Why the average of many estimates beats almost any single person, and when it stops working.
In 1906, at a livestock fair in Plymouth, England, there was a contest: guess how much an ox would weigh once slaughtered and dressed. Nearly 800 people entered, most with no experience of cattle. The scientist Francis Galton, who believed ordinary people were hopeless estimators, collected the tickets to prove it.
He found the opposite. The median of all the tickets, the value right in the middle, came within 1% of the true weight. Closer than almost any single entrant, butchers and farmers included. Galton published it in Nature in 1907 under the title Vox populi.
Everyone gets it wrong, but each in their own way: some too high, some too low, for different reasons. Pool many estimates and those errors cancel each other out, leaving what they all have in common: the bit of truth each person saw.
It has been repeated many times. One of the best-known versions was done by professor Jack Treynor with a jar of jelly beans in class: the class average beat all but one of the students.
There's an important detail. When people estimate very large quantities, a few wild answers (someone adding three extra zeros) can sink the ordinary average. The median, the middle value, isn't dragged by those extremes. And when answers range from 10 to 10 million, the fairest center is found in times, not units, as we explain in why we're so bad with big numbers.
The magic needs one condition: everyone has to estimate on their own. In 2011, a team at ETH Zurich (Jan Lorenz and colleagues) repeated the experiment but let participants see what the others said. The answers grew more alike, people felt more confident… and the group got worse. When people copy each other, errors stop canceling and start adding up.
That's why in GutGuesser you don't see where everyone landed until you've locked in your answer. If you saw it first, your number would drift toward the others and the crowd would stop being worth anything.
In 2008 Edward Vul and Harold Pashler found something curious: ask a person the same question twice, some time apart, and the average of their two answers beats either one. Each time you estimate you draw on slightly different memories and reasoning, and combining them works like the crowd, on a small scale.
A trick that follows: once you have your number, assume it's wrong and ask why. Make a second estimate reasoning the other way, and take the middle of the two, in times.
Every challenge is a small Galton experiment: thousands of people estimating the same thing on their own. On the past days pages you can see how people did on each question, without the answers. Some almost nobody nails; on others the whole crowd is off by ten.
Four questions a day, one minute. Nobody knows the number; whoever gets closest wins.
Play today's challenge