Hi everyone. I’m posting in this forum because I have invented a completely (almost) new system of pairing chess tournaments. I wrote an article about it here:
The article is near finished, but the system is not final. I’m improving it interatively based on feedback I’m receiving, and based on code updates.
The Swiss System is over 100 years old, and used almost universally. Yet few have attempted to invent an alternative to it. The Swiss System has many flaws, mostly because of the notion that points are the be-all end-all of how well you do. But not all points are created equal!
For example, if you are rated under 2200, then a draw against a 2200 is worth more than a win against a 1200, but because we judge performance universally based on points, most would agree that the opposite is true.
The general consensus seemed to be that the idea is most likely bad, with a noticeable minority opinion that it could be tried out to see if it really is bad or if there is any value to it.
It can result in the top two players drawing in round one and never having a chance to be paired with a tournament leader even while winning every other game.
This is not a new idea. An organizer in Pgh tried something similar in the '80’s. His system, also based on performance ratings for prizes, virtually guaranteed that the top players would not win the event. It was heavily biased to favor slightly lower rated players. After a couple of tries, it was not successful at drawing players as the Masters, Experts, and Class A’s caught on that they had little chance of winning a prize, so stopped come to the events. By the by, the author of that system really hated to see high rated players in his events. He got what he wanted.
One of the flaws of this old system was that if there was a large rating difference between #1 and the rest, the top player could go 5-0 and still lose out as only one prize was offered. A player scoring 4-1 might win the event.
That problem is fixed by adding a game where you draw against yourself. It also fixes the problem of calculating performance ratings for both perfect scorers and zero scorers.
I neither understand the value of Mr. Olson’s system nor why a computer program is necessary for pairing any other than the top one or two groups. I also don’t understand how the “draw against equal rated player” helps.
I don’t particularly like the main idea, but “draw against equal” possibly does help the idea.
Without =draw
2350 beats 1250, 1250, 1200, 1200, 1100 (performance rating 1650 - or 1600 if you add in all the results instead of just the top one for a perfect score)
1250 loses to 2350 and beats 1250, 1200, 1200, 1100 (performance rating 1660)
with =draw
same 2350 result averages in 2350 with 5x1600 resulting in 1725
same 1250 result averages in 1250 with 5x1660 resulting in 1592
PS Note that if the 2350 is paired against the 1100 in the final round, forfeiting that game (showing up more than an hour late) will results in a PR without =draw of 1625 (if you ignore the otherwise perfect score) and with =draw of 1770. That means that the 2350 facing those rating ranges may have an incentive without =draw of forfeiting the final round and almost always will have an incentive with =draw (forfeiting the final round against even a 1250 would give a PR, with =draw, of 1740 instead of the maximum of 1725 that would be obtained by winning the last round).
A system that sometimes gives performance and prize-winning bonuses for taking a forfeit loss or zero point bye may not be the best.
It “helps” in the sense that the best estimate of a “performance rating” would be colored by previous information about how good the player is since the likelihood function for rating based upon performance is often extremely flat. That, of course, is why you don’t give prizes based upon performance rating. While FIDE norms are (in effect), based upon performance ratings, they have both rather substantial minimum games requirements and other requirements to prevent results which aren’t very informative to earn norms. To try to use PR’s for (a) pairings and (b) awards for short tournaments is a fool’s errand.
It “fixes” the problem but creates a different one—if you have two players with effectively identical performances (same results against similarly rated players), the higher rated one “wins” which is probably the “wrong” result.
I hope that you are doing this because you want experience with tricky programming problems and not because you actually intend to use this for a tournament. The fact that you have already come up with a “fix” which pretty much defeats the whole purpose of the original concept should tell you that this is not a good idea.
The more I look at this, the more it looks like a “Rube Goldberg machine”. Complications are added to achieve the goal. What is the goal? In all types of competitions we try to balance the need to have a winner chosen balanced by a desire for fairness to all the competitors.
So far, we have tried several different ways to run tournaments:
Round Robins - Pretty fair for the most part. Really fair if there is a double round robin format giving each player Black and White. Unless there are major upsets, the top players should show their class and earn the top place prizes. The downside is that it takes a long time to run these events and they can be expensive to run. Besides international tournaments, I have seen many club championships and club ladders run this way. The tournament is easy to run as there are tables for use for virtually any size, including what to do if someone drops out.
Knock out - Definitely will determine an ultimate winner. Fair? Not in most knock out events I have seen. In tennis, the top players get to bunny bash for a couple of rounds as they play the lowest ranked players. Seedings and rankings might pair evenly or create terrible mismatches. In the first chess events, the fear was that the top two players by skill and strength might be paired too early leading to the rest of the event being an anticlimax. Someone with a lucky set of pairings might rise to the final match before getting crushed. He might also win because of the exhaustion of the opponent who had a tougher schedule. We don’t see this form in chess tournaments too often. It might be exciting for the fans, but terrible for the players. Pretty easy for an organizer to run as half the field disappears every round.
The Swiss System - Pairing system devised to determine a winner. Fair? Partly so. Top down pairings favor the higher rated. This is mitigated by having players play others with a similar score each round. Chances can level out as one meets others with similar ratings and scores. The Swiss System allows more players to compete over a shorter time frame. As tournament entries increased for some events, the use of round robins to determine qualifiers for a final round robin broke down because of time and cost. The US Open used to use this type of round robin sectional format. It became too unwieldy for use as no one could afford to be away for three weeks to compete. The 1946 US Open used a Swiss System entirely as its format because of the number of players. This became the norm for subsequent US Open and all other weekend tournaments from the 1950’s onward. The creation of rating systems was a boon to TDs to help determine pairings. Pairing rules were a way to limit or eliminate bias in pairings. If using ratings for pairings led to some unfairness, it was equally unfair for all by rule. Besides prizes, improving one’s rating became another incentive to play in tournaments. Relatively easy for organizers to set up and can accommodate any size of tournament entry.
I am sympathetic to the idea of increasing fairness for the players. However, we also need to find a winner. Complicated mechanisms which add to fairness make it difficult to find a winner and increase the chances for anomalous results. If you want fairness, then use a Round Robin system. If you want a winner, the Swiss System is a pretty fair test of the champion’s skills and cheaper as if not easier for an organizer to run.
You omitted quads, i.e. several small round robin sections.
There, the idea isnotto determine a clear winner at all. Rather, it is to give competitive games to all players. It’s just the opposite of a single large round robin, a Swiss, or a knock-out, all of which are designed to produce a single overall winner.
Having two 3-player sections at the bottom – or, for that matter, a 6-player Swiss – is nothing more than a method of solving a minor problem with quads, namely, the possibility that the total number of players may not be a multiple of 4.