Author: Chris Carson
Date: 08:03:06 04/13/00
Go up one level in this thread
Yes, you make a couple of good points. The error of measure is 100 points for all the human vs computer games that I have. The error of measure for the SSDF data is around 30, so two error of measure for each sample and you can explain at least 95% of the data. :) This is true for sampling theroy. However, when FIDE or the USCF calculates a TPR or a rating, I have not seen an error of measure associated with it (though it could be calculated). Human ratings (FIDE and USCF for example) are not stable and fluctuate up and down as well. I am not surprised that computer ratings behave the same way (as sampling theory suggests). :) I would love to see GK's rating with +- 100 points next to it (small number of games). Showing error's of measure (or 2 for a 95% confidence) would throw the "who is number 1" question into a big debate. :) Best Regards, Chris Carson
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.