Friday, December 2, 2011

About That Florida-Syracuse Projection

Astute readers will notice that the projection for tonight's Florida-Syracuse game to be played at the Carrier Dome has Florida as a six point favorite. That seems weird, right? I mean, the Vegas line is the reverse of that, with Syracuse a 6.5-point favorite as of this writing, and my own future games report has the Orange as almost a 5-point favorite. So what gives?

My first thought, of course, was that I'd screwed up somewhere. Maybe I'd called Florida the home team, or had reversed home-court advantage in my calculations somewhere. I went back and looked at the code, and everything was right. So I double-checked by running a projection of the teams around Florida in the rankings to see if they'd also be favored at the Carrier Dome. But #8 Marquette would be a 5.5-point underdog according to the model and the Cuse would be favored by 10.5 over #10 North Carolina. So there must be something unique about the Florida matchup that will give Syracuse problems.

The future games report, for the next two days at least, is still based on a simpler method for predicting games that uses only the teams' adjusted offensive and defensive efficiencies, adjusted pace of play, and a home-court advantage constant as its inputs. This is because the preseason TAPE ratings only included this information, and the current TAPE ratings still include at least part of those preseason ratings through this weekend. This much less sophisticated projection model allowed for season projections from the preseason all the way up to now to be at least reasonably valid. It's done pretty well, too. Through yesterday's games, it's been correct on 712 of the 919 games projected based on that method, and that 77.5% success rate is slightly better than the 76.4% predicted success rate.

As more games have been played, though, the current season's data is coming into focus. For any game in which both teams have played at least five games against D-I competition, the projection in the Today's Games post has been based solely on the current season's data. (You can tell which games are which by hovering over the score predictions for each team; if the popup text has projected Four Factors numbers, it's a current-season-only prediction.) Those projections are based on a true projection model, which includes much more granular data than simply offensive and defensive efficiencies. It takes each facet of the game--shooting, rebounding, turnovers, etc.--to model the expected outcomes of each possession, which then gets aggregated into a game projection which includes a full line score. The projected line score for Florida-Syracuse looks like this:

70.9 possessions
(9) Florida
67.1% 86.4 29.8 65.7 16.1 34.9 10.7 17.5 18.0 40.2 16.9 19.0 14.3 9.8 1.6
(7) Syracuse
32.9% 80.2 31.1 63.5 7.7 20.2 10.3 16.4 13.5 35.0 17.5 16.1 13.0 10.6 7.3

There are a couple important things to watch for in this game that a simple possession analysis would not have picked up on that makes this a favorable matchup for Florida. First, check out the three-point shooting. Florida has been the fifth-best team in the nation shooting threes so far this season, while Syracuse's matchup zone defense concedes more three point attempts than most defenses do (while allowing a success rate right in line with the national average). That's where the almost absurd prediction of over 16 made threes comes from. The second area of concern for Syracuse will be their defensive glass. Florida is a very good offensive rebounding team, and the model predicts that they should be able to board 45.5% of their misses against Syracuse's zone.

That level of granularity is what makes the TAPE system different from any others that I'm aware of. Starting on Monday, the Today's Games report will include projected line scores like the one above for every game.

No comments: