Merry Andrew wrote:Actually, no. I've heard several news commentators today lament on how off the mark most of the exit polls were, making this election a true 'cliff-hanger.' Exit polls have gotten to be a huge joke.
For a "huge joke" they sure have the numbers extremely close ...
If you check the link I gave to the actual
final exit poll data, you'll find that it mirrors the actual election results in detail.
No, it's mostly two other problems that came into play.
One: the commentators you reference yesterday were using preliminary exit poll data that was being updated as the day progressed. The preliminary data, leaked out to folks like us by Slate and Drudge, was further off from the actual results than the final exit poll data. Which is only normal: we are talking about incomplete, 'raw' data.
Two: commentators often don't know how to use a poll - no offence. The preliminary exit poll data, for example, had Kerry 51, Bush 49 in both Florida and Ohio, in a later 'edition' revised to Kerry 50, Bush 49. Some commentators will have taken this as meaning that Kerry was winning those states, and then later, when it turned out to be Bush 51 Kerry 49, blamed the exit polls for having "wrongly called FL and OH for Kerry" and being "useless" for predicting the winner.
But that's simply not how it works. A poll has a margin of error. For each candidate the MoE of a regular poll is often up to 3%. So if a poll says Kerry 51 Bush 49, it means that Kerry could have gotten something between 48-54%, and Bush 46-52%. Just to give an example. (I think I got that right.)
So yes, exit polls - and preliminary exit poll data all the more so -
are a bad instrument to predict a winner in close races with. That's not what they're made for, as fbaezer pointed out elsewhere: they're made "to study the demographics and the reasons of voters".
As it is though, even the raw, preliminary data wasn't all that far off - in some 3/4 of states they had both candidates within 2% of what the actual result turned out to be. And the final exit poll data I am quoting in this thread is from overall totals that have the actual election results down to a point, so no reason for dismissal of the numbers here.
Anyway, I had a looong discussion with Fox about all this today already, and tried to summarize the whole thing in 10 points in
this post, and fbaezer explained some more in the next. Lemme know what you think.