Presumably because Bayeselo has some advantages over Elo- https://www.remi-coulom.fr/Bayesian-Elo/

Edited 5/19/2016 15:52:31

Beating someone and then if their rating drops yours does also is so stupid.

1. You played a 2000 rated and he drops to 1750, You beat him at a peak of his skill, and you still beat a 2000 rated player.

2. Abuse potential. Season 14 was the shakiest example of this. Gnuffone played MoD like 5 times total (or maybe red did, and then dropped out to help Gnuffone). MoD also used two accounts.

3. It adds complexity for no reason, I'm sure someone can argue for it, but it's complex for no reason.

1. You played a 2000 rated and he drops to 1750, You beat him at a peak of his skill, and you still beat a 2000 rated player.

Bayeselo assumes that people won't intentionally tank their ratings. And it's a Bayesian approach, which means that it has a "guess" distribution for your actual skill and updates it after every result. So it first thought that your opponent was at a 2000-rating level of skill, but after more games Bayeselo realized it had initially overrated them and looks like they really play at a 1750-level. And then it realized you'd actually played someone who's more like a 1750 than a 2000, so it updates its model for you as well.

According to Bayeselo, your opponent didn't really decline in skill- the model's estimate just got better.

3. It adds complexity for no reason, I'm sure someone can argue for it, but it's complex for no reason.

It's not very complicated- just a simple Bayesian approach to Elo...

Edited 5/21/2016 17:48:38

There's two major issues with Bayeselo:

a) It is complicated. People are confused when, for example, they win a game and lose rating because a past opponent simultaneously lost. I completely understand the system and I think it could have its uses and advantages in other places (pro tennis, pro golf, pro chess, basically anything where people aren't anonymous). However, that's only because I have an interest in various rating systems and because I took the effort to learn it.

b) It is easily exploitable. We've seen examples in this thread of 'team play' where one player on a 'team' tanks their own rating after achieving a victory against a shared enemy and just blatant 'multiplaying' where the two players on the team are really one player.

Both of these problems are eliminating by switching to a simple Elo or Glicko system. They systems are less complicated, and more to the point people are much more familiar with how these work from other places (NCAA football, chess, most online video games). They are also much more difficult to exploit since the 1200 rated player won't be able to get a match against the 2200 player.