Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: ELO interval and confidence levels

Author: Bas Hamstra

Date: 14:21:25 05/28/04

Go up one level in this thread


On May 28, 2004 at 17:05:52, Dieter Buerssner wrote:

>Bas, I fear Elostat error margins are not always reliable. Peter Fendrich and me
>had some discussion here before, and we looked at some extreme results. Elostat
>error margins seemed wrong.  From my memory: we compared something like 1010-990
>under 2 different conditions. Once the better engine won 20 games and 1980
>draws, and once 1010 wins and 990 losses, no draws. At least in one of those
>cases, the error margins reported by Elostat were rather suspicious.
>
>Regards,
>Dieter

Maybe under extreme conditions. But I must say that so far in this little blitz
match I am running the statistics equal exactly what Shredder 5 produces. Same
ELO difference between 2 versions, and the same intervals. But I will keep an
eye on it...

I use version 1.1 here, if I am not mistaken I indeed read somewhere something
about a 50% bug, probably on the web.


Bas.






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.