Author: blass uri
Date: 08:12:01 08/09/98
Go up one level in this thread
On August 09, 1998 at 10:47:58, Danniel Corbit wrote: >On August 09, 1998 at 09:59:47, Christophe Theron wrote: >[snip] >>NEVER. 10 games is NOT ENOUGH. >> >>I have found that a 30 games match has a +/- 5% error margin. That is a 50% >>result can mean 45% to 55%. >[snip] >The book "Statistical Methods" by Pfaffenberger & Patterson [ISBN 0-256-03664-0] >makes the following quote on pages 504 & 505: >"Hypothesis tests concerning the population variance or the population standard >deviation are based on the X^2 distribution first introduced in Chapter 9. These >tests assume that the population random variable X is normally distributed. If X >is not normally distributed, but is approximately so, then these tests should be >used only for "large" samples (n = 100 or more is often required). The degrees >of freedom for tests concerning the population variance and the population >standard deviation are df = n - 1." > >So we might take 100 as a _minimum_ number of trials to get a result about an >unknown population. Interestingly, they supply detailed formula for >determination of the smallest sample size you should use for testing hypothesis. > On pages 503 & 504 the sample sizes worked out in the example problems were 451 >and 266. > >So you are very, very correct. A sample of ten games probably shows us very >little. It depends on the result If the result is 10:0 after 10 games it is significant result. The best test is not to decide about the number of games before the test but to define the possible significant results and to stop playing when you have significant result(of course you have to decide about the maximal number of games before the test). Uri
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.