Confidence Is Not a Competence Signal: Reading Dunning and Kruger Properly

What the finding does and does not say

Popular usage has flattened this into "stupid people think they are clever," which is neither accurate nor useful. The effect is domain-specific: the same person is well calibrated in their area of expertise and poorly calibrated outside it. It describes a relationship between skill and self-assessment, not a category of person.

It is also worth noting honestly that the finding has attracted serious statistical criticism — some researchers argue that regression to the mean and measurement noise account for part of the pattern. The core practical implication survives that critique: self-assessed competence is a weak signal, and it is weakest precisely where you can least afford to rely on it.

The interview problem

Unstructured interviews are, in large part, confidence-measurement instruments. They reward fluency, composure and articulate self-narrative — traits that correlate with expressed certainty far more reliably than with actual capability.

The second half of the finding compounds this. Genuinely expert candidates underestimate their standing and describe their skills with appropriate hedging. Set against a confident, less capable candidate, they present worse. Your process can systematically prefer the weaker applicant while everyone in the room feels the decision was obvious.

Measuring capability instead of confidence

Use work samples. Ask candidates to do a representative piece of the job and assess the output, not the account of it. Structure interviews with the same questions in the same order and defined scoring anchors fixed before any candidate is seen.

Internally, the same logic applies to promotion and development. Self-assessment is worth collecting as data about calibration, not as evidence of capability — and a candidate's awareness of the limits of their own knowledge is itself a meaningful signal, one that unstructured processes routinely penalise.