of the results was that people compensated more for overconfidence when the tasks were not too difficult. But I don’t see that, looking it over now.
I think there’s a less perplexing answer: that at some level we realize that our performance is not 100% reliable, and we should shift down our estimate by an intuitive standard deviation of sorts. That way, we can under-perform in this specific case, and won’t have to deal with the group dynamics of someone else’s horrible disappointment because they were counting on you doing your part as well as you said you could.
First, welcome to Less Wrong! Be sure to hit the welcome thread soon.
Doesn’t your hypothesis here predict compensation for overconfidence in every situation, and not just for easy tasks?
Yes it does.
...
Is there some implication I’m not getting here?
Um, I don’t actually remember now– I thought that one of the results was that people compensated more for overconfidence when the tasks were not too difficult. But I don’t see that, looking it over now.