I’m curious as to what people have thought about Greg Shahade’s column on the Swiss system. Has anyone tried a tournament with random pairings by scoregroup? If you did that, and you had an odd number, would you have a random player drop to the next group and play a random member of that group, or would you have the lowest rated play the highest rated?
A few other comments:
Regarding Ivan X. Cruz’s tournament in which seven of the eight U1500 players went 0-4, I believe that this http://www.uschess.org/msa/XtblMain.php?200707284011-13539478 is the tournament that he is talking about. Please note that he’s the only player to go 0-4.
I’m surprised that no one has yet suggested what we might call the NCAA method of pairing, with 1 vs. 64, 63 vs. 2, etc. I think that the advantage that a Swiss has over this, and the random pairings for that matter, is that as much as possible it equalizes rating differences within the scoregroup. For example, in the first round, each board will be different by about 700 points (or 500, or 50, but the same amount). Upsets and color considerations make that less likely in subsequent rounds, but it is still the goal. With random pairings, you could have a 1200 point difference, and be playing next to someone with a 50 point difference. I fail to see how that is more fair.
Finally, about the issue of what I call whipsawing. One unfortunate thing about the Swiss system is that players who find themselves in the middle scoregroups round after round (+1, even, or -1) will have a hard time finding someone close to their own strength to play against. I think that’s largely because those will be the biggest scoregroups, and I’ve had that problem myself. Perhaps the best solution is to move the section breaks around, so that people will find themselves near the top or the bottom of their section more often.
In any case, I think that the solution is obvious. If there are no tournaments in your area that are doing something you’re interested in, organize one yourself. TDs can’t be that hard to find.
I thought Greg’s article was very interesting, and it has certainly generated some discussion (the most I think I’ve seen for a CLO article).
It would be nice to see someone actually run a tournament with this pairing system, but I don’t know if the players would prefer it or not. It would be hard to convince some to even participate the first time. All the players that do show up might like the system, but if enough don’t make it a financially viable tournament, I don’t know if an organizer would/could risk running it.
The other problem might be pairing programs. Off hand, I don’t believe WinTD can generate “random” pairings. So if the tournament is small enough pairing cards will work. But if it’s small, the Swiss System would probably have worked anyway. So it’s difficult to have a “test run” tournament to generate feedback that makes a difference.
I’d think that the easiest way for WinTD (possibly also SwissSys) to approach handling it is to put the rating in the team code, the prize class in group, and then give everybody the same rating (or leave all players unrated). You can do the reports by team code to get a standard wall chart sequence, or by group, score and team code to get a cross-table by class. This may require some intervention to avoid giving a truly unrated player a bye in a four-round tournament if there is a viable alternative.
One concern would be whether or not it would sequence by entry order when doing the pairings. I don’t know the answer to that.
Swiss-Sys randomizes players with the same rating. It also has a toggle for “player ineligible for bye.” However, for the size of tournaments likely to be involved (I don’t see Bill using this for the World Open), pairing cards would work just as well. We used to run quite large tournaments (like the U.S. Open) without computers.
Come to think of it, with WinTD you could always give all of the unrated players a zero point bye in round N+1 in an N-round tournament, and that would make the program avoid giving them a bye unless absolutely necessary.
A couple of decades ago I used pairing cards for a 450 player section National JH (Peoria in the late '80s - in our section we entered the pairings after I did them manually, then we entered the results into a spreadsheet so that we could get our tiebreaks done and all of the section’s trophies awarded within about an hour and a half after the final game finished), and also for a 7-round 50-player section state primary with 25 players from the same school that were on average less experienced than the other 25 players ('85 - nobody from that school played more than two teammates over the seven rounds and all of the pairings made sense). I feel quite comfortable with pairing cards, but I still have to admit that I am happy to use a computer nowadays.
If this was something like a quick event, it would reduce the pairing/printing/posting time to use a computer.
Many of the more recent and computerized TDs would have difficulty following the normal swiss rules with pairing cards, but randomized pairings means that all they’d really have to keep track of would be scores and colors-due.
The time-consuming part is writing the pairings. Everything else can be spread out with proper organization. Slightly off topic (Shudder! Sign against evil!), my personal opinion is that using a computer does not make a bad TD into a good one, nor does it really allow a good TD to do a better job. What it does is allow a good TD to do a good job with much less physical effort.
Back on topic, I think it’s going to be a while before this is a major factor in “Shahade-variant” tournaments.
Question: Do we have anyone here who played in European swisses in the days before ratings caught on over there? Well into the 80s, most European swisses were non-ratings-controlled – all pairings were made at random. This isn’t quite what Shahade has suggested, but it’s close. From what I’ve been told, the results were not very satisfactory, at least to players who had competed in “real” swisses in the U.S.
D2. Rating differences would vary widely, from near zero in the middle of the score group to perhaps thousands at the two ends.
Advantages:
A1. This system would end the discontinuities inherent near the middle of the score group.
Discussion:
D1 is always a sticky wicket, and can prevent some really good ideas from ever being implemented.
D2 may not really be a disadvantage at all, it’s just not what players are used to (so it’s really part of D1 after all).
A1 deserves a more detailed look, so here goes:
With standard pairings, if the player just above the middle is rated 1601, and the player just below him is 1600, then the 1600 will be paired against a master while the 1601 will get an opponent rated 900. This huge discontinuity would be avoided entirely if the “NCAA method” were used.
On top of that, what if a transposition is made (e.g. to improve colors)? With standard pairings, it’s legal to switch the bottom player in the top half with the top player in the bottom half, so that the 1601 would get the master while the 1600 gets the 900 opponent. (The rulebook calls this type of transposition an “interchange” between the top and bottom halves of the group.) Such an interchange can result in hard feelings, or even howls of protest (unjustified, of course, but some players who seldom look at the rulebook think they have a God-given right to “top half vs bottom half”). The “NCAA method” would eliminate this problem entirely.
In short, the NCAA method may be a great idea whose time will never come, because the existing method is so entrenched in everybody’s minds.
A disadvantage you don’t mention is that this gives an even larger edge to the top-rated players than they have now. The number one player will probably get a free point, while the number 10, say, might face someone nearly equal to him in strength. As I understand it (it’s not one of my major interests), the “NCAA system” is designed as a knockout, to get rid of the weakies and produce the most competitive matchups in the semi-finals and finals. This has been tried in the U.S. Championship with mixed success, but I doubt it’s adaptable to a swiss.
I think that the basic goal of the Swiss system is to insure that all the games are as competitive as possible, while minimizing the number of players with perfect (all wins or all losses) scores. Clearly the “NCAA” system fails on the first goal, although it may enhance the second. Also, as a practical matter it might be difficult to get top players to come out for a five round tournament where they play, say, 900, 1300, 1700 on the first day, as opposed to playing, say, 1600, 2000, 2200. At some point you would want your games to be at least a little competitive, as the paycheck just isn’t worth it.
The Shahade random system seems to fail on both counts, with the only (dubious) benefit giving weaker players a better chance to win several games. If one were going to try this, I’d suggest doing it with a plus-score format. Who knows, it could be an interesting experiment.
I don’t think it would make much difference to the 2300 player whether his first opponent is rated 1600 or 900. It’s a 99% winning expectancy vs a 99.7% winning expectancy, or something like that.
I really don’t see why either system – NCAA or traditional Swiss – is inherently either more or less preferable for a knock-out than for a Swiss. Either way, you’ll get rid of the weakies and produce competitive contests in the later rounds.
If this was the goal of the designers of ratings-controlled Swisses, then they made the poorest choices possible. Any time you pair top half vs bottom half, whether “rightside up” like a Swiss, or “inverted” like NCAA, or random, you have maximized, rather than minimized, the average rating difference. If competitive games had been the goal, a 1-vs-2 system would have been chosen (1-vs-2, 3-vs-4, 5-vs-6, etc).
Any system will do that, as long as it pairs winners vs winners.
You’re assuming a pure open (like the U.S. Open). Those are actually quite rare these days.
Tell that to the NCAA (or the NBA or NFL, for that matter. The point is that they care primarily about the finals, which are (ahem) televised for a lot of money. They want the weaker teams eliminated as quickly as possible. Upsets are considered [ii]very bad[/i]. (Who wants to watch Montana vs. Rhode Island?) The swiss is designed to give everyone a (relatively) competitive paring in each round.
The goal of the swiss is to produce an approximation of a round-robin when the field is large, while being equally fair to all the players. It’s not perfect; it’s just the best that’s been found so far. Shahade’s non-ratings-controlled swiss has been tried, by the way – it was quite common in Europe into the 1980s. The fact that it was immediately discarded as soon as ratings became available ought to tell us something.
Isn’t that a bit of a quibble? You seem to be saying that the Normal Swiss does not reduce perfect scores as quickly as an Accelerated Swiss, and therefore therefore it does not “minimize” them. Normal Swiss pairings reduce the number of perfect scores by at least half each round, so most non-mathematicians would consider “minimize” a fair description.
What the Swiss does not necessarily do is determine a clear winner, but that’s a different question.
You’re right, of course. What I meant to say was that any “top-half-vs-bottom-half” pairing method – rightside up like traditional Swiss, upside down like NCAA, or random – will be about equally efficient at reducing the number of perfect scores.
The same might even be said of any “winners-vs-winners” system, whether top-half-vs-bottom-half or not. 1-vs-2 pairings would be an example of a winners-vs-winners system which is not top-half-vs-bottom-half. Come to think of it, though, 1-vs-2 might produce a few more draws, thus reducing perfect scores slightly more effectively.
Of course, as Prof. Sloan points out, if you want to really reduce perfect scores, you can always go with some form of acceleration. In another thread a while ago, I mentioned an extreme version of acceleration once used by Richard Verber to reduce perfect scores from about 30 (after 3 rounds) to 0 (after 5 rounds).
I was under the (perhaps mistaken) impression that randomized pairings work well when you have more rounds than would be necessary in a single elimination tournament. I know swiss system pairings are used in other things besides chess, and I’ve been told (though I haven’t completely justified it to myself) that it works out in the end if you have enough rounds.
So five rounds for sixty players wouldn’t work so well (though that’s what we had at the last Michigan Amateur), but five rounds for 16 players would work fine. (For ~60 players you’d need at least 7 rounds, since 2^6=64 gives six rounds and we add one.)