Opinions and proposals in the ECU performance debate

by ChessBase
4/16/2011 – The 2011 European Championship generated controversy owing to the unusual tie-break system used. There was a formal protest by GM Peter Heine Nielsen and a thoughtful article by the statistician Jeff Sonas. Naturally we received a large number of letters, and among them two from GMs David Navara and John Nunn. Their proposals are included in the feedback from our readers.

ChessBase 18 - Mega package ChessBase 18 - Mega package

Winning starts with what you know
The new version 18 offers completely new possibilities for chess training and analysis: playing style analysis, search for strategic themes, access to 6 billion Lichess games, player preparation by matching Lichess games, download Chess.com games with built-in API, built-in cloud engine and much more.

More...

Feedback from the ECU performance debate


Eric Peterson, PhD., Slovakia
GM Peter Heine Nielsen is of course completely correct in his protest. The tiebreak method used by the ECU in the European Championship is completely absurd and random. Someone did not think it out carefully. The absurdity of the ECU method can be concisely shown by: "If you beat the highest-rated opponent you faced, this is horrible for your tiebreaks. But if you lost to your highest-rated opponent, this is a tremendous benefit." Obviously a tremendously flawed system.

David Hodge, UK
As a mathematician I was pleasantly surprised to see a very nicely written critique by PHN of how ridiculous the tie-break system implemented is/was. And why the tie-breaks calculated should be scrapped and redone properly. It is clear to me, and should be clear to anyone who sees what has happened that what has been implemented is not a "performance rating" in the sense that anyone would accept it. Furthermore, as Peter points out (though he wrote it less persuasively) there is even a serious problem with the rules as they are written as the tournament's eventual second place competitor would have won the tournament if only his first round opponent (whom he beat) had been rated two points lower!! If his first round opponent was rated one point lower then God knows what the organizers would have done because they would have had to choose which of his first two games to delete, the first one or the second one. The choice would result in deciding who won the tournament. I guess in the scenario that their "performance rating" was uncalculable they would have moved onto tie-break system B. This is also ludicrous due to the fact that the "system" they used isn't even a valid and comprehensive system to cover normal possibilities (like two players having the same rating). One would normally only consider a "second" tie-break system if the players drew on the first system, in this case they would have had to move onto the "second" tie-break because the first one isn't even defined! Maybe Judit would even have won! I would appreciate to hear if there is any precedent for organizers ever to have used such a nonsense system before. There is certainly no way this rule could ever be used again to decide tie-breaks, now that organizers realize what it does, and it would only be fair (if possible) to recalculate performance ratings in a normal fashion. Still ignoring top and bottom ratings as the rules intended is fine, as long as you don't actually pretend the games never happened too!

GM David Navara, Prague, Czech Republic
I read the interesting article from GM P. H. Nielsen on the tie-breaks used at the European Individual Championship. I think that his proposal is much better that the tie-break which was used, but still I think that I have a better idea. I agree with the objections against the actual tie-break rules, which GM Nielsen summed up. Nevertheless, I believe that his proposal is also not completely fair. If someone beats a player rated 1400 in the first round, why should this game count in his percentage without influencing the average rating of his opponents?

I think that it would have been much better to take the "classical performance" and then to remove all the won games of the player, which deteriorate his performance. I believe that such a rule would have helped to provide relatively fair tie-break rules. It should ideally be supplemented with a similar rule, which would not count into the performance the games, which a player lost, but which improved his performance. (I believe that such cases would have been extremely rare at the tournaments like the European Individual Championship. It could be useful in cases when one opponent of a certain player has significantly higher rating than the remaining opponents of the very same players.)

The only difficulty with adding this complementary rule is that the order of applying both rules would need further specification, because it could happen that the different orders of application of these two rules could produce different results because in one case it could be better to cross out some game, whereas in the other one it could be better to keep it. (I could provide some advice how to calculate the performance then.)

Nevertheless, it should not be too difficult to calculate the modified performance on this basis. It would have been very logical, because no win could have deteriorated player's performance and no loss could have improved it. It also should be relatively simple to calculate this. My proposal is just a sketch and needs to be completed, but I believe that it could provide a relatively fair way how to calculate the performance. It is not ideal, but I cannot see any better way.

GM Dr John Nunn, Lyne, England
It's interesting how complicated this discussion has become. The problem is the use of 'performance ratings'. Instead it's much simpler just to use the average of the opponents' ratings. If one player has scored 7/9 against an average of 2500 and another player has scored 7/9 against an average of 2520 then the second player should be ahead of the first on tie-break. It's simple and it doesn't depend on the player's own rating.

At first sight this is equivalent to using 'performance ratings' but that isn't the case due to the 400-point rule, and especially not if one starts eliminating certain results from the procedure. It also has the advantage that it is independent of the FIDE procedure for calculating performance ratings, which is of course subject to change.

I accept the point that if one player is unlucky enough to be paired against an opponent with a rating of 1200 in the first round, then he will effectively have dropped half a point because he will always be last on tie-break, so it makes sense to drop the lowest rated opponent from the calculation.

Paul Lillebo, Asheville, NC, USA
Shouldn't someone make the obvious point that predicting outcomes is NOT a purpose or function of Elo's rating system, or of any useful rating system. Sonas' silly contest to find a system that will "outpredict Elo" has no value for chess. The Elo system does fairly well what it's supposed to do, namely report how strongly the player has played, with emphasis on more recent games. Whether the player's past results are predictive of future results is a whole different question, independent of the rating system unless the rating only takes into account the player's current form, say the past few months. Understandably, elite tournament organizers would love to know how strongly a player will play a year from now, but I'm afraid they'll just have to continue to trust their instincts.

Eddy Fong, Kampar, Malaysia
Perhaps they could consider removing the win against the lowest rated player and the loss against the highest rated player, and calculate the performance rating (Thompson method) based on the set of nine remaining games. This seems a fairer method. All players affected have a point removed. If all games are drawn, then the respective drawn games would be used instead of the win and loss. Only other issues is where a player has one win (with no loss) or one loss with (no win).

Iman Khandaker, Watford, UK
It is bizarre to the point of perversity, that a tie-break method should discard actual RESULTS. Discarding highest and lowest ratings makes more sense – though simpler measures like fewest draws or most wins with black seem equally reasonable.


Articles on the current performance debate

Nielsen protests ECU performance calculations
12.04.2011 – The 2011 Individual European Championship was also a qualification tournament, with the top 23 finishers getting a seat in the 2011 World Cup. Problem was that behind the first 23 players there were 29 with a tied score. So the performance was used to break the tie. However the system produced some bizarre results, as GM Peter Heine Nielsen explains.

Sonas: Assessment of the EU performance calculation
16.04.2011 – The 2011 European Individual Championship left 29 players with a tied score vying for eight places in the next World Cup. To break the tie the ECU used performance ratings, but calculated them in a way that led to some bizarre results – and to a formal protest by at least one player. Jeff Sonas introduces us to other, more logical systems. As usual his report is presented with exceptional clarity.

The Elo rating system – correcting the expectancy tables
30.03.2011 – In recent years statistician Jeff Sonas has participated in FIDE meetings of "ratings experts" and received access to the historical material from the federation archives. After thorough analysis he has come to some remarkable new conclusions which he will share with our readers in a series of articles. The first gives us an excellent overview of how the rating system works. Very instructive.

Reports about chess: tournaments, championships, portraits, interviews, World Championships, product launches and more.

Discuss

Rules for reader comments

 
 

Not registered yet? Register