This is an interesting measurement problem. According to the central
value theorem, if the process used to count is susceptible to random
errors, then the error ("noise") is some number times the square root of
the number counted. The question is, what number should that square root
be multiplied by? One would have to do some statistical testing of each
process to determine it. This has probably been done in the case of
counting machines and voting machines, but I wonder if this has been
studied for manual counting procedures?
Jim
"Hooper, Bill and or Barbara" wrote:
>
> I agree with Scott's observation (below) that our Florida election problems
> can be viewed as a signal-to-noise ratio problem. In my earlier suggestion
> that there is ALWAYS some error in ANY measurement or count, that error
> would be the noise in Scott's analysis, while the real difference in total
> votes between candidates is the signal we are trying to see. If you can't
> reliably see the signal for the noise, call it a tie.
....
--
Metric Methods(SM) "Don't be late to metricate!"
James R. Frysinger, CAMS http://www.metricmethods.com/
10 Captiva Row e-mail: [EMAIL PROTECTED]
Charleston, SC 29407 phone/FAX: 843.225.6789
