The NCAA men’s basketball tournament – commonly known as March Madness – begins March 14. Millions of college basketball enthusiasts will be filling out their brackets, hoping to predict the outcome of all the games. Computer science professor Sheldon H. Jacobson discusses patterns he’s observed in the tournament and how data analytics can provide guidance for fans filling out brackets.
Working with University of Illinois students, he also oversees the BracketOdds website, a STEM learning lab for his students, to provide fans with useful tips in filling out their brackets.
How can data analytics provide insight into the March Madness tournament?
The tournament analytics are a figurative candy store for the analytically inclined. After the First Four, the outcome of the remaining 63 games can be modeled based on data analytics. In fact, you can represent every possible bracket as a unique string of 63 zeros or ones. That reduces the tournament to looking for patterns in how these zeros and ones occur, which translates into patterns in how seeds advance in the tournament.
Does this make predicting upsets easy?
The patterns provide guidance on the distribution of upsets in the different rounds, but say little about which pairings will result in an upset. For example, out of 132 games since 1985, a No. 3-seeded team has lost to a No. 14 seed 21 times in the Round of 64. The chance of a No. 14-seeded team pulling such an upset is around 16 percent, which is quite small, yet it does happen around two times every three tournaments. That is how the laws of chance express themselves in the tournament. Very extreme upsets are rare events but occur with regularity, given a sufficient number of opportunities. We are still waiting for a No. 16-seeded team to beat a No. 1 seed in the round of 64. It will happen eventually, perhaps even this year; we just do not know exactly when.
How useful are the seeds when filling out a bracket? Upsets happen, as you’ve said, but isn’t it still the safest choice to always pick the higher seed to win?
Picking the higher seed, called Pick Favorite, in the Round of 64 through the Sweet 16 is a reasonable strategy to get a good bracket, but not a great bracket. Since 1985, using the ESPN bracket-scoring system, Pick Favorite would have given you an average score of 874 out of a possible 1920, with a high of 1620 in 2008 and a low of 550 in 1990. Upsets bust many brackets. For example, in six of the past seven years since 2010, one or more teams seeded No. 14 or No. 15 won a game in the Round of 64.
So every bracket should have an upset or two. Any recommendations for picking upsets in the Round of 64?
Upsets are exceedingly difficult to predict with any great consistency. The best way to minimize wrong upset picks in your bracket is to have as few as possible. Pick all four No. 1 seeds and all four No. 2 seeds to advance. Pick zero or one total No. 3 or No. 4 seeds to be upset. Pick one, two or three total No. 5, No. 6 or No. 7 seeds to be upset. The No. 8-No. 9 game is a coin toss.
How about power conferences – Big 10, Pac 12, Big 12, SEC, ACC, Big East, and AA – versus mid-majors? How do they do in the tournament?
Power conferences have enjoyed a significant advantage in recent years. Since 2006, there is a noticeable trend in power conference teams getting a larger share of the at- large bids. For example, in 2016, 91 percent of the at-large bids went to power conference teams, compared with 83 percent in 2015, 81 percent in 2014, and 70 percent in both 2013 and 2012. Perhaps 2017 will break that trend. In addition, power conference teams perform exceedingly well in the tournament. Since 2006, 86 percent of the Elite Eight and Final Four teams have come from power conferences. Moreover, all the national champions since 2006 have been power conference teams.
Warren Buffet is a huge college basketball fan and is known for issuing bracket challenges. This year, he is offering $1 million per year for life to any Berkshire Hathaway employee who correctly picks the Sweet Sixteen. How likely is he to pay out?
More likely than one would first think. If all games are tossups, then there are 4,294,967,295 possible Sweet Sixteens. However, some combinations are more likely than others. Optimistically, using historical performance data, the odds are closer to 182,000-to-one. There are 367,000 eligible Berkshire Hathaway employees, so if one-third of them participate, I would estimate that the chances are around 10 to 20 percent that Buffet will be writing some checks.
What is the best strategy for filling out a bracket?
The best suggestion for people who want to fill out a single bracket by hand is to pick which seed you want to win the national championship, and then move backward with the seeds you want to appear in each round. When the teams are announced on Selection Sunday, fill in the teams based on the seeds you chose, limiting the number of upsets in the Round of 64. History suggests that you will be no worse – nor better – than trying to do it based on knowledge of the teams. If you need help, our BracketOdds website has a bracket generator so that people unfamiliar with the tournament can have a bracket created for them that contains the right mix of upsets and favorites advancing.
Contact Sheldon H. Jacobson, Department of Computer Science, University of Illinois at Urbana-Champaign, 217/244-7275, email@example.com.
Writer: Liz Ahlberg Touchstone, biomedical sciences editor, University of Illinois News Bureau, 217/244-1073, firstname.lastname@example.org.
Photo: L. Brian Stauffer.