Tournament Structures

Discuss anything you like about chess related matters in this forum.
Post Reply
Roger de Coverly
Posts: 17941
Joined: Tue Apr 15, 2008 2:51 pm

Re: Tournament Structures

Post by Roger de Coverly » Sun Apr 08, 2018 10:33 pm

Andrew Zigmond wrote:
Sun Apr 08, 2018 10:12 pm
I've generally been in pretty rotten form this season but the one exception was Hull Congress where I somehow managed to score 4/5 despite being the third lowest rated player in the section.
Beating players graded 20 points higher isn't that exceptional.

I think my personal record is something like +70, but that's playing in Opens where the mismatches and potential for upsets is greater.

Andrew Zigmond
Posts: 1682
Joined: Tue Nov 22, 2011 9:23 pm
Location: Harrogate

Re: Tournament Structures

Post by Andrew Zigmond » Sun Apr 08, 2018 10:46 pm

Roger de Coverly wrote:
Sun Apr 08, 2018 10:33 pm
Andrew Zigmond wrote:
Sun Apr 08, 2018 10:12 pm
I've generally been in pretty rotten form this season but the one exception was Hull Congress where I somehow managed to score 4/5 despite being the third lowest rated player in the section.
Beating players graded 20 points higher isn't that exceptional.

I think my personal record is something like +70, but that's playing in Opens where the mismatches and potential for upsets is greater.
Well it isn't but normally I go the other way (start near the top on seeding but drop like a stone into the vanguard). My point was that if the tournament was run on the McMahon system I would have remained in a lower band. The swiss system allowed me to punch (modestly) above my weight and be part of the overall fray.
Controller - Yorkshire League
Chairman - Harrogate Chess Club
All views expressed entirely my own

Alison Bexfield
Posts: 22
Joined: Mon Nov 02, 2015 12:30 pm

Re: Tournament Structures

Post by Alison Bexfield » Sun Apr 08, 2018 11:06 pm

[quote Andrew Zigmond"
I've generally been in pretty rotten form this season but the one exception was Hull Congress where I somehow managed to score 4/5 despite being the third lowest rated player in the section.

My point was that if the tournament was run on the McMahon system I would have remained in a lower band. The swiss system allowed me to punch (modestly) above my weight and be part of the overall fray.
[/quote]

Andrew
I am not quite sure you understand the McMahon system. I think you are referring to entering an intermediate graded section at Hull which had a grade limit of 140? In a McMahon system if you were winning lots of games you end up playing stronger players - you are not stuck in a band. So if you won your first three games, under Mcmahon you could potentially have been playing people from the section above with grades over 140 (as there would be no separate sections under Mcmahon).

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Sun Apr 08, 2018 11:10 pm

I should mention at this point the real advantage of Swiss: it maximises information gained per result.

If you want to find out who is the best player out of a field of 32, there's no point pairing someone against the five worst players in the field. If they score 5/5, what information does that give you? It tells you that they're somewhere between the best and the sixth-worst in that field. Basically, no information at all.

Whereas if they score 5/5 in a Swiss, you can say with 95% confidence that they're actually the best player in the field.

I think it's been mathematically proven that you can't improve on the information-per-result of Swiss.
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

Paul Dargan
Posts: 512
Joined: Sun May 13, 2007 11:23 pm

Re: Tournament Structures

Post by Paul Dargan » Sun Apr 08, 2018 11:16 pm

@Chris - while swiss systems are good at finding winners, they are not good at finding other rankings - there is a huge correlation between final position and result in the last round. Your score in a swiss is typically a less good indicator of relative performance than rating performance - but obviously rating performnce is harder to visualise for players and calculate on the fly.

In the past many European opens used to split the prize pool in half and award one half for your score position and one half for your Buchholz score.

Paul

NickFaulks
Posts: 4910
Joined: Sat Jan 02, 2010 1:28 pm

Re: Tournament Structures

Post by NickFaulks » Sun Apr 08, 2018 11:17 pm

Chris Goodall wrote:
Sun Apr 08, 2018 11:10 pm
Whereas if they score 5/5 in a Swiss, you can say with 95% confidence that they're actually the best player in the field.
Words fail me.

Andrew Zigmond
Posts: 1682
Joined: Tue Nov 22, 2011 9:23 pm
Location: Harrogate

Re: Tournament Structures

Post by Andrew Zigmond » Sun Apr 08, 2018 11:45 pm

Alison Bexfield wrote:
Sun Apr 08, 2018 11:06 pm

Andrew
I am not quite sure you understand the McMahon system. I think you are referring to entering an intermediate graded section at Hull which had a grade limit of 140? In a McMahon system if you were winning lots of games you end up playing stronger players - you are not stuck in a band. So if you won your first three games, under Mcmahon you could potentially have been playing people from the section above with grades over 140 (as there would be no separate sections under Mcmahon).
Alison - apologies, I was misunderstanding and was still in the mindset of narrow swiss style bands.

I think it does come back to the issue of there being no perfect tournament structure. It's also perhaps worth noting that all weekenders were run on McMahon (and assuming the same general entrants) you will ultimately still have the outliers at either end (the one GM at the top and a couple of beginners at the bottom).
Controller - Yorkshire League
Chairman - Harrogate Chess Club
All views expressed entirely my own

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Mon Apr 09, 2018 12:13 am

Paul Dargan wrote:
Sun Apr 08, 2018 11:16 pm
@Chris - while swiss systems are good at finding winners, they are not good at finding other rankings - there is a huge correlation between final position and result in the last round. Your score in a swiss is typically a less good indicator of relative performance than rating performance - but obviously rating performnce is harder to visualise for players and calculate on the fly.

In the past many European opens used to split the prize pool in half and award one half for your score position and one half for your Buchholz score.

Paul
I've been told that you need 2 extra rounds for every additional place you want to be accurate? So a 32 player Swiss that takes 5 rounds to find a winner, takes 7 rounds to find an accurate 2nd place.

I didn't think Buchholz took the grades of your opponents into account?

I guess it's impossible to award prizes based on grading performance in a way that isn't pairing-sensitive, by which I mean that you could lose out on the possibility of a prize because your last round opponent isn't graded high enough and you can only beat him 1-0.
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

E Michael White
Posts: 1327
Joined: Fri Jun 01, 2007 6:31 pm

Re: Tournament Structures

Post by E Michael White » Mon Apr 09, 2018 12:19 am

Chris Goodall wrote:
Sun Apr 08, 2018 11:10 pm
I should mention at this point the real advantage of Swiss: it maximises information gained per result.
I disagree with that statement if you mean with standard pairing methods.

Chris Goodall wrote:
Sun Apr 08, 2018 11:10 pm
I think it's been mathematically proven that you can't improve on the information-per-result of Swiss.
I doubt that one too.

The science of Information Theory states that the information contained in a response is inversely proportional to the probability of receiving that response. Most approved pairing methods make too little use of the tri-state result that a game of chess produces. So the best pairing methods should produce roughly equal numbers for expected numbers of higher rated player wins, lower rated player wins and draws giving maximum opportunity for results against expectation. That is assuming stronger and weaker players get a fair proportion Ws and Bs and we know that that doesn't quite work with approved pairing methods.

Roger de Coverly
Posts: 17941
Joined: Tue Apr 15, 2008 2:51 pm

Re: Tournament Structures

Post by Roger de Coverly » Mon Apr 09, 2018 12:23 am

Chris Goodall wrote:
Mon Apr 09, 2018 12:13 am
I've been told that you need 2 extra rounds for every additional place you want to be accurate? So a 32 player Swiss that takes 5 rounds to find a winner, takes 7 rounds to find an accurate 2nd place.
That's a traditional belief perpetuated by nearly forgotten figures such as W Ritson Morry. I'm not sure I've ever seen a convincing proof of the assertion. In addition, whether pairings are random or seeded should have some bearing.

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Mon Apr 09, 2018 12:33 am

E Michael White wrote:
Mon Apr 09, 2018 12:19 am
Chris Goodall wrote:
Sun Apr 08, 2018 11:10 pm
I should mention at this point the real advantage of Swiss: it maximises information gained per result.
I disagree with that statement if you mean with standard pairing methods.

Chris Goodall wrote:
Sun Apr 08, 2018 11:10 pm
I think it's been mathematically proven that you can't improve on the information-per-result of Swiss.
I doubt that one too.

The science of Information Theory states that the information contained in a response is inversely proportional to the probability of receiving that response. Most approved pairing methods make too little use of the tri-state result that a game of chess produces. So the best pairing methods should produce roughly equal numbers for expected numbers of higher rated player wins, lower rated player wins and draws giving maximum opportunity for results against expectation. That is assuming stronger and weaker players get a fair proportion Ws and Bs and we know that that doesn't quite work with approved pairing methods.
Are you maximising potential information there, or expected information? It's easy to maximise potential information, just choose the pairings with the largest grading difference regardless of score, but since the better players will probably win, the actual results you'll get will be low in information.
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

Paul Dargan
Posts: 512
Joined: Sun May 13, 2007 11:23 pm

Re: Tournament Structures

Post by Paul Dargan » Mon Apr 09, 2018 12:35 am

There's some heavy-weight research on whether swiss or random pairings are more likely to result in the "strongest" participant winning an event and also on the "top k" players really being the k strongest players - it was from Wharton I think, while since Ilooked at it. You need better than A-level stats and an understanindg of Spearman rank correlation co-efficients, etc.

I have a feeling that GM Smerdon (a serious aceademic economist, when not playing/writing about dodgy Bg4 Scandi lines) has written about the efficiency (or otherwise) of swiss systems... perhaps on chess.com?

Paul

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Mon Apr 09, 2018 12:37 am

Roger de Coverly wrote:
Mon Apr 09, 2018 12:23 am
Chris Goodall wrote:
Mon Apr 09, 2018 12:13 am
I've been told that you need 2 extra rounds for every additional place you want to be accurate? So a 32 player Swiss that takes 5 rounds to find a winner, takes 7 rounds to find an accurate 2nd place.
That's a traditional belief perpetuated by nearly forgotten figures such as W Ritson Morry. I'm not sure I've ever seen a convincing proof of the assertion. In addition, whether pairings are random or seeded should have some bearing.
Does it have to be proven theoretically or would you be convinced by 100,000 simulations?
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

User avatar
Chris Goodall
Posts: 331
Joined: Sun Oct 10, 2010 6:40 pm
Contact:

Re: Tournament Structures

Post by Chris Goodall » Mon Apr 09, 2018 11:11 pm

Paul Dargan wrote:
Mon Apr 09, 2018 12:35 am
There's some heavy-weight research on whether swiss or random pairings are more likely to result in the "strongest" participant winning an event and also on the "top k" players really being the k strongest players - it was from Wharton I think, while since Ilooked at it. You need better than A-level stats and an understanindg of Spearman rank correlation co-efficients, etc.

I have a feeling that GM Smerdon (a serious aceademic economist, when not playing/writing about dodgy Bg4 Scandi lines) has written about the efficiency (or otherwise) of swiss systems... perhaps on chess.com?

Paul
Found it:
https://repository.upenn.edu/cgi/viewco ... h_scholars

Thanks for the pointer! The Swiss system they've considered isn't what I would call a Swiss system though, because it pairs randomly for the first two rounds. Also they've used an unbounded exponential function for the distribution of strengths, which you would never get in chess. Also we know from Elo's original research that chess results are distributed on a logistic curve and not the straight line that you get from Bradley-Terry. But for a spherical cow in a vacuum...

Actually, the big advance here is in the modelling of Swiss pairings, which has always been too hard to simulate quickly enough to run thousands of simulations, but now possibly isn't any more.
Chris Goodall, formerly known as Chris Wardle. ECF Grader for the ancient kingdom of Bernicia (or Northumberland and Durham, if you prefer).
Newcastle is not in Scotland!

Roger de Coverly
Posts: 17941
Joined: Tue Apr 15, 2008 2:51 pm

Re: Tournament Structures

Post by Roger de Coverly » Mon Apr 09, 2018 11:37 pm

I don't think he did his research terribly well .
There are many variations on the general structure of the Swiss tournament, but the key idea is that the first few rounds are randomly paired, and the remaining rounds are power-matched.
That view was obsolete fifty years ago when Stewart Reuben and others devised any number of deterministic systems for making pairings based on the rankings of players.

What they didn't do was follow up on a system devised for the British Championship whereby they ranked the top half of the draw in the first round, but randomly assigned their opponents from the second half. That was how, for example, in 1970, the defending champion got a pairing against the future 1972 champion and lost.

It's a necessity that pairings can be reproduced to avoid accusations of bias. Given the existence of pseudo-random numbers, that condition could still apply even where pairings were partly determined by lot.

Given the discrepancies between rankings determined by ECF grades and those by FIDE rating, arguably a tournament using FIDE ranking orders has already been partially randomised.

Post Reply