Zogby predicted in his last pre-election poll that Gore would win 48-46. The actual race was a virtual 48-48 tie. Most pollsters, not just Zogby, had the race virtually tied or within the margin of error.
But the real problems with his current poll are that (1) he is using the 2000 turnout model (D39, R35, I26), which seems unrealistic after 9/11 and the following elections and (2) no one else is finding similar results. Right now they point to a Bush lead, although it is only 3-4 points.
Moral of the story: don't believe any poll in isolation. The best approach is to do what realclearpolitics.com does, which is to use an average of all the major polls.
Perhaps. It's comfortable that way...but the problem with all of the polls is there is something (or several somethings) odd happening for which there is no model. Frankly, we won't be able to determine an accurate model without the benefit of hindsight.
Is the last 5-6% of Bush support really soft? Is there 4% of the electorate that didn't vote before that is going to vote Republican? ...there's a 10% uncertainty right there, quite aside from peoples moods on election day.
Different polls are using different turnout models...and the key IS going to be turnout. We can say this poll or that poll is accurately polling, but are they properly predicting turnout? That we won't know until after the election.