Confidence and competence are different things

The two are treated as the same quality and are separable, which has consequences for hiring, for teams and for how you read other people.

The gap

Research on confidence calibration finds the correlation between confidence and accuracy is positive and considerably weaker than people assume.

Which means confidence carries some information about competence and much less than it is treated as carrying.

Both mismatches are common. Highly capable people who present tentatively, and confident people whose judgement is unreliable.

Why the wrong one is rewarded

Competence is expensive to assess and confidence is free. Evaluating whether someone's judgement is sound requires domain knowledge and time. Observing how certainly they speak requires neither.

Which makes confidence the default proxy in exactly the situations where it is least reliable — brief encounters, interviews, meetings with people outside your field.

Confident statements are more memorable and are repeated more.

Hedged accuracy is punished. A correct forecast stated with appropriate uncertainty is remembered less well than a wrong one stated definitely, which creates a professional incentive against calibration.

Groups defer to confidence. Studies of group decision-making find that confidently expressed views carry disproportionate weight regardless of the speaker's actual accuracy.

Which is a mechanism by which teams reliably make worse decisions than their best member would alone.

The consequences

For anyone competent and not confident: capability is systematically undervalued, and the correction is presentational rather than substantive.

Not manufactured certainty — stating conclusions clearly, once, without preamble and without appended qualification.

"I think we should do X because Y" rather than "I might be wrong but perhaps we could consider X, though I'm not sure."

The content is identical and the reception is not, which is unjust and is worth working with rather than around.

For anyone assessing others: confidence should be discounted, deliberately, because it will otherwise be overweighted automatically.

The correction is to ask for reasoning rather than conclusions. Someone confident and correct explains why fluently. Someone confident and wrong frequently cannot, and the difference is visible within two questions.

Calibration as the actual goal

Rather than more confidence or less, the useful property is confidence proportional to evidence.

Research on forecasting has found that calibration improves with practice, feedback and explicit probability estimates — the work on superforecasting being the most cited example.

The practices that improve it are unglamorous: recording predictions with numerical confidence, checking them, and noticing the pattern of over- or under-confidence.

Almost nobody does this, which is why almost nobody is well calibrated.

Expressing uncertainty well

Since calibration only helps if it can be communicated without sounding weak.

Be specific about what is uncertain. "I'm confident about the numbers, less so about the assumption behind them" is more useful and more authoritative than a general hedge.

Give a probability where you can. "About seventy per cent" carries more information than "probably" and reads as considered rather than evasive.

State what would change your view. This signals that the position is reasoned, and it is the single most credible thing anyone can say in a disagreement.

Do not apologise for the uncertainty. The uncertainty is a property of the situation, not a deficiency in you.

The version worth holding

Confidence is a claim about your knowledge, and like any claim it can be accurate or not.

Someone worth listening to is not the one who is most certain but the one whose certainty tracks the evidence — and who is visibly less certain when the evidence is thinner.

That is a rarer quality than confidence and considerably more valuable.

Hiring and the interview problem

Where the gap costs the most.

Unstructured interviews reward fluency and confidence, and their predictive validity for job performance is consistently poor in the research literature.

Structured interviews with the same questions, defined scoring and work-sample tasks perform considerably better, largely because they measure something other than presentation.

Which is a well-established finding that most organisations continue not to act on.