On Sun, 12 Nov 2000, Kevin Elliott wrote:
This is why people who don't know statistics should not be allowed to think... By no means is that number, by itself, of any significance whatsoever. How many got canceled last election- one number I heard said 14,000. If so then 19,000 is about what one would expect considering increased voter turnout and normal statistical fluctuations.
Quite. The problem here is what happens when the mean expected error of the estimate given by the ballot starts to get significant with respect to the mean popularity difference being measured. There is always some error, but it is not often that the actual difference in votes given to the main participants shrinks too low for the error to have any relevance. Simply put, we are faced with the scourge of binary decision problems based on noisy data. Sampo Syreeni <decoy@iki.fi>, aka decoy, student/math/Helsinki university