I play much better on 19x19 Go than on 9x9. I'm sure that a KGS 2k who has studied 9x9 theory a little could beat me easily. On 19x19 a KGS 2k would have very little chance.
Strangely, different size Hex games do have different ratings.
This has been mentioned many times. There are many reasons why it is a bad idea, including:
a) Cheating: Bots are near-pro in 9x9 while 1-2 dan at most in 19x19. Probably less with slow time settings. I am similar strength in KGS and lose to bots because of mistakes on fast games. In a slow game, I feel horrified by the blunders of the bot and would never lose a game paying enough attention to it. The result is: The best rank in LG is sm9 who has made public he is using software and plays only 9x9.
b) Different games have different depth in terms of standard deviation. Go has near 40 levels, chess has about 13, tic-tac-toe maybe 2. There is no way one can make these Elo ratings of one 40-level game map the ratings of an 8-level game, since Elo rating defines a probability of winning. It just doesn't work. The intention was probably trying to stabilize the rating system with more games, but the result is just the opposite.
c) Players may prefer one game to the other and may have different skill level.
I agree with almost all your comments. My opinion - stated again some months-years ago - is that would be nice to have separate rating for different sizes/variations/timecontrols and combined rating(s).
(the almost goes to "maybe 2 levels for tic-tac-toe". I'd say about 6 levels :)
I mean tic-tac-toe in 2D and 3x3 like it is played by kids. It has basically 2 levels of players: "those who may lose" and "those who don't lose" ;-) but that's not exactly Elo rating. Of course, the game has many more interesting variants.
I'd also prefer separate ranks for each board size. (We had this before, so I don't think it is a technical issue; perhaps it is just a matter of Richard needing to find time to make the change, and he has more interesting things he wants to work on?)
P.S. mirror_man, I don't remember you ever being weak at 9x9 ;-) Glad to see you back here.
I disagree. My record against those who are at 2-3 ranks better than me (and who finish their games) is quite bad. Any mo record against those who are 2-3 ranks worse than me is pretty good.
maybe you select your opponents more careful, or you are lucky. I have periods where I see this stability, but for example in the last months, there were half a dozen new/sandbagger players that came in at 15 kyu, and really are single digit or dan, and I lost every game against them (now who is surprised), resulting in me going from 8.4 to 13.2. (they are now all single digits, thanks to me and some others feeding them).
Am I really five ranks worse now? I might have better days, but I don't think so.
Probably I will be called 'sandbagger' now, when I play (real) 10-11 kyu and they give me handicap and I should really give them. Whatever.
I love that there is a ranking system, but it is extremely unstable, because if various effects.
Yeah, I fluctuate pretty wildly on here in the 7-12 range, whereas my rankings elsewhere are solidly 4-5. Maybe there are more clusters than I listed, but anybody looking for a proper go-style rating system should look elsewhere. Other rating systems take into account the players' histories (and future games), and it's impossible to fluctuate an entire rank level after one or two games (at least not with a body of finished games behind you).
If you want to go really nuts, check out the handicap games. They're rated as if there were no handicap. So if you lose a 9-stone game against someone 10 ranks weaker, it will read as if you lost a straight game, which is a HUGE hit. That's why it's hard to get out of the 9-15 range, because new players come in and just crush you with handicap as they move up the ranks. Heck, I was probably responsible for a few angry fists shaken at the heavens myself.
But hey, it doesn't bug me. And not like the outcome is foreseen. The point of the handicap system is to play an even game. The close battles are always the most exciting. Everybody relishes a 0.5 win.
I think this is because 19x19 game consume too long time to make rating really flexible.
So if you are good - you could improve rating quicker by playing more 9x9 games...
Imagine new player who starts at 20k and playing only 9x9... Then reach for example 10k and want to start also bigger boards... (has to start from 20k?)
-----------
Another reason to divide rankings is strength of computer progams on 9x9
I agree with diference ratings - but probably there are several aspects which has to be analysed
I'm goiing to play a real-life go tournament for the first time next month. Because I'm not sure what my strenght is I'd like to know how the LG-rating is compared with the 'real' rating. If I'm 10k here, can I say I'm 10 k in the real world as well?
Little Golem ratings are highly inconsistent. I would suggest playing some games on a go server like KGS to get a real rank. Even only 10 games or so would give you a good idea.
PS: Don't play blitz games. They won't help your go at all.
I agree that they are inconsistent, but I think in general they are a bit tough since all players start low, so my tip would be to add a couple of ranks (to 8k).
Even if there is a transfer skill between various sizes of Go games they are still different games (mathematically). It follows that they should have different rankings. 9x9 is a very special game because of the border influence. The whole game takes place in the corner so to speak. On 13x13 there are separate corners but hardly any center. And 19x19 is just huge. Three different games.
Slightly off topic: I would like to see 11x11 Go on LG. If 9x9 is viable game, 11x11 would be even more so while being considerably shorter than 13x13. And there is not much software that plays on 11x11.
My current KGS rank 4k; best rank so far KGS 1k; IGS (not active) had different accounts 1d-5k.
I opt for different rankings for different board sizes ( and game variants like toroidal go of course).
GoRanks on LG are highly volatile.
Yet another question ( @ Richard) i wonder about is wether the ranks at the game start are used to calculate the new rankings or the ranks at games end.
It should of course be the first case.
Generally ranks should not be taken too serious here.
There are limits on Richard's time, so the community should be looking for ways to add support and features for themselves. For example, mongoose produced some very interesting statistics for Empathy, and i wrote a script to harvest the HTML pages and compute Empathy player rankings in several ways.
If all of the past Go results are all available, then there's nothing stopping someone from harvesting the results (e.g. with curl, a newer version of wget), computing separate ratings for 9x9, 13x13, and 19x19 using the standard Bradley-Terry model, and making them available to the community.
Using separate accounts for each game isn't really a solution unless everyone does it, which isn't going to happen.
Mirambel: strong Go programs can play 11x11 with the simple change of a constant, and would play quite well (maybe 3d or 4d, as an educated guess).