I don’t understand the first possibility. The “method” in question would be to multiply the sample by the known demographic proportion, no? Not sure what could go wrong there.
And the second possibility isn’t an error in the adjustment, it’s an error in the sample. Obviously, there’s always the metaphysical possibility that the sample isn’t capturing a representative sample, but that’s a different problem from the one being discussed in the post you quoted.
Perhaps. I can certainly see arguments either way. I’d be worried about applying the likely voter screen to such a small sub-sample, particularly if it resulted in a poll with what seemed way too low of a minority turnout (say, below 2004 levels).
I guess maybe it depends on which you think is more likely: a large decrease in minority turnout or a large change in voter party self-identification. Historically I think party ID has proved to be more variable.
When was the last time a Presidential candidate won the popular vote while losing the independent vote?
That’s another thing that’s weird about the polling. Jay Cost also pointed out that the media polls are the ones that seem to be oversampling Democrats, while the non-media pollsters are getting different results. Is the media creating their own house effect?
Fair enough. I would think both things quite likely–i.e., that the black turnout would be lower than 2008 and that more voters would be self-identifying as Democrats.
But it seems to me that if most polls are adjusting demographic numbers after the likely voter screen based on 2008 turnout numbers, then there could be a kernel of truth to the conservative critique, just on completely different grounds than those currently expressed.
My understanding was that most polls use census numbers to make demographic adjustments before doing the likely voter screen. But it’s hard to check that since very few polls report enough about their internal process to say.
Though evidently FL 2004 was 12% black turnout based on the exit polls, so maybe my assumption is wrong.
It is not so simple. Again, here is an article detailing the methods. You can ask and define race and apply the filters in ways that unintentionally skew the data. The result is that Gallup underweights for Blacks and Hispanics relative to the rest (according to that detailed look).
Also likely voter screen is not based on making the demographics match 2008.
The second error is still a methodology error. For example, what if you have an easier time reaching rural middle aged Blacks with landlines than urban young Blacks with cell phones? Is it necessarily accurate to overweight the first to make up for not reaching the second to fit the Black demographic number?
I, for one, am relieved that Romney officially no longer trusts external polls, preferring to rely on data generated by his campaign. 
Ah, I understand both points much better now. Thanks for taking the time to explain.
Republicans are solidly invested in the myth that they are the majority party, that most American are aligned with their views. Any evidence to the contrary must be explained away. Like, for instance, losing.
I can certainly see why conservatives think there might be oversampling going on. On the other hand, even the most Romney-friendly polls don’t seem to bring him within the margin of error.
Rush’s head is asploding today about the recent polls that give Obama leads in crucial states.
Rush said it’s the fault of mainstream media, who is/are in the bag of Obama, and who just want to make conservatives depressed and not turn out to vote that are posting these wrong poll results.
His proof is that if Obama really did have a healthy lead in Ohio, then Obama would not be campaigning there today! So there.
Well, with the daily Gallup showing O+6 and at 50% he certainly needs something to hold on…
This from Mark Blumenthal today:
This sure doesn’t seem like what Quinnipiac was doing based on the internals. Is he right?
Hmm…
I don’t see how we could confirm or deny since they didn’t release any information about their total sample, only the likely voters part. It’s certainly possible, but it does seem kind of suspicious that their weighted results ended up matching the 2008 exit polls.
I caught a little of his act today too. My favorite part is how the polling companies are all purposely skewing the polls toward Obama, so that when Romney wins the rioting can commence.
I’m not sure, what sort of stake pollsters have in riots, but there you go.
Why would pollsters deliberately set themselves up to be wildly wrong when the election takes place? What are they supposed to gain from that?
Not pollsters…
…
…liberals.
The ability to riot, apparently. I don’t think OMGABC is saying they are deliberately skewing things, though, just Limburger.
Like the GOP DC staffers did in Miami? :rolleyes:
Seems like a good time for the Kaiser Chiefs’ “I Predict A Riot.”