At Serendipity 3 in New York City, a $69 hot dog changed how we think about decision-making. Topped with truffle butter and foie gras, the outlandishly priced menu item held the Guinness World Record for most expensive hot dog—and it worked exactly as planned. By anchoring customers' sense of value, the $69 hot dog made the restaurant's $18 cheeseburger suddenly seem like a bargain, and sales soared. But this anchoring effect, which nudges our judgments in predictable directions, has long been considered an inescapable trap of individual thinking. A landmark 1994 study by psychologist Daniel Kahneman—who won the Nobel Prize in Economics for this work—showed how pervasive and unavoidable anchoring bias is.
Now researchers at the University of Pennsylvania have discovered something encouraging: in networks, anchoring bias loses its grip.
In a study published in the Journal of Social Computing, Professor Damon Centola and doctoral candidate Calvin Isch of the Annenberg School for Communication tested whether groups could escape anchoring bias through peer conversation. They recruited 1,600 people for an experiment with a deceptively simple task: guess the number of pennies in a photograph of 246 coins. Half the participants worked alone; the other half worked within peer networks. Before making their estimate, all participants were primed with an anchor value—either a low anchor of 118 coins or a high anchor of 353 coins.
The anchors worked as expected. Without any anchor, participants guessed a median of 185 pennies. When primed with the low anchor, their guesses dropped to 141 (moving further from the truth). When primed with the high anchor, their guesses rose to 200 (moving closer). But what happened next revealed something striking.
After sharing their initial guesses, network participants exchanged opinions with their peers and revised their estimates based on what others had guessed. Those working alone were given the same chance to revise, but with no information from others. The difference was dramatic: participants in the network groups reduced their errors by 22 percent across both the low- and high-anchor conditions, while those working alone showed no improvement at all.
"Social influence in egalitarian networks allowed errors to cancel out, undoing the harmful anchoring bias, and making individuals much more rational in their estimates," Centola explained.
The mechanism behind this success was unexpected. Centola and Isch discovered that participants who were most accurate paradoxically reported less confidence in their own judgments, but greater confidence in others. It was this faith in peers—not faith in oneself—that enabled groups to collectively overcome biases that trap individuals in isolation.
The implications extend far beyond academic curiosity. As more of our decisions flow through online platforms and social networks, this research suggests those spaces could be redesigned to help people think more clearly together. The key is creating egalitarian forums where opinions circulate freely and people feel genuine confidence in one another's judgment.
"Sharing opinions in an egalitarian way enables even explicitly biased populations to overcome biases that are impossible for individuals to escape on their own," Centola noted. The findings hint at a more hopeful picture: we are not doomed by the biases wired into our minds. The right social structures—ones built on mutual respect and open exchange—can help us see more clearly together than we ever could alone.