Author: KarinsDad
Date: 08:50:01 10/07/99
Go up one level in this thread
On October 07, 1999 at 02:26:44, Dave Gomboc wrote: [snip] >> >>Say two 1600 players are supposed to play, but in the last round, one gets in a >>car accident (he is not injured, he just cannot make it to the tournament) and >>forfeits. Nobody knows why he forfeited, so the TD sends in the pairings to the >>federation. > >This isn't a good example, because once it became known that the person didn't >show up for no fault of their own (it would have been illegal for them to leave >the scene of the accident without permission of police), the ratings auditor >would re-rate the tournament, not counting the forfeit as rated. More usually, >the tournament report is not sent in the same day, so the TD will probably get a >call explaining the situation. > >My chess club has touraments that are played at the rate of one game a week, >every Thursday evening. The tournaments are 4-7 weeks long, depending on the >tournament. So we'll keep the two 1600 players who are supposed to play, one of >whom occasionally forgets that it's Thursday and doesn't show up to play. I >hope this situation seems realistic to you. Yes, so since the guy forgot to show up or because it was his Wedding anniversary or because his boss asked him to work late should NOT affect his rating. Ratings should not be used to attempt to predict results between potential games, ratings should be used to attempt to predict results between actual games. > >>These two players are both 1600 (let's assume that they both maintained their >>rating in the first x rounds of the tournament). They are fairly close in >>strength. But after this tournament, one of them is rated 1616 and the other is >>rated 1584. Is it true that the 1584 player is REALLY 32 points lower than the >>other? Of course not. But, that does not stop him from getting this rating. > >>How can one state that it is MORE accurate to predict results between the 1584 >>and the 1616 player when in reality, they should both be 1600 players? The >>prediction should be off. Please explain with more detail since I am clueless on >>this (and yes, I realize that if someone has a propensity to forfeit, this would >>lower his rating in the long run). > >Let's call the fellow who jumped to 1616 "A", and the fellow who dropped to 1584 >"B". It's clear that player B deserves a lower rating than player A. The two >players have the same level of skill, but B is less likely to show up, so his >expected score over a series of games against A would be less than 50%. This is an assumption on your part (i.e. noise). Ratings should be a method to predict probabilities of results of actual games, not probabilities of game results (i.e. the probability of a no-show should NOT be part of the equation). > >>Thanks, >> >>KarinsDad :) > >The rating system just follows what you ask it to follow. If you want to model >"games where both people showed up on time", that is legit enough, I suppose. >Canadian ratings model "all games", which seems more valid to me, personally. >Of course, some people don't want to rate the games that the GM played when they >were sick, or the games where the program lost because the hardware was >overclocked too much, et cetera. I guess it's different strokes for different >folks... > >Before anyone objects too loudly about rating games where a fellow doesn't show >up, consider: not too different is the case of another fellow at my chess club >who does show up for his games reguarly. He is a software developer working on >Y2K code. Once in a while, he gets a call on his cell, stands up, and resigns >his game, because he received a call from work and he has to go in. It might >seem "unfair" to rate that game... at least if he was winning at the time. But >he lost the game! What can you do? And you know, he's more likely to lose >games in the future for the same reason. Worst of all, sometimes people play on >in fairly hopeless positions against him, in the hope that his phone will ring! However, the difference is that he starts the games. He may not finish them due to his "problem", but then again, that would be reflected in any rating system. > >CFC ratings predict results, period. No, they don't. CFC ratings predict results of potential games (whether they are played or not) as opposed to results of actual games. I realize that there are different strokes for different folks, but I have to tell you that this seems really strange to let random noise into the equation. If the USCF wants to predict "results of >games when both people show up before 1 hour has went by on their clocks" >instead, I guess that's their business. > >Dave Guess so. To tell you the truth, I have not even thought about this subject before this week, so I do not think that I have a preconceived notion on how it should work. However, I have a feeling that the reason you are so comfortable with the CSC system is that is what you are used to. And, I can see one major advantage of the Canadian system. It enforces the "gentleman's agreement" to play the game. If you don't play and your opponent still had to show up, your rating suffers. That is probably the REAL reason for the rule. And, looking at it in this light, it makes total sense to use it. KarinsDad :)
This page took 0.01 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.