My good friend Rich Burnham, knowing my interest in off-beat science and stats, drew my attention to this video by YouTuber Michael Stevens (aka “Vsauce”) on an experiment that failed to confirm a phenomenon called the “wisdom of the crowds.”

Normally, as demonstrated by Sir Francis Galton in 1906 from data collected at a country fair on 787 guesses at the weight of an ox,* groups of people exhibit a high level of collective intelligence via a simple median (the “middlemost estimate”)—being off by only 9 pounds for the 1,198 pound ox. This amazes me—blowing away my mindset that the wisdom of a crowd degrades to the ‘lowest common denominator,’ that is, the people with the least knowledge.

Experts agree with Vsauce’s hypothesis that the complete failure of his crowd to correctly guess the number of jelly beans in his jar stemmed from the estimates being shared, rather than gathered with no cross-talk.

“The wisdom of crowds requires that people’s estimates be independent. Studies have found that when people can observe the estimates of others, the accuracy of the crowd typically goes down. People’s errors become correlated or dependent, and are less likely to cancel each other out. We follow our peers, to the detriment of the performance of the group.”

  – Psychology professor Tania Lombrozo, No Man Is An Island: The Wisdom Of Deliberating Crowds, posted 3/12/18 by WGCU, a National Public Radio-member station on Florida’s Gulf Coast

I made the same mistake in a 2019 contest for my Anderson clan. While vacationing together at a lakeside resort, I gathered individuals’ estimates on the number of aluminum-can pull-tabs I’d collected for donation to the Ronald McDonald House in Minneapolis. See the picture below of my wife Karen (holding Bertie) working with our oldest grandchild Archer do the count. I asked the participants to write down their guesses on a clipboard by the jar, which created more fun via the gaming aspects of going just above or below a competitor, but violated the statistical requirement for independence.

An interesting workaround that allows collaboration for tapping the “wisdom of the crowds” is to first break the group into a number of teams and then average out their consensus estimates. See the research, based on results from a group of 5,180 people asked to estimate the height of the Eiffel Tower or the like, at this 2018 Letter by Nature Human Behavior on Aggregated knowledge from a small number of debates outperforms the wisdom of large crowds.

To keep things simple, the next time my bottle of pull-tabs fills up for another contest to guess the total, I will go with the simpler approach for crowd wisdom by banning cross talk and then seeing if the median estimate wins. If it doesn’t work, I will blame it on our family group being too small (though it does exceed 20—all in one cabin!).

* “Vox Populi,” Nature, 1907