"Two heads are better than one," the saying goes. But why?
A common explanation is that groups collect several opinions and average them. One person may guess too high, another too low, and the errors supposedly cancel each other out. The final answer lands somewhere in the middle — and, with luck, somewhere near the truth.
New research suggests that successful groups are doing something much more interesting.
Instead of merely mixing together their first guesses, they often build a new answer through discussion. They divide a difficult question into smaller, easier questions, estimate each part, and then put the pieces back together. Researchers call this process collective Fermi estimation.
The Fermi method is named after physicist Enrico Fermi, who was famous for making surprisingly good estimates with very little information. Imagine trying to guess how many steps there are in a large monument. Rather than pulling a number from thin air, you might estimate the number of floors, work out how many steps connect two floors, and multiply the figures.
None of the smaller estimates needs to be perfect. Together, however, they can produce a reasonable answer.
To investigate whether groups naturally use this strategy, researchers recruited 900 people for three experiments.
In the first experiment, participants entered online chatrooms in groups of four. Each group had to answer numerical general-knowledge questions, but received no instructions about how to reach an agreement.
Some groups simply exchanged their initial guesses. Others began taking the questions apart. They identified useful pieces of information, estimated those pieces, and checked one another's reasoning.
The second kind of group was consistently more accurate.
This advantage did not appear to come from one unusually knowledgeable person taking control. Participation was fairly balanced. What mattered was the kind of conversation the group had.
The second experiment tested whether this approach could be taught. Some groups were told to share their guesses and calculate an average. Other groups were instructed to divide each problem into smaller parts before combining their estimates.
Once again, the groups that broke the problem down produced better answers.
But perhaps the Fermi method itself — not teamwork — was responsible for the improvement. The third experiment examined that possibility. This time, everyone was told to use the method, but half worked individually while the rest worked in groups.
Individuals became more accurate when they divided the questions into smaller parts. Even so, the groups still performed better. Discussion gave participants a chance to challenge weak assumptions, catch mistakes, and improve one another's estimates.
The researchers also analyzed the language used in the chats. Stronger groups tended to mention ideas closely connected to the structure of the problem. When estimating the number of steps in a monument, for example, they discussed staircases, floors, and buildings. Their words revealed that they were constructing an answer rather than voting on one.
There is an important limit to the finding: the experiments focused on numerical estimation. The same method may not work equally well for moral choices, creative judgments, or decisions shaped by personal values.
Still, the study offers a useful lesson for meetings, classrooms, and project teams. A group does not become wise simply because more people are sitting around the table. Nor does a discussion improve merely because everyone gets a turn to speak.
The real advantage appears when people stop defending their first answers and start rebuilding the problem together.