The , as the 1999 paper put it, is that poor performers rate themselves too highly, because the knowledge they lack is the same knowledge that would let them see it. The chart behind it is persuasive. Then someone fed the same method data with nothing in it at all. What came out?
The pattern appeared anyway. Comparing a test score against a self-rating that is itself calculated from that score produces a correlation out of the arithmetic, and it shows up in numbers generated at random with no psychology in them. The chart cannot serve as proof, because it would look the same in a world where the effect did not exist.
The original chart is genuinely persuasive. Split the participants into quarters by test score, show each quarter’s self-assessment beside its actual performance, and the lowest group sits far above the line while the highest sits slightly below it. It looks like a picture of a mind.
The critique is not about the data but about what can be read from it. Comparing a score against a gap computed from the same score ties the two together mathematically, before anyone asks a psychological question. pulls in the same direction on its own.
The test for that is . Generate data at random, where no relationship between knowledge and self-assessment can exist, run the original method, and see what happens. The familiar shape appears, which means the chart is a at least in part.
What follows is narrower than either side usually says. The graph cannot stand as proof, because it would look that way in a world without the effect. It also cannot prove the effect is absent. The dispute is live, some of the original authors hold that it survives correction, and the useful lesson is about the method rather than the verdict.
wat. Data built at random, with no relationship in it at all, still produces the Dunning-Kruger curve.
wat wat
A convincing graph does not always prove what it seems to. The measurement procedure has to be checked before the psychological conclusion is adopted.
How do we know?
Researchers built entirely random data, with no relationship between knowledge and self-assessment, ran the original method on it, and watched the familiar shape appear.
What is still unknown
This is an active dispute. Some of the original researchers hold that the critique overreaches and that the effect survives statistical correction.
Learn something?
Where to next
A player has scored three in a row. Is it really worth passing to him again?statistics
Why did the people who paid full price show up to more plays?bias
What was left of the marshmallow test when they ran it on ten times as many children?replication