Channel · confidence vs. evidence
Overconfidence Effect
A well-documented miscalibration in which people's subjective confidence in their judgments, estimates, and plans systematically exceeds their objective accuracy.
Free-run conditions
In classic calibration studies, when people say they are virtually certain they are right, they are right far less often than they claim: Fischhoff, Slovic and Lichtenstein (1977) found subjects who set the odds of being correct at 100:1 were actually right only about 73% of the time. Moore and Healy (2008) showed that "overconfidence" is really three separable phenomena — overestimation (thinking you did better than you did), overplacement (thinking you beat others), and overprecision (excessively narrow confidence intervals) — that do not always co-occur and can even reverse. Overprecision is the most robust form, but the broader effect is contested: Juslin, Winman and Olsson (2000) and the ecological school argue much of the lab "overconfidence" on trivia questions is a methodological artifact of non-representative item sampling that shrinks dramatically with representative samples.
Channel readings
Knowing with certainty (extreme-odds calibration). Confidence tracked accuracy up to ~3:1, then decoupled; at stated odds of 100:1 subjects were correct only ~73% of the time, and even at 10,000:1-1,000,000:1 odds they were right only ~85-90%.
Fischhoff, B., Slovic, P., & Lichtenstein, S., 1977 · 528 paid volunteers across 5 studies
The Trouble With Overconfidence (three faces). Three distinct constructs: overestimation appeared in ~64% of empirical studies, overplacement in ~5%, overprecision in ~31%; on hard tasks people overestimate self yet underplace vs others, on easy tasks the reverse — explained by regressive estimates of self and (more regressive) others.
Moore, D. A., & Healy, P. J., 2008 · Literature review (365 hits) plus original lab sample
Exploring the planning fallacy. Students predicted finishing an academic project in ~33.9 days but actually took ~55.5 days; only ~45% finished by the date they were 99% certain about — overprecision/optimism in plans.
Buehler, R., Griffin, D., & Ross, M., 1994 · Undergraduate samples (multiple studies)
Naive empiricism and dogmatism in confidence research (ecological critique). Once methodological artifacts are controlled, the hard-easy effect is 'almost eliminated,' implying much measured overconfidence reflects non-representative item sampling rather than a stable cognitive bias.
Juslin, P., Winman, A., & Olsson, H., 2000 · Meta-analytic / re-analysis of prior calibration datasets
Why the signal misleads
There is no single agreed mechanism. The heuristics-and-biases tradition attributes overconfidence to biased information search (selectively recruiting confirming evidence), insufficient adjustment, and an 'inside view' that ignores base rates (planning fallacy). Moore & Healy (2008) offer a more deflationary account: people have imperfect, regressive knowledge of their own performance and even more regressive knowledge of others', so apparent over/under-confidence falls out of how estimates regress toward priors. The ecological school (Gigerenzer, Hoffrage & Kleinbölting 1991; Juslin et al. 2000) argues people are largely well-calibrated to their natural environment and that lab overconfidence is an artifact of experimenters selecting unrepresentative, misleading trivia items.
Cognitive-bias account (genuine miscalibration) vs ecological/artifact account (representative sampling removes most of the effect) vs Moore-Healy regression/error account (overconfidence as a statistical consequence of noisy self- and other-estimates). These remain actively debated; overprecision survives more of these critiques than overestimation or overplacement.
Calibration verdict
Partial lock — robust in its narrow form, unresolved in its broad one; the trace still jitters.
Overprecision (overly narrow confidence intervals; 90% intervals often capture the truth <50% of the time) is one of the most reliably reproduced findings in judgment research and survives many critiques. But the broader 'overconfidence on general-knowledge questions' is genuinely contested: Gigerenzer et al. (1991) and Juslin et al. (2000) showed that representative item sampling sharply reduces or eliminates the classic effect and the hard-easy effect, and Moore & Healy (2008) demonstrated that overestimation, overplacement and overprecision dissociate and even reverse by task difficulty. So the honest picture is a robust core (overprecision, planning fallacy) wrapped in a contested, partly artifactual periphery.
Recorded over-runs
CEO overconfidence and value-destroying acquisitions · 2008
Malmendier & Tate, using 1980-1994 data on large US firms (with overconfidence proxied by CEOs holding deep-in-the-money stock options to expiration), found overconfident CEOs make more, and more diversifying, acquisitions; the market's announcement reaction was significantly more negative for overconfident CEOs (about -90 basis points vs about -12 bp for others).
Physician diagnostic overconfidence and diagnostic error · 2008
Berner & Graber review evidence that physicians systematically underappreciate the chance their diagnoses are wrong; autopsy series across decades show diagnostic error rates of roughly 10-15%, and in one cited test experienced dermatologists confident of a melanoma diagnosis in over half of cases were wrong about 30% of the time.
Damping
Take the outside view: replace the inside-view estimate with the base rate from a reference class of similar past cases, and widen stated confidence intervals (e.g., deliberately ask 'what would surprise me?' and pre-mortem how this could fail). Calibration feedback and using representative, not cherry-picked, evidence also shrink the effect.
Buehler et al. (1994) showed inside-view reliance drives the planning fallacy; Juslin et al. (2000) and Gigerenzer et al. (1991) showed representative sampling sharply reduces measured overconfidence.
Reading the trace in the wild
Look for confident point predictions or narrow ranges stated without base rates or track records: 'we'll ship in six weeks,' '90% sure,' 'no way this fails.' The tell is precision and certainty that outrun the evidence — especially on novel, complex, or self-relevant tasks where no one is citing how similar past efforts actually turned out.
Adjacent channels
- Planning FallacychildA specific overconfidence/optimism in time and cost estimates for one's own plans (Kahneman & Tversky 1979; Buehler et al. 1994).
- OverprecisionchildThe most robust face of overconfidence: confidence intervals that are far too narrow (Moore & Healy 2008).
- Better-than-average effect / illusory superioritysiblingClosely tied to overplacement (believing you outrank others); often confused with overconfidence but is a distinct construct.
- Hard-easy effectmechanistically-linkedOverconfidence rises with item difficulty and inverts on easy items; central to the Juslin et al. artifact debate.
- Dunning-Kruger effecteasily-confusedAbout low performers overestimating their skill specifically; an instance of overestimation, not the same as the general overconfidence effect.
- Optimism biassiblingOverly rosy expectations about future outcomes; overlaps with but is distinct from confidence in current judgments.