Thursday, November 19, 2015

Introducing G-Score

Here's the way T-Rank works, basically:

  • For every game, T-Rank calculates an adjusted offensive and defensive efficiency for each team. 
  • Then, it averages those numbers to come up with overall offensive and defensive efficiency for each team. 
  • Then it uses those numbers to calculate a Pythagorean expectancy, which I jokingly call the Barthag. 

That's the T-Rank.

But if you go back and look at the first step, you can see it's actually possible to calculate a Barthag for each team for each individual game, since adjusted offensive and defensive efficiencies are calculated game-by-game.

Because I can, I did. I'll call that game-by-game rating the G-Score. It goes from 0 (bad) to 100 (perfect). The average is 50. It will appear on the schedule portion of each team's T-Page. For example:

That number -- e.g., the 95 for the Siena game (let's focus on the positive) -- is the team's individual game Barthag multiplied by 100. So Wisconsin's performance against Siena is what you'd expect from a team with a Barthag of .9500. (Like Kentucky!) On the other hand, the performance against Western is what you'd expect from a team with a Barthag of .2300. Like Howard. :(

At this early stage of the season, the preseason ratings are still heavily influencing the T-Rank. A nice thing about G-Score, then, is it indicates how the teams are playing if you leave that influence partially** aside. For example, the Badgers' average G-Score is 58, so their overall performance is akin to #139 ranked UC Santa Barbara. A loss to Western Illinois will do that.

Some other nice things about G-Score is that it will allow us to see trends in a team's performance over time, and also to get an idea about how consistent a team is. Once more data is in, we could even calculate fancy standard deviations and such.

G-Scores are not set in stone. They will fluctuate as we learn more information about the teams, and as teams improve or get worse. So, for example, if it turns out that Western Illinois is actually the class of the Summit league, that loss will start to look better and the G-Score will rise.

Finally, a caveat. Individual basketball games are small-number events. Crazy stuff happens over a single 40 minute period, and we shouldn't make too much of any single game, or any single G-Score. For example, Wisconsin's G-Score of 56 against North Dakota is a bit deceptive since the scrubs lost by 11 in their two minutes of play. If the scrubs had maintained the 25-point lead they inherited, the G-Score would probably be more like 85 or 90, which is a pretty big difference. So that's something to keep in mind.

*Specifically, this would leave the preseason influence aside only for the team whose G-Scores you're looking at, while assuming the T-Rank is correct for all their opponents.

No comments:

Post a Comment