[mrdct]

A few years ago I thought I saw some modelling by hrothgar on the merits of swiss movements which took account of the volatility of results in match-ups and produced a confidence level that the "correct" winner will be found under various scenariois. I'm keen to apply that sort of methodology to this problem.

I'm looking at field of about 20 pairs with quite significant variation in standard, ranging from open national team representatives to fresh out of the beginners course (the latter comprising as much as 25% of the field). There would be 5 pairs with a realistic chance of winning, 7 pairs with a realistic chance of top 3 and 9 pairs with a realistic chance of top 5. The event objective is to obtain a winner and places 2nd to 5th who progress to a further stage.

We have two and a bit days to get to the point where the winner and 2nd to 5th are decided and can play up to 70 boards per day so a movement comprised of around 140 to 160 boards is required (9 x 16 boards is my current thinking).

Some of the issues:

1. The internally generated datum will likely be quite volatile due to the variation in standard so I imagine it would be optimal to exclude a couple of outliers at each end and do some arrow-switching.

2. Movements which involve contending pairs playing a lot of boards against non-contending pairs may be inferior as they can become a test of how well you beat-up the weak pairs rather than how well you perform against the good pairs.

3. It would be a poor outcome if a "flying bunny" emerged in the last couple of rounds of a swiss movement to grab 5th place without ever playing any of the contending pairs.

4. If a swiss movement is used, it would probably need to be 8 or 9 rounds which could become "over-swissed".

5. I am interested in the theory that if a field can be accurately seeded, you should play the first round as 1v2, 3v4, 5v6, etc. Potentially the first session or two could be used to do some seeding with a short-round mitchell followed by howell to get everyone to play 3 boards against everyone and then take some carry-over into a swiss.

[NickRW]

I'm not sure why you're so worried about this - it seems strange to be having a 2+ day event where you have such an extreme mix of abilities...

Anyway, if it is really important to find the best 5 and this matters for some other ongoing later competition - like this is a trials or something - and you expect 25% of the field to be poor and randomising the results, then it seems to me that you'd be best off spending day 1 with a Howell and chuck out the bottom 6 or so. Then use the remainder of the time to sort out the best of the best with another Howell.

Nick

[TylerE]

Maybe the best thing to do is a hybrid? Something like this... break the field into 5 groups of 4, each group plays a 3 round RR within their group. Top 2 from each RR group going into the swiss, which could be say 5 rounds. Basically a nice way of pairing the field down to those in actual contention, and then swissing them.

[TimG]

mrdct, on Aug 8 2010, 07:00 PM, said:

There would be 5 pairs with a realistic chance of winning, 7 pairs with a realistic chance of top 3 and 9 pairs with a realistic chance of top 5.

I'm surprised at these numbers. As I'm reading it, there are two pairs who have a realistic chance at finishing 2nd or 3rd, but an unrealistic chance of finishing 1st. Is that right? That seems a very fine line to me unless there is an overwhelming favorite to win the event. But that doesn't seem to be the case since there are five pairs with a realistic chance of winning.

[mrdct]

On second read, yes they do seem to be cutting it a bit fine, but it is the general gist of the scenario that's important.

The main thing that I'm after is the mathematical modelling that can been done on how accurate a swiss movement is to product a winner and the top n places. My understanding is that such models will have variables based on the volatility of the field and accuracy of seeding. The place I recall seeing Richard Willey's analysis was the now defunct OzOne forum where there was a discussion about the optimal format to reduce 200 teams down to 20.

My gut feel is that a swiss movement will largely achieve the outcome of having the contending players mainly playing amongst themselves, but this may be compromised in a small field which could get "over-swissed".

I note that the recent Australian Open Butler had a first stage with about 60 pairs playing 10 rounds of swiss to produce a top 20 to progress to a second stage played as a complete round robin. But the beginner/inexperienced portion of the field was probably less than 5%. Is that structure mathematically sound?

[Mbodell]

What about a RR where everyone plays everyone 4 boards, then divide the field into a top half and a bottom half where everyone plays 9 boards against each other in the same half. That gives you 76 (4*19) boards in the first half and 81 (9*9) boards in the second half, and as long as all of your "top 5" pairs can manage to stay in the top half after the first RR you should be fine. You can decide what, if any, carryover you want from the first RR to the second (full carryover of results against other "top half" teams and half carryover of results against non-top half teams?).

[foo]

If you want to improve swiss movements, the single best thing that should be done first is to make sure that either
a= every match uses the same sequence of board numbers or
b= have 8 board matches.

Why? because when you use less than 8 boards per match, some sets are more "swingy" than others due to having unequal amounts of each vulnerability ratio.
Compare board sets 1-6, 7-12, 13-18, 19-24, 25-30, and 31-36 for an explicit example of the problem.
As we all know, 6 (or 12 where the same 6 are used each 1/2) board matches are very common at IMPs.

[Oof Arted]

You could of course play all 6 board matches all Swiss Teams thus you would need 1 set of boards for every 3 tables and just circulate the boards around the loop or loops of tables

[foo]

Oof Arted, on Aug 9 2010, 02:39 AM, said:

You could of course play all 6 board matches all Swiss Teams thus you would need 1 set of boards for every 3 tables and just circulate the boards around the loop or loops of tables

That would slow the pace of a round to ~ 1/2 - 1/3 present speed.

IMHO, the "solution" is worse than the "problem"

OTOH, there is nothing stopping us from having removable board numbers for boards and prepping for an IMP match by making as many sets of the same board numbers as needed for the size of the field.

Protecting against the boards from one match being mixed up with boards from another match could be implemented by the simple expedient of having different colored boards or differing patterns/colors/etc for the removable board numbers.

Thus every table would play only the boards that belong in their match and every match would play the same sequence of board numbers.

8 boards rounds, of course, are the other solution. OTOH, most seem to feel that 8 board rounds would result in events taking too long to play.

...and the era of "some board sets are better than others" and attempts at gaming the system for the kind of board set you want for your match would be over.