> dumb people always think they're smarter than they actually are

This is a really bad summary of the original D/K paper, both misleading and incorrect. Maybe this illustrates nicely why the article was needed. They didn't knowingly test any dumb people; it wasn't a random population sample, it was all Cornell undergrads volunteering for extra credit. The results are relative, not absolute, they were being asked to rate themselves against the population (their classmates), not actually rate their ability. That involves guessing how good others are. We don't know how good they all were in absolute terms, only their relative scores. The results did not conclude anything close to "always", it only showed an average bias. The results show that, on the tasks they tested, the people with the best skills under-estimated their scores relative to their peers.

the results are indeed relative. the smarter one gets the same thing still applies. it's a theory not to point at others and say 'he's on mount stupid'.. but rather one to be wary of yourself being on this curve, which in some sense or aspect one always is, as one is never a master of everything in their lives. it's a theory about evaluation of self only applicable in that domain.

I wonder if that's because "best skills" entails "most suspicious" too.

> “dumb people always think they’re smarter than they actually are”

But that makes it sound like only people at the bottom overestimate themselves, when in reality 80% of people overestimate themselves. How about this rewrite: "People who aren't experts think they're smarter than they actually are." And get rid of the word "always", we're talking about averages, not about certainty in every case.

I think you're falling into exactly the same inaccuracy that you're trying to combat. The issue is not that everyone below the top 20% of performance over-estimate their performance, it's that over-estimating performance is strongly correlated to incompetence. Your summary makes it sound like someone who is of average competence is over-estimating their performance to the same extent as those with the worst performance, which is untrue. In fact when we talk about the DK effect almost no one is talking about the average person who thinks they're smarter than 65% of other people because that effect is almost impossible to qualitatively observe.

I agree, not unimportantly the effect is much more pronounced in the bottom group.

When someone overestimates their abilities by 20% it is far less drastic than a whopping 500%

500%? I don't think you can compare percentiles like that.

Let's say someone estimates his percentile to be the 3rd percentile (aka 97% of people are better) and in reality he's in the 1st percentile (99% of people are better). Did he overestimate his ability by 200%? I wouldn't say so, I would say he was very very close.

Even at non-extreme percentiles I don't think such a comparison works. For example in some area it might be that people at the 10th percentile know nearly the same amount as people at the 50th percentile. If it's some area that most people know little about. So the abilities of someone at the 10th percentile would be nearly the same as the abilities of someone at the 50th percentile. So overestimating your percentile by a lot is not the same as overestimating your abilities by a lot.

What I see in practice is that the effect is often used to discredit the competency of people like the WP author did.

> which itself, ironically, is part of the DK effect

I think the effect can help you come to the conclusion many experts in fields come to after a while: The larger your knowledge, the more you are aware of your ignorance. I am sure there is an English idiom that describes it more gracefully. Maybe that is the intersection to the 20%.

Enthusiastically pointing out the discrepancy on the other end of the curve is most probably an indicator that you might have fallen "victim" to the effect itself.

It reads like no one had any sense of how well they would do on the test. So people guessed they'd be slightly above average which is a good prior for a group of Ivy league students. This lead to the lowest being "overconfident" and the best being "under confident". Designing good studies is hard.