Ack - I'm having trouble separating my threads. Sorry for the cross-posts. But I just posted this on the results-thread in re: to some remarks by Fox about the value (or purported loss of value) of exit polls. But since it is s all about exit polls, I guess it belongs better here.
Foxfyre wrote:The news media should fire their exit pollsters as this organization has proved to be pretty close to 100% incompetent.
Foxfyre wrote:Roger, can you remember any year when they were off as much as they were this year though? It's hard not to smell a rat.
Incompetent because the data was leaked or incompetent because of the numbers they came up with?
The numbers weren't actually so bad. I mean, the final exit poll numbers as I'm
looking at them now aren't very off-target. Pretty much on-target, in fact.
The ones that came in during the day, through Slate, compiled incrementally, were sometimes further off - so you have a good point about leaking data thats still incomplete and unreliable. But not even that far off.
The problem is that they only needed to be, say, 3% off in order to end up calling the winner wrong in a bunch of races - thats what you get with a close race.
But click that link above, for example, and find what the complete exit poll data says:
National: Bush 50, Kerry 49. Real results: Bush 51 Kerry 48. Thats close enough.
Florida: Bush 51 Kerry 49. Real results: Bush 52 Kerry 47. Close enough.
Ohio: Bush 51 Kerry 49. Real results: Bush 51 Kerry 49. On-spot.
Iowa: Bush 50 Kerry 49. Real results: Bush 51 Kerry 49. On-spot.
Nevada: Bush 50 Kerry 48. Real results: same.
New Mexico: Bush 50, Kerry 49. Real results: same.
I mean, I get your point: I've seen the numbers come in here via Slate too as election day progressed. Florida and Ohio triumphantically
announced here as both going to Kerry 51/49. Iowa going to Kerry 50/49. Nevada going to Kerry 49/48. New Mexico going to Kerry 50/48.
All wrongly called winners.
Yet even looking at those data, the actual
percentage points weren't far off. In all of these cases (just checking the closest races that were called wrong), even those preliminary exit polls weren't off more than 2% on each candidate's percentage.
In fact <checks some more states>, even those preliminary exit poll data ebrown
posted from Slate, they got both candidates at 2% or less from their actual score in 12 out of 16 states. Only in Michigan (3% off on Bush), Colorado (3% off on both candidates), New York (4% off on both candidates) and New Hampshire (completely FUBAR) was it worse. And that was the preliminary data; as said, the
complete exit poll data has all those data practically spot-on.
So the lesson here is:
- Don't trust a
preliminary exit poll to tell you the right winner when the race is within a 5% margin
- But trust it to be within 2-3% on each candidate's numbers 4 out of 5 times
- And trust the
complete exit poll data to give you a pretty-near complete reflection of what people actually voted. Which makes these exit polls still by far the best assessment we have of voter breakdowns and motivations.
Foxfyre wrote:The exit polls are now suspect, but if they got anything right it was what motivated people to vote in the way they did. High on the list in all exit polls was moral values.
Can't cherry-pick which of the questions in one and the same exit poll you deem to be reliable and which "suspect" - you gotta buy 'em for what they are (see above) - and they come together.