Why do some people stay at the same rank?

[HermanHiddema]

See also pages 6-9 of https://www.yumpu.com/en/document/read/ ... oforbundet. Figure 4 is very close to what gennan is suggesting.

[Uberdude]

My EGD graph so X axis is time not effort. If X axis was "games played on KGS" or "number of tsumegos solved" it would probably be quite a bit closer to linear. With "Words written on online Go forums" is pretty similar to time now.

[John Tilley]

Looking at my time at the Winchester Go Club from 2015 to 2018, the social aspects of meeting friends and playing Go in a good pub were really important. Perhaps just two people went to occasional tournaments out of about ten players. Yes, becoming stronger, would have been nice, but it's all about priorities.

Age and commitments such as work and family life play a big part.

I think many players had deeply ingrained bad habits, which would require a fair amount of unlearning. One or two players actually got upset if you tried to point out basic mistakes.

I put in a lot of effort to become stronger in the early 1970s but hit a wall around 1-2 dan and stopped playing. I recorded lots of games, had them reviewed, but perhaps above all I lacked a teacher. I also studied the wrong things and didn't play enough.

John

[Uberdude]

Shin Jinseo says logarithm-schmogarithm, asymptote-schmasmyptote:

[dfan]

Improvement is logarithmic. The longer you perform, the slower you improve. Only fundamental changes in the amount of time, the quality of training, or preferably both, can increase the slope again. If you have played for 25 years like me, there's so much bad habit ingrained in your playing, that it's almost impossible to still improve.

I think that improvement is linear (well, affine) but rating scales are logarithmic, which comes to the same thing but is significantly less depressing. If you use a Bradley-Terry model (if my rating is a and your rating is b, my chance of winning is a/(a+b)), then you hit diminishing returns at some point; it makes a much bigger difference in your results to go from a rating of 5 to 6 than to go from 50 to 51, and going from 500 to 501 is basically imperceptible.

Most rating systems, like Elo or dan/kyu, are effectively the logarithm of Bradley-Terry. The nice thing about taking the logarithm is that a difference of some particular value between two ratings always means basically the same thing (in Bradley-Terry, you have to look at the ratio). The downside is that it appears that your rate of skill increase slows down as you improve.

[jlt]

My EGD graph so X axis is time not effort. If X axis was "games played on KGS" or "number of tsumegos solved" it would probably be quite a bit closer to linear.

Wow, imagine if you had been studying go as hard as at the beginning, you would be 80 dan by now.

[gennan]

I think that improvement is linear (well, affine) but rating scales are logarithmic, which comes to the same thing but is significantly less depressing. If you use a Bradley-Terry model (if my rating is a and your rating is b, my chance of winning is a/(a+b)), then you hit diminishing returns at some point; it makes a much bigger difference in your results to go from a rating of 5 to 6 than to go from 50 to 51, and going from 500 to 501 is basically imperceptible.

Most rating systems, like Elo or dan/kyu, are effectively the logarithm of Bradley-Terry. The nice thing about taking the logarithm is that a difference of some particular value between two ratings always means basically the same thing (in Bradley-Terry, you have to look at the ratio). The downside is that it appears that your rate of skill increase slows down as you improve.

The difference between Bradley-Terry and Elo is not very big. You can even "derive" the Elo formula from the Bradley-Terry formula.

Both work as P_win = 1/(1 + b/a) where b and a are measures of "skill" of both players. The rating is usually expressed as r_a = ln(a) [with some additional scaling constants], so that P_win = 1/(1 + exp(r_b - r_a))

So instead of using ratios of "skill", you get to use differences in rating, which is more convenient.

So it's true that ratings are a logarithmic scale for "skill", similar to the Decibel scale, the Richter scale and the Kardeshev scale, where a difference in those units actually means some ratio of the underlying quantities.

But when you make a plot like the one that Herman pointed to, but this time using "skill" on the vertical axis, it comes down to just transforming the vertical axis to a logarithmic scale. The plateau effect stays. So I think this rescaling doesn't really matter for the "depressing" implications of that plot.

[gennan]

If you're interested, I got this for the relation between Bradley-Terry "skill" (a in my previous post) and EGF rating from my previous statistical analysis of EGD data (note the logarithmic scale on the vertical axis).

The plot is not linear, because EGF ratings are based on ranks instead of P_win (like Elo ratings), but when you zoom in a bit to a smaller range of ranks, it's close to linear (especially in the kyu range):

[gowan]

Learning curves often follow a logistic pattern, with a slow start followed by a rapid increase with an increasing rate of growth. Then at some point the curve has an inflection point after which the curve continues to increase but the rate of growth decreases. The curve eventually approaches a horizontal asymptote at a level approximately twice the level of the inflection point. If learning go follows this pattern you could predict your maximum by observing the change in the rate of increase.

[Gomoto]

multiple reasons

[SoDesuNe]

I've seen individuals stay at the same rank for over thousands of games over multiple decades?

If you are that consistent, what are some of the reasons you would not improve naturally?

Are they playing rote moves without thinking deeply each game? I doubt that it's a matter of talent since kyu ranks are hardly the limit of improvement for anybody (I could understand as you get into the mid-dan ranks, where margin of improvement decreases). Over multiple years, you are bound to improve slowly but surely, if you put in the effort. Perhaps, many people are simply happy where they are and just want to enjoy playing the game.

But that is just pure speculation on my part. What do you think?

Playing is not studying. It's more like writing a test but unless you review it, you'll never know where your mistakes were. Feedback-loops are essential for improving. Measuring wins, losses or your rank won't give you meaningful feedback to improve.

The perfect example for me is typing. We pretty much all do it and have invested a lot of time in it. But how many of us are actually proficient in typing (fast and steady, using every finger, not looking at the keyboard)?

[Gomoto]

Nobody dares to say the main reason why some are not able to increase their go abilities over their current level. ( genan did in a polite way )

While political correct, it serves nobody. Respecting players at all levels does.

It is nice that go can be enjoyed by people at all levels.

[dfan]

We agree (in fact we are saying exactly the same thing) up until here:

But when you make a plot like the one that Herman pointed to, but this time using "skill" on the vertical axis, it comes down to just transforming the vertical axis to a logarithmic scale. The plateau effect stays. So I think this rescaling doesn't really matter for the "depressing" implications of that plot.

Instead of using the log of the skill (in the Bradley-Terry sense) for the y axis, I prefer to use actual skill for the y axis, so the plot of skill as a function of effort is affine, but with the horizontal lines (representing ranks or Elo) that have exponentially increasing spacing.

Of course this illustrates exactly the same information. But to me it is much less depressing to think "I'm grinding, and of course it's getting harder to hit new ranks because they're farther apart" than to think "My ability to improve has decreased".

[jlt]

Nobody dares to say the main reason why some are not able to increase their go abilities over their current level. ( genan did in a polite way )

While political correct, it serves nobody. Respecting players at all levels does.

It is nice that go can be enjoyed by people at all levels.

Gennan said more or less that go rank depends on several factors:

• Talent
• Time
• Motivation
• Learning environment

You seem to be suggesting that talent is the main factor. No problem with that, I've accepted long ago that I have no particular talent for go, but this doesn't explain what is talent. For me, talent is a combination of

1) Doing (almost) the right thing right away.
2) Being able to identify mistakes.
3) Being able to correct them.

People with ability #1 make quick progress, but it doesn't mean that other people cannot reach a good level after a lot of effort.

People without ability #2 can use the help of a teacher, but a teacher is useless if they don't have ability #3.

Coming back to my analogy with learning English as a second language, people with good listening abilities can pronounce "the" almost correctly from the very beginning. Others will go through a stage of pronouncing "ze", realize by themselves it's wrong and will improve the pronunciation after several months. Others need the help of a teacher who points out the problem and tell them to repeat "The thirty-three thieves thought that they thrilled the throne throughout Thursday" ten times. But some people, even with the help of a teacher, continue to pronounce "ze" for decades, thinking that making the effort to correct the pronunciation is not necessary and that communication is efficient anyway, even with the pronunciation defect.

If (say) some of my stones get captured at the dame-filling stage, someone could tell me "hey, didn't you count liberties? Can't you read what happens if your opponent cuts or peeps?" Well, yes I can do that. Yes I know I should do that. But no, I don't pay attention at every move of every game.

[Gomoto]

It depends not only on effort, but also on learning ability

If somebody really lacks learning ability, it is fair to respect the limits. Go can be enjoyed on many levels. This is one aspect of Go I love.

(To be clear: I am not arguing that all the people not improving on go servers lack the learning ability to do so.)