
BGonline.org Forums
Number of games necessary to get approx PR
Posted By: Maik Stiebler
Date: Tuesday, 11 February 2014, at 10:22 p.m.
In Response To: Number of games necessary to get approx PR (Bob Koca)
Agreed. What would be a better technique? I'd submit the following: Distribute all the games you have into a number of baskets, such that all baskets have an approximately equal total number of decisions D. Maybe something on the order of D=1000 is a good number of decisions per basket, because that is large enough to be approximately met by a well chosen set of games and, if you have 4000 games, means that there will still be enough baskets to do statistics with. Randomize the distribution process to reduce the influence that the temporal development of skill might have. Calculate average PR's for the baskets and the variance V of those average PR's. The number of decisions needed to calculate a PR with a confidence interval of +0.2 is approximately D*V/0.01.

BGonline.org Forums is maintained by Stick with WebBBS 5.12.