Improving Swiss Teams events

[hrothgar]

Swiss Team type events are one of the most popular formats in bridge. We argue that the accuracy of Swiss Team style events can be improved significantly if a Strength of Schedule adjustment is used to complement the normal scoring system.

 

This hypothesis was tested using a series of Monte Carlo simulations. A computer program generated 128 bridge teams with known strength. These teams competing against one in a Swiss Teams type event. At the conclusion of the event, the sample statistic – the ranking produced by the Swiss Teams event - was compared with the population statistic (the objective/known ranking of the team strength). We consider event event formats in which the sample statistic closely mirrors the population statistic superior to formats in which the sample statistic deviates significantly from the population statistic.

 

Monte Carlo simulations can be used to test a variety of different hypotheses. For example, are tournaments with a large number of short rounds more accurate than tournaments with a small number of long rounds. (None too surprisingly, the answer depends on the fixed cost associated with the break between rounds) Alternatively, is there a relationship between the number of teams entering a tournament and the number of rounds necessary to accurately identify the winner. Our most striking result involved using a Strength of Schedule adjustment to the normal Swiss Team scoring system. We determined that a Strength of Schedule adjustment allows tournament organizers to significantly improve the efficiency of their events. Hypothetically, an event organizer could reduce the time required to stage an event without compromising the accuracy. Alternatively, an organizer could hold the length of an event constant and significantly improve the accuracy of the event.

 

Strength of Schedule adjustments can implemented in a variety of ways. For the purpose of this study, we used a very simple SoS adjustment.

 

1. Run a normal Swiss Teams event

2. Calculate the total number of Victory Points won by each team

3. Sum all of the Victory Points won by each team that team i played

against, excluding the head to head competition between team i and

team j.

4.The Team's final rank is determined by adding the Victory Points that

Team "i" won in head-to-head competition and some fraction of the total

VPs won by all the teams that team "i" competed against. (This fraction

is a function of the number of rounds in the tournament)

 

We certainly don't claim that the SoS adjustment just described is by an optimal implementation. However, even this very crude implementation has a dramatic impact on the accuracy of the event.

 

Consider the following tournament format:

 

* 128 teams competing in a Swiss format

* The event consists of "N" 20 board rounds

* The primary statistic used to measure the accuracy of the event is

the percentage chance that the strongest team will land in any of the

top eight places at the close of the event. (We used other metrics

including the Spearman rank coefficient and how many of the top eight

teams placed in the top eight slots. Results were consistent

across metrics)

 

With no SoS adjustment, tournament organizers need to run twelve 20 board rounds to have a 95% chance that the strongest team will place in any one of the top eight slots. If we add an SoS adjustment, tournament organizers can run nine 20 board rounds while still achieving a 94.9% chance that the strongest team will place in any of the top eight places. Tournament organizers can reduce the length of the tournament by 25% without impacting the integrity of the results. (In comparison, if the Tournament Organizers were to run an event with nine 20 board rounds without any SoS adjustment, the accuracy of the event would drop from 95% to 92.3%)

 

At this point in time, the primary value of this study is identify the fact that significant improvements can be made to the traditional Swiss Teams type format. Over time, we hope that it will be possible to make more concrete recommendations regarding the best implementation for an SoS correction as well as an executable that could be used to optimize events formats based on time constraints.

 

Steve Willner was responsible for the original insight that an SoS correction would have a impact the accuracy of the Swiss Team format. All of the coding and simulation work (read this as the "real" work) was done by Alex Ogan and Gerben Dirksen.

[cherdano]

This is an interesting compromise between a simple scoring system and an optimal one. For the latter, I would suppose everyone who has thought about this seriously would agree that a maximum-likelyhood algorithm is best, as I believe the one described in Gerben's paper is.

 

What assumptions did you make on the distribution of strengths? (Without that, your figure of 95% by itself is meaningless, of course.)

[keylime]

Richard,

 

I have a question (Arend you too).

 

Wouldn't a smaller field combined with more variable skill levels and smaller rounds, like a sectional Sunday swiss, required a much stronger correction factor to be implemented so that it would both reward the strong teams' maintaining at or near "par" and the weaker teams that are "above par"?

[mikeh]

While my only exposure to statistics was when I studied engineering 35 years ago, I was (I think) able to follow your post... and I applaud the work done.

 

A couple of comments: my subjective experience suggests that there is one very significant factor, present in most swiss formats, not covered by your approach: randomness of the boards.

 

Take two equally skilled strong teams, who have drawn two equally inferior teams to play a 7 board match. One team has a match with 2 slams, 4 games, and difficult bidding and play problems. They clobber the opps by 56 imps. The other team, playing in an even more superior fashion, have to cope with a passout, 4 hands on which the auction goes 1N 3N on 29 hcp and so on... they eke out a victory by 6 imps.

 

I appreciate that over a large field with numerous matches, these issues fade to some degree, but in real life we rarely have huge fields outside of National events.

 

 

In practice, this is a self-correcting problem if it arises early in the event: the team that drew the flat boards will play teams with fewer VPs than the team that blitzed, and so, on average, will play inferior teams until it 'catches up' by blitzing these weaker teams. But your SoS approach undermines this catching up ability by discounting these wins against weaker teams. So an early flat match (or two) will substantially handicap a good team compared to its peers who have wild hands early.

The solution is, of course, to play duplicated boards but that is a logistical nightmare...given that entirely new boards must be put in play each round.

 

There is also a luck of the draw issue. Last week, on the Sunday Swiss, my team struggled early but finished strong: going into the last match we had a mathematical shot at winning, and ended 3rd. My strong suspicion is that we would have been demoted on the SOS analysis, because we never faced either of the teams that finished ahead of us.... nor did we play the 4th or 5th place teams.... .

 

While the SOS might afford a more reliable indicator of who was playing well that day, I can assure you that winning the 3rd highest number of VPs and then being awarded 5th place while never having a chance to play the teams that finished ahead of us while winning fewer VPs would rankle.... we would, I am sure, have felt that this wasn't fair... and that is nothing to the way we would have felt had our mathematical shot at winning it all come through....

 

Let's look at this in terms of a late-stage matchup. Three contending teams, all with the same number of VPs. Let's say they are leading.

 

Team 1 plays team 2 and battles to a draw. Team 3 gets to play a team several notches lower (in the late stages, teams often play other teams quite distant in the standings due to conflicts arising from earlier schedules).

 

If team 3 wins, but does so narrowly (perhaps due to the nature of the hands), it will end up 3rd, not first... because SoS factors reduce the impact of its win compared to the ties achieved by teams 1 and 2 when they played off.

 

Or consider that both teams 1 and 3 blitzed: now they are tied but team 1 wins the event on SoS factors: even tho the final match assignment was random, and team 3 played a perfect match. In both scenarios, team 3 would feel ripped off... not a situation that tournament promoters should encourage.

[Finch]

The solution is, of course, to play duplicated boards but that is a logistical nightmare...given that entirely new boards must be put in play each round.

All English Swiss events - even many locally organised ones - use duplicated boards.

 

Come on guys, join the 21st centruy :lol:

[AlexOgan]

This is an interesting compromise between a simple scoring system and an optimal one. For the latter, I would suppose everyone who has thought about this seriously would agree that a maximum-likelyhood algorithm is best, as I believe the one described in Gerben's paper is.

 

What assumptions did you make on the distribution of strengths? (Without that, your figure of 95% by itself is meaningless, of course.)

Our view is that to be accepted by players, the scoring system must be transparant/simple. It's not clear that people would even accept something as simple as this -- I think that more complicated systems like Gerben's don't stand a chance.

 

We used normally distributed team strengths with standard deviation 1, with units of IMPs/Board.

[Finch]

While I don't (necessarily) dispute your conclusion, I feel fairly strongly that it would never be adopted in practice. Just think of it - you turn up, play a load of Swiss Teams matches, get the most VPs out of everyone there and suddenly discover you haven't won...

 

Or put it another way: most Swiss Teams events are played for fun. The really top teams events in any country are generally not Swiss, they are round-robin followed by a KO, round-robin in groups followed by a KO, straight KO, or double-elimination KO (or possibly have some form of repechage). People who play in Swiss Teams events are more interested in enjoying themselves, and winning masterpoints against fairly equal teams, than they are in having the best mathemtical chance of the best team winning. Anything that dilutes that pleasure - and I assure you that adjusting your VPs at the end of the event will dilute the pleasure - would be unpopular.

 

I preferred the idea of different match lengths.

[cherdano]

Btw, another problem is that your system will tend to make the last rounds less exciting, as the winner may already be almost certain the last round. This is almost a guaranteed problem with any multiple rounds sports event that is accurate enough that the best team wins almost always: to be that accurate, it can't rely on having the winner be decided by the last round very often.

[skjaeran]

The solution is, of course, to play duplicated boards but that is a logistical nightmare...given that entirely new boards must be put in play each round.

All English Swiss events - even many locally organised ones - use duplicated boards.

 

Come on guys, join the 21st centruy :lol:

All Norwegian Swiss events too. And all other events. Only at club-level you might come across non-duplicated boards. In fact about half of our clubs (I guess) use duplicated boards.

 

And that's not even 21st century Frances, it's late 20th century. :P

[david_c]

While I don't (necessarily) dispute your conclusion, I feel fairly strongly that it would never be adopted in practice. Just think of it - you turn up, play a load of Swiss Teams matches, get the most VPs out of everyone there and suddenly discover you haven't won...

 

Or put it another way: most Swiss Teams events are played for fun. The really top teams events in any country are generally not Swiss, they are round-robin followed by a KO, round-robin in groups followed by a KO, straight KO, or double-elimination KO (or possibly have some form of repechage). People who play in Swiss Teams events are more interested in enjoying themselves, and winning masterpoints against fairly equal teams, than they are in having the best mathemtical chance of the best team winning. Anything that dilutes that pleasure - and I assure you that adjusting your VPs at the end of the event will dilute the pleasure - would be unpopular.

Yep, I agree with Frances. It's interesting to look at "accuracy" from a mathematical point of view, but if I was actually going to play in an event I would prefer it to be scored by straight VPs.

 

(Though, SoS is a good tie-break for teams finishing on equal numbers of VPs. I think it may already be used for that purpose in some cases.)

[jdonn]

I don't understand the desire to achieve this goal of accuracy. If all we want is for the best team to win as often as possible lets not even play, lets just take a vote on who is the best team and give them some masterpoints. And then why would anyone but the best team even show up?

[Gerben42]

Not the best team, the team who played best in the tournament.

[jdonn]

Not the best team, the team who played best in the tournament.

No, exactly what I said. Read the original post.

 

"We consider event formats in which the sample statistic [swiss team results] closely mirrors the population statistic [skill level or ability of the teams] superior to formats in which {this is not the case}."

 

The goal was a format in which the best teams win as often as possible. I am saying that I don't consider accuracy to be a superior format, in fact I consider too much accuracy much less appealing than the status quo.

 

I am not interested in accuracy as you state it either. The elements of luck (including regarding who you draw), randomness, and simplicity of the scoring system are all important. No one will play in something where they can't easily understand the scoring system.

 

It may be interesting to study, but going down this road is a huge mistake that would thankfully never happen. Anyone ever heard of the BCS?

[pclayton]

It would seem that paring the field would help achieve this end, like the NA Swiss. As the field improves, the opportunities to have multiple blitzes in the final rounds against mediocre teams is reduced.

 

I like the ideas of duplicated boards in swiss, but security would need to be heightened.

[awm]

To some degree this has to depend on the importance of the event. In a typical tournament event it's probably not desirable for the best team to "always win." I've frequently heard this reason given for the demise of BAM events at US Regional tournaments -- there used to be a lot of BAMs, but BAM is a highly non-random format, and the same teams won too consistently, causing people to lose interest in competing.

 

However, take an event like national team trials. There is a strong desire for the best team to win, in order to represent the country well and give them a chance at the Bermuda Bowl or Olympiad. It would be undesirable to have a highly random event used to select the national team, as the odds of inferior players "getting lucky" would be too high. On the other hand, simply selecting the team based on who some committee votes to be "best" is subject to a lot of arbitrariness as well, as players who are better known or better liked will often be selected (determining "how good" a particular pair or individual might be is extremely subjective in bridge, and it's easy to point to seemingly knowledgable individuals with wildly different opinions on this). In any case it creates the perception that no matter how good a pair might be, it's hard to break in because of the strong status quo in the selection process.

 

It seems desirable for a team trials type event to have a format where the team is selected by actually playing bridge (rather than by committee) while simultaneously minimizing the chance of an upset due to luck. This particular issue seems to crop up a lot for the US junior team trials, since (for whatever reason) these trials are held over only two days instead of a week. It seems that every cycle a new process is chosen for this selection.

 

As for swiss teams, while I understand Mike's point, wouldn't it be frustrating to be in first place with a round to go, having played virtually all the top teams and won, then play a tight match with the second place team ending in a draw, only to have the third place team pass you because they got an easy opponent in the last round (and never had to play any of the other teams in the top five)? I'd think this also would leave a bitter taste in the mouth of some competitors.

[Finch]

It may be interesting to study, but going down this road is a huge mistake that would thankfully never happen. Anyone ever heard of the BCS?

errmm... only in the context of being the British Cohort Study, or the British Computer Society. I doubt either of those is what you meant.

[Finch]

* 128 teams competing in a Swiss format

* The event consists of "N" 20 board rounds

* The primary statistic used to measure the accuracy of the event is

the percentage chance that the strongest team will land in any of the

top eight places at the close of the event. (We used other metrics

including the Spearman rank coefficient and how many of the top eight

teams placed in the top eight slots. Results were consistent

across metrics)

Consider this tournament format:

 

*128 teams competing in an multiple teams event, scored as total IMPs

*The event consists 1 board rounds, 2 board rounds, or 3 board rounds organised so that it's as close to an all-play-all as you can manage given constraints on the number of boards (you could even play a combination of 1-board, 2-board and 3-board rounds to make sure it's an all-play-all; the teams you play more or less boards against picked at random)

 

How do the results on this format do compared to a Swiss?

 

The thing is, I'm not convinced by the 'intuitive' feeling that a Swiss is inherently more accurate. Yes, you get the teams in contention playing more boards against each other (which is good), but you also waste a load of boards during mismatches. I've played Swiss events where I've won 8-board matches by 60+ imps - I don't think we gained any useful information by playing so many boards against that one particular team rather than playing 2 boards against each of 4 teams.

 

I've seen it claimed that it's better to play an n-round 8-board match Swiss (say) than an all-play-all 2-boards a round. I don't know if that's true or not, and I'd be interested in finding out. I suspect that it may depend a bit on the distribution of strengths of teams present. With a very low variance, the all-play-all I'm sure is better.

[hrothgar]

Not the best team, the team who played best in the tournament.

No, exactly what I said. Read the original post.

 

"We consider event formats in which the sample statistic [swiss team results] closely mirrors the population statistic [skill level or ability of the teams] superior to formats in which {this is not the case}."

 

The goal was a format in which the best teams win as often as possible. I am saying that I don't consider accuracy to be a superior format, in fact I consider too much accuracy much less appealing than the status quo.

 

I am not interested in accuracy as you state it either. The elements of luck (including regarding who you draw), randomness, and simplicity of the scoring system are all important. No one will play in something where they can't easily understand the scoring system.

 

It may be interesting to study, but going down this road is a huge mistake that would thankfully never happen. Anyone ever heard of the BCS?

Few quick comments here

 

1. I think that most people would agree that tournaments need to contain elements of both luck and skill. If an outcome is deterministic and pre-ordained then there is no reason to hold a contest. Correspondingly, if there is no element of skill involved we might as well simply cut cards to determine a winner.

 

2. Intelligent people can differ regarding where one should draw the line between luck and skill. However, I would argue that that regardless of where one chooses to draw the line its desirable to be able to accurately describe one's design choice. In some ways, the value of this experiment has less to do with recommending any one specific format than being able to describe the various trade-offs that are inherent in the choice of conditions of contest. If one doesn't have appropriate vocabulary and methodology, one is reduced to blind platitudes about tradition....

 

3. From my own perspective, I prefer a tournament format that favors skill over luck. I think that its important to note the following: Consider some of the statistics generated here: If we run a tournament with twelve 20 board rounds, there is still a 5% chance that strongest team won't place higher than 9th. This tournament requires close to 4 days to run, however, the rub of the Green still plays an enormous role. The figures for a more traditional tournament with eight seven board rounds are horrific.

 

4. Simplicity of the scoring system was an explicit design criteria. I suspect that we could have (easily) arrived at some much more accurate SoS adjustments at the expense of adding significant complexity. The metric that we suggest is extremely simple.

[jdonn]

4. Simplicity of the scoring system was an explicit design criteria. I suspect that we could have (easily) arrived at some much more accurate SoS adjustments at the expense of adding significant complexity. The metric that we suggest is extremely simple.

It's simple to you. It's simple to me. Do you think it's simple to my grandmother, or even my mother?

 

If you believe so, I'll let you be the one to try and make her understand why she scored the most victory points and didn't win.

[hrothgar]

* 128 teams competing in a Swiss format

* The event consists of "N" 20 board rounds

* The primary statistic used to measure the accuracy of the event is

the percentage chance that the strongest team will land in any of the

top eight places at the close of the event.  (We used other metrics

including the Spearman rank coefficient and how many of the top eight

teams placed in the top eight slots.  Results were consistent

across metrics)

Consider this tournament format:

 

*128 teams competing in an multiple teams event, scored as total IMPs

*The event consists 1 board rounds, 2 board rounds, or 3 board rounds organised so that it's as close to an all-play-all as you can manage given constraints on the number of boards (you could even play a combination of 1-board, 2-board and 3-board rounds to make sure it's an all-play-all; the teams you play more or less boards against picked at random)

 

How do the results on this format do compared to a Swiss?

 

The thing is, I'm not convinced by the 'intuitive' feeling that a Swiss is inherently more accurate. Yes, you get the teams in contention playing more boards against each other (which is good), but you also waste a load of boards during mismatches. I've played Swiss events where I've won 8-board matches by 60+ imps - I don't think we gained any useful information by playing so many boards against that one particular team rather than playing 2 boards against each of 4 teams.

 

I've seen it claimed that it's better to play an n-round 8-board match Swiss (say) than an all-play-all 2-boards a round. I don't know if that's true or not, and I'd be interested in finding out. I suspect that it may depend a bit on the distribution of strengths of teams present. With a very low variance, the all-play-all I'm sure is better.

Hi Frances:

 

I was wondering much the same thing (Its possible that my inspiration was a bit different).

 

One of the basic results that arose quite early in the study had to do with how make the most efficient use of a fixed amount of time. We found that tournaments that used a relatively large number of short rounds produced more accurate results than tournaments with relatively large small number of large rounds. The main countervailing force was the fixed cost associated with round breaks. The more time that people spend stretching their legs/drinking/smoking/peeing/what have you between rounds, the few rounds you want to have.

 

It occurred to me that a formal movement might be a better way to handle the whole situation. I suspect that a BAM type movement might be structurally more efficient. Its easier to force people through some kind of structured movement than run a barometer...

 

However, its unclear whether the advantages of having people play all of the other pairs outweigh the the (considerable) advantage that a Swiss / Danish teams event uses a Barometer type system.

[hrothgar]

There is also a luck of the draw issue. Last week, on the Sunday Swiss, my team struggled early but finished strong: going into the last match we had a mathematical shot at winning, and ended 3rd. My strong suspicion is that we would have been demoted on the SOS analysis, because we never faced either of the teams that finished ahead of us.... nor did we play the 4th or 5th place teams.... .

You also need to consider the flip side of the equation:

 

You placed third in a Swiss Teams event without ever facing the teams that placed first, second, fourth, or fifth. If I placed behind you in that event I'd be bitching about the "Swiss Gambit" which allowed you to place while never facing any strong opposition....

[pclayton]

There is also a luck of the draw issue. Last week, on the Sunday Swiss, my team struggled early but finished strong: going into the last match we had a mathematical shot at winning, and ended 3rd. My strong suspicion is that we would have been demoted on the SOS analysis, because we never faced either of the teams that finished ahead of us.... nor did we play the 4th or 5th place teams.... .

You also need to consider the flip side of the equation:

 

You placed third in a Swiss Teams event without ever facing the teams that placed first, second, fourth, or fifth. If I placed behind you in that event I'd be bitching about the "Swiss Gambit" which allowed you to place while never facing and strong opposition....

Thats exactly the argument the NCAA posed before setting up the BCS.

[hrothgar]

It's simple to you. It's simple to me. Do you think it's simple to my grandmother, or even my mother?

 

If you believe so, I'll let you be the one to try and make her understand why she scored the most victory points and didn't win.

I started playing around with this topic because the Australians were interested in improving the format of their team trials system. Currently, the Aussies hold a VERY big Swiss teams event called the South-West Pacific Teams which has (approximately) 256 teams organized in two pools of 128. This is an event in and of itself.

 

The top 8 teams from each pool advance into a second event called the NOTs (I think that this stands for the National Open Teams). These 16 teams compete in a Knockout type event to determine the Australian representative for major international events.

 

When I posted these results the Oz-One forum, I suggested side stepping the entire issue that you raise. Use a traditional Swiss Teams style scoring system for the SWPTs. All the master points and trophies and such that the bunnies care about will use this scoring system.

 

The Strength of Schedule correction get applied in parallel. It is only used to identify the teams that qualify for the NOTs and to arrive at seeding. In theory, the folks competing in the NOTs are much more interested in the accuracy of the event, possess more technical sophistication, and would be more willing to accept a bit more complexity in the scoring system.

 

Comment 2:

 

Your mother and your grandmother accept the VP scales without questioning them. The math underlying Victory Points and IMPs is every bit as complex as what we're talking about here. You mother and grandmother accept this because they've been shielded from its complexity. I certainly agree that change the system would be quite difficult, however, I suspect that this has much more to do with tradition and hidebound thinking than "complexity".

[jdonn]

The imp and victory point scales convert a score to a different scale without changing the relative rank of the scores. It may not be any less complex than what you suggest on the level at which most people think, but at least they know if they scored the "most", it will still be the most all all the scales. Scoring +500 when the other table scores +100 in your seats is "good" and everyone knows it. The fact that some people might not understand how this will directly translate into imps or victory points doesn't matter, since they know they are going to get some amount of those things to reward the good score. If you told them this score would sometimes be worth 10 imps, sometimes 5, sometimes 18, sometimes -2, you don't think they would find this more complex?

 

You may be right that complexity is the wrong word to use, but I don't think tradition is right either. I think logic is. Which is to say, if you use a format in which a team scores the most victory points and doesn't win, this will go completely against most peoples' basic sense of logic and will leave them unsatisfied.

[mikeh]

As for swiss teams, while I understand Mike's point, wouldn't it be frustrating to be in first place with a round to go, having played virtually all the top teams and won, then play a tight match with the second place team ending in a draw, only to have the third place team pass you because they got an easy opponent in the last round (and never had to play any of the other teams in the top five)? I'd think this also would leave a bitter taste in the mouth of some competitors.

well, yes and no :)

 

Of course, seeing the third or fourth place team slide by me in the last match (and it has happened to me) is annoying... but, and this is a very big but.... I had control of the outcome. All I had to do was win, and usually in this situation, not by a huge margin, thanks to the compression of big wins by the VP scale usually in use. But the SoS adjustment is out of my control, and may indeed be ourely random... altho I suppose that, properly implemented, the match pairings should be according to current SoS rankings, not merely current number of VPs.

 

However, this also gives rise to (likely) perceived fairness issues: if we rank teams not by pure VPs but by SoS rankings, then we may get the seemingly wierd result that the top two teams, in terms of VPs won, NEVER meet... even on the last round. Heck, if I were on one of those, and I saw my counterpart play a weaker team than I did, because I have a good SoS, and score a blitz while I have a tough match, I am surely going to be very annoyed!

 

At its root is the fact that there is too much randomness inherent in short matches, even with duplicate boards. The current method of matching by VP scores is simple and reasonably useful..... in the context of the events run as Swiss matches.... maybe your Australian event would be an exception to this proposition, but even there, so long as the swiss is merely a precursor to more stringent metrics of ability, then the good teams will usually make whatever the cutoff is for the next stage.

 

In the meantime, for the 'bridge for the masses' fields for which Swiss was adopted, my advice is don't fix what ain't broken. The masses like and understand VP swiss without SoS metrics... and never, ever put in place a scoring system that can be applied only by a computer utilizing data unknown to the players.