[tournament-org] The Bar in general
Geoff Kaniuk
geoff at kaniuk.co.uk
Fri Nov 9 14:51:43 GMT 2018
Thanks Toby,
There were a couple of typos in the spreadsheet and I have re-issued it.
Geoff
33 Ashbury Close, Cambridge CB1 3RW 01223 710582
On 08/11/2018 15:55, TobyManning via tournament-org wrote:
> Geoff:
>
> Thanks.
>
> *Supergroup*
>
> First of all, I agree with you about the supergroup. Indeed many years
> ago I put forward the view that, having set the bar, the starting
> strength of those above the bar should represent their *average*
> strength not simply "bar+1".
>
> Thus, for example, if the entry is 3d 3d 2d 1d 1k 1k in a 3 round
> tournament then the bar should be at 1d but those above the bar should
> start at +1 (if the 1 kyus start at -1), and not 0. I ran a few
> tournaments on this basis probably 15 - 20 years ago, but I was told I
> was wrong and the idea never really caught on.
[[
I think supergroup is a useful tool when:
a. There are a lot of strong players within a narrow grade range.
b. There is an excessive grade gap to the next group down.
I am not sure about using just the average grade. You might want for
example to isolate the top group from the rest for the first few rounds
but allow mixing for the last two say.
One of the issues with this may be the effect on Stacey points. One
solution for this and the small bar or handicap problem, could be to
have a Pairing Bar for actually doing the draw and a possibly different
Stacey Bar for awarding Stacey points.
But again that is another debate that would need to happen.
]]
>
> *Spreadsheet Comments*
>
> When I look at your spreadsheet, there are 4 tournaments where the
> number above the bar is less than the number recommended in the Table.
> Of these:-
>
> a) I don't understand No. 12 (4d, 5*1d) where the bar was at 1d but
> there was only 1 player above the bar. The correct answer is for the bar
> to be at 1 dan with 6 above the bar
[[ Agreed - this was a bar change after the event ]]
>
> b) Nos 16 and 19: why did the 1k not play both 3k in No 16?
[[
There were quite a number of uneven games after round 2. The 3k were
losing most of their games.
]]
> Were handicaps used in No 19?
[[
Handicaps are indicated in the opponent list (h)
In this tournament all players were below the bar.
]]
Both cases are difficult ,
[[
Not difficult once one has the full story!
]]
and there might be a
> case for handicaps (consult the players?)
[[
I know consultation has been mooted. I am not comfortable with this
idea, as there can be problems if one player refuses.
]]
>
> c) No 25 The bar should have been at 1k with a supergroup.
[[
This would have made the bar depth 5 with a 1k vs 5d a possibility.
]]
>
> No 1 should have had the bar at 1 dan , when the depth would have been 3.
[[
The maximum grade was 5d.
With the bar at 1d, the bar depth is 5d - 1d = 4
In this tournament we have 4 rounds with exactly 16 above the bar. I
assume this was to get a unique winner.
]]
>
> In addition, No 13 should be a supergroup (see above) with the bar at 1d
[[
The bar *is* at 1d
]]
>
> Finally, of course, Nos 7 and 21 have too many above the bar. For No 7,
> I would have put the bar at 1 dan, No 21 is difficult and a fudge may be
> required.
>
[[
For no 7 the TD had the option:
Bar at 1d pop 5 so 3 below maximum
Bar at 1k pop 9 so 1 above maximum
I can imagine TD choosing least deviation, but I agree with 1d.
For no 21, sorry my mistake: the number of rounds is 6.
]]
> *Use of Organiser's Handbook Tables*
>
> You state:-
>
> /There is plenty of evidence to show that you cannot set the bar based
> on the number of rounds alone. For a 6 round tournament, the BGA tables
> advise you to have a population range from 7 to 15 players, and in some
> tournaments that would lead to a range from 5k to 2d or worse/.
>
> Sorry, I am confused: I don't see this evidence. In a 6 round tournament
> the ultimate winner will (ideally) end up playing people ranked 2nd to
> 7th, and this will be the case irrespective of where the bar is placed.
> And the winner's oppponents should all have a chance of winning the
> Tournament:
[[
No chance!
Take the case of no 21 - a 6 round tournament. The table allows a
maximum of 15 players above the bar. In this tournament the grade range
of players above the bar could be from 6k to 2d. A 6k beats a 2d with
probability 0.1%.
Or take number 6, a 5 rounder where the tables allow a population of 12.
In this case once could have a grade range from 5k to 4d. Winning chance
is now less than 0.01%
I realise slightly unfair data as you do not see the full picture but
any of the longer tournaments will have this possibility.
]]
imagine the outcry if a maverick player gets 6/6, but one of
> the people that he beat is said to have won the tournament. (This would
> have happened last year at the 3 peaks if James Richards had beaten me
> in the last round, see http://www.britgo.org/results/2017/threepeaks).
>
[[
The bar was at 2k, James was 3k on 4 wins last round, you on 3 wins.
Both have same mms of 1. James wins. MMS James = 2, MMS Toby=1.
James gets the chocolate!
No outcry, everybody cheers!
]]
> I reiterate my stance that the Table should take priority, and that if
> there is a choice of values within this range then bar depth is an
> important consideration.
[[
The lower range of the table is based on the idea that the top players
engage in an all-play-all: players = 1 more than number of rounds.
However all-play-all is a closed system. McMahon is open in the sense
that after round 1 there are players from below the bar now able to play
losers from above the bar. So the top players do not get an all-play-all.
The upper range of the table is a compromise. It starts with the rule
N=2^R for R = 3 and 4 rounds. This is the rule for a perfect Swiss
guaranteeing a unique winner. Note also that exactly 8 or 16 players
are needed and often we do not have exactly that number. So no guarantee
of a unique winner.
For R=5, practicality sets in, and I am guessing that our founding
fathers realised that to get 32 players above the bar would be difficult
so it was set at 12. This would have been based on some experience or
just guesswork.
We now have a much better model to determine where the bar should be
set. A modified table + the bar-depth filter is a first step in
implementing that model.
]]
>
> Perhaps I should shut up and let others comment.
[[
I welcome your comments and I hope to hear from others.
]]
>
> Toby.
>
>
> On 08/11/2018 14:28, Geoff Kaniuk via tournament-org wrote:
>> I agree with Toby's points a,b,c below. I even agree that placement
>> of the bar cannot prevent the disadvantage.
>>
>> Where we disagree is with the use of the BGA table in setting the bar.
>>
>> There is plenty of evidence to show that you cannot set the bar based
>> on the number of rounds alone. For a 6 round tournament, the BGA
>> tables advise you to have a population range from 7 to 15 players, and
>> in some tournaments that would lead to a range from 5k to 2d or worse.
>>
>> It is essential to take account of the grade distribution at the top,
>> and the bar depth idea is just a first step in trying to do this. I
>> have created a spreadsheet showing bar-grade data for all our
>> tournaments this year.
>>
>> http://www.kaniuk.co.uk/articles/pairing/bga-bar-grades-2018.xls
>>
>> It has a table on the sheet 'pwin' showing the probability that player
>> with grade Glo (column A) beats player with grade Ghi (row 3)
>>
>> In the sheet 'tours-pub' we have an anonymised table of tournaments
>> presented in bar-depth order. It shows who the winner played and I
>> have detailed the few cases where the winner dropped a game. This
>> happened in just 5 out of 25 tournaments and in nearly all such cases
>> the grade difference was just one.
>>
>> There is data showing the probability that the player at the bottom of
>> the bar (with grade Gbar) beats the maximum graded player. There were
>> 7 tournaments where the bar-depth varies from 4 to 7. In these
>> tournaments the Gbar player has a winning chance against the strongest
>> in the range 0.1% to 8.2%
>>
>> For the 10 tournaments with a bar-depth of 1 or 2 this probability
>> lies in the range 18.1% to 34.1%
>>
>> For the 8 tournaments with bar-depth = 3, the range is 10.7% to 26.1%
>>
>> The spreadsheet also contains a useful plot of these probs vs bar-depth
>>
>> Remember that these probabilities are calculated for just one game.
>> The actual probability of a Gbar winning the tournament is usually tiny.
>>
>> My conclusion is that when the population at a bar-depth of 3 is
>> small, (varying from 1 to 8) the bar can only be lowered if you are
>> prepared to consider using handicaps above the bar.
>>
>> If we do not want to do that then one possibility is to isolate the
>> top group by boosting the initial MMS by a few points - in other words
>> create a super group. This protects the players below the bar from
>> hugely unbalanced games.
>>
>> Geoff
>>
>> 33 Ashbury Close, Cambridge CB1 3RW 01223 710582
>>
>> On 08/11/2018 09:47, TobyManning via tournament-org wrote:
>>> Geoff:
>>>
>>> Thanks for the extra information.
>>>
>>> However, I revert to my intiial question:
>>>
>>> Which is more inportant, to*restrict the bar depth* or to have
>>> the*number of people above the bar compliant with the Table* in the
>>> handbook (http://www.britgo.org/organisers/handbook/tournament4).
>>>
>>> It is still my view that the Table limits shuld be paramount, and the
>>> bar depth should be used to determine the bar within these limits.
>>>
>>> I have re-read your article in BGJ #173, which discusses how
>>> effectively the MacMahon system gives people an even spread of
>>> opponents.
>>>
>>> We need to recognise that, *irrespective of where the bar* is set:-
>>>
>>> a) those at the very top (the 4 dans) will have more "easy" games and
>>> we expect them to have an above-average result
>>>
>>> b) those well below the bar (the 5 kyus, say) will have a 50:50
>>> result on average
>>>
>>> c) there is a cohort of people - in the 3 peaks case the 2d/1d - who
>>> will have a below-average result as they will each have to play the 4
>>> dans at some time.**
>>>
>>> So placement of the bar cannot prevent this disadvantagement; it
>>> merely alters the make up of the cohort in my group (c) above.
>>>
>>> In the 3 peaks example, with entry at 4d/4d/2d/1d/1d/1k/1k/1k, this
>>> disadvantagement is effectively the same whether the bar is set at 4
>>> dan, 3 dan, 2 dan or 1 dan. This is because the actual games played
>>> will be unaffected (each 4 dan is expected to have opponents 4d 2d 1d
>>> 1d 1k irrespective of the bar setting). With the bar at 1 kyu the
>>> disadvantagement is slightly more widespread and the total
>>> disadvantagement starts to increase, and this then falls off a cliff
>>> with the bar at 2 kyu and below.
>>>
>>> In fact, the disadvantagement is essentially constant while the
>>> number of people above the bar is less than (n+1) where n is the
>>> number of rounds. As the number of people above the bar increases
>>> from (n+1) to 2**n this total disadvantagement increases - the amount
>>> of the increase depending upon the bar depth. So if the bar depth is
>>> shallow the number above the bar should tend towards n**2, if it is
>>> deep it should tend towards (n+1).
>>>
>>> There is therefore no benefit from having the number above the bar
>>> being less than (n+1); and indeed it would prevent the (rogue) 1
>>> dan/1 kyu winning the tournament, irrespective of their results
>>> against the 4 dans.
>>>
>>> *Manual Overrides*
>>>
>>> You are quite right to emphasise that TD's can override GoDraw's
>>> defaults. However, my experience is that many TD's - particularly the
>>> inexperienced ones - are reluctant to do this as they are concerned
>>> about possible unintended consequences.
>>>
>>> I think this emphasises the importance of getting the GoDraw defaults
>>> as good as we can.
>>>
>>> Toby
>>>
>>> **I speak from (not really bitter) experience.
>>
>> _______________________________________________
>> tournament-org mailing list
>> tournament-org at lists.britgo.org
>> http://lists.britgo.org/cgi-bin/mailman/listinfo/tournament-org
>
>
> _______________________________________________
> tournament-org mailing list
> tournament-org at lists.britgo.org
> http://lists.britgo.org/cgi-bin/mailman/listinfo/tournament-org
>
More information about the tournament-org
mailing list