Actual v Expected

General discussions about ratings.
Post Reply
Roger de Coverly
Posts: 21291
Joined: Tue Apr 15, 2008 2:51 pm

Actual v Expected

Post by Roger de Coverly » Thu Aug 22, 2013 2:34 pm

A few years ago, the grading team got it into their heads that the grades were wrong and needed to be reworked. In my mind they failed to produce much convincing evidence for this, particularly when applied to the 170 + range. What they did eventually produce was what could be termed "the Grist graph". This was a plot of actual v expected results across grading differences.
http://www.ecforum.org.uk/viewtopic.php ... 012#p15012

With access to the FIDE data, Jeff Sonas has now plotted similar graphs.
http://www.chessbase.com/Home/TabId/211 ... 20813.aspx

What he investigated, that I don't think the ECF did, was whether the graph varied by rating of the players and by the year of investigation. His conclusion was that the FIDE ratings worked better as rating predictors, the greater the strength of the players. Whether he will conclude that rapidly improving players, juniors in particular, are the ones that can distort rating systems remains to be seen.

Going right back to first principles, what was the justification for using the distribution underlying the International Elo tables in the first place? The USCF use a slightly different formula and the ECF linearise it, but the practical chances of the seriously out-rated player seem rather better than predicted by any of these. This is not least because of ratings lagging improvement.

Sean Hewitt
Posts: 2193
Joined: Sat Mar 10, 2012 8:18 pm
Contact:

Re: Actual v Expected

Post by Sean Hewitt » Thu Aug 22, 2013 4:29 pm

Roger de Coverly wrote:What he investigated, that I don't think the ECF did, was whether the graph varied by rating of the players and by the year of investigation. His conclusion was that the FIDE ratings worked better as rating predictors, the greater the strength of the players. Whether he will conclude that rapidly improving players, juniors in particular, are the ones that can distort rating systems remains to be seen.
He already has concluded that. Or, more accurately, agreed with my conclusion that that is the case.

Post Reply