Does KGS rank system encourage users to make new accounts?

[Mef]

This has more to do with Tygem than KGS, but to inject a little data into the conversation about volatility of ratings, I ran a short simulation

Interesting! This may be enough of a tangent to merit its own thread in the tygem forum. A couple questions:


1: Does this include the double promotion rule for if you win 90% of your games?

2: How much would we expect this to change if we allowed a feedback loop (e.g. if 25% of the time you are incorrectly rated, this also means that 25% of the time your opponent is incorrectly rated as well...these could play into each other)

Edit: it looks like you may have already answered #2 above

[skydyr]

There are also more reasons for multiple accounts than concern about rank. I know people with "anti-social" accounts for when they just want to play and not chat to anyone (the lack of a Do Not Disturb tag for KGS being a core reason for this), people with "drunk" accounts for when they can't stop themselves from playing after a few drinks (how well this works seems to vary ) and so on. Both of these are accounts made about rank concerns but not really because of KGS's ranking system.

I think separating blitz and non-blitz accounts is also relatively common.

[MagicMagor]

I don't think the system encourages this behaivour, but people think it does.

Now i'm a fan of the KGS-system but let me explain my point with two "extreme" players.

Player A and Player B both have a stable real life rank, say at least SDK. Player A plays a lot on KGS, he has a huge game history and his KGS-rank has converged to his real life rank.

Player B rarely plays on KGS, so his game history is pretty empty and his rank is ?.
If Player B would start playing on KGS more, his rank will rapidly converge to his true rank.

Now consider both players improve their rank "offline", through study. Player Bs rank will converge rapidly to his new rank, the one of Player A not, because he has a huge game history. This is what people percieve as the systems problem. But i think the problem lies in how Player A percieves the system. If A is playing an average of 10 games/day a single game is not so important. What is a single loss if you play 9 other games that day? If A is not improving that is how he sees the games i presume. It's only for fun and the single game doesn't have that much weight on him.
But if he thinks he has improved this changes. Now he sees each win as a proof of his improvement and is annoyed by the fact that KGS doesn't weight these games as much as he does. So he creates a new account, where each new game has a much heigher weight on the rank.

The problem doesn't lie in the system but in the users. Also improvement is a gradual process it is not discrete (even if the ranks are). Especially in the SDK and Dan-ranks it takes a long time actually improve one stone. If A would keep his 10 games/day during his months of study his winning rate would rise over time. (From 50% to 55% to 60% and so on) And then when he finally self-promotes himself in real life, or has earned enough rating to start with a higher rank at a tournament, KGS has already learned that A has improved a bit. Not enough yet to also promote him, but it may take fewer games than if A had stopped playing on KGS during his improvement (what all these thought experiments of "offline" improvement seems to assume).

Player B on the other hand (an experience i had several times) has no such problem but enjoys the benefit that the system can give him an fairly accurate rank after just a few games.

I don't want to use that "2d?" account, since I feel that I don't deserve to play anyone of that strength without handi stones

I think here is where you make a mistake. First of all, the ?-mark after the rank shows everyone that this rank is not guranteed to be accurate, so your opponent knows you may not be a real 2d. You could also just open new games and state your assumed "real" rank so people know what to expect and play with the right handicap.
And if you - like me - prefer automatch, just play. Because the ?-mark also means KGS will rapidly change your rank after each game. So you will only have to play 1-3 games at fairly wrong ranks.

Since i had times were hardly played at all, i was often in the position of having each game having a heavy influence on my rank. And a normal loosing streak could easily change my stable 2k-rank to 3k or even 4k. But i knew, it would just take me one win against a 5k (which was to be expected from the next game, as my real strength was still 2k) to correct it. Sometimes i feel sorry for the 5k which would get me as their next opponent, but i always hoped i may provide him with a good learning experience.

Inaccurate ranks can always happen in any system, as no system is perfect. But KGS is good at correcting these mistakes rather quickly, which i find the best possible effect.

I also have an account on IGS but haven't played there for years. I'm a 7k there and i don't play there because it would literally take me dozens of games to correct this rank. And every lost game (which still happen from time to time) would push me further back. I really prefer the KGS-system over IGS.

[mitsun]

1: Does this include the double promotion rule for if you win 90% of your games?
No, I did not include the double promotion rule. I suspect this would have almost no effect, assuming you are already reasonably close to your correct rank.

2: How much would we expect this to change if we allowed a feedback loop (e.g. if 25% of the time you are incorrectly rated, this also means that 25% of the time your opponent is incorrectly rated as well...these could play into each other)
This should introduce more noise and hence make ratings more volatile, but I tried it and saw no difference.

[Mef]

1: Does this include the double promotion rule for if you win 90% of your games?
No, I did not include the double promotion rule. I suspect this would have almost no effect, assuming you are already reasonably close to your correct rank.

Well, the situation I was thinking more about is that if you were over/underrated then got corrected, you are already very close to the double promotion/demotion line...you would expect overcorrections maybe 10-15% of the time? (because if you have already won enough games to promote, now it's just a matter of winning 3 of the next 4...or maybe 4 of the next 5). I think you would expect this more at the kyu levels (which use only a 15 game history and lower winning percentage cut offs).

[Boidhre]

There are also more reasons for multiple accounts than concern about rank. I know people with "anti-social" accounts for when they just want to play and not chat to anyone (the lack of a Do Not Disturb tag for KGS being a core reason for this), people with "drunk" accounts for when they can't stop themselves from playing after a few drinks (how well this works seems to vary ) and so on. Both of these are accounts made about rank concerns but not really because of KGS's ranking system.

I think separating blitz and non-blitz accounts is also relatively common.

Well, it's fairly common for chess servers to separate your ratings according to standard/blitz/bullet or whatever on the one account. Why go servers don't do this is odd to me.

[SmoothOper]

There are also more reasons for multiple accounts than concern about rank. I know people with "anti-social" accounts for when they just want to play and not chat to anyone (the lack of a Do Not Disturb tag for KGS being a core reason for this), people with "drunk" accounts for when they can't stop themselves from playing after a few drinks (how well this works seems to vary ) and so on. Both of these are accounts made about rank concerns but not really because of KGS's ranking system.

I think separating blitz and non-blitz accounts is also relatively common.

Well, it's fairly common for chess servers to separate your ratings according to standard/blitz/bullet or whatever on the one account. Why go servers don't do this is odd to me.

I don't know about the chess part, but I wholely agree on the rankings. Rankings should only apply when people are playing the same game, or there is some requirement that each individual plays both blitz and non-blitz, because ultimately the rankings will assume there are players that are good at both, which is non-sense. It is also annoying on Tygem that you have to negotiate the time in auto-match mode.

[Boidhre]

This has more to do with Tygem than KGS, but to inject a little data into the conversation about volatility of ratings, I ran a short simulation with the following assumptions:

win 50% against equally ranked opponent
win (55,60,65,70)% against opponent with rank-1
win (60,70,80,90)% against opponent with rank-2
promote/demote after winning 70%/30% of previous 20 games
reset win rate counter after promotion or demotion

As an aside none of these would hold their rank as a mid-high dan in the EGF system, you would be expected to do better than 70% against rank-1 and against rank-2 80% would be far too low. The EGF system expects more consistent win rates the stronger you are because the assumption is that players play more consistently. 70% against rank-1 is 18-20k territory. Dan level would be closer to 75% by comparison. (For example a 5d would expect to win less than 7% of the time against someone on paper 200 GoR higher where 100 GoR is approximately 1 stone in strength)

Not that the above numbers are wrong, just that the difference is interesting.

Edit: I should point out what rank-1 means in my numbers here. It is not a 2d vs a 1d it is two players whose GoR difference is 100. This is important as 4d facing a 2d could mean a GoR difference between 101 and 399 depending on how strong either of them are for their rank (rank like 1d doesn't mean that much there is a rather large gap between a very strong and very weak 1d in the EGF system never mind that some player's declared rank and their GoR say rather different things).

[quantumf]

I don't know about the chess part, but I wholely agree on the rankings. Rankings should only apply when people are playing the same game, or there is some requirement that each individual plays both blitz and non-blitz, because ultimately the rankings will assume there are players that are good at both, which is non-sense. It is also annoying on Tygem that you have to negotiate the time in auto-match mode.

Forgive the slight hijack, but one thing about KGS that is becoming more and more annoying to me the more I have to endure it (and is forcing me to tygem) is that utter randomness of the time systems in the games on offer. I can get used to a few basics, such as the ones in automatch, but the offered games are insane. Sometimes 1 byoyomi, sometimes, 3, sometimes, 5, sometimes 100. Sometimes 8 seconds, sometimes 10 seconds, sometimes 20 seconds, sometimes 22 seconds. It's ludicrous, and I have to play the clock as well as my opponent. It's a small thing, I know, and I suppose it has it's place, but it's abuse is really starting to affect my enjoyment of the system.

[oren]

Unless I'm playing someone I know, I always use automatch. That way I don't need to check times, how someone has the handicap set up, people changing rulesets, etc... All I need to check is did I get a "fast" game or "medium" game.

[hyperpape]

Unless I'm playing someone I know, I always use automatch. That way I don't need to check times, how someone has the handicap set up, people changing rulesets, etc... All I need to check is did I get a "fast" game or "medium" game.

Making the time settings be "Blitz/Fast/Medium/Custom" in the open games window and presenting things the same way in the list of open games might be a nice change.

I wonder if that's inconsistent with wishing that KGS would implement Fischer time.

[quantumf]

Unless I'm playing someone I know, I always use automatch. That way I don't need to check times, how someone has the handicap set up, people changing rulesets, etc... All I need to check is did I get a "fast" game or "medium" game.

Automatch is cool, but quite often I can't get a game on automatch, so have to resort to the open challenges.

[Nathan]

I agree with Kirby's thoughts about the volatility of the ranking system. I feel deterred from playing many games with one account because the rank will feel "stuck", even if it is just in my head.

I feel a possible solution would be to limit the number of games that affect the rank. For example only the past 30 games count towards your ratings. This would add some volatility to ranking for those players who play many games. I am not sure how this how this would affect convergence. I am interested on others thoughts.

[hyperpape]

With just 30 games, ranks will be volatile and therefore quite inaccurate (if you have an account you use infrequently, you can have probably observed this).

That's not a tweak, it's just giving up on the system, in favor something much more like Tygem.

[oren]

On the opposite end, I just played on Pandanet IGS yesterday for a change of pace. I have a really old account and decided to check how many games it would take to 1d. My account is at 8k+ right now. Even if I win every game in a row, it would still take me 100 games to get the account to 1d.