Calculating playing strength comparisons

[towforce]

I recall in 1999 John Nunn ran Carlsbad 1911 througha fritz 5 and came to the conclusion that the tournament average was 2100. It was controversial at the time.

Whilst I appreciate the fact that computer analysis doesn't account for human long term planning, I don't think that people should be surprised that the general standard of chess was much poorer 100 years ago: it is my sincere opinion that if you took the best of the 19th century chess players (I would be surprised if any members of this forum couldn't name one or two) and put them in a modern high ranking tournament, they would finish near the bottom. They certainly produced some exciting "blood and thunder" style chess - but it would be loosing chess by today's standards.

[Hart]

Somebody did something similar here for several top players:

http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=5850

Rybka is limited to 14 ply but the data seems to follow a sensible path. I would be curious to know what kind of conclusions a statistician would draw from this.

[Uri Blass]

I think that before trying to evaluate players of the past you need to have good way to evaluate rating of players of today only based only on the pgn without data about names and rating of the opponents.

I do not know about a program that exists that is good in this task
and first test for the program can be to give the program games of 100 players when every player has 10 games in the database so 1000 games (50 of the players are going to be with rating 2600-2800 and 50 of the players are going to be with rating 2100-2300)

next step is going to see if the program detect the 2600-2800 players only based on the pgn.

I suspect that there are 2100-2300 players who never missed mate in X in the relevant 10 games (because even when they won their opponent resigned clearly earlier) so missing mates is going to fail.

I suspect that methods like using difference between computer evaluation and the move can also fail because 2100-2300 players with positional style can get positions when it is easy not to make tactical mistakes when there are 2600-2800 players with more tactical style who may get in their games positions when it is hard not to do tactical mistakes.

I do not claim that it is impossible task to have a program that identify the 2600-2800 players as players with higher rating but I think that the task is not easy.

[icander]

Have you checked this site:

http://www.truechess.com/web/champs.html

[Anil]

Also, how about http://db.chessmetrics.com/CM2/PeakList.asp

Code: Select all

 	      Player Name   	   Average Rating   	   3 year peak range   
 #1 	   Garry Kasparov   	   2874   	   1989-Jan through 1991-Dec   
 #2 	   Bobby Fischer   	   2867   	   1971-Jan through 1973-Dec   
 #3 	   José Capablanca   	   2857   	   1919-Jan through 1921-Dec   
 #4 	   Emanuel Lasker   	   2855   	   1894-Jan through 1896-Dec   
 #5 	   Mikhail Botvinnik   	   2852   	   1945-Jan through 1947-Dec   
 #6 	   Alexander Alekhine   	   2841   	   1930-Jan through 1932-Dec   
 #7 	   Anatoly Karpov   	   2833   	   1988-Jan through 1990-Dec   
 #8 	   Viswanathan Anand   	   2822   	   1997-Jan through 1999-Dec   
 #9 	   Vladimir Kramnik   	   2815   	   2000-Jan through 2002-Dec   
 #10 	  Harry Pillsbury   	   2806   	   1900-Jan through 1902-Dec