What people think of Dunning-Kruger is almost never the same as what it is. The Dunning-Kruger hypothesis is that non-experts are (as a whole) not good at esti

What people think of Dunning-Kruger is almost never the same as what it is. The Dunning-Kruger hypothesis is that non-experts are (as a whole) not good at estimating their performance. This does not mean that they think that they are good, just that they think they are better than they are. They still tend to rate themselves as worse than an expert would rate themselves. My favourite explanation for this effect is not that people are bad at estimating their own performance, but that they are bad at estimating the range of performance of the population. They scale and shift their performance into a range they think is reasonable. Imagine a hypothetical test that uniformly scores a population from 0-100; everyone is spread out evenly across the range of scores. The average and median are both 50, and the same number of people get 32 as get 95, and every other score as well. Now imagine everybody is absolutely accurate at picking their own ranking on this test out of the population. Take two people, one who knows they are exactly average at this test, better than 50% of people; and one who is an expert, better than 90% of people. Based on the set up we know that these people will score 50 and 90. Now we 'scale and shift' the score axis. Say for example, that both people assume that the range of test scores will be between 40 and 90; the worst performer will get 40, and the best will get 90. Under this assumption the two people will predict their scores as 65 (vs 40, +25 over actual!!) and 85 (vs 90, -5 under actual!!) respectively. The worse performer grossly overestimated their performance, and the expert mildly underestimated their performance, even though both knew perfectly what their relative performance was. This shows that your estimation of your performance on a test can be wildly skewed by incorrectly estimating what the range of scores will be.