Plackett–Luce sigma decreasing very slowly #1
Description
I was previously using TrueSkill to rate player performance for a FFA boardgame. It's typically played with 6 players and the end game has a single winner. The rest of the players can't accurately be ranked in order, so I am marking them as ties between "close to winning" and everyone else (e.g. ranks of 1, 2, 2, 2, 3, 3).
I decided to give OpenSkill a shot and am finding that sigma is decreasing much slower. With TrueSkill, after 40 or so games I'd see sigmas of ~1, from a start of 8.3334. With OpenSkill, 40 games are getting people down to ~7.
As an aside, there's also a much wider range of μs, despite providing the same starting values as I did to TrueSkill. With TrueSkill, the highest ratings I saw were in the lows 30s, from a start of 25. With OpenSkill, some players have rocketed up to 50+. That may just be apples to oranges comparison there, so I'm not too worried, as it's generally the same players at those ranks, but it was curious because I expected the two to be pretty analogous down to the final ratings.
All my settings are default, using PlackettLuce, with a starting sigma of 25/3 and μ of 25.
BradleyTerryFull on the other hand gives me a very different leaderboard than TrueSkill did, and I don't find it to be accurate.