Summer is ending and curling is back. And hey, it looks like curling season will be closer to normal. I think it’s wishful thinking to expect total normalcy. On the one hand, last week’s Baden Masters returned to a traditional 20-team field. On the other hand, none of the Scottish teams, including world #1 Bruce Mouat, were able to compete due to COVID-related travel complications. We hope for the best but I suspect is a phrase that will occasionally come into play this season.
But fortunately the curling community seems to be embracing vaccines more than most sports. And the schedule for the slams and worlds are set. Canada has kind of opened up for foreign visitors, so the prospect of cross-continental matchups is much greater than it was at this time last season.
We’ll be back to regular posting shortly. In the meantime, here’s what we spent the summer working on:
Ratings v. 1.1: This will require a separate page with the gory details, but the ratings methodology has been updated to improve a few things. Maybe most importantly is handling teams with just a few games in the books.
This has allowed us to reduce the minimum number of games to qualify for the ratings (basically just seven in the last 12 months). Admittedly, this still needs some work. I mean, congrats to Daniel Casper’s team on rolling through the US Junior Championships without a loss earlier this month, but his team debuts at 43rd after that performance and I’m going to say his team isn’t the 43rd best in the world. I mean, their performance was promising, so probably someday!
Fortunately, they play in the US Open Contender event in Minnesota this weekend, so we may get a quick correction as they battle some of the best grown-up teams in the US. But I get annoyed when an algorithm is imperfect and the owner says “just wait for more data”. With more data, everything is easier. However, the point is to do the best with the data you have and we could do a little better here and will continue working on it. Still, it’s more important to get as many serious teams rated as possible, so we’ll err on the side of rating new teams inaccurately rather than not rating them at all.
True win probability: In order to estimate win probability for a particular game state (margin/end/hammer) we can look at past cases of that state and see how often a team won. However, even with thousands of games each year, some game states don’t have enough cases to get a reliable estimate, especially if you want to stratify cases by skill level (which you absolutely should!). So I’ve simulated a bunch of games to get a better estimate of win probability in all cases.
I arrogantly call it “true win probability” although we can never know the true win probability in a given situation. Still, these numbers are closer to reality than what you’d get from just looking at raw cases. If you’re vying for a gold medal, or just a serious curler going to a few cashspiels every year, you should know these numbers. I’ll detail the methodology further in a separate entry. For now, you can go here to see the values for every game state and how a score in that situation affects one’s chance of winning.
It looks great on a mobile device, so don’t be afraid to bring your phone on the ice for handy reference!
Shot data acquisition: I think I have grabbed all of the shot data that is available somewhere on the web, dark or otherwise. That’s like 400,000 shots! I have no further comment except to say this will provide a whole bunch of interesting content in the coming weeks.
Those are the biggies. There were also some other technical achievements that were more time consuming than any of these things and those efforts are the main reasons I am a little late getting started. But the details on that are not very interesting for a general audience.
Anyway, I still need to clean up some stuff around here, but I’ll be back to regular posting shortly. We appreciate your patience.
