eXTReMe Tracker
Front - Tournaments - Search - Rating - Reports - Antichess - Link
  Who is Jeff Sonas? Chess Metrics or Mere Tricks?

Who is Jeff Sonas? Chess Metrics or Mere Tricks?

10.05.2013 /   Andrejic, Vladica (2212)

"He gives you illusion that has the appearance of truth. I give you truth in the pleasant disguise of illusion." — Tennessee Williams, The Glass Menagerie

Wikipedia asserts: Jeff Sonas is a statistical chess analyst who invented the Chessmetrics system for rating chess players, which is intended as an improvement on the Elo rating system. He is the founder and proprietor of the Chessmetrics.com website, which gives Sonas' calculations of the ratings of current players and historical ratings going back as far as January 1843. Sonas writes that, "Since the summer of 1999, I have spent countless hours analyzing chess statistics, inventing formulas and other analysis techniques, and calculating historical ratings." He has written dozens of articles since 1999 for ChessBase.com and other chess websites. He was a participant in the FIDE ratings committee meeting in Athens in June 2010. Jeff Sonas graduated with honors with a B.S. in Mathematical and Computational Sciences from Stanford University in 1991.

People who use the Chessbase website likely know of Sonas as their undivided expert on chess ratings, and statistics in general. He emerged there in 2002, when he published an article, The Sonas Rating Formula – Better than Elo?, which he then followed by churning out numerous other articles for Chessbase.

Two years ago there was a controversy about which players qualified for the World Cup, and I offered my correct answer to the world, and then released what has been done wrong at the European Championship 2011. My article were first reported by Chessvibes and Chessdom, and then almost all important chess media reported this news. However, Chessbase was not among them, although they were properly informed. Ten days later, their expert, Jeff Sonas wrote his article and publish it on Chessbase website. Look at the dates and links to published articles at that time:

05.04.2011 Perpetual Check: "European Championship Scandal"
06.04.2011 Chessvibes: "World Cup qualifiers: how to calculate?"
06.04.2011 ChessDom: "Who qualified for the World Cup? Tiebreak issue also explored"
16.04.2011 Chessbase: "Sonas: Assessment of the EU performance calculation"

If you compare "the original" Sonas article with my article published much earlier, you will easily see quite too many similarities. The pinnacle is the table which alternately rates the standings of the 2011 European Championship, if some other decent rating performance systems were being utilized. Sonas suddenly had my ideas and my table, but not my name, which was not mentioned anywhere! Ethically, such behavior is unacceptable, but I nevertheless choose to refrain from using some words that have long been borrowed to describe such people!

Let us talk about the performance rating system that I created for my own use, and which (in the absence of a good name) I named after myself: Andrejic performance. Sonas expounded a cute little story regarding the origins of this performance system. In his article, he wrote: "Ken Thompson has previously suggested an eminently reasonable calculation for performance rating," and named it the Thompson performance rating. It doesn’t really matter how the thing is called, but the approach is significant. I’ve read almost all of Sonas's articles, and he had never mentioned such a thing before, nor can one read anywhere else that Thompson suggested something similar. The point is that I gave out these nice ideas in my articles and explained why such methods of calculation are good and reliable, and included some numbers and calculations. Suddenly, when he read my article, this concept was good, and as abruptly, Thompson had said something, sometime...

I can say that it’s absurdly humorous for me to see "Copyright Jeff Sonas / ChessBase" at the bottom of "his article".

The main components of a scientific paper are the formulation of ideas, the essence of understanding, the proof of a logical concept, and the inclusion of the data which confirm or refute the theory. If you read my articles about rating e.g. The Truth about Chess Rating and Chess Rating for Laymen, you will see that ideas abound. However, all of Sonas’ articles start with the setup of concrete formulae, some calculations follow, and the end-results are pleasing to the eye. There, you cannot find a logical explanation of why, exactly, we started calculating things in this way.

Generally, Sonas has a good concept, in that chess rating should be such that future outcomes can be more accurately predicted. However, that is objectively impossible — there is no absolute mathematical solution which can definitively rank those predictions. Selection of such a definition is a subjective thing, and beside the three accepted methods (mean squared error, mean absolute error, or simple zero-one loss function), we can create infinitely many different meaningful loss functions. In his article, Sonas claims that his chessmetrics system allows better prediction for chess results than some other rating systems. However, this is because he created his system in such way that it better meets his definition of better predictions.

Sonas began his rating expert journey with a linear function of rating expectations, and he based his chessmetrics rating system upon this. However, in one of his later articles, he concluded that the logistic curve is better. Do we really have to wonder if it is a coincidence that I published The Truth about Chess Rating half a year before? In my article, I consider logical arguments, calculate, and finally show with statistics that we are far away from any linear function.

You can see all sorts of miracles in his "works." One of them is a metric-mania finding the strongest performance in a single event in all of the history of chess. The final conclusion is that Karpov, with 11 of 13 from Linares 1994, has a Sonas performance of 2899, which is better than Robert James’ 6-0 win over Larsen in 1971, with the "modest" Sonas performance of 2895. Of course, there are no logical explanations to be found of why he used specific formulas, but at least we have nice numbers. Should I add that no creature in the universe would be ashamed with Fischer’s 6-0 victory, and that his performance is positive infinity? Consider this "paradox" — I could go on ad infinitum, yet I'm rendered speechless...