Accelerated Pairing Systems

[Jonathan Bryant]

It is highly questionable whether it is the optimum system for the British Championship.

I know nothing whatsover about pairing systems but after Tuesdays 8th round has finished...

Played by Howell but not Jones....
GM Williams (2527, currently 5.5/7)
GM Gormally (2490, 5/7)
IM Littlewood (2358, 5/7)
GM Wells (2498, 4.5/7)
GM Hebden (2468, 4.5/7)


Played by Jones but not Howell
IM Palliser (2413, 5/7)
FM Eggleston (2341, 5/7)
GM Summerscale (2454, 4.5/7)
IM Greet (2443, 4/7)
FM Sowray (2341, 4/7)


To me this is not looking like a system that's working well. Perhaps that's just the way it goes sometimes but if I was Howell I'd be a bit narked by now I think.

[Jonathan Bryant]

Still the most amusing interpretation of the accelerating pairing system always used to be (and probably still is) at the Kings Head Rapidplay ...

I played at the Kings Head Rapidplay a while back.

Reached 2/2 (something I rarely ever did even in the graded sections of Adam Raoof's Golders Green events) and was feeling good until I realised that Mr. BLANK was there and he also had 2/2. Mr. BLANK, I should tell you, is not only not a terribly good chess player he would be strong contender for the title of the worst player who ever lived. A walking contradiction of the claim that practice makes perfect.

... I know nothing about pairing systems but that one certainly seemed to be a bit odd.

[Alex Holowczak]

It is highly questionable whether it is the optimum system for the British Championship.

I know nothing whatsover about pairing systems but after Tuesdays 8th round has finished...

Played by Howell but not Jones....
GM Williams (2527, currently 5.5/7)
GM Gormally (2490, 5/7)
IM Littlewood (2358, 5/7)
GM Wells (2498, 4.5/7)
GM Hebden (2468, 4.5/7)


Played by Jones but not Howell
IM Palliser (2413, 5/7)
FM Eggleston (2341, 5/7)
GM Summerscale (2454, 4.5/7)
IM Greet (2443, 4/7)
FM Sowray (2341, 4/7)


To me this is not looking like a system that's working well. Perhaps that's just the way it goes sometimes but if I was Howell I'd be a bit narked by now I think.

I think the problem is that they played so early. They're both tied at the top, but Howell is seeded higher. Am I right in thinking that therefore Howell plays the first person on the score below them, then the Jones gets the next best after that. As a result, Howell always has the stronger opponents, and will do for the rest of the tournament?

[Jonathan Bryant]

Am I right in thinking that therefore Howell plays the first person on the score below them, then the Jones gets the next best after that.

Dunno. I can see why Howell got Williams for round 8 - couldn't be anybody else - but there were a bunch of people on 5/7. Could have been a lot worse for Jones than Richard Palliser (no disrespect to him).

btw: just to clarify I'm not suggeting the pairings have been 'incorrect' just that they're working out a little harsh for Howell.

Also, I agree with you about the problem being they met early.

[Alex Holowczak]

Am I right in thinking that therefore Howell plays the first person on the score below them, then the Jones gets the next best after that.

Dunno. I can see why Howell got Williams for round 8 - couldn't be anybody else - but there were a bunch of people on 5/7. Could have been a lot worse for Jones than Richard Palliser (no disrespect to him).

btw: just to clarify I'm not suggeting the pairings have been 'incorrect' just that they're working out a little harsh for Howell.

Also, I agree with you about the problem being they met early.

Well, there are 7 players tied on 5/7. Normally, Jones plays the 4th highest-rated of the 7 (Palliser) in his game. 1-3 play against 4-6, at least, as far as possible. I can see how Palliser and Jones are playing one another.

Another person to feel sorry for in the draw is the new GM, Stephen Gordon. Only 1/8 of his games have been against someone not in the top 10. Not quite as bad as Howell's draw, but still none too cheery.

[Stewart Reuben]

E Michael White >Stewart Reuben is correct that a paring system requires an accurate grading system.<

I very much doubt I have ever written that. An accurate pairing system which is based on grading requires a fairly accurate grading system. I think an Accelerated Pairings will work if about 75% of the players are rated. Plain seeded perhaps 60%. This assumes that most of those without grading data are weaker than those graded. Dubov 90%. Burstein 50%.

Jack has it more or less correct. Pair top quarter v second quarter and third against fourth. Then winners from the top half play against each other. Winners from the bottom half play against the best players who did not win in the first round. The remaining players are paired ' normally'.

But say you had 488 players in a 6 round Swiss and wanted no more than one player with 6/6.
Divide the players into 8 groups. 61 in each.
Top 61 play second 61. Say 52 winners.
From the other 362 players let us postulate 160 winners.
From the top percentile you might get 20 winners. The remaining 160 players are meeting opponents considerable stronger than themselves. In my experience this might leave 16 winners.
Round 3 you get 8 winners from the top group. 2 from the remaining group. (Actually I have seldom known even one.)
Round 4 3 winners. None from the remain ing group. For such a large group, I would probably postulated 4 rounds of Acceleration. Sadly I have never had so many. Thus the downfloat would meet the highest with 3.5, not one of the weaker lucky players.
Round 5 at most 2 left with 5/5. Mission accomplished.

I will leave somebody else to try it with 10,000 players. Just have more and more sections. The Evening Standard opens actually had 10 groups to avoid players more than 40 grading points apart meeting in the first round.

Of course if a weaker player keeps on winning, he will meet much higher graded opponents.
In an early Islington I put Les Blackstock down as ungraded by mistake. He was over 200 strength. The system coped.
Stewart Reuben

[Stewart Reuben]

Concerning the fact that David Howell has met significantly stronger opposition than Gawain Jones.
Dubov is the only system I know of which tries to balance off players so that they get to meet similar opposition. It has been tried in England, but wasn't liked except by me.
I have often observed that seeded system of any type always favour the second highest rated player.
I presume though you do realise David has an advantage in the tournament, and this applies to all tournaments for the highest rated, particularly round robins. Since he cannot play himself, he has a built-in advantage.
Stewart Reuben

[Jonathan Bryant]

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

So I'm now emperor of the universe and I've decided that we want a system for the British Championship that will give the best chance of avoiding a situation where somebody who finishes in the top 2 or 3 has played (on average) much stronger or weaker opposition than the other 1 or 2 people at the head of the field.

So what's the best system now assuming 11 rounds and a similar number of players as used to enter before I took over my new position?

Also, would it matter if there was a rating threshold to pass before somebody was able to enter the main tournament (say 2250 to pick a number at random)?



oops - posted all that before seeing you're last post.

I had thought of Howell's advantage as you explain it although since that seems to come down to he has an advantage because he's the best player that seems kind of fair!

[E Michael White]

But say you had 488 players in a 6 round Swiss........ Divide the players into 8 groups. 61 in each. Top 61 play second 61. Say 52 winners

Lets assume an open grades 2000 - 2800. So players packed approx 6 per 10 FIDE points.Thats 100 points per group so players play against oppos 100 points lower. Thats not too different from my suggestion of oppos 100- 200. You estimate 52 wins 9 draws. I suggest more like 24 draws and 37 wins. I dont have the stats just samples off a database.

The remainder of your example also is optimistic about the draw rate. I know this might seem to make the chance of 6/6 by 2 or more, less likely but I did say

ordering the middle of the tournament table closely

ie instead of the usual middle order swiss mixup.

For a very large event you run the risk of tightly packed groups which increases the chance of a multiple tie for first place as the draw rate becomes too high and you wont have enough rounds left for non-acceleration. This will look very strange when combined with a confused middle order.

The patterns you prefer suggest that you probably started with a double colour in the Mastermind game from about 20 years ago with coloured pegs if you ever saw that. That strategy also loses information with each turn.

[E Michael White]

Of course the reason for seeking a pairing which aims for a 33% drawn game rate is that it leaves a 66% decisive result rate. Decisive results in the right % separate out the competitors to show the right order and the winners but too many wins, as other have pointed out, give no additional information as many of them would be entirely expected. The accelerated pairing systems in use at present do this in an informal way which does not usually reach the optimum.

[Stewart Reuben]

I rather think an 800 point spread is inadequate for a huge number of players in order to employ a super-accelerated pairing system. Evening Standard opens had something like 2600 to 1200. Probably 400 is the minimum for normal quartered Acceleration.
33% draw rate in Swisses is very high even for a relatively small rating band.

Jonathan Bryant >So I'm now emperor of the universe and I've decided that we want a system for the British Championship that will give the best chance of avoiding a situation where somebody who finishes in the top 2 or 3 has played (on average) much stronger or weaker opposition than the other 1 or 2 people at the head of the field.
So what's the best system now assuming 11 rounds and a similar number of players as used to enter before I took over my new position?<

Dubov is the best to avoid that particular problem. By the way, if David and Gawain continue to be on the leading scores throughout, the bias against David will continue to the end of the tournament.

>Also, would it matter if there was a rating threshold to pass before somebody was able to enter the main tournament (say 2250 to pick a number at random)?<

That might even exacerbate this particular problem. Accelerated Pairings would lose most of its value.

I feel an aphorism coming on.
All Swiss pairings systems are defective but some are more defective than others.

Jonthan Mestel partially disagrees. He wrote a long article in the BCM many years ago. He postulated that no pairing system is necessarily better or worse than another. The problem lies in how we assign the places. The most obvious solution is to use the TPR (tournament perfomance rating) to decide the position. That is not good enough because a player might have preformed particularly well, or badly. Thus use the rating at the end of the tournament and iterate the TPR based on that.

Imagine trying to explain that to a member of the press, and players who would have to await the final computation before knowing where they stood.

I believe all rating prize should be determined from the TPR, not the raw score. That is how it is done for the British, Hastings and Gibraltar (where there is in 2010 a £3000 rating prize for players 2500-2599). One year Xu Chen got £2000 rating prize with a point less than somebody else in the same rating band.

Stewart Reuben

[E Michael White]

In the British so far the draw rate is about 25%. If the grading system is to be used to determine the pairing, its not the actual draw rate, its the expected draws suggested by the grading system that counts. Tournaments are won by good performances that turn expected draws into wins. If too many of the expected results in the accelerated period are wins/losses not enough information is generated to split competitors. An extreme example would be if Howell and Jones played the 22 lowest rated players between them.

[Stewart Reuben]

E Michael >An extreme example would be if Howell and Jones played the 22 lowest rated players between them.<
I think, if the scenario you suggested had worked out like that, it would be the end of the Accelerated System!

I diagree with you about theory and practice. The Fractional Accelerated System to give it its correct FIDE name, to distinguish it for the Bonus Point Accelerated System, is based on observed results, not theory. Its employment depends on the prejudices of the controller (including me) and the players. the same is true of other systems.

For example Almog Burstein is biased in favour of his system. That uses Bucholz after the first 3-4 rounds. If there are 8 players, arrange them in order of Bucholz. Then 1-8, 2-7, 3-6, 4-5. That too will probably avoid the Howell, Jones bias of the 2009 British. The problem is I fear that it is too late, generally bias may have already started in rounds 2-3. So he used Accelerated for the first 2 rounds in Dresden for the Olympiad. But it should have been 3 rounds and the results were all over the place.

Stewart Reuben

[E Michael White]

I disagree with you about theory and practice

The practice is non-existent for my suggestion but the science of information theory, upon which it is based is well established since about 1952. I have noticed that my method would allow norm seekers to be paired higher up within a score group where possible without devaluing the approach; I dont think other systems can cater for the needs of norm seekers as well as separating potential prize winners.

A possible explanation for the existence of so many alternative Swiss draw methods is that the original system of top half v bottom half is probably the most inefficient at establishing comparators in a 3 result (W/L/D) game.

I dont think we will agree at this time so we dont need further discussion.

[Stewart Reuben]

But there is the point that my system has never been computerised. Thus simulations are impossible. Of course one would have to establish what objectives to look for.
Stewart