Accelerated Pairing Systems

[Richard Bates]

Most players i think believe it works even worse at Hastings (my only recent experience). It takes several rounds for the effects of 3rd/4th quarter players having more points than 2nd quarter players after round 1 to work its way out of the system and actually tends to favour norm seekers from the 3rd quarter (who are generally long shots) at the expense of 'expected' norm seekers in the second quarter (unless they have an exceptional start - even a good result in round 1 is undermined by weaker games subsequently if one starts dropping too many 1/2 (or full) points).

The reason is because the effect is not restricted to first round winners. After round 1 you have too many 3rd quarter players on 1/1. After round 2 you will find some of these sneaking through to 1.5 or 2, and meanwhile a lot of "4th quarter" players will get to 1/2 through getting easier games than expected. (In a "perfect" round 1 scenario the second round will see the 4th quarter playing games exclusively among themselves). The effects continue to be felt for 4 or 5 rounds at least, which in a 9 round tournament is way too long.

Still the most amusing interpretation of the accelerating pairing system always used to be (and probably still is) at the Kings Head Rapidplay, where they accelerated for one round only! Creating the bizarre scenario of the first quarter playing the second qtr in round 1 and the the first quarter playing the 3rd quarter in round 2

[Simon Spivack]

Still the most amusing interpretation of the accelerating pairing system always used to be (and probably still is) at the Kings Head Rapidplay, where they accelerated for one round only! Creating the bizarre scenario of the first quarter playing the second qtr in round 1 and the the first quarter playing the 3rd quarter in round 2

If Richard wants I can pass on his comments about the pairings to Colin M. Colin does listen.

A non sequitur: this may be the moment to mention that the plan is to hold the Kings Head Rapidplay at a new venue. One has been found which looks promising. This was discussed at the club's AGM a few weeks ago. Last year it was necessary to cancel when the landlord drastically raised the rent.

[Richard Bates]

Last time i played in it i won it, so i'm not sure i'd want to rock the boat...

(there may of course have been a cunning idea behind it which i failed to spot - eg. creating a larger than expected number of draws in round 1 thereby removing the possibility of multiple winners at which point the acceleration was abandoned)

[Peter Constantinou]

There is a serious chance that he could have played all the other contenders by round 9 and a couple of other will have KO'ed themselves, so he might end up with relatively weak opposition in the last couple of rounds, which could lead to an anti-climactic finish. Why not have a couple of rounds of mismatches to start off with and build to a crescendo.

If we're considering just an entertainment perspective, then I think David VS Goliath mismatches in the first couple of rounds have some intrinsic value, as well as saving the big GM fights for an exciting finish. Everyman vs Grandmaster is always good fun, especially if the amateur wins a game every now and then!

[Mike Gunn]

I think there is a strong argument for using random/lottery swiss pairings rather than seeded ones. There is an argument that seeded pairings always discriminate against the same group of players, e.g. those at the top of the bottom half. These players will be regularly unlucky in their pairings in every tournament they play. If instead you use the lottery system then you may be "unlucky" in one tournament but there is a chance you will be "lucky" in the next one.

[IM Jack Rudd]

The big problem with random pairings is that nobody can then replicate the pairings; if you have a deterministic method, someone can challenge the draw and say "this is what the draw should have been", which is not the case with random pairings. Thus, random pairings can be manipulated to an extent that deterministic pairings can't.

[Roger de Coverly]

I think there is a strong argument for using random/lottery swiss pairings rather than seeded ones

Stewart Reuben once suggested that if you wanted lottery pairings then you should do a random draw for seeding and then apply "normal" seeded system rules. Hopefully you would use a seeding system which minimised the "maybes" of arbiter discretion. In fact just use a computerised system.

There was a thread some while ago about a congress in Yorkshire where in former years, the random pairings always seemed to pair the visiting IMs/GMs early on, thus giving local players an easier run to the prize money. So I think you would want lottery pairings to be done by a program. Whether you would really want the first round of the British to throw up, say Conquest v Howell and Jones v someone at the back of the room in the first round remains to be seen.

[IM Jack Rudd]

There is an argument that seeded pairings always discriminate against the same group of players, e.g. those at the top of the bottom half. These players will be regularly unlucky in their pairings in every tournament they play.

That's only the case if they're regularly seeded in the same place in every tournament they play; this is an argument for non-uniform sets of grade boundaries.

[Richard Bates]

There is an argument that seeded pairings always discriminate against the same group of players, e.g. those at the top of the bottom half. These players will be regularly unlucky in their pairings in every tournament they play.

That's only the case if they're regularly seeded in the same place in every tournament they play; this is an argument for non-uniform sets of grade boundaries.

I think Mike means "those at the top of the bottom half" as a generic group, not that the individual players in that group are always the same. In the same way as it's not worth bothering to vote, coz the Govt always gets in.

[Mike Gunn]

If tournament organisers were far-sighted enough to have oscillating (from year to year) grade boundaries then Jack would be right but I see no sign of this. Stage 1 of my annual club championship is a 7 round Swiss (with accelerated Swiss pairings) has roughly the same players each year (and guess who is always on the border of the top and bottom halves).

[Stewart Reuben]

Swiss Pairing Systems have advantages and disadvantages. There is some discussion of this in my 'Chess Organiser's Handbook'.
All tournaments with an odd number of rounds favour players who get white in round 1. They score about 4.7 to 4.3 in 9 rounds over another player with the same rating. All tournaments with an even number of rounds will be likely to result in some players having plus two of one colour. The only way to eradicate that is for each player to meet the same opponent twice in a round. I have only known that to be done for blitz tournaments. I would like to see it for a FIDE Rated Swiss, but it would be hard to get title norms - but easy to get some FIDE money.

Seeded top half v second half on all score groups throughout is definitely biased against the players at the top of the second half.

Accelerated Pairings require Gaussian distribution of ratings to be at its most effective. Evening Standard London Chess Championship in about 1980 had 250 players. After 4 rounds there was only one person with 4/4. Bojan Kurajica won the tournament with 5.5/6 and met the strongest opposition. Mission accomplished.
It is highly questionable whether it is the optimum system for the British Championship. But any discussion needs the parameters first to be set.
Stewart Reuben

[Stewart Reuben]

Swiss Pairing Systems have advantages and disadvantages. There is some discussion of this in my 'Chess Organiser's Handbook'.
All tournaments with an odd number of rounds favour players who get white in round 1. They score about 4.7 to 4.3 in 9 rounds over another player with the same rating. All tournaments with an even number of rounds will be likely to result in some players having plus two of one colour. The only way to eradicate that is for each player to meet the same opponent twice in a round. I have only known that to be done for blitz tournaments. I would like to see it for a FIDE Rated Swiss, but it would be hard to get title norms - but easy to get some FIDE money.

Seeded top half v second half on all score groups throughout is definitely biased against the players at the top of the second half.

Accelerated Pairings require Gaussian distribution of ratings to be at its most effective. Evening Standard London Chess Championship in about 1980 had 250 players. After 4 rounds there was only one person with 4/4. Bojan Kurajica won the tournament with 5.5/6 and met the strongest opposition. Mission accomplished.
It is highly questionable whether it is the optimum system for the British Championship. But any discussion needs the parameters first to be set.
Stewart Reuben

[Alex Holowczak]

Accelerated Pairings require Gaussian distribution of ratings to be at its most effective. Evening Standard London Chess Championship in about 1980 had 250 players. After 4 rounds there was only one person with 4/4. Bojan Kurajica won the tournament with 5.5/6 and met the strongest opposition. Mission accomplished.
It is highly questionable whether it is the optimum system for the British Championship. But any discussion needs the parameters first to be set.
Stewart Reuben

I suppose that needed it though. log 250 / log 2 = 7.96ish. For a normal system to work, you'd need 8 rounds. So in that situation, the acceleration was justified. The British Championship doesn't need the acceleration though, because it has plenty of rounds to make the top players play each other.

[Stewart Reuben]

I should have mentioned Accelerated Pairings also require accurate ratings/grades.
I should have written: But any discussion needs the objectives first to be set.
I well remember David Eustace leading the discussion on the London Chess Association Pairing System in the 70s. You can find that in the Kazic book, 'Chess Competitor's Handbook'.
I believe Acceleration was first used, quite recently, for the British because a seeded System top half against second half was not coping to enable people to get title norms. I was not at that time involved. Whether it is still necessary for that purpose, I don't know.
Another reason for using Acceleration is that it avoids this bias against the top of the second half, as happened for example to Keith Arkell in the 1984 British. A later British had 2 players of very similar rating. One was at the bottom of the top half. the other at the top of the second half. They both drew all four first games. The first met no rated players at all. The second opponents all over 2300. Which of these players was disadvantaged?

Few of you will know there is another Accelerated System devised by Phil Haley of Canada before me. Mine only first existed because I did not understand the explanation I was given.
Of course I believe the system used in Britain is better. But I am in no way biased!

Other systems recognised by FIDE include Burstein (used for the Olympiad) and Dubov.

Stewart Reuben

[E Michael White]

Stewart Reuben is correct that a paring system requires an accurate grading system. At the present time that presents two problems firstly both FIDE and ECF are rapidly losing cred and secondly even if they were neither deflating nor inflating they wouldn’t give an accurate estimate of who might win.

I think Jack Rudd described the process at Hastings on this forum here for accelerated pairings. From what I remember the justification stated elsewhere is not based on mathematical reasoning but along the lines of something like “..the losers of the first round top half wipe out the winners of the bottom half in round 2 etc.” Such a statement requires 100% wins by players graded maybe 200 FIDE points higher than their oppos. Not even the grading system assumes those results.

The optimum pairing method, which can only be a theoretical goal, for establishing a sole winner and ordering the middle of the tournament table closely is to use the grading system to ensure that the number of players expected to win, lose or draw within each score group for each round is roughly equal. Which means arranging for 1/3rd of the games to be drawn ! I haven’t checked the frequencies of draws but I suspect that pairing players in the same score groups where possible against players between 100-200 FIDE points apart might be the best bet. My whole database has a 30% draw rate for all games which possibly rules out random draws within score groups but for a sample within 0 – 100 points the draw rate is too high at 49%.

This system has some justification in Information Theory which roughly says to obtain maximum information, questions should be asked where the possible replies have near equal frequency. For pairings the arbiter asks questions which have 3 possible answers W/D/L. If it were possible to arrange a draw in this way a 6 round tournament could be expected to establish a sole winner from 486 players.