What kind of tournament system would allow dropping rounds?

[Matti]

In another thread there was discussion about the EGC weekend tournament, where some strong players decided to leave the tournament before the end, even it is not allowed by the EGF tournament rules. Can we design a tournament system, which allows also the strong players to skip rounds, if they wish to? Assume there is interesting amount of prize money, the tournament is played over a weekend and we have at least 100 players, so that we need several McMahon groups or so to handle the tournament.

[RobertJasiek]

Necessary:
- Distribute top places and prizes only to players having played all rounds.
- Do not let depend places on whether opponents have played all rounds.

Achieving this:
- By definition, the first placement criterion is "Having played all rounds Yes / No".
- Do not use any opponent-dependent tiebreaker (e.g., do not use any tiebreaker).

[Celebrir]

The problem with not using any tiebrakers is that it only can be used if there are splitable prices. If e.g. a tournament qualifies the top x players for something there has to be a decision.

In the other thread I mentioned that I helped with a 3600 player event. Maybe I should extend on this a little bit: I'm a judge for the cardgame Magic: The Gathering that some of you might know and we had a tournament with over 3600 players over 2 days with 15 rounds (plus top8). I know this are numbers that sound more than weird for go, but maybe we can learn something from that because droping is quite normal in this tournaments. But before that we have to see what the both games have in common and what not:

-Magic games are faster. The time limit for a round is 50 minutes (for both players, not for one) + extra time.
-All players start with 0 points
-Magic matches are played as first who gets 2 points. Results range from clear 2:0 wins to 0:0 draws. So it is possible to draw and to win like in go. Every win gets you 3 points, every draw 1 point.
-The tiebrakers are in this order: Opponents-match-win-percentage, Game-Win-Percentage, Opponents-Game-Win-Percentage (the last two can obviously not be used in go) This percentages can't go under 0,33 to prevent to much influence from extreme bad performances.
-There are more rounds, but also more players.
-The events and players are far more competitive than any european go tournament I have seen until now.
-After round 3 players start to drop

The tournament structure for this type of events is the following:
-First day is 9 rounds. To qualify for day 2 you have to make a result of X:2 or better.
-Second day is 6 more rounds plus a top8 single-ko.

The tie-braker system itself is used for every size starting at 8-player tournaments

What I think we can learn from this:
-Using a percentage number might help to reduce the noise. A 1:2 drop as opponent is always as good as a 2:4 drop.
-A threshold for how low a tiebraker can get minimizes the influence of extreme cases (like a 0:3 drop player)
-The mentality to allow players to drop does not lower the seriousness at all.

[Matti]

Necessary:
- Distribute top places and prizes only to players having played all rounds.
- Do not let depend places on whether opponents have played all rounds.

Achieving this:
- By definition, the first placement criterion is "Having played all rounds Yes / No".
- Do not use any opponent-dependent tiebreaker (e.g., do not use any tiebreaker).

Your criteria would mean that losing a game on a round is better than not playing a at all on that round. Why this?

[RobertJasiek]

It is possible to modify my simplistic suggestion if you dislike its radical consequence for players not playing all rounds. The suggestion is meant as a proof of concept, not as the only valid solution. Feel free to modify it for alternative solutions.

[Matti]

If have witnessed a tournament game where the players where reluctant to play and four moves where played until black resigned. If skipping a round is worse than losing we might get this kind of behaviour. If a player is allowed to skip a round and chooses to play, it should be because he wants to play and not beause of other reasons.

Placement criteria:

Number of wins (+ starting McMahon Score)
optionally: smaller number of losses
optionally: CUSS or other tie breaker depending on how early ones wins occur.

Prizes should be divided with players with same number by the first criteria. Subsequent criteria may introduce smaller differences. For example if three players are to share prizes: 300, 200 and 100, one could give each 200 or by tie breakers give 250, 200 and 150.

[Celebrir]

If have witnessed a tournament game where the players where reluctant to play and four moves where played until black resigned. If skipping a round is worse than losing we might get this kind of behaviour. If a player is allowed to skip a round and chooses to play, it should be because he wants to play and not beause of other reasons.

That is exactly what I fear with a no-drop-policy

Placement criteria:

Number of wins (+ starting McMahon Score)
optionally: smaller number of losses
optionally: CUSS or other tie breaker depending on how early ones wins occur.

Prizes should be divided with players with same number by the first criteria. Subsequent criteria may introduce smaller differences. For example if three players are to share prizes: 300, 200 and 100, one could give each 200 or by tie breakers give 250, 200 and 150.

I think the distribution of prices is nothing that can be decided while creating a scoring system. This will and should also ways be the decision of the tournament organizer.

Why don't we use opponents-match-win-percentage as tiebraker? If you win early this will be automatically better because you get players with more wins for a longer period compared to the person who loses first and then wins. Besides this is not influenced by drops because the percentage of those players is frozen at the percentage they had when dropping. CUSS would become worse each round because it is as if the dropped player loses every round for your tiebraker score

[Matti]

If have witnessed a tournament game where the players where reluctant to play and four moves where played until black resigned. If skipping a round is worse than losing we might get this kind of behaviour. If a player is allowed to skip a round and chooses to play, it should be because he wants to play and not beause of other reasons.

That is exactly what I fear with a no-drop-policy

Placement criteria:

Number of wins (+ starting McMahon Score)
optionally: smaller number of losses
optionally: CUSS or other tie breaker depending on how early ones wins occur.

Prizes should be divided with players with same number by the first criteria. Subsequent criteria may introduce smaller differences. For example if three players are to share prizes: 300, 200 and 100, one could give each 200 or by tie breakers give 250, 200 and 150.

I think the distribution of prices is nothing that can be decided while creating a scoring system. This will and should also ways be the decision of the tournament organizer.

You are right. Prize structure and tournament system are separate things.

Why don't we use opponents-match-win-percentage as tiebraker? If you win early this will be automatically better because you get players with more wins for a longer period compared to the person who loses first and then wins. Besides this is not influenced by drops because the percentage of those players is frozen at the percentage they had when dropping.

It seems a possible tie breaker.

CUSS would become worse each round because it is as if the dropped player loses every round for your tiebraker score

No, opponents do not affect CUSS.

[Celebrir]

No, opponents do not affect CUSS.

True, I mixed the different tiebrakers up in my head here, sorry. The problem is that the CUSS doesn't involve your opponents strength at all which I don't like that much.

So what about the following ranking criteria:
1) (McMahon) points
2) Opponents win percentage (OWP) with minimum at 0.33
3) Opponents OWP

Opponents OWP seems a little bit lucky, but ultimately it shows more about how good your opponents were. I don't see anything better here.

I think OWP needs a minimum because 0-X (drop) would be hard to compensate in the score otherwise and 0.33 seems to be a mark that has proved itself good at Magic.

[HermanHiddema]

@Celebrir: I think OWP doesn't combine well with McMahon. Is it better to have met an 1d with 4/5 (80% OWP) than a 3d with 2/5 (40% OWP), given that they have the same McMahon score?

[RobertJasiek]

Even a 5 minutes sudden death blitz tiebreak tournament is a more meaningful tiebreaker than any opponent-dependent tiebreaker. Nobody tell me there would not be enough time for such a tiebreaker if there are indivisible prizes important enough to be indivisible.

[Matti]

@Celebrir: I think OWP doesn't combine well with McMahon. Is it better to have met an 1d with 4/5 (80% OWP) than a 3d with 2/5 (40% OWP), given that they have the same McMahon score?

We could design a new tie breaker: sum of opponents expected scores, give that they would have played all the games in the tournament (SOES).

Of course Robert's idea to play lightning games is also possible.

[Celebrir]

@Celebrir: I think OWP doesn't combine well with McMahon. Is it better to have met an 1d with 4/5 (80% OWP) than a 3d with 2/5 (40% OWP), given that they have the same McMahon score?

True, we should adjust that... Maybe giving one win for every rank difference to the player whos OMW you're calculating? This way at the moment they play the OWP would be the same, but future performance can change it. The question is whether it would be worth the effort.

Even a 5 minutes sudden death blitz tiebreak tournament is a more meaningful tiebreaker than any opponent-dependent tiebreaker. Nobody tell me there would not be enough time for such a tiebreaker if there are indivisible prizes important enough to be indivisible.

I don't agree with that at all. Another game would only show your strength against that one player, who might be a good or bad match up stylewise for you. An opponent based tiebraker takes into account the whole tournament and a lot of players.

[Matti]

Even a 5 minutes sudden death blitz tiebreak tournament is a more meaningful tiebreaker than any opponent-dependent tiebreaker. Nobody tell me there would not be enough time for such a tiebreaker if there are indivisible prizes important enough to be indivisible.

Having indivisible prizes and allowing players to drop rounds when they wish do not work in the same tournament. What would you do, if there is left two players eligible for the indivisible prize and both decide to drop the next round? Also, players might choose to skip a tie break lightning game.

[RobertJasiek]

Even a 5 minutes sudden death blitz tiebreak tournament is a more meaningful tiebreaker than any opponent-dependent tiebreaker.

I don't agree with that at all. Another game would only show your strength against that one player, who might be a good or bad match up stylewise for you. An opponent based tiebraker takes into account the whole tournament and a lot of players.

A tiebreak tournament does not necessarily equal only one additional game. E.g., it could be a mini-round-robin. It is only one game only if the number of mutually tied players is two and they must play only one tiebreak game against each other.

Style is a non-issue. A player who cannot cope with a particular opponent's style must improve his skill.

Usually, an opponent-dependent tiebreaker does not take into account the whole tournament; it is not a tiebreaker measuring the tournament results of all participants (other than the player himself) of the tournament.

A first-order opponent-dependent tiebreaker (such as SOS) does not take into account a lot of opponents. In many tournaments with relatively few rounds and an intermediate to big sized top group, a first-order opponent-dependent tiebreaker often takes into account an only small percentage of those opponents involved in the tie. Those players shall be distinguished but such a tiebreaker often depends much on other opponents.

Your opinion reformulated can be made a bit more meaningful to being worried about tiebreak games measuring performance AGAINST usually fewer opponents than some first-order opponent-dependent tiebreaker measures performance BY usually more opponents. Such an objection is more concerned with the number of measured opponents than with the number of the player's measured own results. IMO, measuring the player's own results is always more meaningful than measuring his opponents' results. This is so because a tournament is a competition about which player achieves his best own results. For this reason, Wins or MMS is the tournament's first placement criterion (and SOS never is the first placement criterion). A tiebreak tournament measures the same kind of thing as the first placement criterion because this is the most meaningful for the player's own performance.

There is, however, an alternative: a tiebreak tournament with games played against new opponents, in which only the tied' players results are interpreted for the prize distribution but the opponents are motivated by, say, rating effects of the tiebreak games so that every opponent takes the tiebreak games seriously. Unfortunately, I doubt that any such motivation is good enough to overcome psychologic preferences by such opponents for wishing that certain of the tied players may get better final places. So the alternative is, IMO, only good as a thought experiment.