Monday, March 07, 2011

picking winners in sports

Here is a great study for my sports fan friends, as well as anyone who is interest in the “Wisdom of Crowds.” A series of studies looked at the wisdom of “ignorant” crowds in predicting sports outcomes. Not totally ignorant, but really semi-informed. What they did was survey people who are semi-informed about a sport about what player and team names they recognized. They didn’t ask at all about who they thought would win, just if they recognized the names. Then they used that name recognition to predict who would win the games. Whichever name was recognized by more people, they predicted that team would win. For tennis, it was just the players’ names. For World Cup soccer, they looked at how many of each team’s players’ names were recognized. In four different studies, the name recognition was a better predictor of who would win than the ATP rankings (tennis) and FIFA rankings (soccer). The more difference there was between the name recognition of the two players or teams, the better the strategy worked.

I think the reason that the semi-informed crowds did better than the experts was that the experts can be biased by old anchors (who was good last season) whereas the crowds’ name recognition was more likely to be based on current information. This depends of course on the better players being more likely to be discussed in the news, rather than because of their outside scandals. It also depends on having a diverse set of people to survey so that their errors will cancel out (this idea is discussed at length in the book “Wisdom of Crowds”). It turns out that the extra information that experts have doesn’t add any value and in fact can take away from their decisions.

Think about how bookies set their spreads. They don’t look at experts’ opinions or other objective facts. They just want half of the bettors (who resemble the semi-informed survey participants) to be on each side of the bet. So they are using this same crowd, but in a different way. The study compared the two and found that the name recognition surveys were significantly better than experts, but a little worse than bookies. But it was close enough that if you use the name recognition method you can be correct in your Vegas bets more than half the time and make money in the long run. Not too bad!!!

They also speculate on how individuals can simulate the name recognition survey without doing all the work. One thing you can do is count how often the names are mentioned in the media. This measures approximately the same thing that leads to name recognition in the public. You can also look at how often the names are searched on Google (which Google publishes now).

If you are totally isolated from data, you can think about a wide variety of populations and estimate how well each of them would recognize a particular name. Then average these together. I think this one is a bit of a stretch, but it may be better than nothing if you need to make predictions. It may still outperform using expert rankings.

No comments: