Using Microsoft’s TrueSkill to Rank Texas Six-Man Football Teams

People have always compared six-man football to a computer game. With its wild scores and fast-paced style, six-man football can be just as exciting, if not more so.

So what if you played the entire six-man season on your Xbox?

That’s why I decided to use the TrueSkill Ranking System created by Microsoft Research for Xbox Live games and apply it to the season.

Microsoft uses this algorithm to track the skill of gamers in order to place them in competitive matches. On the Microsoft Research website it explains that the ranking system is characterized by two attributes: a players average skill level (ranking) and the degree of uncertainty in the gamer’s skill (this would be the variation in the players level). If you would like to read more, it can be found at the Microsoft Research website, http://research.microsoft.com/en-us/projects/trueskill/

Heck, gamers love this thing. If you want to really know more about the math behind it, a guy named Jeff Moser has written a bunch on the subject.

Very nerdy links—→
Here http://www.moserware.com/assets/computing-your-skill/The%20Math%20Behind%20TrueSkill.pdf

And here
http://www.moserware.com/2010/03/computing-your-skill.html

As with any algorithm, there are several ways you could implement it. I decided in the first trial I would start of with no priors. This means all teams started the season with the base ranking. Since I also program primarily in Python, I decided to use the Trueskill package developed by a South Korean game developer named Heungsub Lee. It seems to be well-written and moderately documented.

Let’s take a quick look at the top-10.

Not bad, but not stellar. Let’s see the next 20

OK, but again, you can see some flaws.

One of the really cool things TrueSkill brings is a pretty good ability to predict the winners of games. In the version I ran, I allowed TrueSkill to predict all of the games, but throwing out the really close ones where it determined the probabilities of a team were between 47%-53%. One of the reasons to do this would be that during all of week one games, with probabilities at 50-50, we can just throw those out.

For the season, TrueSkill predicted 832 of 1108 (75.1%) games, which isn’t bad. For the playoffs, it is even better at 82-20 (80.4%).

It will be interesting to see how it does this week.

Matchup: Zephyr (11-2-0) at Abbott (11-2-0)
Rating Before: 370~28 / 369~29
Match Quality: 0.82
Zephyr odds winning: 50.9%
Prediction: TOSS-UP –> Zephyr

Matchup: Crowell (9-3-0) at Borden County (13-0-0)
Rating Before: 381~28 / 425~33
Match Quality: 0.68
Crowell odds winning: 27.6%
Prediction: Borden County

Matchup: Follett (12-1-0) at Buena Vista (13-0-0)
Rating Before: 370~31 / 409~37
Match Quality: 0.68
Follett odds winning: 30.2%
Prediction: Buena Vista

Matchup: Midland Trinity (11-2-0) at Houston Emery-Weiner (11-2-0)
Rating Before: 377~28 / 381~28
Match Quality: 0.83
Midland Trinity odds winning: 47.4% [predict: TOSS-UP]
Prediction: TOSS-UP –> Houston Emery-Weiner

Matchup: Guthrie (11-2-0) at Richland Springs (13-0-0)
Rating Before: 351~29 / 432~35
Match Quality: 0.44
Guthrie odds winning: 13.8%
Prediction: Richland Springs

Matchup: San Marcos Hill Country Christian (11-1-0) at WF Notre Dame (9-2-0)
Rating Before: 366~35 / 341~30
Match Quality: 0.75
San Marcos Hill Country Christian odds winning: 63.3%
Prediction: San Marcos Hill Country Christian (this one was played this morning-correct)

Matchup: Austin Veritas (12-0-0) at Waco Live Oak (10-2-0)
Rating Before: 408~37 / 339~32
Match Quality: 0.51
Austin Veritas odds winning: 81.7%
Prediction: Austin Veritas (this one just finished-correct)

I plan on making some adjustments to this and seeing what comes up. Stay tuned.

Just for fun, here’s the entire list