March 2013 marks the 75th anniversary of the NCAA March Madness. It also signals the start of countless office guessing games, online pools, and a nationwide fan-fueled frenzy for all the latest tips and tools to build the best brackets.
Generally, people tend to base their picks on a wide array of (unreliable) variables, such as fan favorites, ill-informed “tips,” and other junk science (i.e. “always pick at least one team with a blue uniform to make it to the Final Four” or “Diaper dandies are beatable”). Building a bracket is easy, but creating a good one takes data science.
Let’s look at the stats. There are 147,573,952,589,676,412,928 different ways to fill out a bracket. That’s 68 possible teams, 64 that end up playing with 63 teams each with a possibility of losing once—or 2 to the 67th power.
That’s quite a small chance for building the perfect bracket. According to Jeff Bergen, a professor of mathematics at DePaul University, if you know a little bit about basketball you can improve your odds to roughly one out of 128 billion. While this may still seem pretty bad, there are a number of data scientists looking at ways to help you craft a winning bracket. Continue Reading…