In a few hours we'll know the 65 teams that will compete for the NCAA men's basketball championship. And of course, over the next 4 days, office workers across America will spend their lunch hours meticulously filling out their bracket. In the past, in preparation for the tournament, I collected a lot of statistics and crunched a lot of numbers; but I've never done particularly well.

(from GoonSquadSarah on Flickr)

As it turns out, my statistical models performed modestly. My bracket was incredible average - it wasn't close to winning, but it was hardly the worst in the pool. It's possible that my model itself was flawed (it was rather simplistic), but the real story is probably that there's just too much variability in college basketball to use past performance to predict future success.

In that sense, using numbers to pick winners in college basketball reminds me of a lot of the statistical models used in the social sciences. Sure, they provide insights into real events, but they're not always bulletproof, especially when it comes to making predictions. The real risk is in assuming that, just because numbers are involved, whatever the "tell us" must be the truth.

The real key to winning a March Madness pool is to think about the tournament as a game theory question, rather than a mathematical equation.