11-30-2004, 05:29 PM
November 30, 2004
Statistical analyses that depend on testing a large number of possible sequences to find a few that work are not valid. If you used your code program to test numerous word sequences and ELS possibilities until you found some that worked, the statistical probability that you assign to your result is not correct. In fact, the chance of finding a sequence that works is good as long as you are willing to use a trial-and-error process until you find one. That pick-and-choose approach is inefficienct but now can be facilitated by computers that allow rapid analyses.
In order for your ELS statistics to be correct you must start with a single a priori hypothesis and single ELS value and and test it alone. If the first one doesn't work, and you keep looking until you find one, the statistical significance is lost.
Otto
Statistical analyses that depend on testing a large number of possible sequences to find a few that work are not valid. If you used your code program to test numerous word sequences and ELS possibilities until you found some that worked, the statistical probability that you assign to your result is not correct. In fact, the chance of finding a sequence that works is good as long as you are willing to use a trial-and-error process until you find one. That pick-and-choose approach is inefficienct but now can be facilitated by computers that allow rapid analyses.
In order for your ELS statistics to be correct you must start with a single a priori hypothesis and single ELS value and and test it alone. If the first one doesn't work, and you keep looking until you find one, the statistical significance is lost.
Otto