You don't put too much stock in polls?

Good Lord, that 2004 polling was awful, how did I ever convince myself that Kerry had a 50%+ chance of winning? But the average of those late '04 polls DO look like they reflected the reality pretty well, which means good things for the present.

I, too, was naive in 04; I'm pessimistic now -- I think something bad is going to happen, and I think the racist factor will be worse than we expect.

There's a guy here on Boston radio named Jay Severin (one of the wingnuttiest people I've ever heard, but he's rather entertaining) who spent 30 years or so as a GOP campaign consultant. Long story short, he says the acid test of poll reliability is whether or not a campaign is willing to pay the given pollster for its results. He claims Gallup is garbage (apparently they don't do paid polling for campaigns?). I forget who is supposed to be reliable.

I think the racist factor will be worse than we expect.

I believe this is possible. Additionally, there's a bit of xenophobia in there--Barack Obama is a decidedly non-Anglo Irish name. And folks making decisions based on those factors are less likely to accurately respond to polls.

But, chin up. McCain is doing everything he can to make sure Obama wins. At this point, only "The Bradley Effect" can keep Obama out of the Oval Office.

However, a Bradley effect of 8 points on a 10 point poll means that Obama still takes it by 2 points. I expect it may be close, but Obama is gonna take it.

I'm betting McCain takes Florida, though. Obama takes PA, VA, MI. And of course New York and California. And enough of the other states to win.

If you honestly "don't put too much stock in polls", why not promise to write no more columns on polls this year?

If half the money spent on polls was instead given to children's hospitals, our country would be 100% better off. Newly registered voters are likely to make a difference in many states and they aren't sampled by the polls. And look at the wildly different poll answers on the bailout when small wording changes are made.

I remember on election day 04, Zogby said JK would lose the pop vote but win FL and OH.

If you look at the '04 polling, Bush was ahead in nearly all the likely voter screens. However, most of the late, registered-voter screens had Kerry slightly ahead. Also, Kerry was generally polling ahead in Ohio, although Florida polling was all over the place.

In hindsight, the likely voter screens in the closing weeks were likely correct -- we kept telling ourselves that they were oversampling Republicans, that the Dems had this massive turnout operation, etc. And even allowing a Bush edge in the PV, most of us leaned heavily on good polling in Ohio and Florida.

That's why on election day, the Conventional Wisdom (among most pundits) -- i.e. most predictions, even from many pollsters -- was that Bush would win the popular vote by a point or two but that Kerry would win the electoral vote by winning Ohio.

a Bradley effect of 8 points on a 10 point poll means that Obama still takes it by 2 points

Yes, but what's the rationale for assuming a 8-point "Bradley effect"? There were five black candidates for major statewide office in 2006 - Michael Steele, Harold Ford, Ken Blackwell, Lynn Swann, and Deval Patrick. There was no notable deviation from the final polls. In Ford's case - and Ford was the guy attacked with that "Harold, call me!" Playboy playmate ad - he outperformed the final polls by 2-3 points.

To be fair, today's obvious disparities in the tracking polls are very much the exception, not the rule.

For a week until yesterday, for example, Rasmussen, Gallup and Diageo/Hotline were almost in exact agreement, showing Obama leads that were within at most two points from each other every day. For the nine days before that, none of them ever showed an Obama lead of more than four points more or less than any of the others. (The Research 2000/Kos poll is another matter - you have to subtract about three points from Obama's lead in that one to get it in line with the other polls.)

So yes, obviously they are all out of whack today, as they were yesterday already when Obama's lead in the Hotline poll crashed. And it's easy to point and wonder, on a day like this, about just how random these poll numbers all seem to be. But on a day to day basis, those three tracking polls have shown pretty reliable trends, roughly consistent with each other.

In fact, we havent seen the numbers for these three polls diverge this much at any time in the past month and a half. So it's a bit unfair to now pounce on today's numbers and use them as a showcase about the unreliability of polls. Obviously one or two of them just happened to stumble on an unrepresentative sample for the day. Check them again in three days' time and they're most likely back in tune with each other.

In the meantime, we're just left to wonder: who's got the unrepresentative sample? Gallup or Hotline?

Yes, but what's the rationale for assuming a 8-point "Bradley effect"?

No rationale. Just saying even if there is a "Bradley Effect", it's unlikely to make a difference, especially if Obama is polling 10 points ahead. Or 8. Or 6.

One thing that probably hasn't been studied sufficiently is whether there is a sort of generalized Bradley affect: i.e., a divergence from the polls best explained by the unwillingness of some poll respondents to truthfully tell who they plan to vote for. The thing is, such a divergence wouldn't occur only when one of the candidates is black, but whenever there was a lot of pressure from one of the concentric circles that make up a voter's community. That could be the evening news, or it could be the guys at the bar or the folks at the church.

There have already been signs, strictly anecdotal, that some white voters who aren't especially enlightened racially (to put it mildly) nonetheless plan to vote for Obama out of fear for their economic prospects. If I were to guess which segment would show large traces of this generalized Bradley effect, it would be among these voters (i.e, white proles who would be derided by their peers as wusses or worse if they copped to voting for the Dem).

Trackers are basically shit, and the Diageo/Hotline tracker is the shittiest, because it has a miniscule sample (~300 RV per day) and thus a high volatility.

Brad DeLong explains.

The Hotline/ Diageo poll also just re-adjusted it's partisan weighting yesterday taking 1% point away from Dems and giving 2% to the Republicans. So Obama didn't crash, the sample just got re-weighted.

I guess I don't understand why the Bradley effect should be significant this cycle. Here's my reasoning: don't racist voters have plenty of reasons to cite for not voting for Obama that are, in their view, socially acceptable? As in: he's a Muslim, or he hangs around with terrorists, or he sat in the pews for 20 years with Wright?

Nobody reasonable views these as valid, mind you, but that's not the point. The point is that while folks are wary of being viewed a racist, they have plenty of other excuses that allow them to mask that racism.

And today Hotline/Diageo is back to O47-M41. It's just a random walk.

From what I recall, a big part of the argument in favor of Kerry back in 2004 was in elections between an incumbent and a challenger, late deciders usually would break for the challenger, and against the incumbent.

So, the thinking went, if Kerry was behind, but by only a couple of points, he could still pull it out.

The other argument was that the Democrats had a really large turnout operation. They had barely lost in 2000, so the thinking went, so they just needed slightly better turnout to win.

I wasn't sure if I should believe either of these theories in 2004, but I really wanted to. In the end, either they were complete baloney or there were other factors pushing things in the other direction.

Overall, I agree with Erza's post. As much as I got my hopes up in 2004, the polling consistently had Bush with a small lead. That turned out the be the final result. This time around, its the opposite. Sure, anything can happen. But I'd sure rather have numbers like these than the 2004 numbers.

Finally, to the folks that are convinced that there will be a Bradley effect, I say hogwash. You sound like bitter old men.

The Bradley effect theory in 2008 is for Republicans what the "undecideds break for the incubent" theory was for Democrats in 2004. It sounds plausible, enough to get their hopes up. But in the end, not enough.