I’m sure this exists somewhere - I’ve just not seen it. Is there somewhere available to users a discussion of the rationale behind the rating algorithm that OGS uses? I’ve always been puzzled that one could be consistently ranked, say, at 12k, over hundreds of games, and then play another well-established 12k player, lose twice, and drop an entire stone because of two consecutive losses (which makes zero sense to me, since two consecutive losses to an evenly ranked player indicates just about nothing about relative strength). Is there a reason why ranks don’t seem to have “staying power”?
This should help: EGF Rating on Sensei’s Library. (I’ve read somewhere on the forum that OGS rating is heavily EGF inspired.)
The thing is that you have exactly 50 % chance of winning playing the opponent of the same rating (not rank, rank is just for fun, the underlying rating makes all the work).
(Oversimplification follows!) So, if you lose and then win with the same opponent, your rating should stay the same, because of these 50 %. If you lose twice, you lose rating points: You’ve won 0 % of games, so you are worse than the opponent (who’ve started with the same rating as you had). This makes sense, doesn’t it?
If you lose twice and then win twice (with the same opponent, presumably), your rating should be same as was in the beginning, because of these 50 %. And so on. There’s a neat table in the link provided above.
There are other rating systems, most notably Glicko and Glicko-2, that work with more variables. If you are looking for a discussion of this type, I’m not aware that there is any.
I have a question though… so when I play an opponent who is near my rank (and I have some 700 games on the system) I don’t expect to lose half a stone for losing a single game. Is my rank not considered “solid” after so many games? I realize of course that my rank does fluctuate… but really … lose two full stones if you go on a 4 game losing streak? A 4-game losing streak against an even opponent should not be outrageously rare… in fact the likelihood of having four consecutive losses against an even opponent is about 8% which means such streaks should happen with perfectly even opponents.
Should there not be some kind of weighting applied for the number of games at rank? Perhaps there is, but maybe it’s too weak?
No, it isn’t. There is no such variable as “solidity” in ELO rating system. (Well, actually, there is something like that, but not in the sense you want it to be. There is a parameter that describes a rating stability. Roughly, the higher your rating is the more stable it is considered.)
I do not know the answer to this one. The used (ELO) rating system does one thing: It predicts your chance of winning against an opponent (of some rating of the same rating system).
I’m not sure how (and why) it should reflect that someone has played 100s of games and did not exceed the rating range of say 800—899. When there is a momentum in play, one could stay in said interval (which equals 13k nominally) even if one is on a winning streak against someone of rating 1300—1399 (8k), which could be pretty disappointing (for both the solid 13k player who could cry misrank and the 8k player who could cry sandbagger).
The other thing that comes to mind may be obscured by the rank/rating distinction. In fact, you do not have 50 % chance of winning against the opponent of the same rank. You can have 30—70 % chance of winning (for DDKs), because the rating range of the rank is 100 rating points.
EDIT:
I do not fully understand this, but I did some math (yay!) and found out that there’s 6.25 % chance of losing to opponents with exactly the same rating (i. e. 50 % every time), and 15 % chance of losing to the same opponent (starting with the same rating and adjusting) 4 times in a row.
This thread is outdated since Jul 2017. At this point the OGS rating system changed to glicko2. The rating system changed in many aspects. All info in this thread about Elo on OGS is obsolete.