Most people, asked whether they are above-average drivers, will say yes. Roughly 80% will, in study after study, across countries and decades. The math obviously does not work. About half of drivers are below average, by definition. The interesting thing is not that overconfidence exists, but that it is so reliable across domains and so resistant to correction. People who know about the bias do not show much less of it. Awareness is not protection.
Overconfidence is structural, not personal
Behavioral economists distinguish three flavors of overconfidence: overestimation of your actual ability, overplacement relative to others, and overprecision about how much you know. All three show up reliably in experiments, even with experts in their own fields. Doctors estimating diagnostic certainty are wrong more often than their stated confidence implies. Investors estimating their stock-picking edge are wrong almost without exception. Engineers estimating project completion times underestimate them in roughly 90% of cases, and continue to do so even after being shown their own track records. This is not a personality flaw concentrated in arrogant people. It is a structural feature of how the brain weights its own evidence, and arrogant people are simply less embarrassed about expressing what most people quietly feel.
The cost is highest where the consequences are delayed
Overconfidence is most dangerous in domains where feedback is slow, sparse, or distorted. Driving is the canonical example. The average driver gets thousands of hours of feedback that says they are fine, because they did not crash today. The few catastrophic data points are too rare to update on personally. The same dynamic applies to investing, health behaviors, parenting decisions, and career choices. You get noisy, encouraging signals for years, and the rare costly signal arrives too late to be useful. Domains with fast, clean feedback, like competitive chess or pilot training in simulators, produce more calibrated confidence over time. Domains with slow feedback do the opposite. Most of life is the second kind.
Calibration is a learnable skill
The good news is that confidence calibration responds to practice, even if it does not respond to mere awareness. Forecasting researchers like Philip Tetlock have shown that people who explicitly score their predictions, attaching probabilities and tracking outcomes, become significantly better calibrated within a year. The exercise is uncomfortable. Most people do not enjoy seeing how often their “90% sure” predictions turn out to be 70% accurate. But the exercise works specifically because it bypasses the intuitive system that produces overconfidence and substitutes a written record. The same mechanic explains why pilots use checklists, why surgeons use timeouts, and why financial advisors who follow a documented process beat ones who rely on judgment. The discipline is not in trusting yourself less. It is in checking yourself more.
The takeaway
Overconfidence is not an enemy you can defeat once. It is a permanent tilt in the system, and the only reliable countermeasure is process. If you find yourself thinking “I do not need to write this down” or “I have done this enough times,” you have found the moment to write it down anyway.
Leave a Reply