Life In 19x19
http://www.lifein19x19.com/

Life-Time Performance and Age-Related Declines
http://www.lifein19x19.com/viewtopic.php?f=13&t=7475		 Page 1 of 2
Author:  ez4u [ Sun Dec 23, 2012 7:30 am ]
Post subject:  Life-Time Performance and Age-Related Declines
While working on the KGS ranking system topic with data for amateurs, I could not help but wonder whether pros were different. Needless to say this led sooner rather than later to having GoGoD/Kombilo open on the desktop and a few of my favorite people under the microscope. First up here are some graphs showing the career winning percentages of Cho Chikun, Cho Hun-hyeon, and Yi Ch'ang-ho. The winning percentage is calculated as a 100-game moving average, with a smoothed line to help see what is going on. For each player the horizontal axis shows their age at the time that the 100th game in the sample was played. Since all the players have relatively few games in GoGoD from their earliest years, the first age that shows may seem rather late (especially for Cho Hun-hyeon). No deep thoughts on this yet.

Cho Hun-hyeon 1900+ games
Attachment:
100-game MA Cho Hun-hyeon v2.jpg
100-game MA Cho Hun-hyeon v2.jpg [ 76.62 KiB | Viewed 11233 times ]


Cho Chikun 1900+ games
Attachment:
100-game MA Cho Chikun v2.jpg
100-game MA Cho Chikun v2.jpg [ 77.01 KiB | Viewed 11233 times ]


Yi Ch'ang-ho 1700+ games
Attachment:
100-game MA Yi Changho v2.jpg
100-game MA Yi Changho v2.jpg [ 74.91 KiB | Viewed 11233 times ]
Author:  jts [ Sun Dec 23, 2012 9:17 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
Wow, interesting analysis. Thanks! One question, though: is this really life-time performance, or is this a secular increase in the quality of the competition?
Author:  ez4u [ Sun Dec 23, 2012 10:02 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
jts wrote:
Wow, interesting analysis. Thanks! One question, though: is this really life-time performance, or is this a secular increase in the quality of the competition?

Well, I guess we would have to ponder what the difference would be. Everyone's competition will change over a life time. At the same time we all grow older. So how should we sort it out?

That said, there are some interesting bumps along the way. The younger crew in Korea came to prominence just after 2000. So we can check the charts to see whether we would want to link that change to everyone's performance or not. (The actual checking will be left to the reader as an exercise :salute: )
Author:  gowan [ Sun Dec 23, 2012 10:25 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
I think there is a general belief in decline of performance as a player ages and these graphs support that. This general trend could be attributed to many things. For example, there is a belief that younger players are "hungrier" for success which leads to higher win/loss ratios. We see this decline in other fields, e.g. mathematics and physical sciences. People in their 50's and older are often satisfied with where they've gotten in their lives and don't have the drive that younger people do.
Author:  topazg [ Sun Dec 23, 2012 11:02 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
gowan wrote:
I think there is a general belief in decline of performance as a player ages and these graphs support that. This general trend could be attributed to many things. For example, there is a belief that younger players are "hungrier" for success which leads to higher win/loss ratios. We see this decline in other fields, e.g. mathematics and physical sciences. People in their 50's and older are often satisfied with where they've gotten in their lives and don't have the drive that younger people do.


I would be interested to see "reached 9d", "won first major title", and perhaps "won 5th international title" as milestones, and then compare pros who hadn't reached these heights to these three. I suspect "won it all, not as motivated" is a much bigger factor until a lot later - if you like the ages on the x axis, they are hardly all that close between lee changho and the other two ;)
Author:  tapir [ Sun Dec 23, 2012 11:59 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
There is also the difference in tournament systems. Cho Chikun has his initial high winning rate drop as he moves through the preliminaries / oteai only to increase again later. No such "waves" in the graph of the Korean players.
Author:  John Fairbairn [ Sun Dec 23, 2012 12:29 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
Yes, interesting exercise! It may be that not all of the decline is age related, though. A common joke among go and shogi players in Japan is that a player's results decline sharply when he gets married. Superficially, at least, results seem to bear this out.

For those who believe the rising quality of the competition may be an important element, Cho Hun-hyeon once told me he thought this was definitely the case in Korea. Japan was different then because olders pros were sheltered somewhat by being seeded in preliminaries according to grade. The upshot was that they spent more time playing each other than playing rising new stars.

For the players cited, though, I suspect sheer exhaustion rather than age is the factor that marks the start of the decline. Having a very high winning percentage implies that they are playing in finals and title matches, and so they are playing the very strongest opponents with great frequency.
Author:  SmoothOper [ Sun Dec 23, 2012 7:59 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
I was thinking that Cho Chikun was washed out, but I reviewed some of his games just the other day, and seems to be just as sharp as ever. To some extent I think he has become the standard bearer for the Japanese school. Outsider's have somewhat of an advantage since opponents don't know what to expect.
Author:  Mef [ Mon Dec 24, 2012 9:32 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
ez4u wrote:
jts wrote:
Wow, interesting analysis. Thanks! One question, though: is this really life-time performance, or is this a secular increase in the quality of the competition?

Well, I guess we would have to ponder what the difference would be. Everyone's competition will change over a life time. At the same time we all grow older. So how should we sort it out?


Just brainstorming ideas to try and split the effects of aging from the effects of time...

1:Compare aging curves for pros with similar career spans across different chronological time spans, check to see if there is better correlation of decline with age or calendar year.

2:Instead of looking at 1 player's winning percentage over time, compare the aggregate group of players winning percentage split out by age of player vs. age of opponent (e.g. How does the winning % of 30 year olds vs. 27 year olds compare to the winning percentage of 35 year olds vs. 27 years olds?).

3:Look at the average age of title holders over the years. If it is consistent, it would suggest that there is a certain approximate age the players are peaking at, if there is a sudden step change it might suggest a shift in the strength of competition (a crop of very strong youngsters).

4:Combine ideas from 1 and 2...once you know the average expected winning percentage of players X years old vs. players Y years old, normalize player's winning percentages accordingly by age, then look at historical trends (where there times when the young players were doing especially well compared to young players of the past? Were there times the old players kept their dominance especially long? These could demonstrate changes in the quality of competition).

...perhaps one of these days I'll have to invest in GoGoD and start trying to look at some of these things...
Author:  topazg [ Mon Dec 24, 2012 10:07 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
John Fairbairn wrote:
Yes, interesting exercise! It may be that not all of the decline is age related, though. A common joke among go and shogi players in Japan is that a player's results decline sharply when he gets married. Superficially, at least, results seem to bear this out.


As a married man, I would be very surprised if this isn't the case, particularly for title holding level players where the need to put bread on the table becomes superceded by other priorities.
Author:  logan [ Mon Dec 24, 2012 1:32 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
Although I think I'm improving, my graph is probably the same as these :lol:
Author:  Uberdude [ Mon Dec 24, 2012 2:46 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
topazg wrote:
John Fairbairn wrote:
Yes, interesting exercise! It may be that not all of the decline is age related, though. A common joke among go and shogi players in Japan is that a player's results decline sharply when he gets married. Superficially, at least, results seem to bear this out.


As a married man, I would be very surprised if this isn't the case, particularly for title holding level players where the need to put bread on the table becomes superceded by other priorities.


Doesn't seem to have hurt Lee Sedol too much (and he has a daughter)! Lee Chang-ho though...
Author:  topazg [ Mon Dec 24, 2012 7:14 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
Uberdude wrote:
topazg wrote:
John Fairbairn wrote:
Yes, interesting exercise! It may be that not all of the decline is age related, though. A common joke among go and shogi players in Japan is that a player's results decline sharply when he gets married. Superficially, at least, results seem to bear this out.


As a married man, I would be very surprised if this isn't the case, particularly for title holding level players where the need to put bread on the table becomes superceded by other priorities.


Doesn't seem to have hurt Lee Sedol too much (and he has a daughter)! Lee Chang-ho though...


Sure, it's not a sufficient factor, but it is a contributary one. The level it contributes I suspect depends on personality type and burning desire to keep claiming titles. For some it will deprioritise things more than others.
Author:  ez4u [ Mon Dec 24, 2012 7:29 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
Round Two. I went to 50-game moving averages because the coverage of some players is so light at the beginning of their careers. For example, GoGoD contains only 39 games by Sakata prior to the age of 30 (I excluded all handicap games, pair go, etc. for all the players). On the other hand, it has 56 games that he played in his 70's.

Hashimoto Utaro 1400+ games (interestingly Hashimoto's career was so long that half the games in GoGoD were played after the age of 50) DOB: 1907-02-27
Attachment:
50-game MA Hashimoto Utaro.jpg
50-game MA Hashimoto Utaro.jpg [ 85.43 KiB | Viewed 10904 times ]


Sakata Eio 1200+ games DOB: 1920-02-15
Attachment:
50-game MA Sakata Eio.jpg
50-game MA Sakata Eio.jpg [ 79.82 KiB | Viewed 10904 times ]


Rin Kaiho 1500+ games DOB: 1942-05-06
Attachment:
50-game MA Rin Kaiho.jpg
50-game MA Rin Kaiho.jpg [ 81.76 KiB | Viewed 10904 times ]
Author:  ez4u [ Mon Dec 24, 2012 7:39 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
And that fourth attachment that is over the limit for a single post... :blackeye:

Seo Pong-su 1200+ games DOB: 1953-02-01 ( :tmbup: two days older than yours truly)
Seo is an interesting case. His life-time winning average is 55.1%. However, he is famous for finishing as runner up to Cho Hun-hyeon in something like 50(!) titles. If we exclude Cho, his life-time winning average against "the rest of the world" is 62.7%. The rather mild decline we see in his average over time may be due to the decline in the number of games he had to play against Cho.
Attachment:
50-game MA Seo Pongsu.jpg
50-game MA Seo Pongsu.jpg [ 80.31 KiB | Viewed 10900 times ]
Author:  jts [ Mon Dec 24, 2012 7:57 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
Question (perhaps one that JF or TMark can answer): if a young gun plays a game that shows amazing skill and/or promise, does that increase the chances that the kifu will be preserved? Or is the transmission down to us basically random?
Author:  TMark [ Tue Dec 25, 2012 2:07 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
jts wrote:
Question (perhaps one that JF or TMark can answer): if a young gun plays a game that shows amazing skill and/or promise, does that increase the chances that the kifu will be preserved? Or is the transmission down to us basically random?


It is more a question of whether the game gets published. In the recent past that would depend on newspapers or magazines having the space but today a lot of material is on the web. However, a lot of preliminary games in tournaments don't get published immediately. Sometimes you have to wait for the "Complete" games of somebody to be issued. Also, when it was running, the majority of the Oteai games were not published.

Best wishes.
Author:  ez4u [ Tue Dec 25, 2012 6:38 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
At first I thought mef's brainstorming was just nuts. But then I thought there must be something, hmmm...
Anyway it's Christmas so here is a little present for those who like numbers...

Recipe:
1. Take Kombilo game list of all the even games from GoGoD (67000+) and stick it in excel.
2. Stick the text file for JF's name dictionary into excel and parse out the year of birth where ever it appears in the form (yyy[y] following the first parenthesis in the info block.
3. Do a vertical lookup to get the YOB for White and Black and calculate their age at the time the game was played.
4. Delete all weird results: missing YOB, negative ages, games by 3-year olds, games by 100-year olds, etc.
5. Delete all games before 1950 (proxy for the age of modern tournament/komi Go).
Due to the quick and dirty nature of 2 (mainly) and 5, end up with 52000+ games in total.

Next:
a. Calculate the difference in age of the players.
b. Calculate the result of the older player in each game.
c. Bucket the games by decade of age difference: decade0 = players up to 10 years apart in age, decade1 = players 10 to 19 years difference in age, decade2 = players 20-29 years difference in age.
d. Ignore the rest since this covers 86% of games and even a 500-game moving average does not pick up enough "decade3" games to make a nice graph. :blackeye:
e. Calculate 500-game moving averages for each "decadeX", smooth the result and graph it over time.

Voilà!
Note that the compression of the years on the left of the horizontal axis and the expansion of the right side represents the different number of games available by year.
Attachment:
500-game MA Older vs Younger.jpg
500-game MA Older vs Younger.jpg [ 106.81 KiB | Viewed 10817 times ]
Author:  TheBigH [ Wed Dec 26, 2012 12:58 am ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
Isn't it also true that, if you're good enough for long enough, people will try to work out ways to beat you?
Author:  lemmata [ Wed Dec 26, 2012 8:39 pm ]
Post subject:  Re: Life-Time Performance and Age-Related Declines
topazg wrote:
gowan wrote:
I think there is a general belief in decline of performance as a player ages and these graphs support that. This general trend could be attributed to many things. For example, there is a belief that younger players are "hungrier" for success which leads to higher win/loss ratios. We see this decline in other fields, e.g. mathematics and physical sciences. People in their 50's and older are often satisfied with where they've gotten in their lives and don't have the drive that younger people do.
I would be interested to see "reached 9d", "won first major title", and perhaps "won 5th international title" as milestones, and then compare pros who hadn't reached these heights to these three. I suspect "won it all, not as motivated" is a much bigger factor until a lot later - if you like the ages on the x axis, they are hardly all that close between lee changho and the other two ;)

Good idea! This seems especially relevant for major titles that have a challenger system so that the title holder only has to win 4-3, 3-2, or 2-1 to keep the title. It is natural then for the title-holder's winning percentage to be lower because he would only get to play the strongest challenger. For people like Cho Chikun or Kobayashi Koichi, this may have kept their winning percentages down over the years. I am kind of talking out of my rear, but it's a theory.
Page 1 of 2 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/