After investigating results of expert picks and algorithmic prediction of 2011-2012 season NFL game in earlier posts, we may have been convinced that systematic study of relevant data may lead to better “expert”. Machine beats human with a wide margin (68% v.s. 65% accuracy rate)?
Now if you move your eyes to the last column of the prediction image, you will notice the word “Pick’em“. It is the average of all predictions by NFL fans who submitted their picks on ESPN.com before the game. A kind of “crowd prediction” by non-experts.
Like Accuscore, Pick’em prediction scored 10-6 in the first week of 2011-2012 NFL season, no more, no less.
However, at the end of the 17 weeks of regular season games, Pick’em tied with accusore with 174 right picks (68%accuracy). Isn’t it amazing? The next table contains the performance of the best four experts, Accuscore and Pick’em.
Aha! Wisdom of the crowd kicks in nicely. A classic example of this phenomenon is often mentioned as:
At a 1906 country fair in Plymouth, eight hundred people participated in a contest to estimate the weight of a slaughtered and dressed ox. Statistician Francis Galton observed that the mean of all eight hundred guesses, at 1197 pounds, was closer than any of the individual guesses to the true weight of 1198 pounds.
How good does the the mind of that crowd do? There were 500 estimates, and the results were:
-The lowest guess was 308 lbs.
-The highest was more than 8000 pounds.
-The average was 1792 pounds.
And the real weight? The ox weighs 1795 pounds. Three pounds off.
People rock! (How about random forests and boosting, if you know what I’m talking about.)