Northumbria Masters

[Alex Holowczak]

I don't believe I ever saw this document. What were their stated objectives?

I've taken the time to trawl through minutes & annexes, as well as SPP's own website, and have found nothing. Can anyone help?

Two papers comparing various systems are at http://pairings.fide.com/documents/118- ... rings.html
The first paper was distributed for discussion between members of SPP before Baku in 2016, while the second one is an update for 2017.

I can confirm those are the documents I remembered reading, including the conclusion "As far as I have analyzed the data the Swiss Dutch + Baku acceleration will work better than normal Dutch Swiss for all types of tournaments with 9 rounds or more."

[NickFaulks]

I can confirm those are the documents I remembered reading, including the conclusion "As far as I have analyzed the data the Swiss Dutch + Baku acceleration will work better than normal Dutch Swiss for all types of tournaments with 9 rounds or more."

But what does "better" mean? I intend to read the document carefully to find out.

[Alex Holowczak]

I can confirm those are the documents I remembered reading, including the conclusion "As far as I have analyzed the data the Swiss Dutch + Baku acceleration will work better than normal Dutch Swiss for all types of tournaments with 9 rounds or more."

But what does "better" mean? I intend to read the document carefully to find out.

When I read it, I thought it meant "minimise the mean rating difference between two opponents throughout the tournament". Intuitively that seems desirable, since the traditional aim of acceleration is to ensure norm seekers don't play a low-rated opponent in Round 1.

[Roger de Coverly]

Intuitively that seems desirable, since the traditional aim of acceleration is to ensure norm seekers don't play a low-rated opponent in Round 1.

They also seemed to suggest that it marginally increased the likelihood of norms being made. That's a little subjective as is their assertion about reducing rating differences as it would depend on the reliability or otherwise of their results model.

For British events there have to be question marks about the reliability of FIDE ratings in creating a correct ranking order below the 1900 to 2000 range. Whilst k=40 for juniors can rapidly change a rating for the most active players, it also introduces instability. The FIDE implementation of Elo lacks what many national systems were forced to adopt, namely a reboot rule. A rule where a rapidly improved player can be treated as if they are new. Scotland have a rule which I think treats a player as new if their performance is 200 points better than their previous rating.

I could see the need for acceleration in really big events forty years ago like Stewart Reuben's six rounders with well over a hundred players per section. Perhaps also Capelle when it had six or seven hundred. For nine rounders even the size of the London FIDE Open or Gibraltar, it just looks wrong to have players in rounds 2-5 not playing others of the same score.

[NickFaulks]

When I read it, I thought it meant "minimise the mean rating difference between two opponents throughout the tournament".

If that's what you want you can just do it, starting with 1v2, 3v4 etc. The point of the Baku system seems to be that it approaches this, while bearing sufficient similarity to a Swiss that you can get away with calling it by the name.

Intuitively that seems desirable, since the traditional aim of acceleration is to ensure norm seekers don't play a low-rated opponent in Round 1.

I don't think that's right. It was originally intended for short tournaments with many players, to ensure that only one could win with a perfect score. The application to norm events came much later.

[NickFaulks]

For British events there have to be question marks about the reliability of FIDE ratings in creating a correct ranking order below the 1900 to 2000 range.

But that doesn't matter, because for norm purposes it will be uplifted anyway. A strong player who has a good tournament is unlikely to play more than one of them.

The FIDE implementation of Elo lacks what many national systems were forced to adopt, namely a reboot rule.

And will continue to lack it. There is an effective reboot rule, since if eighteen games are played in one rating period it does reboot ( for young players ) and if they are played over a few months it comes close. For most federations this works well, and it would be wrong to distort the system for the benefit of those which continue, for whatever reasons, to be reluctant to submit games for rating.

There is a case for some different treatment of rapid and blitz ratings, where many players do have gaps of several years, and this is under consideration.

[Roger de Coverly]

But that doesn't matter, because for norm purposes it will be uplifted anyway.

I was thinking more about pairings which depend on rankings rather than ratings. Adding extra points randomly relative to playing strength would presumably have some effect, although I don't know quite what. Still if they have a moderately convincing results model, the effect on having rankings not correlated to playing strength could be investigated.

[Brian Valentine]

The FIDE implementation of Elo lacks what many national systems were forced to adopt, namely a reboot rule.

And will continue to lack it. There is an effective reboot rule, since if eighteen games are played in one rating period it does reboot ( for young players ) and if they are played over a few months it comes close. For most federations this works well, and it would be wrong to distort the system for the benefit of those which continue, for whatever reasons, to be reluctant to submit games for rating.

There is a case for some different treatment of rapid and blitz ratings, where many players do have gaps of several years, and this is under consideration.

Nick,
You kindly confirmed how the k factor reboots under Elo a long time ago in another thread. From how I now understand it, 18 results over any period other than one month, tends to only around a 2/3rds reboot very quickly. Also if one looks at the formula for actual results in the papers Shaun pointed to compared with Elo expectation, it suggests a huge under-performance of the highest rated player at lower ratings. This would suggest FIDE has a problem. With absolutely no evidence my first point of investigation would be the FIDE treatment of juniors.

[IM Jack Rudd]

For British events there have to be question marks about the reliability of FIDE ratings in creating a correct ranking order below the 1900 to 2000 range.

But that doesn't matter, because for norm purposes it will be uplifted anyway. A strong player who has a good tournament is unlikely to play more than one of them.

Oh, it can easily happen. All you need for it to happen is to be a norm-seeker at the bottom of the first quartile. You beat a weak player in round 1, and then in round 2 you get paired against another weak player who performed a giant-killing.

[NickFaulks]

Oh, it can easily happen. All you need for it to happen is to be a norm-seeker at the bottom of the first quartile. You beat a weak player in round 1, and then in round 2 you get paired against another weak player who performed a giant-killing.

I did look at that precise scenario, but in Northumbria it seemed unlikely to damage the chances of a norm seeker. The very worst case is that your first opponent is 2100 and the second is 1900 ( uprated to 2050/2200 ). Beating both of them is a perfectly good start to a nine round campaign.
As a general point, having an opponent whose rating is so low that it gets uprated is "free money", so it is statistically advantageous to meet one of them.

Sorry to go on, but I think the point can be summarised as follows. For an IM performance norm you need two things.

1. A rating performance of 2450. You can do that against any opposition, you just have to score heavily enough.

2. Opponents' average rating at least 2220. That is the key, and in Northumbria and events like it I cannot imagine that anyone performing at 2450, so scoring at least 7/9, would fail that test.

[Stewart Reuben]

Nick >I don't think that's right. It was originally intended for short tournaments with many players, to ensure that only one could win with a perfect score. The application to norm events came much later.<

There were and are two other reasons for Acceleration. Both of these were addressed very early on, long before 1977 when it became apparent that it was possible to get a norm in an open Swiss.
1. To diminish the number of encounters between vastly different grades. This was defined as 40 grading points. A disadvantage of acceleration is that you get a few such pairings later on in the tournament.
2. To diminish the 'bouncing' effect. I played in an open Swiss in Bermuda in 2001, rated about 2200 US. Round 1 I had White against a 1400 player and won. Round 2 I had Black against a GM and lost. Round 3 I had White against a 1600 player and won. Round 4 I had Black against an IM and lost. Round 5 I had White against an 1800 player and won. That was the end of the tournament. I regarded that as unsatisfactory.

[NickFaulks]

I played in an open Swiss in Bermuda in 2001, rated about 2200 US. Round 1 I had White against a 1400 player and won. Round 2 I had Black against a GM and lost. Round 3 I had White against a 1600 player and won. Round 4 I had Black against an IM and lost. Round 5 I had White against an 1800 player and won. That was the end of the tournament. I regarded that as unsatisfactory.

So did I. I played in loads of them and generally knew in advance that I was likely to get 3/5, and that my best chance of 3 1/2 was to concede a draw to a weaker player. Drawing with a stronger player was equally effective but not so easy.

However, a couple of times we asked the players ( local and overseas ) whether they would prefer a different system and they were adamant that they would not.

[NickFaulks]

Also if one looks at the formula for actual results in the papers Shaun pointed to compared with Elo expectation, it suggests a huge under-performance of the highest rated player at lower ratings.

I can't find anything of that sort. Where should I be looking?