See slides here
Since 2006, ESPN the Magazine and ESPN Insider’s “Giant Killers” project has used statistical analysis to forecast big upsets in the NCAA college basketball tournament. We define a Giant Killer as any team that beats an NCAA tournament opponent seeded at least five spots higher in any round. Our statistical model isolates Giant Killing opportunities due to mis-seedings; more important, it indicates styles of play that help generate (or ward off) upsets. In general, we find that underdogs should employ high-risk/high-reward strategies, while Giants must minimize chances for their opponents to score. We have refined our model as we have been able to add more data, and have predicted success for a series of low-seeded, unheralded teams, such as Cleveland State in 2009, Old Dominion in 2010, VCU in 2011, Ohio in 2012 and Oregon last year.