Lessee if we can break this down real simple for ya, gozmo.
Polls are nothing more nor less than a means by which an outcome's probability may be estimated at a given level of confidence to lie within a given range ... the "Margin of Error" at the "Confidence Level". Strictly speaking, any separation less than twice the margin of error is "statistically too close to call"; if the MOE is 3%, for instance, and the Confidence Level is 95%, candidate A could poll 51% and candidate B could poll 49%. Now suppose candidate B receives 2% more actual votes than had been projected, while candidate A recieves 2% fewer. Though "Leading by 2 points going into the election", candidate A loses by 1 point, yet the poll projection remains "accurate within the margin of error" ... there was a 95% probability the election would mirror the poll, plus-or-minus 3%, and it did. In fact it did pretty well; it was off by half the margin of error. Without getting into the math of it, there was a 50% chance the election could go either way ... "too close to call".
Now consider that the recorded election results themselves are static; they are as they are. They are recorded, regardless any other consideration. That, by default and by definition, makes them the benchmark. If the polls and exit polls and the recorded results all are congruent with one another within the margin of error of the polls, all likely is well, to a statistical probability approaching certainty. If there is disparity beyond the margin of error, there is cause for closer examination of things.
Exit polling data is dynamic, and cumulative, the collection of the data occurs while the event is in progress, and is subject to skew or error from any number of causes, from weather to the news of the day ... even the time of day ... and the sites, and individuals, selected for the sampling. Lets again suppose early exit polls show candidate A with a 1% advantage, those a bit later the advantage becomes 2%, a while after that its 3%, then its back to 2%, then back up to 3%, then as evening wears on drops to 2%, then 1%, then turns to a 2% disadvantage, and finishes at a 1% advantage. When the actual vote is counted, however, it turns out candidate B actually received 2% more votes than had candidate A. Does that indicate anything untoward? No, not at all; the exit poll data missed the recorded result but remained well within the margin of error; there was always the statistically very significant possibility the race could go either way.
Now, should one candidate have gone into the election with a significantly-beyond-margin-of-error polling lead, and the day's exit polling reflected similar results, yet at the end of the day, the opponent eked out a narrow victory, or worse, be brought a crushing victory through a last-minute surge, it would be right to raise an eyebrow ... and a helluva fuss.
In our most recent election, what the exit polls showed was that, just as the pre-election opinion polls showed, by all the polls the difference between the candidates was too narrow to allow any better than a gut guess at the eventual outcome. From the internals of the exit polls, one could determine within statistically significant probability that
x number of
y demographic who actually voted
APPARENTLY chose one candidate or the other for set
z of reasons .... and not a damned thing more. To assume more could be made of the exit polling data ... that a probable winner could be called well before the polls closed, would, to an experienced gambler, be a stupid bet; the uncertainty zone was way too wide for a straight-on unhandicapped wager, one way or the other.
As has been said, "Its not the exit polls; its the stupid". Not understanding polls or probabillity, hopeful, even overconfident dilletantes and wonks lept to encouraging, though unwarranted, conclusions, and found themselves inconvenienced by reality.
But nevermind ... I've got a great idea to smooth things between us. Wanna play cards?
:wink:
