Changing Swiss Pairings

Topic 3 from yesterday's PoB show:

1. 3. Should swiss tournaments use ratings in making pairings?

dpruess took the position that players should be assigned random pairing numbers for the sake of pairings, rather than being in order of rating.

yay! you agree!!

Yes, ratings are necessary for pairings.

Imagine a large club-level swiss tourney (with 1000+ participants) in which Magnus Carlsen and Vladimir Kramnik participate. Let's say that Carlsen and Kramnik are randomly paired with each other in Round 1. Suddenly, one or both of these GMs has a good chance of not taking first place, when theoretically they should have tied for first (won all their games).

While this would give lower rated players a better shot at winning money, I think that the tournament result should reflect the players' performance as opposed to the luck of the draw.

That's why I specified a large club-level tourney.

If there are enough rounds for Carlsen and Kramnik to play each other eventually, then the pairings don't matter at all, since the top players will end up facing against each other regardless of how they are seeded.

But in a really big tournament like the World Open, there are usually not enough rounds for the top players to play each other, and that's where seeding counts.

The current standard Swiss system certainly has its problems...although it has the benefits of an established tradition.  But I wonder if a random-pairing system is much better.

Perhaps the McMahon Pairing system should be looked at for chess.  This system is currently the standard for Go tournaments in the New World and Europe.  Rather than having me explain it imperfectly in my own words, here is a link the KGS Go site with a full explanation.

Amy

In usual Swiss tournaments with pairing by rating, the top players will usually have to try to win the first couple of rounds relatively easily, then around the middle of the tournament, the best players will meet on the top boards, and finally in the last rounds, they will keep facing more or less the strongest opposition.

If pairings were done in a random way, the first boards would not be determined by strength and performance, but also to a large degree by luck in the pairings, because now also players can reach the top by having the luck of being randomly paired with relatively weak opponents. To make such a tournament fair, so that in order to win it, you will need to have faced really strong opposition, you would need more rounds in tournaments with random pairings than in tournaments with sorted pairings.

In large Swiss tournaments, some rounds will have to be used to roughly sort the players by strength, resulting in pairings with great playing strength difference for some rounds, until the later rounds will feature more evenly matched pairings.

In tournaments with random pairings, the number of pairings with great difference in playing strength will be a lot bigger, and it will take much longer until the field has eventually somewhat been sorted according to playing strength and performance, and until all the pairings start to be more even and interesting.

This is why random pairings can only be chosen for small tournaments in which the number of rounds is relatively large compared to the number of participants (for example for little hobby player tournaments in which you do not have ratings for most people), and why all big tournaments I know of use sorted pairings. Saving time (fewer rounds), making it more interesting (less imbalanced pairings) and being much fairer (The winners will have to face the strongest opposition).

Mac Mahon is basically giving the higher rated players a head start that they are allowed to keep. This has the function of doing the sorting of the field not mainly during the first rounds, but even partly before the first round. The assumption seems to be that playing strength in Go seems to have an even bigger significance than in chess.

Accelerated Swiss pairing is similar to Mac Mahon in that it gives the higher rated players a virtual head start which affects only the determination of the pairings of the first rounds. After these rounds, the stronger players will already have been able to play against each other, giving more interesting games (and better norm chances), and the strongest players from the lower rated pool (who have been allowed to score many points against relatively weak opposition during these first rounds) are sure to face enough strong players during the rest of the tournament (in which the pairings are done without the virtual head start points), so they will only remain up there if they really prove themselves against very tough opposition.

So both of these systems have the purpose to accelerate the sorting of the field according to strength. Random Swiss pairing decelerates this sorting.

Cogent analysis, ilmago !

I also think McMahon pairing is the best way to go, particularly for tournaments in which there is a wide range in player abilities and where you don't have enough time for enough rounds to optimally sort the number of players involved.  The major difference between McMahon and accelerated Swiss is that with the McMahon, the point given at the beginning is real and counted to the end of the tournament, while with accelerated Swiss it is taken away after a round or two.  What the McMahon does is basically let you get in an extra round in your tournament without having to take the time to play it, by making the assumption that the first round results are pretty much a given, since the abilities of the players in the Swiss first round are so far apart.  So one advantage is a whole round-worth of better sorting for a large tournament in which you don't have the time to play an optimal number of rounds.  The second advantage is that lower ranked players that don't really have a chance to finish at the top, will have a more pleasant experience in their tournament, not facing a blowout loss in the first round, but having a better chance to get another win against a closer matched player during the day of playing.  Holding the lower half back a point at the start actually gives them all more actual wins on the day.  In both Swiss and McMahon, the matches continue to get more even with each round.  With the McMahon you get a head start at the more even match ups.  At seems that with accelerated Swiss in which you take away the point after a round or two, the more even early pairings created by the acceleration point are immediately followed by one or more blowout rounds when you take away the virtual point.  I don't see any real benefit to that.  In fact, if you set up a sample tournament and compare pairings (assuming that game results end up as expected with the higher rated player winning the first round), you find that the only difference between a regular Swiss and a one round accelerated Swiss is the round in which the big blowout occurs.  The pairings end up identical, except with rounds one and two swapped.  With one round acceleration, the exact blowout pairings you avoided in the first round happen in the second round instead.  Better to let the point be kept as a real one (McMahon pairings) and avoid the biggest mismatches altogether.  The only possible potential drawback could come from a very incorrect initial rating.  But even that will be sorted out, unless a player that was actually the top in ability, but ranked in the bottom half.  Their perfect score would be beaten by a perfect-scoring top-half player, who had the given point plus a perfect score.  But it would be appropriate since the top half perfect score would be against much harder competition than the lower half perfect score.

@dpruess: But doesn't that ruin the whole point of Swiss tournaments? A big part of them is based on built-up suspense, which keeps growing as the top few boards continue to do well. However, randomly assigned pairings could lead to a quick end for even some of the top players, forced to play the best in round one.

that is possible ChessiGood, but i think there will still be suspense as the number of players tied for first dwindles. and to me it's important for competitions like this to put the competitors on somewhat fair ground. for professional players in the U.S. these big swisses are an important part of their work, and it's not fair that higher rated players should get a much easier ride event after event after event.

you'd still be able to do it, using the randomly assigned pairing numbers all participants would have.

I don't agree with numbers. Rating is actually good.

Because then a 2200 would play a 1300 and same for other tables. That wouldn't be fair, would it?

Randomly assigned pairings goes past ridiculous.  I suppose the whole idea of it is that now anyone can win the tournament, instead of, you know, the winner of a tournament being decided by who the best player in that tournament is.  I'm just imagining the World Open being won by someone untitled who got paired against 1800s for seven rounds due to random pairings...  In fact, if we're going to pair randomly, why don't we just abolish the whole rating system and say that we're all winners if we try?  The Swiss system is logically designed to eliminate the weak players from the field early and progress to create the best chances of the strongest player winning the tournament.  Random pairings are an excellent way to ensure that far fewer strong players will have any interest in joining tournaments.  Here's more food for thought: What if we made the seeding random for March Madness, instead of based on record?  That way everyone would have an equal shot instead of the "higher rated teams getting an easier ride".  At the risk of this turning into a bannable rant, I'd say that the main reason chess.com has still not established itself (and possibly never will) as a place conducive to strong chess is that it's built for low level players in every way.  Whether it's ludicrous things like stackable premoves or five minute long bullet games due to poor lag management by the server, insistence that the site can prevent engine usage when the exact opposite is the case as long as they understand how to manipulate t3 (which 95% don't, in all fairness) or the general idea of bending over backward to appease all the class players at the cost of losing the smaller group of less profitable higher level players, the only thing that keeps chess.com able to in any way feign relevance in the chess world is a valuable domain name and widespread advertising.  "Chess.com - where the grandmasters play a few blitz games and write articles if we pay them."

It's more frustrating because this site has a fair amount of good things going for it that another site that I won't name lacks - excellent articles, active forums with a nice community, and strong correspondence chess - so for some reason despite all of the shortcomings I still come back (until I go on a rant like this one and get banned).  Despite knowing that chess.com is a business that cares significantly more about revenue and sheer numbers than trying to get a legitimately strong base of players (not just a few GMs who pop in now and then who are paid by the site) I still always come back here - there's something I really like about the articles, the correspondence, and the atmosphere here despite loathing how the site is managed.  Oh well, I've had my rant, time to go to bed.

@IrrationalTiger -- Does it matter much whether chess.com has many top players on it?  I personally don't care if they're there or not.  I'm not going to be playing them.  I think it's entirely appropriate that the features of this site match up with the playing abilities of its users.  But that's a different topic.

Regarding random pairings vs something like the McMahon system, I think it depends on what is the composition and aim of your tournament.  If you have a small number of fairly evenly matched players or if the ratings are not likely to be accurate, random pairing within point groups should be fine.  For tournaments with many participants of widely varying ability, I think McMahon (and no random pairings) is the way to go.

no. at the beginning of the event, when everyone has 0 points, you'd assign a pairing number to each participant. right now the pairings numbers are just assigned in order of rating. so if there were 6 players with 2.0, they'd have numbers like 1, 3, 9, 13, 27, 28; and if your number was 1, you'd know you were to be paired against 13 (if colors matched) just like in the current pairing system.

Tiger, you're obviously not going to get banned for explaining what you think are the shortcomings of the site.

the idea behind that alternative pairing system is actually to produce a more fair winner, not to randomize the result. a weak player won't win the world open by getting paired with 7 consecutive 1700s unless those 1700s have beaten GMs. i'm no expert on march madness, but i suspect that there is some kind of a "regular season" involved, and the higher seeds are getting an easier ride in recognition of the achievement of their overall season. but each chess tournament is an individual competition, and should thus be fair within itself. that's just my opinion.

i don't quite understand in what way we appease class players at the expense of stronger players. i can tell you that we try to make it a great site for players of all levels. and one of my constant foci is bridging the gap between class players and professional players in order to improve the chess world.

i will ask some people their opinions about whether the pre-moves should be stackable or not. i don't see much else to improve in your rant, but if you have other suggestions, i (or other staff) would always appreciate a message.

imo random pairings are a bad idea. Using ratings for pairing makes it more likely that the strongest player will win which is what tournaments should be about. Random pairings only make it more likely that a weak player will score highly or a strong player badly, and intoduces an element of luck. This is bad imo.

@the anti-chess.com rant

imo chess.com is more appealing to the lower-rated players because- the training software is geared at weaker players, also at higher levels there are more cheaters due to the lack of an option to make a computer account.

also i think premoves should not be stackable but i dont care much either way

I think that rating should definitely be a factor in first round and every other round pairings.  The #1 player should get paired with the guy in the middle the #2 player should gets paired with the (middle - 1) a.s.o.  Here the lowest rated player is expected to loose and get an easier game and the highest rated player to win and get a harder game.  The only time you get a harder game is if you win.  With this systems if you keep winning the games it should get harder and harder as the games progress.  If you think that those with a higher rating get an easier games well this is a misconception.  For instance if your in the top half of the draw you should theoretically get an easier game in the first round than the top board.  In-fact if the games go 'to form' the highest rated player should get the hardest game in every round (except those in the lower half of the draw in round one, the ones in the bottom half of the top half of the draw in round 2 [anyways you'll only get one game harder than the top board]).  The only time you play harder opposition than the top board is when you keep winning when you are expected to loose.  So you should only get harder games than the top board when you're improving.  If you are improving you'll have a higher rating in the next draw, with all the perceived benefits this brings.