Trend in Resignations

[ez4u]

In another topic, which suffered from a bad rodent infestation...

As usual I exaggerate for the purposes of stimulating discussion, but I think there is a bigger trend going on. It's to do with time limits.

In the good old days, the aim of the fuseki was to get to the endgame. Nowadays the aim is to get to the middle game.

...

Behind all of this, obviously, is also the observation that resignation is much more common in modern games with short time limits. More blunders or more accent on fighting? Probably both.

This did not quite pass the smell test for me. Today I had the chance to grab some data from my brand new Summer 2011 GoGoD CD (thanks JF and T Mark ). Here are the results of searching for "B+R" and "W+R" using kombilo and then dumping the numbers into excel. I do not think that the trends we see are tied to the recent introduction of short time limits in tournaments. The major increase for both Black and White looks like it is more likely related to the original introduction of clocks. The steadiness of Black's ratio speaks rather against the influence of changing time limits. Meanwhile the gradual increase for White looks like it is more likely related to changes in komi over the years than time limits. Anyway YMMV, as always!

[oren]

A second thing to add for interest would be what move did the resign occur?

In the past, pros could easily play to endgame and resign when down by 3-5 points vs resigning during fights after 100 moves today. So maybe add a new chart with 'average final move' from resigned games and see if that has trended downward. That would still indicate lack of endgames.

[Dusk Eagle]

I've certainly noticed a trend - I've had 33 consecutive games and counting on KGS without one ending by scoring, with all but a couple ending by resign .

[snorri]

Before 1900 or so, there were a lot more records of games that were unfinished, which exaggerates the effect seen on that end.

[judicata]

Before 1900 or so, there were a lot more records of games that were unfinished, which exaggerates the effect seen on that end.

Good point. Are unfinished games removed from the data?

Also, it probably makes more sense to limit the graph to games that white/black actually won. In other words, out of all games that white won, what percentage of them was by resignation? And the same for black, of course.

[ez4u]

OK, OK, maybe you guys got lucky and raised some good points. Hmmm...

So I went back to the drawing board and did the following with kombilo:
Searched for "B+" and "W+", thus finding all games won by Black and White respectively. There were 33,003 games and 29,867 games.
Within each of the previous selections, searched for "B+R" and "W+R" for games won by resignation.

In all cases, I also searched on the empty board, which will exclude all handicap games and also all the old Chinese games that began with two White and two Black stones on the board.

The result is this graph.


I also searched for "B+ " and "W+ " (i.e. with an empty space rather than an "R" or a number) to isolate the games with an unclear result. There were 58 such games for Black and 64 for White. How to treat these is, by their nature, unclear. The limiting case is to assume that they are all games decided by resignation. Doing so results in this graph.


There are a few games that end up double counted when there is a comment like - W+ (but another source gives B+). This will be picked up by both the "W+" and "B+" searches. There seemed to be around 45 such games but I could not tell exactly. Most periods had no such games and the highest incidence was 4 out of 964 games in 1850-1899. I ignored the problem since I could not think of a way to exclude them in an automated fashion.

I think we can see a tendency for more games to be decided by resignation over the last ten years. However, we also see that the increase is from a long-term base rate of around 60% up to most recently around 75%. I do not see that this supports the idea that games were primarily played out to the end in previous periods. But as usual YMMV

Edit: Here is the First Graph data adjusted to exclude the games played under Chinese rules (issue raised in later posts below). I searched for "RU[Chinese" to identify the games. I am placing it here for easy comparison. It does not alter the basic trends greatly.

[snorri]

There is also a bias towards more resignation if you are playing by Chinese rules. (John F. usually blames this on the nuisance of Chinese scoring. I am not entirely convinced.) Since there are relatively more games available in GoGOD in recent times that are played with Chinese rules (in the 50s there were only 5!) this may account some of the trend.

Winter 2010 GoGod Stats, 2010 games

900 Chinese Rules

346 B+R 38.4%

340 W+R 37.8%

Total: 76.2%

1333 Not Chinese

465 B+R 34.9%

443 W+R 33.2%

Total: 68.1%

ez4u, not to make more work for you, but it would be interesting to see your graphs with Chinese rules games filtered out.

Also, in defense of John F., time limits do seem to be a significant factor. If we compare the Agon cup games since 2002 with the Kisei, Meijin, and Honinbo games in the same period, these are all with Japanese rules. But the resignation stats are markedly different:

Agon since 2002: 286 games

97 B+R 33.9%
103 W+R 36.0%

Total: 69.9%

Kisei, Meijin, Honinbo since 2002: 1575 games

466 B+R 29.6%

438 W+R 27.8%

Total: 57.4%

So we have several factors increasing the frequency of resignation over time. Decreasing time limits is one, the increase in the number of games played with Chinese rules is another, and komi is yet another. Since komi and time limits also affect style, we might consider style a derivative effect.

[Mef]

Also, in defense of John F., time limits do seem to be a significant factor. If we compare the Agon cup games since 2002 with the Kisei, Meijin, and Honinbo games in the same period, these are all with Japanese rules. But the resignation stats are markedly different:

Agon since 2002: 286 games

97 B+R 33.9%
103 W+R 36.0%

Total: 69.9%

Kisei, Meijin, Honinbo since 2002: 1575 games

466 B+R 29.6%

438 W+R 27.8%

Total: 57.4%

So we have several factors increasing the frequency of resignation over time. Decreasing time limits is one, the increase in the number of games played with Chinese rules is another, and komi is yet another. Since komi and time limits also affect style, we might consider style a derivative effect.

Not to weight in on either side of the "Resignation as a function of time" argument, but isn't he Agon cup a Pro-am event, as opposed to the other titles which would have professional leagues filtering out the competition? I would think a larger number of resginations would be expected from a field with wider skill disparity. The number of confounding variables continues to grow....

[ez4u]

There is also a bias towards more resignation if you are playing by Chinese rules. (John F. usually blames this on the nuisance of Chinese scoring. I am not entirely convinced.) Since there are relatively more games available in GoGOD in recent times that are played with Chinese rules (in the 50s there were only 5!) this may account some of the trend.

Winter 2010 GoGod Stats, 2010 games

900 Chinese Rules

346 B+R 38.4%

340 W+R 37.8%

Total: 76.2%

1333 Not Chinese

465 B+R 34.9%

443 W+R 33.2%

Total: 68.1%

ez4u, not to make more work for you, but it would be interesting to see your graphs with Chinese rules games filtered out.

Also, in defense of John F., time limits do seem to be a significant factor. If we compare the Agon cup games since 2002 with the Kisei, Meijin, and Honinbo games in the same period, these are all with Japanese rules. But the resignation stats are markedly different:

Agon since 2002: 286 games

97 B+R 33.9%
103 W+R 36.0%

Total: 69.9%

Kisei, Meijin, Honinbo since 2002: 1575 games

466 B+R 29.6%

438 W+R 27.8%

Total: 57.4%

So we have several factors increasing the frequency of resignation over time. Decreasing time limits is one, the increase in the number of games played with Chinese rules is another, and komi is yet another. Since komi and time limits also affect style, we might consider style a derivative effect.

I'll have to think about the Chinese rule part. I can't do it directly in Kombilo so it would mean gathering the info and backing it out in excel.

I hesitate to go into finer detail by tournament. When the numbers get smaller any competent statistician (i.e. not including me) would warn us the results become less and less reliable. Also we start to cherry pick our results, e.g. just to oppose your findings:
NHK (TV tournament) since 2002: 459 games
129 B+R 28%
123 W+R 27%
Total: 55%

Essentially unchanged from NHK since 1953: 1,108 games
317 B+R 29%
283 W+R 26%
Total: 54%

But who's counting?

BTW, we also have to struggle with social issues. One pro in Japan told me that on the television tournaments, the losers will sometimes play the game all the way out because they do not want to have time left over for analysis. They do not want to discuss their loss in public!

[snorri]

Not to weight in on either side of the "Resignation as a function of time" argument, but isn't he Agon cup a Pro-am event, as opposed to the other titles which would have professional leagues filtering out the competition? I would think a larger number of resginations would be expected from a field with wider skill disparity. The number of confounding variables continues to grow....

Nice point. Some major international tournaments are like this now, too. More than in the past...

I tried looking at KBS vs. Kuksu, but got very similar numbers between the two (close to 70% end in resignation), despite one being 3 hours and the other an obvious quick play at 5 minutes + 5x30s. KBS is televised as well and I've seen players analyze after the KBS games, but they cut out in the middle of that. I'm not sure the Koreans care that much about being embarrassed.

If we speculate that the time limit effect kicks in above 3 hours (i..e., Mickey Mouse is more patient that previously suspected ) it's harder to find fair comparisons. Suggestions?

In my own experience, in very fast games I may be less likely to resign because I don't have time to estimate the score properly when in byoyomi. I doubt this effects pros, though.

[snorri]

I'll have to think about the Chinese rule part. I can't do it directly in Kombilo so it would mean gathering the info and backing it out in excel.

Drago makes it a bit easier, because it gives you some SQL-like queries on top of the Kombilo database.

[snorri]

I hesitate to go into finer detail by tournament. When the numbers get smaller any competent statistician (i.e. not including me) would warn us the results become less and less reliable.

Yes, some care is required. That's why I stopped looking at the win rates in played variations. The numbers quickly get very low and pathological. I'd see these great results on some variations only to look further and find out that the high win rates were mostly because it was Sakata playing them.

[ez4u]

I'll have to think about the Chinese rule part. I can't do it directly in Kombilo so it would mean gathering the info and backing it out in excel.

Drago makes it a bit easier, because it gives you some SQL-like queries on top of the Kombilo database.

I have not used the data functions in Drago yet. Maybe it is time I figured them out. Anyway I excluded the Chinese-rule games (14,300+ games) the old fashioned way and posted a new graph with the earlier ones above. The overall situation is not changed greatly although the lines shift somewhat.