Version 1.3 has been implemented as of September 12, 2022. I’ve changed how the influence of each result decays with time. Results now decay at a constant rate each day. The previous method did not do this so a team’s rating could kind of randomly fluctuate when they weren’t playing for reasons that were difficult to explain. That won’t happen as much now. A team’s rating can still change while they aren’t playing, but that will be mostly do to the influence of their opponents’ performances that week.
The drawback to this approach (I guess) is that the most recent week carries more weight than it used to so a team can make a bigger move with a good or bad event. I haven’t had the time to do enough testing to know if this helps or hurts the predictive ability of the system. In the end, it probably doesn’t make much difference.
The other thing I’ve added is a measure of dominance in wins. It’s pretty simple for now. Basically, if a game is close going into the last end, the winning team gets credit for 80% of a win (and the losing team gets 20%). Otherwise, teams get full credit for a win or a loss. How do I define “close”? If you’re up 2 or less without hammer or up 1 or less with hammer going into the last end.
It doesn’t change the ratings much, but since I coded it up, I might as well include it. Maybe the resulting stats on close games will be interesting once I find a way to display them.
Version 1.2 has been implemented as of November 1, 2021. The main update is an improvement on how to regress the ratings of teams that have played a few games. In the Bradley-Terry algorithm, we adjust the team’s winning percentage by adding two wins and two losses to its record.
This seems to handle brand new teams a bit better that the method used in v. 1.1. It especially improves the rating for teams that have a poor winning percentage in their first event or two. In v. 1.1, those teams were given unrealistically poor ratings.
I have also changed the minimum games threshold so that teams that have not played games in the current season are more quickly removed from the ratings. All inactive/retired teams are now removed after January 1st.
If you want the basic nuts and bolts of my ratings philosophy and some of the design choices in v. 1.0, along with the differences between my ratings and the World Curling Tour’s rankings, go here.
A description of updates made in version 1.1 can be found here.