One of the smartest guys I know was involved in a discussion just after the debates. I wish I could credit him with this, but at least we can bask in his knowledge:

I've worked fairly frequently with polling organizations, and the entire industry is in an upheaval. By and large, the trend in polling has dramatically turned over the past 12 years. They used to run 2-4 points more liberal than reality, and about 5-6 points less racist than reality -- 10-15 points in the South.

Consider the polls just prior to the 2000 Presidential election:

BUSH GORE
% %
CBS ^ 6-Nov 44 45
CNN/USAT/GALLUP #@ 6-Nov 48 46
CNN/USAT/GALLUP # 6-Nov 47 45
IBD/CSM/TIPP ^ # @ 6-Nov 47.9 46
REUTERS/MSNBC ^ # @6-Nov 46 48
VOTER.COM ^ # @ 6-Nov 50 45
VOTER.COM ^ # 6-Nov 46 41
ABC # * 5-Nov 48 45
HARRIS ^ @ 5-Nov 47 47
HOTLINE ^ 5-Nov 45 42
ICR ^ 5-Nov 46 44
NBC/WSJ 5-Nov 47 44
PEW @ 5-Nov 49 47
PEW 5-Nov 45 43
WASHINGTON POST * 5-Nov 48 45
FOX/OPINION DYN 2-Nov 43 43
MARIST COLLEGE 2-Nov 49 44
NEWSWEEK 2-Nov 45 43
AVERAGE 46.7 44.6
ACTUAL 47.9 48.4

The models applied by the various polling agencies are surprisingly simple in most cases, and in many cases, as dumb as turnips. Many of them were developed in the '50s, and the assumptions used to create the models are often just plain loopy. Add in the non-response factor, and the fact that these companies don't really know what tweaking to do for young voters, and the fact that cellphones have totally confused these guys, and I trust the polls less than I trust Bud Selig. At one major polling organization, I have heard the COO say, in a rather large meeting, "It's a good thing that we're slightly less inept than Gallup."

Right now, polls run conservative compared to reality, at least in individual races. And don't spend too much time worrying about whether the poll is surveying registereds or likelies. The distinction's questionable even before the survey design, and the methods for determining likelies are akin to predicting exactly when a 40 HR guy will actually hit the HR.

They're pretty much Win Shares.

Unless Bush has an average 5 point lead in the polls on November 1, I expect the big news story on November 3rd to be the surprising margin by which Kerry won. The Democrats have a better ground game nationwide this time, which is surprising. If African-American turnout is high, it could even be perceived (or, more accurately, spun) as a mandate.

Of course, the GOP is far better at actually executing, so it's not inconceivable that Kerry could be down 15 points on November 1st, especially if his off-message celebrity proxies like Aaron MacGruder continue their brainless performances on the cable shows.

But I'm a data guy and the data strongly suggests that the polls are more conservative than reality.