A day or so before the NCAA Tournament began, I used The Daily Cardinal's brand-new flashy win probability calculator to look at teams that would provide the best long-run return on a bet to win the championship. That is, I compared each team's odds of cutting down the nets with the odds the betting public gave them.
All told, there were only seven teams that projected to give a positive return in the long run. None of them were the top-seven deemed most likely to win the tournament by ESPN's bracket-makers (Kansas, Michigan State, UNC, Oklahoma, Kentucky, UVA and Oregon all had negative long-run expected returns).
One team that did make the cut, however, was Villanova, which of course ended up winning it all in dramatic fashion.
An important part of understanding probability is recognizing that, by definition, it's imperfect. The goal is not to correctly predict every game every time, because no numbers could have told you Middle Tennessee would beat Michigan State. What they could tell you, though, is the Blue Raiders had about a four percent chance of winning, so not as rare as you might initially suspect.
The Daily Cardinal Calculator had its ups and downs, but thankfully it had mostly ups. It finished with a 48-19 record, which is about what you'd expect based on the probabilities it predicted. The geeks over at FiveThirtyEight finished 50-17, while Ken Pomeroy led the way at 51-16.
At first glance, that might look bad. But again, probability isn't about perfection. It's about long-run consistency. And there's reason to believe that this year was just a fluke.
For starters, the DC Calculator may have been significantly less lucky than Pomeroy and FiveThirtyEight. The average win probability for a favorite by our estimate was about 72 percent. That means there's about a 60 percent chance of correctly predicting 48 of the 67 games, as it did. So the Calculator really didn't benefit from any lucky bounces.
Pomeroy, meanwhile, was extremely fortunate. He gave an average win probability of 68.7 percent, meaning there was only a 10.9 percent chance he would correctly predict 51 of 67 games. There are two explanations for this: his prediction model may need a tweaking and is giving favorites lower win probabilities than they deserve, or he really was just very, very lucky this year.
The numbers seem to point to the latter. In games that were expected to be decided by three points or fewer (about 65 percent win probability), Pomeroy correctly predicted 68.8 percent of games. That's way higher than you would expect from a group of games where the favorite had an average expected win probability of just 57.4 percent.
I looked back a season to see if Pomeroy was getting consistently lucky, and found that it really was just one good run in 2016. The Calculator finished 1.5 games ahead of Pomeroy's system in 2015 (it refused to predict the Notre Dame-Wichita State game, as the teams had identical ratings) at 50-16-1 to 49-18.
So what did our Calculator do poorly and what did it do well? For starters, our rating system really underrated Notre Dame. The Fighting Irish made it to the Elite Eight without being favored in a single game by the Calculator. But again, this is a perfect snapshot of how unlucky it was this year. Notre Dame could have easily lost against Stephen F. Austin and Wisconsin, as each game came down to wild finishes with the Irish overcoming a late deficit.
The Calculator did, however, do a really good job identifying upsets in the First Round. There were an amazing 13 seed-upsets in the first round (this tournament really was great), which tied a tournament record. But the Calculator didn't even see a lot of those games as upsets. It actually had six of those underdogs as favorites with more than a 50 percent chance to win the game.
It correctly foresaw three No. 11 seeds winning (although one of them was Michigan, not UNI), and it correctly identified the two No. 10 seeds that advanced.
While this was a very good debut year for the Daily Cardinal Calculator, there are still many kinks to work out. But there's a lot of time between now and March, and those kinks will get worked out. Keep an eye out for a new-and-improved version of the Calculator in 11 months, when we'll correctly predict all 67 games. Based on our probabilities from this year, there's a 0.0000000079 percent chance of doing that, but never say never, right?
