I’ve got a bot tending a google sheet containing team ELO ratings for the season, as well at ratings from previous seasons. Teams like things like this for scouting, so I thought I’d share it. Here’s a link to the shared folder with all the sheets, including an archived one for today (Apr 15, 2019) which I added in case something goes wrong with the live sheet: https://drive.google.com/drive/folders/1zNNNHjfZuxf3VJYvG6fIxQnEQOIR9Gdu
The ratings generally are pretty accurate at predicting the success of teams at the world championship. If you look at the historical data, one of the top 2 teams on my ratings wins the high school World Championship almost every year, with the exception of one or two years, where the winning team was in the top 4 I believe.
Also, the specific high school/middle school ratings aren’t necessarily accurate, since vexDB can’t properly classify all middle school teams. So if you’re a middle school team improperly classified as a high school team, I’m aware of this.
You can ignore the rest of this post if you aren’t interested in how it works or the parameters I’m using for the rating calculations. If you are interested but don’t know how ELO works, then I suggest doing a quick google search before reading the rest of the post. I’m glad to answer questions as well.
These ratings are using a pretty standard ELO system with a few provisional matches, as well as a proportional weighting system for alliance partners. That is, if you have more points than your partner, you will earn/lose more points than them, as a reward for “carrying them”. For example, team A has 1000 points and team B has 500, and together they earn 60 from winning a match. Team A has 1000/1500 of the alliance’s total ELO score, so they would receive 1000/1500 (2/3) of those 60 points, meaning they get 40 points added to their rating, and team B would only get 20 points added to their rating. Their opponents would lose points in the same manner, as a ratio of the sum of their points.
I’m using a k-factor of 64, which is divided between the two teams on an alliance (equally matched partners would each effectively have a k-factor of 32, whereas a team carrying another team would have a higher effective k-factor due to the system described above). Things like DQs and no-shows are not currently factored in to the system.
Let me know what you all think, and happy scouting.