The college basketball game predictions and tourney win probabilities come from rankings that have 3 components. These ensemble numbers informed my bracket advice for members of The Power Rank.
1. Team Rankings
The Power Rank started with team rankings that take margin of victory in games and adjusts for strength of schedule with an algorithm I developed. The first college basketball rankings appeared for the 2010 tournament.
With a small sample size of 30 some games, these rankings aren’t perfect. The rating for each team, which gives an expected margin of victory against an average D1 team, has uncertainty associated with it.
These uncertainties seem larger for a few of the primary contenders this year. Based on these numbers, top ranked Gonzaga had a 31.5% chance to win the tourney before it started. As much as I like that team, this win probability is too high.
Now I combine these team rankings based on points with other rankings.
2. Market Rankings
These rankings come from taking closing point spreads in the markets and adjusting for strength of schedule with the same algorithm. I wrote these rankings before the start of the tournament.
While the markets consider analytics, these numbers also have a subjective component based the opinions of those willing to wager money on games. This led to a higher rank for North Carolina in the market rankings than the team rankings.
3. The Preseason AP Poll
This might surprise you, but the preseason AP poll is a powerful predictor of postseason performance. Nate Silver found the higher ranked team won 72% of tournament games from 2003 through 2010.
The wisdom of crowds gives this poll its predictive power. No one pollster has a perfect ranking, but the collective wisdom of many sports writers is a powerful assessment of team talent.
Once the season starts, the pollsters react to wins and losses in predictable ways, e.g. drop a team when they lose, even if its a close game against a better team. The predictive power of the poll decreases as the season moves along.
I’ve found the same for preseason college football polls.
Michigan versus Oregon
The team rankings favor Oregon by 0.8 points over Michigan. Over the course of the season, adjusted margin of victory views Oregon as the slightly better team.
The markets more strongly favor Oregon, as they make the Ducks a 4.3 point favorite in this games. These rankings weight recent games more, which should help both teams.
The preseason polls also strongly favor Oregon, as the Ducks ranked 5th compared with 42nd for Michigan. Back in November, most viewed Oregon as the more talented team.
The ensemble of these 3 predictions makes Oregon a 3.5 point favorite over Michigan.
The injury to Oregon forward Chris Boucher, one of the nation’s best shot blockers, makes predicting this game more difficult. However, this factor doesn’t explain how the market favors Michigan by 1.5.
It seems like the markets have considered the recent form of Michigan quite heavily.
Ensemble rankings
The rating next to each team gives a predicted margin of victory against an average D1 team.
1. Villanova, 19.0
2. Gonzaga, 18.9
3. North Carolina, 18.6
4. Kentucky, 18.4
5. Kansas, 17.8
6. Duke, 17.4
7. West Virginia, 16.7
8. Oregon, 16.4
9. UCLA, 16.0
10. Louisville, 15.9
11. Baylor, 15.7
12. Florida, 15.5
13. Virginia, 15.4
14. SMU, 15.4
15. Arizona, 15.3
16. Wichita State, 14.9
17. Wisconsin, 14.8
18. Purdue, 14.7
19. Florida State, 14.2
20. Cincinnati, 13.6
21. Iowa State, 13.5
22. Butler, 13.3
23. Michigan, 12.9
24. St. Mary’s, 12.7
25. Oklahoma State, 12.6
I’ll be posting these ensemble predictions for the remainder of the 2017 tournament.
It isn’t just the “form” of Michigan which sways the market. It is their “blessed runway story” which captures the imagination of the “gaming public,” making them think of Michigan as a team of destiny, at least until they aren’t. (You and I both know Vegas markets are a function of their betting equilibrium, not that dictated by the mathematics!) Even if Oregon is the better play, do you want to be the one tempting the “Destiny Gods” for what I have as a 50.5% chance to be right? Probably not me.