March Madness Brings Out the Analytics
The luck of the Irish is projected to be strong today, which is St. Patrick’s Day. But luck can only carry you so far in the NCAA Tournament, the first round of which also begins today. To float among the basketball gods with a winning bracket, intelligent fans are turning to AI and analytics. Luckily for you, analytics tools for March Madness abound on the Net this year.
One great thing about the NCAA Tournament is the ridiculously steep odds of having a perfect bracket–that is, correctly calling the winner of 63 straight games. It is so difficult that it has never been done since in the tournament was expanded to 64 teams in 1985.
According to the NCAA, the record for the most verifiable wins in a bracket was 49, which was set in 2019 when an Ohio man correctly predicted the entire 2019 NCAA tournament going into the Sweet 16. But it all came apart for him in the regional semi-finals, when Purdue beat Tennessee in double overtime.
The odds of getting a perfect bracket by randomly picking winners is one in 9,223,372,036,854,775,808 (that’s 9 quintillion and change), according to DePaul University mathematics professor Jeff Bergen, who posted a YouTube video with the math 10 years ago. If you factor in the First Four games–which are play-in games for the 64-team tournament that were played Tuesday and Wednesday this week–the odds shoot up to one in 147 quintillion.
However, you can bring those long odds down considerably by doing one simple thing: considering the seeds, which reflect official oddsmakers’ analysis of the quality of the teams. “If you know something about basketball,” professor Bergen said, “then your chance of getting a perfect bracket is one in 128 billion”
The chances of a 16th seed beating a number one seed are exceptionally low. In fact, it had never happened in the men’s NCAA tournament until 2018, when the 16th seeded University of Maryland Baltimore County Retrievers soundly trounced the top seed, Virginia, by 20 points. Lightning struck again in 2021, when the University of Illinois was badly upset by Loyola-Chicago, pleasing Sister Jean but taking the fight out of the Fighting Illini.
But one in 128 billion are still pretty long odds. In fact, you have a much better chance of being struck by lightning this year (one in 500,000). That’s why billionaire Warren Buffet felt comfortable offering anyone in the world a billion dollars to anyone could fill out a perfect bracket back in 2014. Buffet, who runs the competition every year inside Berkshire Hathaway, kept his money that year, and likely will never pay out on that bet, since a perfect bracket is practically an impossibility.
But as the old saying goes, “Perfect is the enemy of the good.” Whether you are just playing for fun or are wagering on the tournament (Forbes says Americans will bet more than $3 billion on the tournament this year), the NCAA Tournament presents fertile ground for testing one’s analytic mettle.
One group that’s emerged over the past few years is Bracketodds. Sheldon Jacobson, a computer science professor at the University of Illinois started his investigation into March Madness as a class project back in 2006. But over the years, it has grown into a project at the university’s STEM Learning Laboratory. And in 2011, Jacobson launched the Bracketodds website, which includes all the win/loss data in the NCAA Tournament going back to 1985.
“The website is basically an opportunity for sports fans like you to have some fun with the analytics and see how it works,” Jacobson tells WCIA.com in Champaign, Illinois. “People want to know, what’s the secret sauce? What is analytics when it comes to picking a bracket? And we try to uncover that and shed some light on how advanced analytics can help people put together their brackets in an informed way.”
The secret, of course, is selecting the upsets. The NCAA selection committee has already given us a lot of information with the seeding, which is based on various factors, such as teams that win their conference championships, which gives them an automatic bid. But there are 350 teams in 32 Division 1 men’s college basketball conferences, which means things get a little more dicey once you move away from the top 25 rankings (via AP, the coach’s poll, NET rankings, etc.).
Bracketodds gives readers an historical analysis of the matchups of various seeds at different stages of the tournament. You are generally safe picking teams ranked one, two, or three, since they rarely lose to teams ranked 16th, 15th, or 14th in their region, respectively. Only two number one seeds have ever lost this matchup in the 36 years of playing since 1985 (no tournament was played in 2020 due to COVID). A team seeded 15th has beat a number two team only nine times (out of 144 games, since there are four regions and hence four number 15 seeds every year). No. 14 seeds are 22-122 all-time, Bracketodds tells us, while No. 13 seeds are a little better – 31-113.
Things start to get interesting with the matchup of the number 5 and 12 seeds, where the lower-ranked team has a 51-93 record. The odds flatten out after that, with number 11 seeds (which play the sixth seeded team) having a 54-90 record while the number 10 seeds (which play the seventh seeded team) are 57-87. The game versus the eight and ninth seeds is, as you would imagine, a coinflip; the number 9 seed actually holds a 73-71 advantage, per Bracketodds.
Taking aim at the game between the fourth and 13th seeds, Jacobson points out that, in the last three tournaments, the number four seed has only won seven of those 12 games. “But the three tournaments before that, the 4s won 11 of the 12,” he told WCIA. “So is this going to be an 11 and 12 kind of year, or is it going to be 7 out of 12 kind of year? We won’t know until they take the court.”
Informed with the historical record, a bracket enthusiast can make reasonable guesses about which games may produce an upset. The odds of creating a perfect bracket, of course, are extremely thin, but who knows? Maybe this is the year.