Huh? Overconfidence May Signal Skill After All

Dunning-Kruger May Have It Backward, Researchers Argue

For decades, the Dunning-Kruger effect has been one of psychology’s most famous findings, holding that the least skilled people are the most overconfident. A new theory in Psychological Review argues the reverse may hold: the most skilled people may be the ones who overrate themselves most, and bragging and cold feet may be a matched pair of traits that evolved to work together.

Researchers Chris Dawson, of the University of Bath, and David de Meza, of the London School of Economics and Political Science, built a mathematical model around two well-known quirks of the human mind. Overconfidence is the habit of rating one’s own ability higher than it really is. Loss aversion is the tendency to feel a loss more sharply than a gain of the same size.

Dawson and de Meza also say the classic Dunning-Kruger pattern is a statistical illusion. Test scores include luck, de Meza said. Someone who scores badly may simply have drawn unlucky questions, which makes their own guess look far too high. Sorting participants purely by score, he said, can “manufacture a false psychological phenomenon.”

Model Shows Overconfidence Can Signal Real Ability

If people are known to exaggerate, why would anyone believe a confident claim? The researchers borrow an idea from economics called signaling. A college degree is the classic example. Employers trust it partly because finishing is costly, and people who are less capable or less driven would struggle to finish.

Overconfidence carries a similar price, the authors argue. Someone who truly believes they are better than they are will act on that belief, and the bad calls add up: a failed business or a poor trade. Those mistakes cost less for people who really are skilled. A highly capable person who overrates herself slightly still makes mostly sound choices, while a low-skilled person who overrates himself by the same amount makes costly ones. That gap keeps the signal honest.

People may not be consciously aware of any of this. The researchers argue that evolution, through a long process of trial and error across generations, could favor such a trait because of its social benefits.

Hidden Loss Aversion Cushions the Cost of Overconfidence

Loss aversion, the cold feet in this pairing, acts as a brake. Losing twenty dollars stings more than finding twenty dollars pleases, a quirk that traditional economic models treat as irrational. In the model, the extra sting steers overconfident people away from the worst gambles.

Here is the catch: the brake has to stay out of sight. If loss aversion were visible during a pitch, listeners would expect bold talk to go unmatched by bold action and would discount it. Because loss aversion only partly offsets the cost of overconfidence, the signal stays costly and believable. In the authors’ words, “There is no need to boast of timidity.” Caution typically shows up at decision time, after the persuading is done. That pattern gives the paper its title, “Talking the Talk, Not Walking the Walk.”

Data From 4,000 Online Test Takers Fits the Overconfidence Theory

To check the idea, the authors used an existing dataset from a 2021 study in Nature Human Behaviour. About 4,000 participants, recruited through Amazon’s Mechanical Turk, a site where people complete small online tasks for pay, took 20-question grammar and logic tests, then guessed how many they had answered correctly. The paper does not report their demographics.

On average, participants guessed more correct answers than they earned: 12.5 versus 10.2 in grammar, and 10.9 versus 9.5 in logic. Still, higher guesses went with higher scores, so confidence carried real information even when inflated. In grammar, 25.35% of participants underestimated their performance, and in logic, 35.42% did. The gap between belief and reality also tended to widen as scores rose, a pattern the model predicts and the reverse of Dunning-Kruger.

None of this proves the evolutionary theory, which the authors concede is inherently hard to test. Nothing in the analysis measured loss aversion, so whether the two traits actually evolved together is still unknown.

Sex differences get speculative treatment as well. Men show more overconfidence than women in many studies, and the authors propose a mating explanation. Women have historically prized resources and commitment in a mate, qualities that are hard to see and need signaling, while men have put more weight on physical attractiveness, which is visible. Evidence on whether men or women are more loss averse is mixed, and the model predicts no clear pattern.

Removing Both Biases Could Leave People Worse Off, Authors Argue

Overconfidence does cause real harm, including failed startups and excessive stock trading. The authors’ point is that judging each trait alone misses how the two work as a system. Nobel laureate Daniel Kahneman advised, in his 2011 book Thinking, Fast and Slow, that organizations eliminate both traits if they can. The authors call that advice possibly misguided and warn it could introduce a new distortion.

Bragging and caution may look like contradictions, but the paper’s closing line puts it this way: “two seeming bads may actually make a good.” Until the idea is tested more directly, treating these traits as flaws to scrub out looks premature.

---

Disclaimer: This article summarizes a published academic paper for general information only. It is not medical, psychological, financial, or professional advice, and the study’s findings are theoretical. Readers should consult the original paper and a qualified professional before making decisions based on its conclusions.

---