Who was the most accurate national pollster this year?

[Mr. Morden]

"Pollster accuracy" is inherently unmeasurable, IMHO.  Yes, you can measure the accuracy of the final poll before the election, but the final poll before the election isn't that useful, as it's coming out just a few hours before you're going to find out the election results anyway.  We follow polls not to figure out who's going to win just a few hours ahead of time, but to track the campaign weeks and months in advance of election day.  Does this kind of "pollster accuracy" comparison really measure that?

Here's a plot of IBD/TIPP, compared to the pollster.com regression line that combines all polls, for the final weeks before the election:



IBD/TIPP leaned several points more towards McCain than the overall polling average during most of the final weeks of the campaign, but they both converged to about the same number on election day.  So who was right?  Was IBD/TIPP right, and Obama only led McCain by 2-4 points or so in the final weeks, but then surged to a 7 point lead at the end?  Or was the overall polling average right, which had Obama's lead over McCain remaining relatively constant in the final weeks?

We can also look at the trendlines for all of the individual trackers:



This is supposed to "prove" that IBD/TIPP is a great pollster, while Gallup is bad?  Uh....OK.

[Mr. Morden]

The IBD and Zogby polls are amusing... note the huge surge in both on the very last day.

With the exceptions of Gallup and Zogby, all the polls kind of converged towards the right answer in the final days of polling.  But those polls that converged on the right answer at the very end were all going in different directions throughout the rest of the campaign.  Which shows how stupid measurements of "pollster accuracy" are.

What you really want to be able to measure is how accurate are the polls weeks or months before the election.  But unfortunately, that's objectively unmeasurable, since you have no way of knowing for sure what the correct answer is.