Author: Robert Hyatt
Date: 06:53:27 01/09/00
Go up one level in this thread
On January 08, 2000 at 23:16:55, blass uri wrote: >On January 08, 2000 at 22:29:17, Robert Hyatt wrote: > ><snipped> >>No it isn't... this is about statistics and sampling theory. The Elo >>rating system predicts game outcomes between players in the _same_ pool of >>players. > >You cannot predict games outcomes between players even if the players are human >by one number. > >suppose there are 2 player who score 50% against 2400 players > >The first player is stable and score 80% against 2200 players and 20% against >2600 players. > >The second player is unstable and scores 60% against 2200 players and 40% >against 2600 players. > >what is the right elo for the 2 players. > >The best guess is 2400 but you cannot predict if their outcome against 2600 >players by one number. > >You need at least 2 numbers(rating and stability) to do a better prediction and >a better rating system > >There are also cases when > >A scores 60% against B >B scores 60% against C >and C scores 60% against A so information about the history games between the 2 >players can also help in prediction of the outcome. > >Knowing who is the white player is also important in prediction the outcome. > >Uri This is known as the "standard deviation" and "variance" sort of statistical analysis. Certainly applies to Elo calculations. But nobody really quotes those as it gets too 'wordy'. ie 2445 with a variance of +/-45 with a 90% confidence interval. It gets left at "2445". :)
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.