Author: José de Jesús García Ruvalcaba
Date: 07:36:07 06/21/01
Go up one level in this thread
On June 20, 2001 at 17:48:35, Dann Corbit wrote: >On June 20, 2001 at 17:36:06, Albert Silver wrote: > >>On June 20, 2001 at 15:58:55, Dann Corbit wrote: >> >>>Using the rating calculator found here: >>>http://wolf.project-w.com/prog/rating/ >>> >>>With a starting ELO of 2600, playing against opponents of 2600 ELO, and scoring >>>1% of the points (a TPR of 1800) would give a new ELO of 2592 (which is a >>>difference of 8 ELO). >>> >>>However, I see that the number of different opponents you face has a big impact >>>on the calculation. So, if you played a 100 game match against a single >>>opponent, it would not hurt much if you lost every game. But if you played 100 >>>single games against different opponents and lost them all, it would cost a lot. >> >>That makes no sense to me. Why would losing 100 games to a single 2300 opponent >>be less detrimental to my rating than losing 100 games to 100 different 2300 >>opponents? > >Even this is (perhaps) just poking numbers into holes. Using the tool from the >above web site, I used the following: > >Opponent rating was set to 2600 >Total score to .01 (which is what an 1800 player would manage) >Your rating to 2600 > >and it spit back: >Your new rating is 2592, you lose 8. > >Perhaps this calculator also does not perform FIDE type calculations. Or >perhaps I am using it incorrectly. That tool does not calculate FIDE ratings or anything close to that. For the FIDE rating it only matters the rating of the opponents and the score of the player, it makes no difference to play the same opponent twice or two different opponents with the same rating. Here it goes a brief explanation. To calculate the rating change for player in a chess competition, take first the average rating of the opponents at every round (so if our player faces n-times the same opponent, that opponent's rating is to be used n-times in the average calculations. What happens normally is that you play every opponent once, or every twice, you face seldom one opponent more times than another. So, for most tournaments this average is the same as "average rating of opponents"). Now make the substraction rating of our player minus average rating of the opposition (in the sense described above). This number can be negative, zero or positive. This rating difference is translated into a "expected scoring percentage". It is always greater than 0% and lower than than 100%. It is an strictly increasing function and is odd with respect to the value 1/2 (in the sense that this function minus the constant 1/2 yields an odd function). So a rating difference of zero gives a expected winning percentage of 50%. Now multiply this expected winning percentage by the number of games. This number is the "expected score" of our player. It does not have to be a whole multiple of 1/2! (i.e. a player can be expected to score 4.4 points, which obviously is imposible). At this point substract the actual score of our player minus the expected score. This again can be negativ, zero or positiv (a player wins rating points when she/he scores better than expected, etc.) Multiply this score difference for the famous "K-factor". I do not know the current regulations, but for FIDE it used to be 10 for players rated over 2400 and 15 otherwise. This last product is the rating change for this player in this event. It will be eventually added to the player's rating at some point in the future, preferably before the next rating list appears. Hope this helps. José.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.