The original D-K papers also found different curves for different subject matter.
I can think of several explanations for this, all of which might be true are definitely at least a little true:
Some subjects have higher variance in performance, resulting in steeper D-K curves.
Some subjects have higher variance in test-ability-to-measure-performance, again resulting in steeper D-K curves.
An actual D-K effect does exist, sometimes, superposed over the statistical mirage; and it’s stronger for some subjects than others.
An anti-D-K effect exists, and it’s stronger for some subjects than others.
Something else is happening I don’t know about.
And they made the unusual choice of dividing their populations into quartiles, throwing away quite a bit of resolution.
Doesn’t seem unusual to me ( . . . or suspicious, if that’s what you’re getting at). I get away with using deciles at my day job because I work on large datasets with low-variance data, and I get away with it here because I can just add zeroes to the number of elves simulated until my plots look as smooth as I want; Dunning & Kruger had a much smaller sample since they were studying college classes full of real round-eared human beings, and sensibly chose to bucket them into fewer buckets.
I can think of several explanations for this, all of which
might be trueare definitely at least a little true:Some subjects have higher variance in performance, resulting in steeper D-K curves.
Some subjects have higher variance in test-ability-to-measure-performance, again resulting in steeper D-K curves.
An actual D-K effect does exist, sometimes, superposed over the statistical mirage; and it’s stronger for some subjects than others.
An anti-D-K effect exists, and it’s stronger for some subjects than others.
Something else is happening I don’t know about.
Doesn’t seem unusual to me ( . . . or suspicious, if that’s what you’re getting at). I get away with using deciles at my day job because I work on large datasets with low-variance data, and I get away with it here because I can just add zeroes to the number of elves simulated until my plots look as smooth as I want; Dunning & Kruger had a much smaller sample since they were studying college classes full of real round-eared human beings, and sensibly chose to bucket them into fewer buckets.