The 2012 edition of the Flat Track Stats ranking system is now live.

You can read more about what's different in this version, and see the docs to catch up on how this system works.

As the new season gets under way keep an eye out on this blog for more exciting features.  There's much more to come -- we've got big plans for 2012!


Comments

Well I see from my snapshots that you guys have been making changes since this article was posted.

For example:
Emerald City was favored to beat Santa Cruz even though Santa Cruz was ranked higher.

And Rage City and Fairbanks are ranked even though neither had played a sanctioned game before.

How about an article explaining some of your most recent changes?

The algorithm has not been modified since this post. I think the minor variations you're seeing are due to some updated bout results from 2009 that propagate through.

As for Emerald-Santa Cruz, they were separated by only 1.8 ratings points going in. Emerald was listed as the "Home" team, so the home/away bias made them favored by the slimmest of margins. While I said that the home/away bias got smaller for the new algorithm, it didn't go away entirely.

Rage City and Fairbanks were both accepted as full WFTDA members (as far as we know) as of March and June of last year, respectively. So while they had not played any sanctioned bouts previously, this was a sanctioned bout. As such, we begin ranking them immediately.

Hi. I'm a skater who likes this site a lot. I have a question about how the system seems to work for games played in quick succession, such as a tournament. It seems that the current system "locks in" ranking changes immediately after each game. Yet we all know that the few days of a tournament can have ups and downs. So say, for example, that a team pulls off a better-than-expected result in their first game of a tournament and gains +25 ranking points, and that catapults them above other tournament opponents. All of a sudden they are expected to win games against those opponents, so what would have been an upset for them going into the tournament now becomes a must-win situation. To compound that, if one of their subsequent opponents had a sub-par first game, dropping their ranking number, the expectations are even more lopsided. Wouldn't it make some sense to use each team's previous week's ranking number for all games played that week, for a bit of stability within each week?

I would say that if a team pulls off a better-than-expected result, then that means FTS was either underrating them or overrating their opponent. If the former, then their next bout should be more accurate with their new value than their old. If the latter, then the next bout should correct their ratings, bringing them back to where they should have been.

I'm going to be mean and nasty, Player 1. But if your team has ups and downs during a tournament, that sounds like your team has a training issue, not a rankings issue.

FTS's model has a distribution of likely performances for any given matchup. This distribution is based off of each teams perceived skill, which is based off of past performances. Depending on your mathematical world view, that distribution can be interpreted as either a representation of the model's lack of knowledge of the team's actual skill, or as a measure of the variance in that team's own performance. In reality, it's most likely due to a combination of both factors.
The point being that FTS's model ASSUMES that all teams are capable of having ups and downs within their performance... from Gotham all the way on down. And this is regularly demonstrated in many examples.

In other words, if you have two teams play each other every day for a month, you would expect them to not always have the same DoS, but you would expect there to be variance in the DoS around some average value. So, I'm going to say ups and downs does not imply any issue.

Thanks for helping me understand this, N8. I'm obviously not a statistician, just interested in how it works (quite apart from questions of training, Southbay ;-) ).