Oh no, I fear you’re going to be upset about more than just the Presidential results.
Do you actually think that the pollsters have method of calling a certain proportion of ethnic minorities when they do their polling?
Here’s a question: if Romney wins, the conservatives on this board (and elsewhere) will feel a certain amount of justified vindication. However, what if (as seems likely to me) he wins by a small margin in various states/nationally, and nowhere near the landslide that many are predicting? What if these margins are well within what the polls say? IOW, what if the polls were reasonably accurate? Will that factor into their thinking or expressions of joy at all?
The ARG poll is actually D+6, and the Rasmussen daily tracking poll, which I assume is the one you’re referring to, is actually D+2, with independents breaking for Romney at either 9% or 11% (weirdly, the linked article says both). The Monmout poll matches what you say.
And yes, the correlation between high percentages of independents in the Romney column and a high D-R gap tends to hold, albeit with the fair amount of scatter ome would predict when looking at smaller samples.
Grab a screenshot. Play it back for him on Wednesday morning. 
So surprise surprise, day before the election Gallup and Rasmussen both converge on 49/48 +/- 2 (i.e. statistical tie).
Thread win.
It is not a statistical tie, it is a statistical 5 point lead for Obama.
RCP has Ohio at +3 Obama right now. Has any state poll in the modern era ever called it for the wrong candidate when that candidate led by 3 in the average?
Not when there have been three or more firms polling the state.
I just want to make sure I understand, because I’ve been having trouble following it.
The OMG/Republican argument is that the “top-line” polls are wrong because when you look at the internals they show “too many” Democrats, AND they show independents voting for Romney.
Therefore they’re “over-sampling” Democrats, and the fact that they show independents voting for Romney means Romney’s going to win.
That’s the argument, right?
And the response is: “Nobody is over- or under- sampling anything. All that’s happening is some I’s are calling themselves D’s, and some R’s are calling themselves I’s.”
In other words, a bandwagon effect.
And the OMG/Adahar counter-response is…?
**OMG **has never been too clear on what he thinks the mechanism is, but **adaher **and others point to the demographic turnout projections (i.e., percentage of white voters, etc.) that pollsters use to weight their samples to account for the fact that some groups are easier to reach than others.
The rebuttal is that these demographics are (1) relatively stable over time; (2) not universally used to weight polls, meaning it doesn’t really systemically explain the results.
I think the counter-rebuttal is that even a stable indicator can change unexpectedly (true but not especially predictive), and that this could explain enough of the polls being off to put the real margin at only slightly Obama.
FWIW, Sabato’s Crystal Ball has just released its election prognostication: Obama 290/Romney 248
What do you guys make of that assessment?
I can see a lot of that scenario playing out; really, the only thing I have a problem with is that they’ve ceded Virginia to Romney. Based on Silver’s model, I think it’s enormously likely that VA will go to Obama.
Is there any evidence ever that “internals” of a poll are a better forecasting tool than the top line?
I agree that it’d look weird on the map, and make zero sense politically. But imagine the reponse from a stereotypical Texan. 
Just as a matter of math, the margin of error for sub-groups is higher.
Because sampling error are randomly distributed around the mean (that is, you’re equally likely to be off on the high side as on the low side), the sampling errors in multiple sub-groups tend to cancel themselves out a bit for the topline number. A poll might oversample R-leaning independents just as a matter of chance, but the same poll is equally likely to oversample D-leaning Latinos, say. That’s why it’s not really proper to say the topline is only the result of some outlier result in a sub-group.
But that’s not the real sin of analysis here.
The error they are making is not so much in looking at the sub-samples, but in removing them from their context. With independents, they want to look at the 2012 sub-samples and plug them into the 2008 polls. But that doesn’t make any sense, since identifying as an independent is a fluid thing, and the current data says that independents are slightly more right-leaning than 2008 because Republican identification is lower.
Actually there are systematic House effects across pollsters. That’s why Nate is saying Romney has a 12-15% chance rather than much lower.
Discussion here. You can see that Rasmussen bends Republican, while Survey USA bends Democrat. Another Look at Survey Bias – VOTAMATIC
OMG et al are essentially saying that they have a better take on reality than polling professionals. I say, you weight the Houses based on their past performance - which is what Nate does.
Another concern is that pollsters, not wanting to get burnt, might be weighting their ending polls to be consistent with the other pollsters. The effect is called “Herding” and it could be statistically devastating. Pollsters May Be Herding – VOTAMATIC
At any rate, after the detailed election returns are in, I will look forward to the discussion about House effects and the like. I doubt whether Party affiliation, which fluctuates routinely, will provide much insight even if Romney wins by 10 EV. But this matter should be susceptible to investigation.
Thanks. I guess I should refine my question. Has there ever been an election where something that would routinely get reported as an “internal” of a larger poll was actually the better indicator, across polls, than the top line number?
Measure for Measure, I don’t think house effects involve the same kind of bias as the skewed poll theory.
House effects are the consistent departure of a polling firm’s results from the means for that race. You could also calculate them based on departure from the election results in the previous cycle, but I don’t think that’s what Nate does (and in any event, would still be distinct from the skewed polls theory).
By contrast, the skewed polls theory is that the particular turnout model used by most pollsters for this election is wrong. This is not captured by either method of calculating house effects.
It’s a good question. I doubt it’s easily answerable, since as far as I know no one has built a database aggregating the internals of polls.