PDA

View Full Version : Expected Score

Miguel
09-07-2008, 06:29 PM
I've been looking at the (Swiss Perfect) results from the Freytag Checkmate Open (http://www.sachess.org/results2008/freytag1.htm), and it seems that the expected scores are based on the average rating of the opposition.

This seems a little strange to me. I would have guessed, because of the linearity of the expectation operator, that the expected score should be the sum of the individual expectations (http://en.wikipedia.org/wiki/Elo_rating_system#Mathematical_details), not N*E[ave]. (e.g., for Dejan Antic: E = 0.96 + 0.88 + 0.79 + 0.64 + 0.93 + 0.82 + 0.54 = 5.56, not 5.88* (or 5.83).)

Can somebody explain?

* I think this value of 5.88 probably comes from using an expectancy table like the one found here (http://everything2.com/?node_id=1118278).

Bill Gletsos
10-07-2008, 12:40 AM
I've been looking at the (Swiss Perfect) results from the Freytag Checkmate Open (http://www.sachess.org/results2008/freytag1.htm), and it seems that the expected scores are based on the average rating of the opposition.Correct.

his seems a little strange to me. I would have guessed, because of the linearity of the expectation operator, that the expected score should be the sum of the individual expectations (http://en.wikipedia.org/wiki/Elo_rating_system#Mathematical_details), not N*E[ave]. (e.g., for Dejan Antic: E = 0.96 + 0.88 + 0.79 + 0.64 + 0.93 + 0.82 + 0.54 = 5.56, not 5.88* (or 5.83).)

Can somebody explain?Up until October 2005 FIDE always did their rating calculations based on the average rating of the players opponents subject to a 350 point cut-off.
Swiss Perfect followed this FIDE model.
This does not give totally accurate expected scores.

This is just one reason why I dont believe the SP rating reports should be published by organisers as it gives rise to misleading iderstanding and expectations.

FIDE now do the calculations as you suggested.

I explained this at the time in post http://www.chesschat.org/showpost.php?p=74093&postcount=15

Miguel
10-07-2008, 05:59 PM
Thanks Bill, that's what I was after.

Kevin/pax's TPR is exactly what I was thinking of. But a second definition I had thought of is the rating which has the largest probability of achieving a score of X. Essentially, a maximum likelihood estimate of playing strength. I'm not sure if they must necessarily be the same (I'm guessing not), but the difference is probably going to be small anyway.

eclectic
10-07-2008, 06:21 PM
don't forget that freytag is not a fide rated tournament but rather one rated using the glicko 2 model (i think)

Miguel
10-07-2008, 07:40 PM
don't forget that freytag is not a fide rated tournament but rather one rated using the glicko 2 model (i think)
Just using Elo should give a reasonable approximation, especially for the open section where most players probably have low RDs.

Bill Gletsos
10-07-2008, 11:10 PM
Just using Elo should give a reasonable approximationUnlikely.

I would recommend Barry Cox's Glicko calculator.

Miguel
10-07-2008, 11:45 PM
Unlikely.

I would recommend Barry Cox's Glicko calculator.
I was referring specifically to the expected score. If RD is low, Glicko's g() is close to one, so that E_Elo is approximately the same as E_Glicko.