Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Program to calculate statistically significant results

Author: Steve Maughan

Date: 14:13:30 01/29/00

Go up one level in this thread


Christophe,

>I'm not sure I understand. But if you don't take draws into account, then your
>results are not OK.

As I said the binomial distribution does not account for half points.
Nevertheless, the program is 100% accurate as long as there is not an odd number
of draws.  For example suppose A v B resulted in +10 -4 =2 then one would say
that the score was A = 11 and b = 5 ie not significantly better at the 95%
level.

If there are an odd number of draw then you can interpolate.  Suppose A v B
resulted in 10+ 5- 3= then A has 11.5 and B has 6.5 then once again the score is
not signigicant since for when the loser get 6 point the winner must score at
least 15 and when the loser score 7 points the winner must score 16;
interpolating would suggest that when the loser scores 6.5 the winner must score
15.5 for the result to be significant at 95%.

Statistics is not an exact science - in ths case the level of significance
required is arbitrary - the whole point of the little program was to show the
order of maginitude of the differences required to give a significant result.

Regards

Steve Maughan



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.