Author: KarinsDad
Date: 15:33:54 01/29/99
Go up one level in this thread
On January 29, 1999 at 17:47:08, James T. Walker wrote: >It seems to me that it's not what Arpad Elo said that matters. It's his formula >that counts. Since it is what determines the ratings it has to be true! The statistical formula does not determine the ratings and does not match the ratings adjustment formula. Read below. > Not >that there is no probability of error but that since the formula gives you the >rating based on the WE (Winning Expectancy) then the WE is correct by >definition. Of course because of statistical probability people or computers >will not always perform as per the WE. The point is that people and computers >get their ratings from the same formula. The only thing imperical data will do >is show that there is of course only probability not perfection. Maybe if you >could gather the statistics from millions of games the the actual data and the >probable data would be equal but don't bet the house on it. >Jim Walker What are the formulas again? Someone posted the statistical formula a few weeks back, but I forget. I believe that if you check, the statistical formula does not correspond to the USCF ratings adjustment formula of +32 points beating someone 700 points higher, +16 points beating someone the same rating, etc. The statistical formula is a crude approximation for WE and is not exactly based on the points handed out (at least in the USCF). Think about it. You beat someone 700 points higher in standard times. That is for all practical purposes next to impossible (without some other factor such as your rating is deflated) and something that would make the newspapers and you get 32 rating points for it. Hello! The descrepancies work both for statistics beyond 300 points and for points gained for beating someone beyond 300 points. The increase in points has no bearing on anything EXCEPT within about +-300 points of your current rating where it closely matches. Any formula that more closely matches the USCF data of millions of games across the board would be better than the ELO statistical formula which breaks down. Now, I realize that the USCF rating is not based 100% off of ELO (i.e. it is about 100 points higher due to the floors for one thing), but it is close enough for discussion purposes. In other words, the point gain/lose formula should be based off of millions of games of data and not on the mathematics. The mathematics should also be based on where you are in the ratings (see below). Additionally, in the USCF, if you have a rating above 2200 (I think that is the number, I'll have to look it up), then the modifications to your rating are cut in half. Why is that? Because otherwise, one good tournament could improve your rating by 150 points and statistically and strength wise, you do not belong there. The mathematical system falls apart at the top of the scale, just as it is easier for a 1200 rated player to beat a 1700 rated player than it is for a 2100 expert to beat a 2600 GM. Why? Because a GM makes very few mistakes and has been playing for years to get to that level (i.e. to be that solid). It is rare for him to blunder to the level that the expert can see it. The 1700 player makes mistakes all of the time and sooner or later, he will make one that a 1200 player can capitalize on. KarinsDad
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.