A pairings philosophy

Running Chess Tournaments
1

After just a few minutes of training, the novice TD can correctly pair the first round of any event relatively quickly.

We all know the formula: Top half vs. bottom half, and within that specification the highest from each of those groups plays each other, the next highest play each other, all the way down to the bottom, with perhaps the lowest rated already having received a bye. Colors, of course, alternated, all the way down.

That is chess. Interestingly enough, that is not playoff basketball (pro or college), pro football, hockey, pro baseball, tiddlywinks, or Missouri bullfrog racing. These have a process where in any given round the top seed is playing what at least is theoretically supposed to be the lowest seed possible.

Why the descrepancy? Did chess in its wisdom find a better way? Or it is just that we like to be different? Or everyone could care less, but one might as well stick with what is tradition? I should like to know who invented the first-round pairing system and what advantages it contains.

One of the big negatives (of course I was leading up to that, wasn’t I!?) that I can see is the fact that it is so stacked against the top seeds. The system is not arbitrary or fair in that they consistently face better competetion than those seeded within a few slots below them. They are penalized with a harder opponent because their rating is higher.

If you must be unfair, it is probably better to reward their chess excellence with easier pairings. However, I have an idea, that would eliminate this discrimination altogether. Simply put, while still within the context of top half vs. bottom half, have the opponents paired at random.

Currently within, say a 16-player group, the first round would sport 1-9, 10-2, 3-11, 12-4, etc. Again, this really unfairly favors seed #8, and unfairly gives detriment to seed #1. If this random pairing process would occur, seed #1 may still get seed #9 by luck of the draw, but he may get seed #15, or someone else entirely.

A further refinement would be to have the top half players due white vs the bottom half players due black, and vice versa. Whenever applicable, the same process could be applied to other large score groups in successive rounds.

The positives, as I see them, and feel free to add to the list are:
*greater sense of equality, rather than the current rating-based determinism (by far the most important reason)
*more excitement as the top seeds are more likely to face each other in the final rounds
*better chances for the top players of the bottom half to score since they may not have to play the local master who is a zillion points higher than everyone else
*the new and exciting you-never-know who-you’re-gonna-get mentality
*greatly increased flexbility in pairing as you don’t have to constrain the pairings of color histories, etc. within the rating limits of today’s pairing rules
*for better or for worse, worse competetion means statistically worse tie-breaks for the higher rated player, (or from the other point of view) but now the lower rated seeds have a chance to catch up in this category.

The negatives, as I see them, and feel free to add to the list are:
*less early round excitement, as the results may become more predictable on the top boards
*computer programs don’t currently have this option; it would have to be paired by hand in the meantime
*a non-random, more predictable system allows players to prepare for an obvious pairing earlier (currently they can go home on day one of a tournament earlier at night and prepare, once they see what the “forced” pairing would be)
*rating-based deterministic pairings is our tradition, and people, in general, are uncomfortable with changes

I’m probably not doing full justice to the lists but this can of worms seems like it has a lot of merit. Unless I can be persuaded otherwise, next time I organize a smallish swiss, I’d probably advertise this as my pairing method, to try it out. Maybe someone else reading this would be so kind as to test-drive it for me in your weekly club tournament!?

I know the Swiss pairings system is somewhat of a sacred cow among chessaholics, but please try to come to this question as objectively as possible. I await your feedback :slight_smile:
Ben Bentrup

2

The difference is that chess is not just a sporting event. It is also about producing good games. The reason for the top-bottom pairing structure is to produce pairings between playerrs of as close to equal strength as possible, and thus increase the chance of well-played games.

In addition, with few exceptions, the difference in playing strength between the top and middle player in a tournament is already so large that most first-round games are gross mismatches. This is not the case in, say, college basketball. Would the NCAA system work as well if any five guys who just learned how to throw the ball could enter a team?

3

I don’t care for the idea of randomized pairings, but I think your original idea of “folding” the pairing list (first vs last, second vs second-last, etc) has a lot of merit – provided that it’s done in every round (not just the first) in every score group, and provided that minor transpositions are allowed to improve colors, etc.

The problem with the current system is that there is a big discontinuity at the mid-point of the score group. If player X is rated 5 points higher than player Y, and both are near the middle, then X might face a beginner while Y has to play the local grandmaster. This also brings up arguments concerning transpositions across the center line (we had such an argument a few months ago). Folding the pairings would avoid all this.

The current system (at least in a single-section event) results in forgone conclusions in round 1. Folding the pairings would mean that, at least near the middle, the matchups would be better. Near the extremes, I suppose, the matchups would be worse, but how much worse is a 1000-point difference than a 500-point difference?

This is another example of an idea so good that it has no chance of ever being adopted.

Bill Smythe

4

Allow me to respectfully disagree with your “few exceptions.” For example in recent years, I’ve usually been able to slide into an U2000 section of a tournament as one of the top seeds where say there is also some kind of novice section…U1600, U1400. I would say that is almost invariably where I would have to face an upper 1700 to low 1800 player in the first round. This is certainly a winnable, but by no means, an easy match-up statistically. Playing the junior that is playing up, or even some random player nearer the bottom of the list would be much preferable than playing the top player(s) of the bottom half. If the rating differences are as large as you proport them to be, then it is no fault of the pairings, but probably an indication to the organizer to create a separate section next time around.

Howsoever, I did think of a huge negative to this pairing philosophy that basically must kill it. Namely, however much you randomize those pairings as I would have liked to do, you can never guarantee randomness. You post the pairings that you either pulled randomly out of a hat (or even had a computer do), and then someone accuses the TD of cheating for a friend (imagine the playing TD giving himself an easy pairing, even if it was completely by luck), being bribed or whatever. As ludicrous as the claim may be, there is absolutely nothing the TD can do to disprove the claim either. The lone exception is to do every random pairing by having unbiased players pick names out of a hat, which is laborious, time consuming and therefore impossible.

That leaves us with going bact to the rating-based predicated pairings. Now, as per my original argument, and further developed by Bill, we could still choose to adopt top seed vs. low seed, second best vs. second worst, etc. while still open to minor adjustments based on color, team, and player history. This system still doesn’t make random and thus “fairest” pairings, but it doesn’t punish players for having a high rating (if anything, I think people would mostly agree that they should be rewarded). This system is, for the most part, the sports way, and disagreeing with rfeditor when he says: "The difference is that chess is not just a sporting event. It is also about producing good games. " I say that good games are produced when higher rated players more predictably play each other in final rounds. However, in an 1600-2000 section (for example), you’re still going to have as good games in preliminary rounds as the wild card vs the #1 seed in the NFL. Both should be roughly equally competitive. Millions still watch those games. I don’t think the few chess spectators that exist will be scared off by this new system.

Believe it or not, my goal is not to help the higher rated player, just to make the system more random and fair. I don’t want people wishing their rating was 10 points less cause that would have given them an opponent 100 points less for a give round. There is one hybrid solution that I can think of where it is semi-random but also, there is a set process that anyone could follow to show that no TD special interference happened in the ratings. Namely: a three or four round cycle of pairing “shifts” could be developed so that within a given score group, the top player of the top half is not always playing the top player of the bottom half.

For example in the first round with top half vs. lower half, top player “A” plays the bottom of the lower half (and all boards are paired likewise). In round 2, he plays the top player of the bottom half (as normal, and all players do so). In round three, maybe, it could be the top 1/4 is paired against the bottom 1/4, and the middle half is paired all top down.

Of course, the actual type of rotation wouldn’t matter, as long as the same players of the same half of a score group don’t always have to play the hardest competition. The only requirement is that it be something that anyone could follow if they ever decided to give a darn (not necessarily easy though), that a computer could pair it, and that it be random.

Here’s wondering if any more new ideas will come out.
Ben Bentrup

5

It is worth noting that “random” pairings were (and perhaps still are) used in most European swisses. It did not work well, at least in the opinion of those who had experienced “real” (ratings-controlled) swisses in the U.S. I suggest you ask the opinion of someone who played in such tournaments in the 70s or 80s.

A number of swiss variants have been tried over the years, such as Jerome Weikel’s “no adjustment for colors” rule. They produce mildly interesting results, but I ahve yet to see any that were demonstrably superior to the basic model.

6

European swisses, huh? I’ll have to keep that in mind. Thanks for the tip! Anyone know anything about those randomly paired swisses, or can ask anyone who probably will have played in a few? Let us know!

7

I think the big reason the pairing philosophies are different is that ALL those others (except the last two which I know nothing about) are elimination events, designed to pick only the winner. Swiss style pairings are designed to find winners too (they are by corollary also designed to find the perfect loser, but we rightfully don’t advertise that aspect), but also place people in approximate order of results.

With the folded pairings you would find people in the middle fighting for their life every game while the leaders and followers haev a relatively easy time of it. Of course this depends on how far off from 2**n you are …

8

This argument overlooks that those who fight for their life “every game” are different people each round.

A player near the middle who wins will play a stronger player next time, since he will now be well into the bottom half of his new score group. Likewise, a player near the middle who loses will play a weaker player next time.

Bill Smythe

9

Thanks, Bill - I considered waht you said, and realized what I meant to say was that the leaders aren’t fighting for their life as quickly. As an over-simplistic example, Take a 5 round swiss with 32 players. Rank them in order and here-s the over-simplistic part: assume the highest rated player always wins.

Both pairing approaches will arrive at 1 perfect score in 5 rounds (pairing number 1). The old traditional way will have #1 play numbers 17, 9, 5, 3, and 2. (or 3 of his 4 closest competitors). The folded pairings result in him playing #32, 16, 8, 4, and 2. That’s a weaker field.

In addition - in this perfect world - the first messy pairings in the old traditional way occur in round 5 (if I did the tedious part correctly) for 4 of the pairings (they match the round 3 pairings for those guys). THe new folded pairing approach has 8 ‘collisions’ in round 3 and 4 in round 5.

Does this prove anything? Of course not … But the harder question isn’t really whether another pairing philosophy is OK, but to measure which one is better, you have to agree on how to measure - which means you have to be able to define what you want to achieve. I’m not sure I can do that yet. (In the basketball setup, I think you can come closer - whittle teams down, preserve the top seeds for maximum TV revenue, preserve a bracket structure which might generate rivalries longer, … and so on).

10

OK. This is indeed an argument in favor of the traditional (chess) system.

One big argument in favor of the folded system (traditional everywhere except chess) is, I guess, that it avoids discontinuities at the mid-point of each score group (as I outlined earlier).

Thanks for your thoughts.

Bill Smythe