View Full Version : NOT format
Lets hear what you think of this year's format.
And what you think the best format possible is (within the time constraints)
I've got a few thoughts on this:
1. Final should be on Sunday, not Monday, so that there are still people hanging around playing in the side event(s) as the winners are decided.
2. Make the SWPT a single field, preferably at a single venue.
3. Shorten the qualifying to three days, perhaps something like 11 x 16 board matches (3 on Monday, 4 on Tuesday and 4 on Wednesday).
4. Have 20 teams qualify for the NOT and then seed them into 4 pools of 5 teams based on finishing position. Each pool will play a round-robin of 16 board matches on Thursday with some carryover, but not so much to have teams that just scraped into the top 16 out of it before they've even started (probably a 10VP spread from 1st to 5th with decimals so that there can't be a tie for progression to QF). Have the top 2 from each pool go to the quarter-finals. 64-board QF on Friday, 64-board SF on Saturday and 64-board Final on Sunday.
5. Possible variation would be to make the SWPT 10 x 16 board matches to accommodate an NOT match on Wednesday night. Key benefit of this is that the QF, SF and Final will all have 1st segment on preceding evening and 3 segments the next day leaving enough time for people to drive home or get early-evening flights when they get knocked out.
6. I'm not a great fan of the evenings-off. I go to Canberra to play bridge, not to sample the local night-life.
7. I think the ABF should seriously look at providing some incentives for international teams and/or players to lift the profile of the event and the standard of the event.
8. For the financial viability of the event, it is important that gross table number are maintained, so an attractive array of side events will need to be developed. However, making the entry fee for the SWPT include/assume playing in side events up Saturday night could underwrite the finances of the event.
It is a good use of this forum, discussing event structures.
Last year I wrote to the editor of AB:
-----------------
The number of rounds needed in a Swiss teams event is determined by a simple mathematical formula. The objective of a Swiss event is to guarantee that all winning teams have met each other before the event is complete. This number of rounds (for the number of teams ?N?) is found by log2N (rounded up). That is, it is determined by powers of 2 so that 5 rounds are needed for 17 to 32 teams, 6 for 64, 7 for 128 etc.
Therefore the VCC with its 10 rounds could handle up 1024 teams and the NOT Swiss 14 rounds up to 16,384!
As you can see the 9 rounds suggested by you (finishing Wednesday night) is more than adequate catering for up to 512 teams. To start a new ?National Swiss Teams? on Thursday concurrent with the 16 team knockout would cater for all players at the event. This event could also have 9 or less rounds, with its own finals concurrent with a consolation event.
As we all know, by the Wednesday night session the endurance factor has already kicked in and as you have pointed out the Swiss from this point on is simply a grind for all involved. ....<snip>
------------------------------
The empirical evidence from this year indicates the number of rounds in the Swiss could have been 9 without much effect on the outcomes.
The top 10 teams were (in number order)
Rydges:
After 9 rounds
1,2,4,5,8,14,21,26,31,39
After 14 rounds
1,3,4,5,6,8,12,14,19,21
Hellenic:
After 9 rounds
1,2,3,5,7,8,10,12,15,21
After 14 rounds
1,2,3,5,7,8,10,12,15,16
As an organiser of such an event I would be entirely happy with the results after 9 rounds. The movement has done its job.
The next decision is what to do after round 9?
Speaking as a member of team 21 from Rydges this year, I would've been delighted if the music hand stopped after Round 9 when we were running 7th rather than Round 14 when we were running 10th!
On the other hand, in 2006 when my team finished 6th at the NCC, after Round 9 we were languishing in 12th place but had just played consecutive matches against 6 teams seeded 22 or less including 1, 2 & 5. At that point of the event only 3 teams had higher VPs of opponents played, suggesting to me that differentiation amongst the "borderline" teams was under-swissed.
As far as the mathematics of swiss movements is concerned, it's my understanding that determining the number of rounds by powers of two is fine for determining a winner, but may not be sound for differentiating the lower places. It may not matter too much because if a seriously contending team can't make the top 4, they probably wont have much chance of winning the event anyway.
Over-swissing really seemed to impact badly on the McManus team (seed 2 at Rydges) this year with them sitting pretty in 2nd or 3rd place for most of the qualifying and then hitting seeds 1, 3 & 10 in the last three rounds. They actually wound up with the higher VPs of opponents played than any team in the field, including the winners.
revert2b
08-02-07, 01:51 PM
All I can say is that you don't have to be a genius to realise that there is something inherently flawed when you play 64 boards to reduce the field from 16 teams to 8, then only 32 more to get down to 4 teams.
I agree. It is patently absurd to play 280 boards to decide 16 teams then play a 32 board knockout. 32 boards is not long enough for any knockout match and it is ridiculous when you have just spent 5 long days playing a qualifying. If this year's final was played in 32 board sets you would have two convincing winners, depending on which 32 board set you chose.
I have been telling the organisers this for several years (see AB editorials) but they always come up with some new and usually stupid variation of their own. I don't know who makes the decisions but the one thing the ABF has to recognise is that those who are charged with the job right now are not the right people.
You don't have to be clever to work out a good format - just do what they do at major events overseas!
3 days qualifying is TONS. If this doesn't seem obvious to you already, read Ian's piece. Isn't that convincing? Once you accept that, everything else falls into place.
paul
DavidStern
09-02-07, 09:45 AM
Our team were delighted to be meeting OzOne over 32 boards as we felt this gave us a better opportunity to beat them than over 64. I certainly felt that giving teams 1-4 a second chance not offered 5-8 as being desireable but question whether teams 1-4 should be initially be playing each other rather than 5-8. I am not dismissive of this year's format but think it could be fine tuned.
Dealing specifically with David Thompson's comments only because he has provided a lot of thoughts:
1. Final should be on Sunday, not Monday, so that there are still people hanging around playing in the side event(s) as the winners are decided.
AGREE TOTALLY
2. Make the SWPT a single field, preferably at a single venue.
THE ONLY PLACE THAT THIS IS POSSIBLE IS THE CONVENTION CENTRE WHICH IS FINANCIAL UNVIABLE. SPLIT FIELDS ARE PROBABLY HERE TO STAY
3. Shorten the qualifying to three days, perhaps something like 11 x 16 board matches (3 on Monday, 4 on Tuesday and 4 on Wednesday).
16 BOARD MATCHES PROVED UNPOPULAR - ALSO 4 MATCHES A DAY IS NOT FAVOURED AMONG 6 PERSON TEAMS - ORGANISING LOGISTICS ETC
4. Have 20 teams qualify for the NOT and then seed them into 4 pools of 5 teams based on finishing position. Each pool will play a round-robin of 16 board matches on Thursday with some carryover, but not so much to have teams that just scraped into the top 16 out of it before they've even started (probably a 10VP spread from 1st to 5th with decimals so that there can't be a tie for progression to QF). Have the top 2 from each pool go to the quarter-finals. 64-board QF on Friday, 64-board SF on Saturday and 64-board Final on Sunday.
DISAGREE TOTALLY. THE BEST TEST OF WHO IS THE LEADING TEAM IS LONG KNOCKOUT MATCHES
5. Possible variation would be to make the SWPT 10 x 16 board matches to accommodate an NOT match on Wednesday night. Key benefit of this is that the QF, SF and Final will all have 1st segment on preceding evening and 3 segments the next day leaving enough time for people to drive home or get early-evening flights when they get knocked out.
6. I'm not a great fan of the evenings-off. I go to Canberra to play bridge, not to sample the local night-life. WHILE I UNDERSTAND THAT POINT OF VIEW IT IS A MINORITY VIEW GIVING JUST ABOUT EVERY OTHER EVENT WORLDWIDE. WHY NOT ENJOY WHAT CANBERRA HAS TO OFFER OR THE COMPANY OF YOUR TEAM MATES? YOU DONT PLAY ANY LESS BRIDGE BY HAVING THE NIGHTS OFF YOU SIMPLY START EARLIER - I DONT GO TO CANBERRA TO SIT AROUND A HOTEL ROOM IN THE MORNINGS EITHER
7. I think the ABF should seriously look at providing some incentives for international teams and/or players to lift the profile of the event and the standard of the event.
AGREE TOTALLY
8. For the financial viability of the event, it is important that gross table number are maintained, so an attractive array of side events will need to be developed. However, making the entry fee for the SWPT include/assume playing in side events up Saturday night could underwrite the finances of the event
So now to my viewpoint.
Two problems have to be considered in regard to the format. Firstly table numbers are dropping
Also as the seeding committee noted the standard of teams past the obvious contenders is dropping away dramatically a lot of this due to bridge professsionals playing in lesser-contending teams (OK I am being polite) think Val Cummings, Kieran Dyke, Murray Green, David Beauchamp, Ted Chadwick, Paul Lavings, Tania Lloyd, Ashley Bach, Michael Cornell, Seamus Browne and others. So we need to look at just how many should be qualified. Maybe 16 teams is not that relevant. Maybe we should qualify the top 2 to the seni finals on Saturday and have the next 4 in each section play a knockout for the other 2 places (noting of course the benefit the top 2 get by having a day off) I don;t know but this forum is a good start.
There is no reason that the event could not be run as follows:
dx Last Train starts on Sunday and Monday
dx Women's and Seniors Qualifying Tuesday Wednesday Thursday
dx Women's and Seniors Finals Friday and Saturday
dx Multi Pairs Friday
dx Swiss Pairs Saturday
dx SWPT Sunday Monday Tuesday Wednesday
dx New 2 Day Butler Pairs, Match Point Pairs or Teams on Thursday Friday (or swap with Mixed Teams and these events on Weekend)
dx 64 board round of 16 Thursday
dx 64 board round of 8 Friday
dx 64 board round of 4 Saturday
dx 64 Board Final Sunday
dx Mixed Teams Saturday and Sunday
From my point of view the priroties in order are:
dx Finish on Sunday
dx Long knockout matches in the final
dx 20 board matches in the qualifying
dx No evening play
dx One venue
The bigger problem than discussing format here is to get the convenor and ABF to buy into the need for change. perhaps the drop from almost 300 teams to 207 is a wakeup call that they should listen to.
Lets not forget there are two parts to the NOT.
The SWPT (Swiss) is played by say 300 teams and the NOT by 16.
A massive rework of the format would affect all those teams and the convenors need to get their input before taking such a drastic step.
Yes, i agree that a shorter 3 day swiss and then long matches would be ideal but it does seem to serve the needs of just the top teams.
If the other teams feel similarly, then sure, the ABF should make such a change.
My attempt was to try and work out a good format within the constraints that we do have today. i.e. Starting Friday and ending Monday.
My first suggestion is a simple one.
Finish swiss off on Thursday evening
Friday, Sat, Sun, Mon : 4 64 board knockout matches
My second suggestion is
Finish swiss off Friday morning (after one match)
Rest of friday : 40 board R/16
Sat, Sun, Mon : 64 board QF/SF/Final
The format needs urgent fixing for everyone, good and average player alike. To play 5 days of qualifying, which for any decent team is a total waste of time, is not an option. 3 days is a huge stretch but I suppose we could live with that for historical reasons. And how does this format suit the average player? They sit there day after day with no hope of any success in terms of the event.
The truth is that this is a crap format for everyone. When I used to complain about the format the administrators would say but it is popular. Well, that is not the case any more. The NOT is heading for extinction as fast as the gold coast is shaping up to be a great event.
I pretty much agree with all the points David makes except I would like to see a prestigious match point event back on the calendar. There is more to life than swiss imps, whether it be pairs or teams!
When you look at the information sent by Ian about the difference between 9 rounds qualifying and 14 rounds, it is very similar. In fact at the hellenic club the top 9 teams stayed the same!! Therefore I think 3 days of swiss is more than enough. Then there should be longer knockout matches. The rest of the feild should either play on or have a different event starting.
Ish
In response to David Stern:
David Stern wrote:
>Our team were delighted to be meeting OzOne over 32 boards as we felt >this gave us a better opportunity to beat them than over 64.
Indeed so. David even used this fact to motivate our team successfully.
David Stern wrote:
>Two problems have to be considered in regard to the format. Firstly table >numbers are dropping
According to many people I have asked, the drop from almost 300 to about 200 teams ia mainly because much of the bottom half of the SWPT field has switched to play in the three day Women's or Senior's events because the five day SWPT is too long. Thus one could poll the players, but they have already spoken by rejecting the SWPT's format and are no longer playing in the SWPT, so any poll would be ineffective and biased.
David Stern wrote:
>There is no reason that the event could not be run as follows:
>Last Train starts on Sunday and Monday
>Women's and Seniors Qualifying Tuesday Wednesday Thursday
>Women's and Seniors Finals Friday and Saturday
>Multi Pairs Friday
>Swiss Pairs Saturday
>SWPT Sunday Monday Tuesday Wednesday
Peter Gill writes:
Australian Open Pairs Championship on Thursday/Friday
Losers in Round of 16 and Round of 12 can go into AOP Final on Friday.
64 board round of 16 Thursday, played in two stages of 32 as in 2007
(or could be 32 boards (16-->12) and 40 boards (12-->8)
64 board round of 8 Friday, 10am start
64 board round of 4 Saturday, 10am start
64 Board Final Sunday, 10am start
Mixed Teams Saturday and Sunday
My two modifications to David Stern's schedule are:
(1) Reintroduce AOP, giving masterpoints to top 90% of field in very generous quantities. As Bill Jacobs in recent VBA Bulletins has pointed out after his research, the main reason why matchpoints is losing out to Swiss Pairs in Oz is that mp pairs gives way less mpts to the punters than Swiss Pairs, for no justifiable reason except some superseded silly ABF formulae. So fix this, give MUCH MORE matchpoints to the AOP (to the bulk of the field , not to the winners) than to any Swiss Pairs events, and the problem will be solved if the matchpt masterpt change is properly publicised. Refer to VBA Bulletins in recent months for more info.
(2) John Brockwell's reply to Sartaj's private email (not on this forum) makes a very convincing case for retaining the 16 --> 12 and 12 --> 8 format used this year to reduce the 16 teams to 8. That was not the problem, in fact it was a good idea IMO. The problem was that two Q'F's were too short (Oz-One vs Brogeland and Noble vs Pepsi/Vainikonis).
I have gone through the qualifiers for the last five years (ten sections) on the Thursday night and two rounds later. There is little difference, except that Ish's Rothfield team who won the NOT a couple of years ago would not have qualified, probably justifiably. Right now, I do not have time to list the data here.
Peter Gill.
Paul wrote:
>I don't know who makes the decisions but the one thing the ABF has to >recognise is that those who are charged with the job right now are not the >right people.
I totally disagree.
(1) I do know who makes the decisions. The ABF Tournament Committee. Not the Convener. The ABFMC approves the TC's decisions, or (rarely) refuses to accept the TC's decision. The names of all such people are on the ABF website. The TC includes (or included) many good thinkers, at least two of whom have been or are on the Oz-One Committee. I think your criticism is thus an inadvertent implied criticism of the selection of people on the Oz-One Committee, and is unjustified in my opinion.
(2) John Brockwell's reply to Sartaj's email (ref my other post, immediately above this post, it is not on this forum) was logical and well thought out, IMO. John argued strongly that qualifying 8 teams from each section is good -and I found his arguments convincing. John argued further that Qualifier 8 should not be on the same footing as Qualifier 1 - and I found his arguments convncing.
>You don't have to be clever to work out a good format - just do what they >do at major events overseas!
I am unconvinced by this. The winners of this week's NEC Cup scraped in through the too-short Swiss by getting a lucky draw and a max win in the last round. There is plenty of evidence that NEC Cup and Yeh Bros Swisses are too short, and that low qualifiers are favoured by the unfair formats overseas. USA does not use Swiss Qualifying so we cannot compare with them.
Off-topic:
To demonstrate that this administrative stuff ain't easy:
At the NOT, my teammate (or megateammate?) Boye Brogeland pointed out to me that successful countries such as Italy and Norway have one selector who selects the National Open team. I told him that Australia has selected the following National teams: 2007 U21 team (made semis of WC), 1991 U26 team (made semis of WC), 1989 U26 team (made semis of WC), 1968 PABF team (came 1st), 1969 PABF team (came 1st). Our other teams have not been selected by selectors, although the 1973, 1979 and 1989 Australian Open Teams which came in the top four in the world did have a pair selected onto the team (Marston-Burgess in 1989). In summary, every single successful Aussie team overseas in our history has had some element of selection by selectors involved!!! When I told Boye, he was not surprised, he thought this was exactly as one would expect.
Do we select the Aussie cricket team by getting Qld to play off against WA?
I was reading some NZ Bridge magazines from the 1970s the other day. A young chap named Paul Marston suggested that the Playoff Winning Team play off against a team selected by a selector (or selectors), with the winner of the match being the National Team. What a great idea. Why has nobody ever done it? Because bridge admin is not easy?
Peter Gill.
(1) I do know who makes the decisions. The ABF Tournament Committee. Not the Convener. The ABFMC approves the TC's decisions, or (rarely) refuses to accept the TC's decision. The names of all such people are on the ABF website.
In my well informed opinion the decisions on the format of the whole tournament are the convener's. All decisions are Sean's and the success of the event is on his shoulders. We should seek Sean's input on this thread.
[QUOTE=Peter]Paul wrote:
>I don't know who makes the decisions but the one thing the ABF has to >recognise is that those who are charged with the job right now are not the >right people.
I totally disagree.>
Are you suggesting that the organisors are doing a good job when it comes to NOT format?
I don't think there would be any bridge player of standing in the world who would agree with you. We sat around playing two days of unecessary swiss only to advance to then play very short knockout matches. How can this be right?
>You don't have to be clever to work out a good format - just do what they >do at major events overseas!
I am unconvinced by this. The winners of this week's NEC Cup scraped in through the too-short Swiss by getting a lucky draw and a max win in the last round. There is plenty of evidence that NEC Cup and Yeh Bros Swisses are too short, and that low qualifiers are favoured by the unfair formats overseas. USA does not use Swiss Qualifying so we cannot compare with them. >
When I said events overseas I was referring to ACBL and WBF. Of course, the Yeh Bros and NEC have flawed formats. Last year the Yeh organisors were forced to change the format because the northern hemisphere players were refusing to come because the KO matches were too short (24 boards). And they were being paid to come!
Note the NOT had 32 board KOs and the gold coast had 24 board KO matches. And it seems the swiss is going to have 2 extra rounds next year!!
For what reason? Look at the teams that got through this year. Could anyone suggest that they deserved to be in the top 6 but were deprived by the lack of matches? Of course not. The top six were seeded in order: 2, 1, 11, 12, 5, 3. But we are being forced to turn up a day early to play pointless swiss matches followed by the same unacceptably short KO matches.
The GCC will lose the overseas players just like the NOT has if they go down the same misguided path as the NOT.
DavidStern
24-02-07, 10:08 PM
Just as a throw away line - how about 10 (or 9 if we have to finish by Friday afternoon) rounds of swiss with 28 board (56 boards per day) matches. Why should we be so hung up on 20 board matches.
This will likely lead to more 'stable' and meaningful results while at the same time keeping the field there for the requisite number of days.
Anyway just some thought for more discussion.
D
Paul wrote:
>I don't know who makes the decisions but the one thing the ABF has to >recognise is that those who are charged with the job right now are not the >right people.
Peter wrote:
I totally disagree.
Paul wrote:
>Are you suggesting that the organisors are doing a good job when it comes >to NOT format?
Not at all. The people on the ABF Tournament Committee right now are not all the same people as 12 months ago.
>I don't think there would be any bridge player of standing in the world who >would agree with you. We sat around playing two days of unecessary swiss >only to advance to then play very short knockout matches. How can this be >right?
You are assuming that the ABFMC and ABFTC personnel are uinchanged - a false assumption.
>Note the NOT had 32 board KOs and the gold coast had 24 board KO >matches. And it seems the swiss is going to have 2 extra rounds next year!!
I have been working on this in the last few days, to prevent the two extra matches at the GC, mainly by talking to Therese Tully yesterday afternoon and last night.
Therese Tully told me this weekend that if we do something within a few days, there's a good chance that she will reverse the decision that was announced (in writing I think) at the GC the other day - about having two extra rounds.
I will email Therese Tully to try to get your preference to happen.
Peter Gill.
David Morgan
25-02-07, 02:12 PM
I agree that significant changes are required. As others have noted, these have to accommodate the needs of serious players and those for whom the event is purely recreation. I think the following would be on the lists of many in both groups:
* finals when others are around (and schedule them so that people are encouraged to watch on Viewgraph; if security issues can be managed, have the field [in whatever event is running] play the same boards as the finalists so they can compare what they did with what the experts do);
* no night matches; and
* a single venue (but not if the cost is excessive).
Issues that are important for the serious players include:
* knockout matches of meaningful length; and
* shorter/better qualifying so that fatigue is less of a factor (it will always be a factor; I'd argue that it should be less of a factor).
I'm agnostic on 4*16 vs 3*20; one advantage of 3*20 is that it is much easier to schedule (not that the organisers got it right this year as the time between matches two and three was too long); another is that it simplifies choice issues for teams of six.
In the days before the proliferation of shorter events in Canberra I proposed using a Rosenblum Cup-type format. (A variation if the final is to be held on the Sunday, 3*20 boards, and reintegration from the Swiss repechage into the Round of 16:
Mon: 64 groups of three/four/five [depending on entries] play a round robin with one team progressing to 40-board knockout matches
Tues: Round of 64
Tues/Wed: R32
Wed: R16
Thurs: either two rounds of Swiss to qualify top 8 or 8 losers from R16 play 40-board knockout vs top 8 from Swiss
Thurs/Fri: R16
Fri: R8
Sat/Sun: 64-board matches.)
While there are many advantages to such a format (including that good teams play long matches [at least 40 boards] against other good teams at all stages after Day 1; and that non-serious players can continue with a tried and true format, although this might now be only three days), there are some disadvantages to such a format. One is that seeding is more of an issue as this influences both the original group a team is placed in as well as its likely future opponents. (This need not be a huge hurdle, though: the ACBL has been doing it, including for overseas players, for decades.) Another is that it really needs the event to take place in a single venue. Personally, I think that the advantages for serious teams outweigh the disadvantages and this would be better than a Swiss of 9 rounds followed by R16/R12 then knockout.
David
Given the threads about Canberra 2007, what about a similar discussion about Gold Coast 2007.
This is an event that actually surveyed players and has a governemnt grant to market the event to make it a better and stronger event for the future for all players (novice to expert). What needs to happen ?
Any thoughts on the venue and a single field. Broadbeach is more classy than Surfers.
Do people like the presntation dinner
I have been working on this in the last few days, to prevent the two extra matches at the GC, mainly by talking to Therese Tully yesterday afternoon and last night.
Therese Tully told me this weekend that if we do something within a few days, there's a good chance that she will reverse the decision that was announced (in writing I think) at the GC the other day - about having two extra rounds.
I will email Therese Tully to try to get your preference to happen.
Peter Gill.[/QUOTE]
Gee, it will be great if you can achieve that.
paul
I wrote:
Therese Tully told me this weekend that if we do something soon, there's a good chance that she will reverse the decision that was announced (in writing I think) at the GC the other day - about having two extra rounds.
Paul wrote:
Gee, it will be great if you can achieve that.
Me now:
At the Playoff Therese said that no decision has been made at this stage.
Therese gets the results of the GC survey today (Tuesday).
She is finding that there are a lot of people - both top players and average players - who would prefer 12 rounds, not 14 rounds. However, I think the ABF would strongly prefer her to have 14 rounds, on the ABF's belief that 12 rounds is underswissed.
So if anyone has any info about 12 rounds being enough Swiss rounds, then please post it here - the info will then reach Therese.
Her other concern is the new Consolation event on the Friday/Saturday. Her current thoughtas are:
- perhaps the Swiss Pairs should use a One-round Delayed Draw, like the Yokohama Swiss Teams - the consolation event at the NEC Cup? This would avoid the undesirable delays between rounds caused by the massive field size.
- there should be two simultaneous Consolation events (like the American Nationals) - a one day Consolation event on the Friday (Swiss Pairs?) and a two day Consolation event over the Friday/Saturday (Mixed Teams?) so that players can plan to leave at times that suit them.
- if so, then what should the session times be? Should the Swiss Pairs be pm/evening or am/pm or am/pm/evening on the Friday? Should the ANA Pairs on the Satuday morning be reinstated as a third option? Then it would be ANA Pairs Sat am and Mixed Teams (Fri 2 sessions Sat am/pm), with Swiss Pairs on Friday. ls that the best way to go? She said she would like to make the consolation events better, but is unsure whether to avoid a late finish on the Friday night or to avoid morning play on the Friday.
These issues might sound irrelevant to you, but Therese's point is that if there are to be 12 rounds instead of 14, then (1) the ABF has to be made happy with the idea and (2) the format of the Consolation event(s) has to be dramatically improved.
A more radical suggestion was to slot the Australian Open Pairs in on the Friday and Satuday. That one is a bit left field and unlikely unless there is public support for the idea.
Feedback most welcome - by posting here. Your guidance and thoughts could help make the GC format better.
Therese and I will have another chance to chat in Banduing in June, when we both play for Australia at the PABFC.
Peter Gill
I have been thinking for some time that it would be great to do some work on the theory of Swiss events ... there seem to be great numbers of opinions but very few facts to back them.
Does anyone with a mathematical background want to get involved in that??
And, does anyone know where we might get hold of historical data from actual Swiss events such as the NOT and the GC?
As examples of the things we don't really know .... Ian McKinnon rightly quotes a formula that a Swiss with N rounds will accommodate (2 to the power N) entrants: so, 5 rounds for 32 teams, 6 rounds for 64 teams etc.
However, this formula is the right answer to a question totally unrelated to bridge - how many rounds to get a single winner under conditions where the same team will always win when two teams play each other?
That's different from the NOT or GC in at least two important ways ...
first, we want more than one winner eg 4 or 8 or 16 teams to qualify for the next stage, and
second, we know for sure that at bridge team A will not beat team B 100% of the time in a short match, even if team A is considerably better.
So, at bridge ... how many rounds do we need to select a winner? How many rounds to select the top 4 teams? 8 teams? 16 teams?
And, what difference does the number of boards per round make to the answers?
Etcetera, etcetera.
I have been thinking for some time that it would be great to do some work on the theory of Swiss events ... there seem to be great numbers of opinions but very few facts to back them.
Does anyone with a mathematical background want to get involved in that??
And, does anyone know where we might get hold of historical data from actual Swiss events such as the NOT and the GC?
As examples of the things we don't really know .... Ian McKinnon rightly quotes a formula that a Swiss with N rounds will accommodate (2 to the power N) entrants: so, 5 rounds for 32 teams, 6 rounds for 64 teams etc.
However, this formula is the right answer to a question totally unrelated to bridge - how many rounds to get a single winner under conditions where the same team will always win when two teams play each other?
That's different from the NOT or GC in at least two important ways ...
first, we want more than one winner eg 4 or 8 or 16 teams to qualify for the next stage, and
second, we know for sure that at bridge team A will not beat team B 100% of the time in a short match, even if team A is considerably better.
So, at bridge ... how many rounds do we need to select a winner? How many rounds to select the top 4 teams? 8 teams? 16 teams?
And, what difference does the number of boards per round make to the answers?
Etcetera, etcetera.
You have raised good points.
I would say they should cut the swiss at the point they can be fairly sure that the winner is in the cut. Look at the NOT and see how many times the winner was not in the cut after 9 rounds. I haven't done the check but I would guess almost never. This is why the next 2 days are a complete waste of time in terms of finding a winner, and that is surely the main point of the event. I might add that you will also find the better teams going better early if there is an early cut. For example, I think my team would have missed the cut this year after 9 but I can tell you that that would have been most unlikely had the cut come at 9. We knew we were going for a sunday drive.
In short, we churn away for 2 extra days only to go into knockouts which would definitely produce different winners quite often if they were a bit longer.
I defy any clear thinking person to convince us that the current format is anything short of plainly wrong from the point of view of winning the event.
I also spoke to Sean about this forum. Where are you sean? I hope you are busy and not hiding :)
However, this formula is the right answer to a question totally unrelated to bridge - how many rounds to get a single winner under conditions where the same team will always win when two teams play each other?
Correct. Swiss events were devised (I believe) for chess where there is a win, lose or draw outcome. If we chose to have events in bridge where you have win,lose,draw then the Swiss is ideal. We do not have this situation and results are converted to VPs. A team winning 16-14 in every match in NOT would not qualify even though they never lost a match.
Swiss teams is a "bridge movement". The organisers choose it together with the number of rounds and qualifiers to define the movement. That is the rules chosen. You get a result and proceed from there.
In my experience using log2N + 2 as the number of rounds will generally give an equitable result. Of course things happen that are undesirable but as you say the best team does not always win in bridge. It is the nature of the game and one of the reasons why we play.
Correct. Swiss events were devised (I believe) for chess where there is a win, lose or draw outcome. If we chose to have events in bridge where you have win,lose,draw then the Swiss is ideal. We do not have this situation and results are converted to VPs. A team winning 16-14 in every match in NOT would not qualify even though they never lost a match.
Swiss teams is a "bridge movement". The organisers choose it together with the number of rounds and qualifiers to define the movement. That is the rules chosen. You get a result and proceed from there.
In my experience using log2N + 2 as the number of rounds will generally give an equitable result. Of course things happen that are undesirable but as you say the best team does not always win in bridge. It is the nature of the game and one of the reasons why we play.
I received this from a US mathematician friend of mine, Richard Willey
Hi Paul
I've been reading the Oz One forums over the past couple months. I haven't seen fit to comment on any of the discussions, however, the recent debate regarding optimal design for the NOTs has peaked my interest to some extent. I'm stuck doing a fair amount of work with math and statistics. If you can get me the right information, I might be able to produce a reasonable paper. Who knows... It might even validate the "log2N + 2" rule of thumb that Ian was discussing.
(Please feel free to post this to the forums. If you prefer, ask Ian to all "Hrothgar" to post)
My natural inclination is to approach this as an applied statistics problem. In theory, we can model bridge match between two teams as statistical sampling. All statistical sampling problems work the same way. You have a population statistic. (In this case, the population statistic is the relative strength between the two teams). You also have a sample statistic. The sample statistic is your estimation of the population statistic based on some experimental trials that you are conduction. If your two teams play a 24 board match, your sample statistic will be derived from these 24 separate samples. In general, the greater the number of samples you have, the closer your sample statistic will be to your population statistic.
Its important to note: Any kind of statistical model is by definition imperfect. Statistics don't prove anything with absolute certainty.
Rather, everything is based on a so-called "confidence interval". At the end of the day, your confidence interval suggests that we're 95% sure that team A is actually strong than Team B. The more samples you take, the more narrow that confidence interval becomes. If you play a seven board match, your confidence interval might (hypothetically) be 68%. Increase the match length to 24 boards, and the confidence interval will shrink to 95%. Increase the match length to 128 boards and your confidence interval is now 99%. (So far, none of this should be overly surprising. I suspect that everything will correspond with folk's intuition)
Now we get to the tricky part: How would one use this structure to try to improve the NOT format. The problem is relatively
straightforward:
We have some basic parameters
The total length of the event is X hours It takes Y minutes to play a board Setting up a round of a Swiss match has a fixed overhead of Z minutes
How should one allocate the number of rounds and the number of boards per round in order to narrow the confidence interval the most?
Unfortunately, solving this problem is going to be quite tricky. In part, there is going to be some real issues getting some of the basic information that we need to fully parameterize the problem. In particular, we need some kind of estimate that we can use to approximate the relative strength of the top teams. If the difference in strength is "large" then you aren't going to need that many boards to find a winner. Alternatively, if the top two teams are closely matched you're going to need a lot more boards to produce the same confidence interval.
In addition, we're going to need some estimates about the variance in board results. Lets assume that we have three different Teams (A, B, and C). A and B are evenly matched. Both are considerable worse than C. On any given board, we expect that team C will beat either team A or B by 1.3 IMPs. However, Team A is playing a high variance / top or bottom type system while team B is playing things down the middle.
The style of play is going to impact the results. The higher the variance of the methods that people are using, the more boards you're going to need to get the same level accuracy.
Finally, there are going to be some issues that can best be summarized as political in nature. Its entirely possible that an optimal tournament design can't be implemented because it looks different from what people are used to. Here's one very simple example: Every swiss team match that I've ever played in uses a constant number of boards per round. Round 1 has seven boards, round 2 has seven boards, round 10 has seven boards, ad infinitum. Think back to some of my earlier comments. The first few matches of a Swiss teams event are very random with lots of strong teams playing against weak teams. Over time, things settle out and teams rise (or fall) to their own level.
Once you've hit the midpoint of the event teams are much more closely matched. By the time you hit the finals, you hope that you have your top teams meeting at tables one and two. I can pretty much guarantee you that the optimal design for a Swiss Teams event will increase the number of boards played per round over the course of the tournament.
Hopefully, this will give you some kind of idea about the type of information that we'd need to do a "proper" analysis of the NOT format. In an ideal world, it might be possible to get some of the raw data from board results from earlier NOTs / SWPT events. I'm not sure if wireless scorepads have made their way into team events or if they're confined to clubs. If they are in widespread use, we probably have access to some fairly good data.
You can't however reduce early matches to too few boards. Even with the current length of matches, there are always a few seeds who don't win their first match. Reducing the number of boards in early matches would probably result in more upsets.
Perhaps, though, a more savage mid-event cut approach is needed. No team should really have the opportunity to have that early 'whoops. Nevermind, we've got plenty of time to get back'. Albeit, the nature of a swiss inherently lends itself to giving an expected 'easier run' for the next round or two.
In an ideal world, it might be possible to get some of the raw data from board results from earlier NOTs / SWPT events. I'm not sure if wireless scorepads have made their way into team events or if they're confined to clubs. If they are in widespread use, we probably have access to some fairly good data.
Richard,
The net IMP result of every Swiss Teams match played in the SWPT for at least the past 3 years is available through www.bridgeunlimited.com or follow the links from the ABF website. However, you'll need to do a lot of clicking and pulling-down to get what you need. Perhaps if you could define exactly what data you are after, I could speak to the ABF-scorer, Martin Willcox about getting an appropriate data dump from his database. If I understand correctly, the fields are seeded down to about 30 or so teams and then ranked by masterpoints thereafter so it should be possible to work out the volatility and expected match result between teams separated by x seeding positions.
Dave.
DavidStern
29-03-07, 01:37 PM
I could speak to the ABF-scorer, Martin Willcox about getting an appropriate data dump from his database
OH Good Luck!
Bruce Neill wrote:
>on the theory of Swiss events
>Does anyone with a mathematical background want to get involved??
Ross Moore used to be the best mathematician in this part of the world at this sort of stuff. Alas he is in Siwtzerland for 5 months, but can be emailed.
Peter Buchen is the other obvious person to contact.
>And, does anyone know where we might get hold of historical data from >actual Swiss events such as the NOT and the GC?
The data is beautifully presented on the ABF Website. You click events, choose your event (NOT or GC) then can easily flick back from year to year.
It takes almost no time at all to gather dtaa such as the following.
Top 8 after x rounds (based on Seeding)
NOT 2007
Rydges Non Rydges Venue
9 10 12 14 9 10 12 14 = number of rounds
5* 1 5* 5*(3) 1 2* 1 1
4 5* 12 8 2* 1 2* 2*(1)
1 4 2 3*(2) 8 8 8 8
26 31 26 1 10 10 4 3*(3)
8 26 1 19 12 3* 3* 12
2 2 3* 6 5 5 5 16
21 7 4 4 15 12 12 16
31 3* 19 14 17 15
* = semi-finalist (1) = winner (3) = lost semi-final
Team 3 at Rydges was Noble, well known for their big Swiss finish, when the four pros play every match at the end.
Allowing for the Noble factor, we already have an indication that any of 9,10,12 or 14 rounds would have been fine.
Can anyone remember who wrote the article in AB magazine in 2003 (or perhaps 2004) about the NOT Swiiss draw and the mathematical analysis of NOT qualifiers. Was it by David Hoffman?
I will see how the spacing comes out, by posting this, then do another post with the data for the four years 2003 to 2006.
Peter Gill.
OK, so the spacing comes out as a mess.
Let's try again, using 2006.
Four columns are 9-10-12-14 rounds
NOT 2006 Rydges
1 1 1 2
2 8 2 8
8 2 8 4
11 6 4 1
6 12 6 12
3 3 12 6
4 9 20 9
7 10 3 3
Team 2 came 2nd and Team 4 came =3rd.
NOT 2006 Other Venue
5 5 1 1
1 1 5 5
2 4 2 4
10 2 4 2
4 6 6 6
6 10 14 44
8 44 44 14
17 11 8 8
Team 1 came 1st and Team 5 came =3rd.
Very striking, the top five being almost the same each time!
Peter Gill.
Much better, here's the other years
2007 (repeated due to mess in first post)
Four columns = 9-10-12-14 rounds.
2007 NOT Rydges
5 1 5 5
4 5 12 8
1 4 2 3
26 32 26 1
8 26 1 19
2 2 3 6
21 7 4 4
31 3 19 14
Team 5 came =3rd and Team 3 came 2nd.
Team 3 = Noble, who finish strongly.
2007 NOT Other Venue
1 2 1 1
2 1 2 2
8 8 8 8
10 10 4 3
12 3 3 12
5 5 5 16
15 12 12 10
7 9 17 15
Rounds 9,10 and 12 look better than Round 14. Teams 16 and 15 swissed it in with big finishes. Seeds 4 and 5 missed out due to tough last round draws.
Team 2 came 1st: hurray! my team :) and Team 3 came =3rd and Team 1
was unlucky to lose a short quarter-final match (ref other threads).
2005 NOT Rydges
1 1 1 1
10 10 5 5
5 5 10 10
3 95 3 3
4 3 13 13
16 115 4 16
20 47 9 20
6 4 11 4
95 = Griff Ware, an underseeded youth team.
115 = M Whibley, an underseeded yotuh team
Rounds 9 and 14 are similar (6 Courtney out and 13 Gosney in)
Team 1 came 2nd
2005 NOT Other Venue
7 7 7 11
8 5 8 7
13 8 5 9
5 3 11 5
3 1 13 3
28 4 9 2
38 11 4 17
11 13 3 6
Team 2 Rothfield won the Final after deciding not to play their two sponsors for more than 25% of the boards in the Finals, transforming the team's strength. Teams 5 and 6 (Wiltshire) ran =3rd. This was the weird year when Team 11 Ewart swissed into top position, and Team 1 Bremner-Moore was rarely in the hunt. Howerver, even in this strange year, the qualifiers after 9 or 10 rounds look like a better group than after 14. Yes, I know - the winners of the Grand Final don't then qualify - but remember what Ish said when they won the Final easily - that their team didn't even deserve to qualify.
I think that a straight mathematical analysis might miss some data such as teams like Rothfield and Noble changing their strength from time to time.
History of Swiss:
First used in chess event in Zurich in 1895, invented by Muller, hence its name. First used at bridge in 1960's, probably in 1967.
My google search found no existing research of the type that interests us.
Also worth a mention is that 11 rounds (omitted from my summary) looks to me like the best producer of Finalists, from my cursory glancing at limited data, and assuming that seeding is a reasonable guide. 9, 10, 12 and perhaps 14 rounds also look pretty good to me.
Peter Gill.
2004 NOT Rydges
2 2 2 2
6 3 4 4
8 4 6 9
4 8 9 3
3 51 8 8
9 5 1 7
5 6 3 1
13 15 7 5
Team 1 was a team of Indonesians, not their top players, who surprised most of us by winning the Grand Final. Whether they should have qualified is moot.
Team 2 ran 2nd, Teams 3 and 8 came =3rd.
2004 NOT Other Venue
3 3 3 3
7 7 1 1
2 2 2 2
1 4 7 6
6 1 13 7
14 13 28 14
12 12 9 9
4 34 6 20
The qualifiers after 9 rounds look pretty good to me.
None of these eight teams made the semis.
2003 NOT Rydges
1 1 3 3
2 11 4 4
3 3 1 1
5 2 18 11
11 4 11 2
12 5 5 18
21 7 7 5
4 45 10 7
Team 2 (Neill) might claim that Round 12 is not a good guide. :)
Team 3 came =3rd.
2003 NOT Other Venue
1 1 1 1
4 101 101 2
101 11 4 4
121 6 7 9
11 4 2 101
5 5 5 5
7 12 26 6
22 16 11 7
101 = underseeded youth team Griff Ware
121 = American Barry Goren playing with Aussie youths
Team 1 came 1st, Team 2 (Noble, swissing as usual) came =3rd with Team 4.
I can easily go back another 4 years to 1999 if anyone wants more such data published here. It only takes a few minutes to put this data here.
My conclusions based on my non-scientific study of the data:
9 rounds = very good
10 rounds = very good
12 rounds = very good
14 rounds = OK
Peter Gill.
I asked heaps of people at the Gold Coast and elsewhere about NOT format.
Paul Lavings, as well as lots of average players, expressed the same view:
- that the SWPT desperately needs to switch from a 5 day format to a 3 day format. Lots of people have switched from the SWPT to the 3 -day Swiss of the Seniors or Restricted the previous week because they prefer the shorter event. They cannot last for a 5-day marathon in the second week, as well as the first week. They all very much want 3 days. No survey would pick this up, as they are no longer present at the SWPT/NOT.
Other views expressed:
- that the name SWPT should be abandoned. The reasons for the name change no longer exist. Call the whole week the NOT, like it used to be when it was huge. This would enable people to go back home and say that they won five matches at the National Open Teams, which sounds cool.
- make M-Tu-W the 9 round qualifying Swiss, with the Australian Open Pairs on the Thurs - Fri, with huge masterpoint allocations for the AOP as advocated by Bill Jacobs et al in the VBA Bulletin in the last 12 months.
The NOT Round of 16 on Thurs (losers into AOP Final maybe), QF on Fri, Sf on Sat, GF on Sunday as it should be.
Peter Gill.
The first few matches of a Swiss teams event are very random with lots of strong teams playing against weak teams. Over time, things settle out and teams rise (or fall) to their own level.
Once you've hit the midpoint of the event teams are much more closely matched. By the time you hit the finals, you hope that you have your top teams meeting at tables one and two. I can pretty much guarantee you that the optimal design for a Swiss Teams event will increase the number of boards played per round over the course of the tournament.
Richard Willey makes a very good point here.
Also the rule log2N + 2 is based on observations over 40 years of organizing, participating and watching Swiss teams events. This number of rounds tends to produce the same outcome in events (with the top positions juggled) as with more rounds.
As Richard says, the confidence in the result of any match increases with the number of boards. Therefore ideally when the match is late in the event (finals etc) the the number of boards should be greater than in the qualifying stages where the strength of the teams is unknown or unreliable. A top team playing a very weak team over 64 boards is boring for both teams and proves nothing.
There is one other major issue that the oganisers need to address in events like SWPT. The bulk of the teams, those that do not aspire to win the NOT, are there because they want to play against a few top teams. The thrill of playing against members of a state or national team is the highlight of the week.
With all these issues (number of boards per match, number of rounds, length of the event, player participation reasons etc) in mind I suggest we approach the problem alonmg the following lines.
The NOT qualifying stage is a 2 day, 12 round by 10 board matches Swiss with all teams participating. Thats 6 rounds a day, 2 per session and 60 boards per day. From this you produce a 32 team field for the NOT. I would expect that all would agree that if you are not in the top 32 at this stage then you do not deserve to win the NOT. It is "over Swissed" but all the weaker teams have had a chance to play a top team or two.
The rest of the teams continue with a new event (say SWPT qualifying, say 7 round Swiss over two days, with 20 board matches).
The NOT starts with 4 groups of 8 teams, grouped 1,5,9,13,17,21,25,29 then 2,6,10,14,18,22,26,30 etc as they finished in the qualifying event. These four groups play a 7 round 20 board match against all teams in their group over two days. Two teams qualify from each group for the 8 team finals. So Friday is quarter finals over 64 board matches, Saturday semi final over 64 boards and Sunday finals also over 64 boards.
The teams that do not make the TOP eight (24 teams) may join the top 8 (or more if all 24 teams do not continue) from the SWPT qualifying and play a two day 7 round, 20 board Swiss final for the SWPT winner.
None of these events exceed 2 days (apart from NOT finals). The events are:
NOT qualifying, two days.
NOT round of 32, two days.
NOT finals 1 to 3 days,
SWPT qualifying, two days.
SWPT finals, two days.
Teams can withdraw after any event and can enter at either qualifying stage. This flexibilty would satisfy many players and removes the major objections to the Swiss in the NOT qualifying stage.
Email received from Theresa Tully an hour ago:
Hi All
We have gone back to the drawing board & after listening to what the people want (not extra days & more choice) we have come up with the following.
Matchpoint Pairs .. same as before .. start Sunday 1pm & finish Tuesday afternoon.
Teams .. same as before.. one field .. 10 rounds of 14 boards ..finishing thursday.. 6 to go through to finals
Thursday night is still a free night for all.
Ivy Dahler Swiss Pairs :
Entries close Thursday night & field split into 2 with hand records after every round
9 rounds of 8 boards - 4 Friday afternoon (deferred draw) 2 Friday night (live) & 3 Saturday (live)
commencing 1.30pm, 2.38pm, 3.56pm, 4.52pm, all deferred draws then 8pm & 9.20 pm (live draw )
Saturday 10.30am , 11.50 am & 1.10 am all live draw.
Mixed Teams New Event Friday only (must be at least one of opposite sex in team & playing)
This will be 6 matches of 9 boards commencing Friday 10.30am & finishing around 7pm (last match starts 6pm)
Saturday New Event Commencing 10Am? 10.30Am? 11am?
We want suggestions, Something people can preenter & play in & finishing around the same time as the Swiss but slightly later.
Opinions please .. I need to then get approval (I have had unofficial ok) & let the people know.
Cheers
Theresa
I guess this type of data is not what people want, because nobody seems to be interested, except me.
Or perhaps, my presentation might be unintelligible?
2007 Gold Coast 248 teams one section
9 10 = number of rounds
2 2
9 1
11 11
7 12
1 5
3 3
5 28
4 6
2006 GC (Home section) 124 teams
9 10 12 = Number of Rounds
1 1 1 = seed of team in first place
3 3 3 = seed of team in 2nd place
9 9 9
6 38 5
13 2 6
15 6 2 = seed of team in 6th pace after n rounds (n = 9,10 or 12)
2006 GC (Away = GCI section) 124 teams
9 10 12 rounds
7 3 3
2 7 7
3 5 5
6 16 12
13 13 2
23 23 6
2005 GC Home
9 10 12 rounds
6 6 6
7 4 4
2 2 2
12 7 18
4 1 12
45 12 7
2005 GC Away
9 10 12 rounds
25 25 25
7 2 2
3 9 10
9 3 9
1 5 15
20 1 4
2004 GC Home = lower level
9 10 12 rounds
3 3 3
4 4 4
11 2 2
2 1 11
1 11 7
5 5 15
2004 GC Away Section
9 10 12 rounds
4 4 4
2 2 2
1 10 1
9 6 10
3 1 8
15 9 6
2003 GC Home
9 10 12 rounds
4 4 4
13 13 13 (Indonesia)
14 6 6
3 14 14
41 41 2
1 5 41
2003 GC Away
9 10 12 rounds
3 3 3
4 2 2
2 6 1
5 1 6
28 4 10
6 8 4
My conclusion after entering and studying all this data is that anyone who thinks that 9 or 10 rounds is inadequate to qualify 4, 6 or 8 teams to the Finals of the NOt of GC should take a look at he data then rethink their position.
Peter Gill.
revert2b
03-04-07, 01:23 PM
my conclusion after studying this data is that i have no idea what it means!
I suspected that I was writing unintelligible stuff. It is a trap which is easy to fall into.
2006 GC (Home section) 124 teams
9 10 12 = Number of Rounds
1 1 1 = seed of team in first place
3 3 3 = seed of team in 2nd place
9 9 9
6 38 5
13 2 6
15 6 2 = seed of team in 6th pace after n rounds (n = 9,10 or 12)
What this means is that in the Gold Coast Teams in 2006, in the Home Section (i.e the ANA Hotel section), 124 teams took part.
After 9 rounds, top seeded Team 1 was leading from Team 3 (the number underneath 1 in the list above), 3rd was Team 9, then the Team seeded 6th then the 13td seeds with the 15th seeds in 6th place.
After 10 rounds, Team 1 led from Team 3, then Team 9, Team 38, Team 2 with Team 6 in 6th palce.
After 12 rounds, Teams 1 still led from Team 3 and Team 9, followed by Teams 5, 6 and 2.
This enables comparision of whether having 9, 10 or 12 rounds in the GC Teams Swiss makes much difference.
My conclusion from scanning the data is that 9 rounds is enough, and it doesn't matter much whether you have 9, 10, 12 or 14 rounds.
It seesm to be that it is better to base one's opinion on data, rather than
to make dogmatic statements without evidence to back up your views.
This was an issue when the ABF were thinking of insisting on the GC Swiss having 14 rounds,. Now that that danger has vanished, the data does not matter much for the GC, but might help the NOT organisers to improve the format of the NOT.
Peter Gill.
2006 GC (Away = GCI section) 124 teams
9 10 12 rounds
7 3 3
2 7 7
3 5 5
6 16 12
13 13 2
23 23 6
2005 GC Home
9 10 12 rounds
6 6 6
7 4 4
2 2 2
12 7 18
4 1 12
45 12 7
2005 GC Away
9 10 12 rounds
25 25 25
7 2 2
3 9 10
9 3 9
1 5 15
20 1 4
2004 GC Home = lower level
9 10 12 rounds
3 3 3
4 4 4
11 2 2
2 1 11
1 11 7
5 5 15
2004 GC Away Section
9 10 12 rounds
4 4 4
2 2 2
1 10 1
9 6 10
3 1 8
15 9 6
2003 GC Home
9 10 12 rounds
4 4 4
13 13 13 (Indonesia)
14 6 6
3 14 14
41 41 2
1 5 41
2003 GC Away
9 10 12 rounds
3 3 3
4 2 2
2 6 1
5 1 6
28 4 10
6 8 4
My conclusion after entering and studying all this data is that anyone who thinks that 9 or 10 rounds is inadequate to qualify 4, 6 or 8 teams to the Finals of the NOt of GC should take a look at he data then rethink their position.
Peter Gill.[/QUOTE]
In Peter's now elucidated sample of eight Gold Coast swiss qualifying sections over a four-year period, the following observations can be made:
1. The leading 6 teams after round 9 occupied on average only 3.625 (60.4%) of the top 6 spots after round 12 and have never occupied all 6 of the top 6 spots.
2. The leading 4 teams after round 9 occupied on average only 2.625 (65.6%) of the top 4 spots after round 12 and have only once occupied all 4 of the top 4 spots.
3. Even looking at the conversion rate of the top 4 after round 9 to the top 6 after round 12, on average only 3.375 (84.4%) make it through.
I don't think the data that Peter has presented supports the argument that 9 rounds in enough. All it tells us is the leading teams after 12 rounds tends to be quite different to the leading teams after 9 rounds.
What is interesting to see is how often semi-finalist in the Gold Coast Teams were not in a qualifying position after round 9. Using the same sample period that Peter used (2003 to 2006) during which the format was two separate fields each qualifying 2 teams to the semi-finals, in three out of those four years a semi-finalist after round 12 was not in a qualifying position after round 9. Of those three teams, two won their semi-final but neither actually went on to win the event.
What is clear to me is that running 9 rounds of swiss will almost certainly result in a different finals composition to running 12 rounds of swiss. As to which is more equitable, meaningful or accurate is very much unclear.
What the 9-round-swiss evangelists need to do is come up with a sensible argument on mathematical or statistical grounds as to why 9 rounds in enough.
I guess the situation that organisers would want to avoid is a team with genuine prospects of winning the event not qualifying to the knock-out stages. I think that the best way of avoiding that is to have more teams going through to the knock-out stages, but of course that leads to either scheduling problems to fit an extra round of KO matches in, or forces organisers to make the KO matches shorter.
The aim of the swiss is to select a group of teams that includes the strongest team. Our aim is not to accurately place the lesser teams. The makeup of the top group will always change if you keep the swiss churning but the point is that the strongest team will almost always be in the group after 9 rounds. However, I would bet that the strongest team would lose a 24 or 32 board KO match about one time in three, even if they are quite a bit stronger. In other words, there is no reason to play more swiss but every reason to play longer KOs.
In Peter's now elucidated sample of eight Gold Coast swiss qualifying sections over a four-year period, the following observations can be made:
1. The leading 6 teams after round 9 occupied on average only 3.625 (60.4%) of the top 6 spots after round 12 and have never occupied all 6 of the top 6 spots.
2. The leading 4 teams after round 9 occupied on average only 2.625 (65.6%) of the top 4 spots after round 12 and have only once occupied all 4 of the top 4 spots.
3. Even looking at the conversion rate of the top 4 after round 9 to the top 6 after round 12, on average only 3.375 (84.4%) make it through.
I don't think the data that Peter has presented supports the argument that 9 rounds in enough. All it tells us is the leading teams after 12 rounds tends to be quite different to the leading teams after 9 rounds.
What is interesting to see is how often semi-finalist in the Gold Coast Teams were not in a qualifying position after round 9. Using the same sample period that Peter used (2003 to 2006) during which the format was two separate fields each qualifying 2 teams to the semi-finals, in three out of those four years a semi-finalist after round 12 was not in a qualifying position after round 9. Of those three teams, two won their semi-final but neither actually went on to win the event.
What is clear to me is that running 9 rounds of swiss will almost certainly result in a different finals composition to running 12 rounds of swiss. As to which is more equitable, meaningful or accurate is very much unclear.
What the 9-round-swiss evangelists need to do is come up with a sensible argument on mathematical or statistical grounds as to why 9 rounds in enough.
I guess the situation that organisers would want to avoid is a team with genuine prospects of winning the event not qualifying to the knock-out stages. I think that the best way of avoiding that is to have more teams going through to the knock-out stages, but of course that leads to either scheduling problems to fit an extra round of KO matches in, or forces organisers to make the KO matches shorter.
From Richard Willey
Hi Paul
I have a couple other good programers working on this topic. We're
still at the preliminary stages, but we already have some interestng
results. I'm forwarding an email from Gerben Dirksen which will given
you a good idea of the methodology that we are using.
To start with, we're creating a set of virtual bridge teams with a
known strength.
The strength of the teams is normally distributed (This is an
assumption. Later on we can use real tournament data to parameterize
the model)
We then have the virtual bridge teams compete against one another a
Swiss teams event. (We have some standard statistical models that
predict what happens when two teams of known strength play a board
against one another)
At the end of the Swiss teams event, we can compare the results of the
event with the known strength of the teams prior to the event. We can
then reach some conclusions about the accuracy of the tournament
format.
Right now, we're still arguing back and forth about the best way to
measure accuracy. We have a couple reasonable approximations along
with a more precise mechanism that we are still coding. (For anyone
who cares, the complex measure is (essentially) R Squared - the
coefficient of determination.
Here's our first (preliminary) result:
A Swiss team format that holds the length of each round constant is
less accurate than one increases the length of later rounds. I'm
forwarding Gerben's original email.
>100000 Swiss team tournaments of 9 rounds, team 1 = strongest team,
>team 2 = next strongest, etc. (from Gaussian distribution).
>Calculated AVERAGE TEAM NUMBER of the winner.
>8-board rounds: 2.86
>increasing round length (4 ... 12): 2.69
>decreasing round length (12 ... 4): 3.10
>
>Sounds like first-order proof that longer matches later on is a good idea.
>Finally I tried 16 rounds (8 board less!) of 4 boards and got: 3.93
>I didnt use a VP table yet but instead a match result was IMPs/board
* sqrt(boards) ,
>the formula on which the VP table is based.
>I think the correlation between result and strength in this way would
be a good measure.
>As a first quick check I simulated the following:
>
>100000 Swiss team tournaments of 9 rounds, team 1 = strongest team, team 2 =
>next strongest, etc. (from Gaussian distribution). Calculated AVERAGE
TEAM NUMBER >of the winner.
>8-board rounds: 2.86
>increasing round length (4 ... 12): 2.69
>decreasing round length (12 ... 4): 3.10
>
>Sounds like first-order proof that longer matches later on is a good idea.
>
>Finally I tried 16 rounds (8 board less!) of 4 boards and got: 3.93
>
>I didnt use a VP table yet but instead a match result was IMPs/board
* sqrt(boards) , the >formula on which the VP table is based.
This is excellent work being done here.
From an organisational viewpoint the changing of match length is not a feasible approach. After all we are trying to get the qualifiers for a final, not the winner of the event.
That raises the second point: with Q qualifiers required shouldn't the test be to get the average of the top Q teams to be close to Q/2?
This is excellent work being done here.
I agree with Ian - this is wonderful work.
From an organisational viewpoint the changing of match length is not a feasible approach. After all we are trying to get the qualifiers for a final, not the winner of the event.
Small changes to match length from round to round are obviously going to cause timetabling problems, and perhaps confusion.
CHALLENGE: think outside the obvious ... how about halving the length of the first few matches, and having some simpler method of doing the draw.
For example if the late matches are to remain 20 boards, the first few rounds might be groups of 4 playing 2 x 10 board matches, with the second round drawn on the basis of winner plays winner, loser plays loser.
That would obviously require careful hand-holding to introduce successfully, and someone might well think of something better.
BUT .. don't assume we have to keep doing what we've always done.
hrothgar
10-04-07, 04:24 AM
The aim of the swiss is to select a group of teams that includes the strongest team. Our aim is not to accurately place the lesser teams. The makeup of the top group will always change if you keep the swiss churning but the point is that the strongest team will almost always be in the group after 9 rounds. However, I would bet that the strongest team would lose a 24 or 32 board KO match about one time in three, even if they are quite a bit stronger. In other words, there is no reason to play more swiss but every reason to play longer KOs.
This is a really useful data point.
Paul is making a very specific assertion here: The over-riding goal of the NOT's is to identify the single best team. Other goals (producing an accurate ranking for teams 2-8, or indeed 2 ? 128) can be considered as secondary to the process. (Admitted, there appear to be some political constraints. Otherwise, the NOT would - probably - be run as single elimination KO)
If Paul's statement is true, we can design an experiment which will (hopefully) produce an near optimal tournament format.
Stage 1 of the tournament will consist of a round robin. The round robin will serve two distinct purposes: First, the round robin will narrow the field from 128 entries down to eight teams. Second, the results of the round robin will be used to ?seed? a KO (Team 1 plays 8, team 2 plays 7, ...)
Stage 2 of the tournament will consists of a KO. The KO will consist of three rounds, which will narrow the pool of eight down to a single winner.
The trade-off here should be pretty obvious. One the one hand we want to make sure the round robin had enough time to identify the best team and create an accurate seeding structure. On the other hand, we want to provide as much time as possible for the KOs to maximize the chance that skill will over come luck. (Things get especially complicated because the accuracy of the seeding will have an impact of the relative length of the rounds in the KO)
Regretfully, I've never had the opportunity to participate in this event. It would be useful to get some (basic) information about the format. In particular, how long does the event last? (The single most important constraint is going to be the total number of boards that can be played)
This is a really useful data point.
Paul is making a very specific assertion here: The over-riding goal of the NOT's is to identify the single best team. Other goals (producing an accurate ranking for teams 2-8, or indeed 2 ? 128) can be considered as secondary to the process. (Admitted, there appear to be some political constraints. Otherwise, the NOT would - probably - be run as single elimination KO)
If Paul's statement is true, we can design an experiment which will (hopefully) produce an near optimal tournament format.
Richard, there are two events.
SWPT and NOT see:
http://www.abf.com.au/events/not/2007/WeekTwo.pdf
Regretfully, I've never had the opportunity to participate in this event. It would be useful to get some (basic) information about the format. In particular, how long does the event last? (The single most important constraint is going to be the total number of boards that can be played)
The current NOT format is:
http://www.abf.com.au/events/not/2007/NOTFormat.pdf
The main objective of the Swiss teams event (SWPT) is to qualify the top 16 teams that will include the ultimate winner. This number of qualifiers Q is one variable that can change.
Note the number of venues may change. In particular there may be only one venue in the future and so only one field in the SWPT, not two.
The bulk of the players play in the event because of the opportunity to play against top teams. They also receive generous quantities of gold masterpoints which are hard to obtain elsewhere.
These players finance the event.
The second objective of the SWPT is to satisfy the needs of the bulk of the players. This should include a predictable routine such as a simple time table, the same numbers of boards per match, a reasonable length match (current 20 boards) and the same scoring method in each match.
Complications can easily be introduced in the NOT part of the event, round robin, KO and different match lengths etc.
Therefore the modelling on a Swiss teams event that Richard is performing should determine:
The minimum number of rounds to produce the top Q qualifiers with a high confidence level
The minimum number of boards per match that is needed to produce a reliable result
hrothgar
17-04-07, 10:17 PM
Alex did another simulation yesterday evening. He was specifically looking at the accuracy of a Swiss Team type event in identifying the top eight teams in a large field. Here's a few key parameters
1. The field consisted of 128 teams
2. The length of a round was fixed at 20 boards
3. The number of rounds varied between two and twenty
Alex ran 1000 "virtual" tournaments for each round length and then calculated a few key statistics
Results
1. The first column indicates the number of rounds (thanks for the catch Ian)
2. The second column indicates the number of times that the strongest team made its way into any of the top eight positions.
3. The third column indicates how many of the top 8 teams made their way into the top eight positions. (It might be better to worry about how many of the top four teams made it into the top eight positions)
4. The last column indicated the Average position of the strongest team at the end of the Swiss
2 545 2.92 13.42
3 706 3.44 7.92
4 788 3.83 6.07
5 815 4.12 5.51
6 862 4.46 4.16
7 879 4.61 4.27
8 917 4.75 3.37
9 914 4.87 2.99
10 932 5 2.9
11 934 5 2.7
12 950 5.16 2.62
13 960 5.23 2.35
14 968 5.38 2.19
15 966 5.47 2.2
16 971 5.5 2.09
17 989 5.67 1.93
18 983 5.79 1.95
19 984 5.83 1.9
20 985 5.9 1.92
Once we're able to digest these types of figures, we can start considering the tradeoffs between the time allocated to the Swiss and the time allocated for the subsequent KO. (At the moment, Alex is investigating whether introducing a simple "Strength of Schedule" adjustment will significantly improve the accuracy of the Swiss teams format)
In answer to one of Ian's questions:
Given our original set of assumptions, twelve 20 boards gives a 95% chance that the strongest team will survive into the KO stage.
I attached a .txt file with all of the results in comma separated variable format.
Thank you Richard and Alex; excellent.
Unfortunately there are a couple of typos.
The first column is "the number of rounds" and all the rows are out by one.
So the results are:
Rnds times/ Qual Ave Posn
1000 top 8 team 1
1 545 2.92 13.42
2 706 3.44 7.92
3 788 3.83 6.07
4 815 4.12 5.51
5 862 4.46 4.16
6 879 4.61 4.27
7 917 4.75 3.37
8 914 4.87 2.99
9 932 5.00 2.9
10 934 5.00 2.7
11 950 5.16 2.62
12 960 5.23 2.35
13 968 5.38 2.19
14 966 5.47 2.2
15 971 5.50 2.09
16 989 5.67 1.93
17 983 5.79 1.95
18 984 5.83 1.9
19 985 5.90 1.92
Therefore 11 rounds give a 95% chance of the top team being in the top 8.
These figures do support the theory that log2n+2 (9 in this case) gives an acceptable result with 93.2% chance. Even log2N could be acceptable.
log2N for 7 or 8 rounds (there is slight anomaly here) gives ~91.6%
log2N + 2 or 9 rounds 93.2%
log2N + 4 or 11 rounds 95%
hrothgar
19-04-07, 12:24 AM
The following post might stir up some controversy. The results are fairly straight-forward however, there are some significant political ramifications. You might decide that implementation is more trouble than its worth.
Steve Willner made a suggestion that implementing a Strength of Schedule correction might have a significant impact on the accuracy of the Swiss Teams format. His logic is that a victory over a "strong" team should be worth more than beating up on very weak teams.
We adopted a very simple Strength of Schedule (SoS) correction that is applied to the scores at the end of the Swiss Teams event. (We haven't tried a dynamic system that impacts the matchings in individual rounds). We calculate SoS by summing the total number of VPs won by each team. When you compete against team X, we award you some fraction of the total number of VPs that that team won during the course of the event (excluding the VPs won during head to head matches).
We tested a number of different fractions (0%, 10%, 20%, 30%, ... 100%) trying to determine which fraction created the "best". Here once again, we're measuring accuracy by determining how often the strongest team will fall within the top eight spots in the tournament.
I've attached the results of our latest output as a txt file in Comma Separated Variable format. However, here's the key data points:
1. The SoS correction has a very significant impact on the accuracy of the tournament. Recall our initial results: Without an SoS correction, you need twelve 20 board rounds to achieve a 95% chance that the strongest team will place in one of the top eight slots. Using an SoS correction, we can hold tournament with nine 20 board rounds that still achieves an 94.9% placement rate. SoS permits us to make much more efficient use of available time.
2. The SoS correction fraction appears to be a function of the number of rounds. The more rounds you have in a tournament, the higher the SoS fraction should be. We only tested SoS fractions between 0 and 1, however, I we need to test values higher that 1. In addition, I suspect that we might want to consider weighting the SoS value by the magnitude of the victory and see whether this improves the accuracy of the measure.
Here's the rub. Machiavelli famously observed that "There is nothing more difficult to take in hand, more perilous to conduct, or more uncertain in its success, than to take the lead in the introduction of a new order of things". We're suggesting a BIG change in the way things work. I'm quite sure that we'll be able to "prove" that changing the format will improve the accuracy of the selection process. Its unclear whether the general populace will believe that the complexity of the new process is balanced by the improvements in accuracy.
One option would be to "hide" the complexity from the rank and file players. The SWPT uses the same scoring system that it always did. Masterpoints are allocated using the traditional system.
However, the SoS correction gets applied to the results of the SWPT team tournament as part of the selection criteria for the NOTs. (There is a much smaller pool of players in contention for slots in the NOTs. I suspect that - by and large - these players are more technically sophisticated and more likely to accept these types of simulations as a valid part for the selection process)
I think the last two posts by Hrothgar (Richard) should be read by everyone.
The data in the first of his posts seems to me to suggest strongly that nine or ten rounds in the SWPT would be adequate. Do others agree?
Regarding the idea in his second post:
I think that correcting the actual scores to make them more accurate is at the very least several years away in the land of Oz where the underdog beating the mighty is a popular tradition of our country.
A problem is that Richard and Steve Willner's idea requires the identity of the top teams to be fairly clear. Is this identification done by seeding or by actual results at the end of the SWPT? If the latter, then progress scores before the last round will not be exact, whcih is not that great a thing in practice.
However,the idea overall does seem to have merit.
Peter Gill.
hrothgar
19-04-07, 04:07 AM
A problem is that Richard and Steve Willner's idea requires the identity of the top teams to be fairly clear. Is this identification done by seeding or by actual results at the end of the SWPT? If the latter, then progress scores before the last round will not be exact, which is not that great a thing in practice.
Hi Peter
I probably should be a bit more clear about whats going on here:
Alex Ogen is running a whole bunch of monte carlo simulations. The monte carlo simulations allow us to have "virtual teams" with a known strength compete against one another and observe the results. At the end of the simulation, we can compare the population statistic with the sample statistic.
The population statistic is our objective data about the relative strength of the different teams. The sample statistic is the result of one specific virtual tournament. We then compare the sample statistic to the population statistic and use this evaluate the accuracy of the sampling system - in this case the conditions of contest used for the tournament.
All the data that we use to generate the Strength of Schedule adjustment comes from the same virtual tournament. We're not using any information about the objective population statistics. Rather, we're making (very substantial) adjustments to the normal scoring system used in a Swiss Teams event. This adjustment essentially states that you want to play as many matches as possible against teams that finish the event with with good scores. The fact that you are playing a strong team is actually more important than how well you score against said strong team. Obviously, you still need to win your matches against the strong teams since this means that you will continue to be matched against strong teams.
This alternative scoring system can be implemented after the tournament without making any changes to the normal Swiss event. Steve and Alex seem to be making a convincing argument that this adjusted scoring system significantly improve the accuracy of the sampling procedure.
Now, it might very well be that you want to run inaccurate tournaments that increases the likelihood that weak teams will win. Thats something that y'all will need to decide. As I noted a couple posts back, you might want to consider a system in which
1. Masterpoint allocations use one evaluation system
2. The selection process of the NOT uses another
The bunnies will get their precious masterpoints, but you'll maximize the chance of identifying the strongest team for international competition....
Hi Peter
I probably should be a bit more clear about whats going on here:
Alex Ogen is running a whole bunch of monte carlo simulations. The monte carlo simulations allow us to have "virtual teams" with a known strength compete against one another and observe the results. At the end of the simulation, we can compare the population statistic with the sample statistic.
The population statistic is our objective data about the relative strength of the different teams. The sample statistic is the result of one specific virtual tournament. We then compare the sample statistic to the population statistic and use this evaluate the accuracy of the sampling system - in this case the conditions of contest used for the tournament.
All the data that we use to generate the Strength of Schedule adjustment comes from the same virtual tournament. We're not using any information about the objective population statistics. Rather, we're making (very substantial) adjustments to the normal scoring system used in a Swiss Teams event. This adjustment essentially states that you want to play as many matches as possible against teams that finish the event with with good scores. The fact that you are playing a strong team is actually more important than how well you score against said strong team. This alternative scoring system can be implemented after the tournament without making any changes to the normal Swiss event. Steve and Alex seem to be making a convincing argument that this adjusted scoring system significantly improve the accuracy of the sampling procedure.
Now, it might very well be that you want to run inaccurate tournaments that increases the likelihood that weak teams will win. Thats something that you all will need to decide. As I noted a couple posts back, you might want to consider a system in which
1. Masterpoint allocations use one evaluation system
2. The selection process of the NOT uses another
The bunnies will get their precious masterpoints, but you'll maximize the chance of identifying the strongest team for international competition....
Well done Richard, Steve and Alex. You have got to the bottom of the matter in a more rigorous way than I have seen before and your adjustment is clever. Ian asserted similar results and the data from long long swisses, like the SWPT backs this up.
I would see no problem doing as richard suggests - scoring the swpt by the actual results but deciding the field for the NOT only after adjusting for sos. Any team that dropped out would only have themselves to blame.
After reading this, no one can reasonably argue that you need more than 9 rounds in the swiss for the integrity of the event.
This all leads to something close to Paul's and my heart.
Player ratings.
Clearly when applying an SofS to the results we need to determine the strength of the teams. The player rating scheme supported by OzOne is the way to go. See Bridge Central (http://www.bridgecentral.com/) for more details.
I have taken the liberty to extract the data from the csv file supplied by Richard and present here for easy reading. I hope I interpreted the data correctly.
Rounds No Sos 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
20 985 986 987 987 987 987 987 987 987 987 987
19 984 984 985 985 985 985 986 988 987 987 987
18 983 984 985 988 988 989 990 990 989 989 990
17 989 991 993 992 992 992 991 991 991 991 992
16 971 972 974 977 980 981 983 984 983 983 985
15 966 968 971 973 973 974 973 974 975 977 977
14 968 973 977 979 979 979 981 980 979 981 981
13 960 963 968 969 970 972 974 973 974 977 979
12 950 963 966 969 968 968 968 970 969 970 970
11 934 939 945 949 951 954 957 961 960 959 960
10 932 934 944 945 948 947 946 946 948 951 953
9 914 924 932 933 935 942 946 947 948 949 948
8 917 924 927 935 936 939 938 934 939 939 937
7 879 883 890 900 905 905 903 905 904 904 902
6 862 865 878 884 888 886 892 886 880 873 873
5 815 829 834 837 843 843 844 831 831 821 810
4 788 801 808 809 816 809 803 787 780 770 752
3 706 712 721 728 730 718 687 652 603 539 490
2 545 558 566 520 435 316 162 56 30 19 14
Also I applogise for interpreting the previous "eventNum" as the number of rounds. The table above corrects this.
The ABF is having a meeting this weekend. I wonder if it would be too rushed for anyone to let the ABF know about the existence of Hrothgar's (Richard's) impressive data and ideas.
Now that evidence exists that 9 or 10 rounds is plenty for the SWPT, and it seems to many of us that the rank and file players would also prefer 9 rounds over three days (so that they can play the Seniors then the SWPT without exhausting themselves, for example), what should the SWPT/NOT format be?
(1)
M-W 9 round Swiss 9 X 20 bds
Th R16 64 bds
Fri QF 64 bds
Sat SF 64 bds
Sunday Final 64 bds
Monday nothing
Th - Australian Open Pairs Qual 2 sessions, special high mater-point scale
Fri - Australian Open Pairs Final and Consolations incl drop-ins from NOT
OR
(2)
something else?
Peter Gill
hrothgar
19-04-07, 10:04 PM
The ABF is having a meeting this weekend. I wonder if it would be too rushed for anyone to let the ABF know about the existence of Hrothgar's (Richard's) impressive data and ideas.
Peter Gill
Few quick suggestions
1. Most of the "heavy lifting" on this project was done by Alex Ogen and Gerben Dirksen, with Steve Willner and I providing some suggestions. Alex and Gerben deserve the real credit.
2. From my perspective, I feel fairly comfortable with our "core" results. I don't believe that you need more than nine 20 board rounds for a Swiss event with 128 pairs. I also believe that we can make a convincing argument that adding a Strength of Schedule adjustment to a Swiss format will significantly improve the accuracy of the event.
3. Its unclear to me how one should best structure the NOT KO that immediately follows the Swiss. A straight KO with four 64 board matches is one option. However, you might prefer to have slightly fewer boards in the early matches and slightly more in the later matches.
If I were presenting this right now, I'd recommend suggesting that you allocate
three days for the Swiss teams event and
four days for the KO that follows
but allow some flexibility in designing the Conditions of Contest.
Sensational work. Thanks guys.
Now comes the task of chanelling this information to people who will be in a position to affect things.
Clearly when applying an SofS to the results we need to determine the strength of the teams. The player rating scheme supported by OzOne is the way to go.In an earlier post Richard said "We calculate SoS by summing the total number of VPs won by each team", so a players rating system wouldn't be needed for the SofS adjustment as you would make the adjustment based on VPs of opponents played.
It would be interesting see how an SofS adjustment would have affected historical NOT fields, what immediately springs to mind is the unlucky McManus team from 2007 Rydges which plunged to 18th place with its last two rounds yielding 1 and 13 VPs, but interestingly they had higher VPs of opponents played than any team in the event, including a margin of 245 VPs of opps more than the the 8th qualifier, Genc.
hrothgar
20-04-07, 09:57 AM
Here's my take on matters:
There are a LOT of different ways in which a Strength of Schedule adjustment can be implemented. We decided to implement a fairly simple SoS statistic which was completely based on results within the same tournament. As David notes, we implemented the SoS adjustment as follows
At the end of the match, we added a fraction of the total VPs earned by each team that you faced during the tournament (Excluding those earned in the head-to-head match). We tested a variety of different fractions trying to determine which yielded the best predictive power. We discovered that the fraction depended on the number of rounds. For a tournament based on nine 20 board rounds, a fraction of .9 gave the best results.
I'm not particularly wedded to this methodology. (I suspect that we'd be able to come up with something better) I will, however, point out one advantage to this system: My impression is that rating schemes are a touchy political issue. Some people really like them. Others really dislike them. The SoS adjust that we're using right now is based completely on results in this one tournament. It can be described / framed as an adjustment to the scoring system, which completely sidesteps the whole debate about ratings.
Some might argue that this is useful in a political sense.
Mikey1nt
20-04-07, 03:21 PM
Very impressive work and data - wow.
I'm not convinced that one of the core asumptions is correct though. It seems to based on one of PM's assertions that no one has challenged, that "The aim of the swiss is to select a group of teams that includes the strongest team. Our aim is not to accurately place the lesser teams."
I agree that it is not the aim to accurately place the lesser teams, but:
1) Why shouldn't it be the job of the qualifying to also select the 2nd best, and 3rd best teams, and so on? Surely that leads to a better event post qualification?
2) Further to 1 above - a note on definition - The purpose of the event is to find the team who "plays the best bridge during the event" (eliminating as much luck as possible) not to find the "best" team. This also suggests that the goal of the qualifying is not just find 1 team. If the goal was just to find the best team you could hold the prize giving after the seeding commnittee finished!
3) If the qualifying format conveys a benefit in, say, finishing first (eg having choice of opponents in next round) then surely one of the qualifying aims is to have the "best" team finishing first, not just being one of the teams? And then so on for the 2nd and 3rd best teams etc?
Michael Ware
hrothgar
20-04-07, 11:54 PM
I'm not convinced that one of the core asumptions is correct though. It seems to based on one of PM's assertions that no one has challenged, that "The aim of the swiss is to select a group of teams that includes the strongest team. Our aim is not to accurately place the lesser teams."
I agree that it is not the aim to accurately place the lesser teams, but:
1) Why shouldn't it be the job of the qualifying to also select the 2nd best, and 3rd best teams, and so on? Surely that leads to a better event post qualification?
Hi Michael
We considered a number of different metrics which could be used to evaluate the accuracy of the ordering produced by the tournament. One option was using a statistic called the Spearman Rank Correlation which measures the sum of the squared deviations across the entire width of the sample. A second option was measuring how many of the top eight teams placed in the top eight slots. Another option was measuring the frequency with which the best team survived the Swiss Team portion of the event.
Our results were (broadly) consistent across the different metrics. For example, when we chose an SoS coefficient that maximized the chance that top team survived the Swiss Team stage, this same coefficient also maximized how many of the top 8 teams placed in the top eight slots and also maximized the Spearman correlation. I'm sure that I find counter examples - Its the nature of the beast - However, by an large, if we do a good with one metric, we'll do a good job with any related metrics.
2. From my perspective, I feel fairly comfortable with our "core" results. I don't believe that you need more than nine 20 board rounds for a Swiss event with 128 pairs. I also believe that we can make a convincing argument that adding a Strength of Schedule adjustment to a Swiss format will significantly improve the accuracy of the event.
I agree, though the number of boards (20) in each match is arbitrary and is based on what is done now, what the players want, the organisation issues etc. This claim is supported by the evidence. An earlier posting states:
>100000 Swiss team tournaments of 9 rounds, team 1 = strongest team,
>team 2 = next strongest, etc. (from Gaussian distribution).
>Calculated AVERAGE TEAM NUMBER of the winner.
>8-board rounds: 2.86
>increasing round length (4 ... 12): 2.69
>decreasing round length (12 ... 4): 3.10
>
>Sounds like first-order proof that longer matches later on is a good idea.
>Finally I tried 16 rounds (8 board less!) of 4 boards and got: 3.93
The average team number of the winner for 9 * 20 board rounds was shown to be 2.9 (cf 9 * 8 and 2.86).
Therefore within the limits of the tests (8 to 20) it can be claimed
the number of boards in Swiss teams matches does not appear to be that critical
3. Its unclear to me how one should best structure the NOT KO that immediately follows the Swiss. A straight KO with four 64 board matches is one option. However, you might prefer to have slightly fewer boards in the early matches and slightly more in the later matches.
If I were presenting this right now, I'd recommend suggesting that you allocate three days for the Swiss teams event and
four days for the KO that follows
but allow some flexibility in designing the Conditions of Contest.
No one would suggest that the number of boards in knock-out or round-robin events are not critical. Knowing what number of boards is needed in these types of events does effect the way you structure the event. The current number (64) may not be the best choice. Once the "correct" number is known then a decision on the recommended structure of the NOT can be finalised.
Ianmac wrote;
No one would suggest that the number of boards in knock-out or round-robin events are not critical. Knowing what number of boards is needed in these types of events does effect the way you structure the event. The current number (64) may not be the best choice. Once the "correct" number is known then a decision on the recommended structure of the NOT can be finalised.
For KO matches:
Well, 32 boards is wrong, as the quarter-finals of this year's NOT showed.
64 has the advantage of being about as many as you can fit into one day.
And as it's the more boards the merrier, 64 boards seems obvious for NOT.
The Playoff is different. For the Finals of both the 2007 Open and Women's Playoffs, 64 boards did seem too short, given that there was not much between the two finalists in each event on paper. I think 96 or 112 or 128 boards would be more sensible for the Grand Final of the Open and Women's Playoff. Other countries use much more than 64 boards for such a match.
Peter Gill.
There are a LOT of different ways in which a Strength of Schedule adjustment can be implemented. We
decided to implement a fairly simple SoS statistic which was completely based on results within the
same tournament. As David notes, we implemented the SoS adjustment as follows
At the end of the match, we added a fraction of the total VPs earned by each team that you faced
during the tournament (Excluding those earned in the head-to-head match). We tested a variety of
different fractions trying to determine which yielded the best predictive power. We discovered that
the fraction depended on the number of rounds. For a tournament based on nine 20 board rounds, a
fraction of .9 gave the best results.
I'm not particularly wedded to this methodology. (I suspect that we'd be able to come up with
something better) I will, however, point out one advantage to this system: My impression is that
rating schemes are a touchy political issue. Some people really like them. Others really dislike them.
The SoS adjust that we're using right now is based completely on results in this one tournament. It
can be described / framed as an adjustment to the scoring system, which completely sidesteps the whole debate about ratings.
Some might argue that this is useful in a political sense.
It is time for ratings. The introduction of improved scoring methods based on ratings will accelerate the use of ratings. Yes, I do believe we can come up with something better.
There is already a political swing towards the use of player ratings (in Australia at least). We are hoping to have this scheme in place within the next year or so. It is this sort of issue that raises the need and the profile of ratings. Assuming we have player ratings available it is feasible to consider a new approach.
If I read correctly the SoS was applied at the end of the event. Therefore at no time during the event would a team know for sure they were a qualifying team. They could only guess. I do not think you can separate the results of the SWPT and the NOT qualifiers.
The model is based on a normal distribution of players ranked 1 to 128. Most Swiss events start out with the seeded list of teams. The seeding of teams is usually arbitrary and often contentious. Using a player rating scheme to determine this rank is desirable. From now on let us refer to the rank of the team as the team rating; to know a team's rank is to know their rating. For modeling purposes this is also Gaussian distribution, with values between 0 and 100 to 2 decimal places.
I suggest we use the difference in the ratings of the two opposing teams in each match as the measure used in the strength of schedule. This is adjusted to the VP scale being used in the event and a fraction applied as in the previous tests. The optimum value needs to be determined, and could be greater than 1. Is it still .9?
This produces a par for the match.
The par is published with the draw and all teams know how well they need to perform against their opposition before the match starts.
Consider this suggestion (there are many ways):
Two teams, one with a rating of 60 and the other 40. Difference 20.
The VP scale has a maximum win at 48IMPs.
20% of 48 is 9.6.
Apply our optimum fraction (say .9) and get 8.63 which rounds to 9 (simplicity for players).
The par for the match is 9.
The team with the lower rating has a 9 IMP start.
An intriguing consequence of this approach is the luck of the draw is reduced if not eliminated. Dare I suggest a random draw is feasible?
A couple of comments in blue
Very impressive work and data - wow.
I'm not convinced that one of the core asumptions is correct though. It seems to based on one of PM's assertions that no one has challenged, that "The aim of the swiss is to select a group of teams that includes the strongest team. Our aim is not to accurately place the lesser teams."
I agree that it is not the aim to accurately place the lesser teams, but:
1) Why shouldn't it be the job of the qualifying to also select the 2nd best, and 3rd best teams, and so on? Surely that leads to a better event post qualification?
If we played the SWPT forever we would be almost certain of placing the top 8 teams in their proper order and that would be ideal. But of course we cannot do that so the question is when do we stop. From Richard's work we see that the long term winner will be in the top 8 after 9 rounds 95% of the time. And adding more rounds does not improve things much. I am sure we would also capture places 2-4 after 9 rounds most of the time.
2) Further to 1 above - a note on definition - The purpose of the event is to find the team who "plays the best bridge during the event" (eliminating as much luck as possible) not to find the "best" team. This also suggests that the goal of the qualifying is not just find 1 team. If the goal was just to find the best team you could hold the prize giving after the seeding commnittee finished!
The only importance of seeding is to properly balance the two fields. It plays no part in the result.
3) If the qualifying format conveys a benefit in, say, finishing first (eg having choice of opponents in next round) then surely one of the qualifying aims is to have the "best" team finishing first, not just being one of the teams? And then so on for the 2nd and 3rd best teams etc?
Michael Ware
Much interesting reading here . . .
A few quick thoughts - while I agree that the proposed longer knockout matches produce a more reliable winner - it is worth noting that only a very small number of teams have a chance of winning the NOT. Other teams enter it with other goals - ie qualification for the last 16.
The SoS adjustment seems like an excellent idea and if Australia is to continue using Swiss as (more or less) its only event format then serious consideration should be given to it.
Which brings me on to a suggestion (perhaps the NOT is not the best event to try this with) the best format for an event that I have played in is double-elimination knockout - used at the Spring Fours in England.
Single elimination knockout is too dependent on seeding in my opinion (it matters if you are seeded 64 or say 53 - team 64 plays the top seeds while team 53 gets to play team 11 in the second round).
It has the advantage of allowing you to play longer matches than a swiss of similar length and size.
Hope all in Aus are well.
MW
Further thought . . .
The problem with the new NOT format wasn't just the short QFs - apart from the evening off there was little advantage for the winners of the top half - even if the matches had been longer Brogeland and Noble might still have won the QFs.
How about the following - at the end of nine rounds (ideally using the SoS adjustments) the top 16 teams from each of the two fields qualify (now there can be little on no doubt that we are getting not only the best team qualifying but the top few surely) - this format allows the 2nd chance teams to retain that chance if they win their first two matches.
On Thursday AM (32 boards)
Top Four from each field play in 2nd chance pool 1st v 4th 2nd v 3rd (from other venue)
5th v 16th, 6th v 15th, etc (from other venue)
Thusday PM (32 boards)
4 Winners from top group play
4 Losers from Top group and 12 Winners bottom group play
Friday AM (32 boards)
2 Winners from top group - Winner advances to Semi-Final on Saturday
2 Losers and 8 Winners
Friday PM (32 boards)
1 Loser and 5 Winners play
Saturday/Sunday - 64 board semi-final/final - the winner from Friday AM should get as a minimum choice of opponent - in the Spring Fours they are also given the right to extend their match if trailing at full time.
vBulletin® v3.8.4, Copyright ©2000-2012, Jelsoft Enterprises Ltd.