Author: Sune Fischer
Date: 08:01:04 06/03/02
Go up one level in this thread
On June 03, 2002 at 08:41:12, Chris Carson wrote: >...you can compare avg GM, Top 10 average, # above 2700, ... >That is my point! Ratings can be compared and descriptive stats from different >era's can be compared. This is not the only type of comparison that is valid, >but this is one valid and objective method to compare people from different >era's. Hypothesis testing can also be done to establish any ratings inflation, >these would include skew analysis. No you can't compare ratings. If you take a pool of 100 people and let them play eachother again and again they will improve, yet their mean rating should stay constant. If you add bonus points (e.g. 0.5 elo for every match played) then how will you know how much elo this strength increase is worth, ie. why not add 0.456 elo)? How will you even measure they have improved, since they have _all_ improved, so there will be no relative strength displacements (in principle). What if all 100 players suddenly stop playing for 1 year, they become rusty yet their rating remains from what it was a year ago. How will you decide on the amount of elo to subtract when a player as been inactive for a given time? Finally, new players will enter this pool all the time and climb in rating on the expence of others, and when old GMs stop playing they do not release their rating back into the pool of active players, so the average must go down. There are no fixpoints to calibrate by, no human performs at a *constant* level, even from day to day there can be large fluctuations. Machines without learning can perform at a constant level, so they would have to enter the pool as players, doing the task of fixpoints. -S.
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.