Is the Swiss Pairing System Fair?

[Stewart Reuben]

It is not the Swiss System in itself that is unfair, it is the pairing system used that may, or may not, be unfair.
I have commented on all this many times, particularly in my book 'Chess Organiser's Handbook'.
1. A lottery system is 'fair'. That is, were you to computer generate the pairings randomly perhaps 1000 times, that is the system that would produce the results most closely corresponding to the rating achievemnts of the players. It was the system used in England until 1965 when I introduced the Seeded Pairing System. Eventually that replaced the lottery system. One actual lottery, might produce terribly skewed results.
2. A seeded Swiss System is highly effective at finding the 'correct' winner. Defining that as the player who has the highest Tournament Performance Rating in the event.
As Nick and others point out, it can be horribly biased, though seldom against the winner. One British Championship, 2 players of similar rating were seeded, one just in the top half, the other at the top of the bottom. They had the same colour in round 1. They drew their first 3 games. After 4 rounds the lower rated player had played only against opponents over 2300. The higher rated player had not met a rated opponent.
3. The Dubov System seeks to rectify this. Basically each round it balances of the ratings of the opponents on a given score group. This has been tried in the British and Gibraltar. David Welch didn't like it and it was abandoned. It is heavily dependent on accurate ratings.
4. Burstein System. That relies on Bucholz (Sum of opponents' scores). 8 players in a group. Put them in order of Bucholz and pair 1v8, 2v7, 3v6 and 4v5 - taking into account colours.
5. Accelerated Pairings. These are very effective for keeping apart players of tremendously disparate playing strength and for resolving a huge tournament with few rounds. It is heavily dependent on accurate ratings. We no longer have huge events such as 250 players in 6 rounds. Its poularity has waned.
6. One reason for the popularity of the Seeded Swiss System is that it is computerised efficiently. But that has the effect that younger pairings administrators just use the seeded system blindly. They don't consider alternatives and thus don't take things further. Many, if asked why there has been a particular pairing, will respond, 'Because the computer says so.'
7. If you want, I could explain another system that has never been used anywhere.

All pairing systems, including all-play-alls should be designed so that the players enjoy themselves. That is the prime objective.

No APA in which Magnus Carlen competes, but not Caruana, can possibly be totally 'fair'. He meets a weaker field than the others contestants.

No tournament where a player meets an opponent once can possibly be 'fair'. Each event should be two games against the same opponent, one with white and one with black.

Matches are fair. Knockouts are fair, but only for finding the winner.
Everything else is a compromise.

[Roger de Coverly]

6. One reason for the popularity of the Seeded Swiss System is that it is computerised efficiently

A cross between seeded and random should computerise reasonably well. So you seed one side of the draw and pick the other side randomly or semi randomly. You would still need rules for floats and colours. There are various techniques that enable "random" pairings to be repeatable on someone else's computer. Done manually there were always suggestions of bias.

The British Championships of long ago used something like this in early rounds. If it was totally random, you could potentially have something like Penrose v Clarke in the first round.

[Michael Flatt]

The Theory and Rules of Seeded Swiss Pairings are simple and easy to understand. It is possible for pairings to made without the use of a computer (i.e. using manual pairing cards) and competitors can check the accuracy of them and question why they have a different opponent than expected.

Anyone making Swiss Pairings should be capable of explaining how the draw was made. Computer pairings can be fiddled and should not be trusted purely because they are claimed to be computer generated. It is also possible that the result of one or more games had been recorded incorrectly.

The test of fairness might better be considered to be whether a player is allocated the opponent that he had expected, based on the pairing rules.

[Stewart Reuben]

You can use a computerised seeded Swiss Pairing System to do the pairings randomly. Simply assign the 'seeding' numbers randomly.
Before designing or choosing a pairing system, you should decide what your objectives are.
One might well be: comprehensible to an intelligent 12 year old.

[Kevin Thurlow]

"7. If you want, I could explain another system that has never been used anywhere. "

Sounds intriguing - pray continue.

[Stewart Reuben]

In the 1970s the London Chess Association, at the behest of David Eustace, considered the objectives of a Swiss. That was he first time I had ever come across management by objectives. Abbreviating what can be found on page 42 of Kazic's book.
1. Players should enjoy themselves as much as possible. Pairing together relatives, members of the same club or from the same distant area, or contestants with a difference of about 300 FIDE, (40 BCF) should be avoided. Entrants generally prefer to play, rather than win by default. (That is not true for GMs, or people aiming to win an event.)
2. A player's score should reflect his performance, e.g. similar scores should have similar Bucholz and similar rating performances.
There is no BEST pairing system. One method may suit an event with few players and many rounds, another with many players and few rounds may require a completely different system.
3. A player should be able to pre-determine his opponent, not have to wait until the pairings are published. The arbiter should not have any choice. Thus two separate arbiters would make the same pairings.
I have also written on these matters in 'The Chess Organiser's Handbook'. You can find that material on the ECF website.
I only thought later of
4. The system should be comprehensible to an intelligent 12 year old.
5. Tiebreaks are not used in Britain unlike most other countries. This is because the late Harry Baines and I believed that a player should know what he had to achieve prior to the start of the last round. In about 1980 I said, 'Sonneborn-Berger' is less accurate than tossing a coin.'

In a seeded Swiss there is absolutely no doubt the second highest rated player has an advantage over the highest.

Many years ago Jonathan Mestel said, 'There is nothing wrong with any Swiss System. It is the method of assessing the results that is faulty.
Thus you could do the following. Use a seeded Swiss Pairing System. Then determine the new ratings based on a k=70 factor. (That ensures the rating turns over in 11 rounds. Use 80 for 9 rounds.) Now feed back the new ratings into the system and determine the TPR of each player. Use that for ranking the players. I did investigate that for one British, but it didn't make much difference and have never again explored the matter.

Kevin Thurlow asked for details of a system that has never been used. It is explained in the second edition of my book in greater depth.

In a knockout, if the lower ranked player wins, he takes up the seeding position of his ousted opponent. Using the same concept for a Swiss.

1. 64 players. 1v33, 34v2... in the usual way.
2, Let us suppose 33 beats 1. Then 33 becomes the new number 1 seed. 1 becomes the 33rd seed.
3. Let us suppose 1 and 33 draw. Then 1 becomes 16 seed and 33 18. It may be better if the player who had black gets the number 16 and the player with white, 18.
3. Obviously 2 or more players may have the same seeding number. Then the highest number goes to the person who had the highest ranking in the round just completed.
I have no idea how this would work out. But it certainly hasn't been computerised. It is satisfyingly complex, for those who like complex pairing systems.

Another system that was used only once, I believe, in Birmingham. Again 64 players. 1 v 64, 63 v 2, 3 v 62 and so on. That principle was followed in every round on every score group. It vastly favoured the highest rated players and was never again used.

[Kevin Thurlow]

Thanks Stewart - Obviously I didn't read your book carefully enough!

"4. The system should be comprehensible to an intelligent 12 year old."

All you need then is players that intelligent...