I’d like to preface this post with saying this project was inspired by some of Caleb Sykes’ recent postings on FRC Elo. They’re really worth checking out if you’re into statistics.
Recently, a little game called Overwatch has been consuming my life, specifically the competitive scene. I decided to try and run some simulations on FRC Matches as if they were ranked like a video-game competitive/ranked mode, and this is the result.
All teams start at 2500, and their rank is adjusted up and down depending on a few factors, including: Win/loss/tie, Win Streaks, and Winning Margin. This is done using the Elo ranking system. The maximum rating is 5000, and the minimum 0.
The tiers are the same as Overwatch, that is:
Bronze: 0 - 1499
Silver: 1500 - 1999
Gold: 2000 - 2499
Platinum: 2500 - 2999
Diamond: 3000 - 3499
Master: 3500 - 3999
Grandmaster: 4000+
The Average Joes feel your pain. We started the season feeling good whenever we cycled gears fast enough to get four rotors turning. But at MSC and Daly, 445 was our most common losing playoff score.
I wonder how many other teams had this experience? It would be interesting to see a scatter plot of Jaci’s ELO vs. FUEL OPR.
I see a flaw with this system (although you will have flaws with any other numerical ranking system). This does not take into account difficulty of schedule. Teams that were at more difficult regionals or divisions at worlds are bound to be lower ranked than others. Or, if a team is paired with a poor alliance partner in playoffs, the team will fall again. Maybe accounting for DOS and difficulty of regional would help. Although this gives a general direction of where a team lies, we still search for the perfect formula.
Another issue I’ve always had with Elo is that match order matters. If you scramble up the order matches, you’ll get different results, which I’ve never been a fan of.
The Elo system, for games based on location, is intended to rank teams against each other in that specific region (with a bit of bleed accounting for championships). You see this in how the system works for FRC, and also for video games as well (Australia is paired with the Americas region for overwatch, but server routing puts our regions in two very different ballparks)
That being said, the Elo system isn’t perfect, and isn’t entirely well suited to FRC for obvious reasons. Limiting the elo to an event-per-event basis would be a better representation, but the amount of games played and sample size doesn’t make it viable.