|
"Error should skew in both directions; if the coin keeps coming up heads, something is wrong with that coin."
Well statisticaly (and realistically) speaking, there isn't something wrong with the coin, there's something wrong with the toss. Coins cannot be biased so as to land heads (or tails) more or less than 50% of the time. The bias comes from the toss.
Cannon's reasoning is absolutely correct but only assuming the exit poll data is a random sample - which I can assure you it certainly was not random.
A better analogy is to have 50 people flip a coin 50 times and then only asking 10 of them to report the results of their first 25 tosses. Basically, it's meaningless. The only exit poll data that IS useful is the final uncalibrated data that Edison-Mitofsky claim will be released in 3 months.
|