Well reality does seem to have a liberal bias …
In other words, everybody else is wrong because they’re not willing to throw out data that contradicts with their preset expectations.
This is a classic example of why nobody here takes your arguments seriously.
Well, since I’m here early, here is more of what I mean. If you click the crosstabs at the bottom of the page and look at Florida, you’ll find it has a split among likely voters of 36D/27R/33I, which would quite possibly be the largest advantage for Democrats in the history of Florida and would simply blow away Democratic turnout in 2008.
My point was that OMG’s version is different from the principal version being purveyed by the likes of NRO.
So what is the cause of the error OMG? If the poll is randomly sampling, what is the mechanism by which it is skewing reality?
It doesn’t necessarily mean Democratic turnout is up- it may just mean Democratic identification is up (and/or Republican and independent turnout is down). It could be the exact same voters in 2008, just a handful of the independents now call themselves Democrats, and a handful of the Republicans call themselves independent.
Also, good job using the correct adjective (“Democratic” versus “Democrat”).
Basically you can’t just throw out data because you don’t like it. The data may be wrong- but you need a better reason than “it’s significantly different from my expectations”.
Another thing that can be looked at, even if the voter splits data is wrong- trends. And the trends have been bad for Romney for weeks- pretty much all the polls have been showing improvements towards Obama’s support.
So whatever the actual voter splits are, over the last several weeks, they’ve been moving towards Obama.
And check this out. If the poll numbers don’t agree with your view of reality, just adjust them! Automagically, Romney is up by 7 points nationwide! It’s so obvious!
Of course this unskewedpolls.com website is totally nonpartisan and absolutely impartial. Nothing but the facts when we unskew these liberal lying polls! Look at our favorite websites … Laura Ingraham, Thomas Sowell, Rush Limbaugh, Dick Morris … impartial Americans all!
If there is any methodological skewing going on, I’m thinking racial weighting is the most likely culprit. In this FL poll, they only got ~8% responses for black and hispanic voters which they then weighted up to 13% each (to match known demographics). This could, potentially, skew the part ID a few percent (and the top-line) if in fact black and hispanic voting is way down from 2008.
So really it comes down to whether you think minority voters will stay the same, increase, or decrease from 2008. Simple demographics would tell you that it should at least stay the same, but some pollsters (and the GOP in general) seem to be hoping that it drops, based on 2010 results.
We have to be careful not to conflate two separate issues: demographic weighing and turnout models. We know that turnout models aren’t the culprit, because the registered voter samples have basically tracked the same swing toward Obama. So the problem would have to be with the demographic weighing before the likely voter model is applied. But this adjustment isn’t based on 2008 turnout. It’s based on census numbers. To think that adjustment is wrong you have to disbelieve the census numbers, right?
Actually, looking at it, they seem to have done the demographic weighing after applying their likely voter screen. That does not seem kosher.
The unweighted frequency is 34.9 D, 28.8 R, 32.4 I, 4.0 Other. That’s not really that different then the likely voter split.
Go right ahead. 
Here’s Gallup’s national numbers on party ID, 1988-2011. They bounce around a bit, don’t they?
And that’s national. Smaller subgroups of the population (e.g. states) tend to bounce around more with respect to practically any statistic under the sun. So if it’s not unusual for the national party ID numbers for a party to change 3-4% from one year to the next, it’s even more plausible that they’d do that in Florida.
No. First of all, I give polls from any reputable pollster the benefit of the doubt, modulo house effects. (E.g. PPP has a noticeable Dem house effect this year (they didn’t in 2008), and Ras has always had a noticeable GOP lean.) Second, I’d expect a poll to control for demographics - race, sex, age, immutable characteristics like that. The only thing I’d use party ID for is if a poll seemed to be way out of line with other polls, I’d look at its party ID numbers to see if they were way out of whack. But 4% off of some aging benchmark? That’s a pretty slight difference in the world of polling. I’d shrug and move on.
But yes, polls confirming other polls are important. There is such a thing as sampling error, and there’s definitely such a thing as nonsampling error. If a poll is the only one to be way out of the pack in some direction, I’d question it even if - particularly if - it supported my biases. For instance, NBC/Marist had a bunch of VERY pro-Obama state-level polls earlier this year. My Too-Good-To-Be-True-ometer pinged quite a bit over them, and indeed, nobody else got similar numbers, and eventually they didn’t either.
Why is it unlikely? And the next poll that someone does can check it. Most all of them ask about party ID.
I’m not ‘people.’ I’m me. Try to keep up.
No, that’s your CLAIM. We’re debating its truth, so we’re looking at evidence to test your claim. Saying you only want to concern yourself with your claim, without considering the evidence, is bullshit.
Gallup said that in 2011, only 27% of Americans identified as Republican. Given how things have gone since last year, that number might’ve even dropped a tad.
And those tools are probably carefully guarded trade secrets.
One of my neighbors years ago was a top officer in the AC Nielson NY office. (HQ was in Chicago). He once explained in general terms how they chose the households they did and how persistent they were in getting them signed up. If they couldn’t get the house they wanted, they’d take three or four neighboring houses as a substitute. It was most definitely not random sampling in the usual sense.
It’s easy to see why there could be a systemic error if you only polled people who answer the phone on the first call. But what would be the source of systemic error if you randomly select a house and if persistent attempts fail, you target the neighboring house? There’s have to be something really weird going on for that to lead to a systemic partisan skew, I would think.
I’m not sure how kosher it is or isn’t. But what seems to be happening (at least in this one poll) is that they are having a hard time getting black and hispanic likely voters to answer the phone and respond to the poll. So you either end up with a sample that has demographic turnouts of 9% black and 8% hispanic or you weight them up to the historical numbers.
At least a couple of the very best polls for Romney have shown these types of low minority response rates (and also, often, a very old sample too).
That decision really comes down to whether you think minorities and young people will vote or not. And, of course, in that light the efforts to increase the difficulty of voting (either early or in person) make perfect sense…
What still makes no sense about this as a conspiracy is why every single polling company (except, I guess, Gravis and maybe Rasmussen) is willing to intentionally damage their reputation for accuracy just to get this president re-elected.
I would think the correct approach would be something like this:
Call 2000 people. 1500 are RV, and let’s say 150 are African-Americans. You ask your likely voter questions, and 1200 are LV, of which 100 were African-Americans.
Then you weight the RV sample by census demographics, increasing the RV for African-Americans to, say, 175. Since you know that 100/150 was the likely voter percentage based on your sampling, you apply that same ratio to 175 and get 116 as the weighted A-A LV.
Wouldn’t that be a more objective approach than assuming 2008 turnout?
You’re being unfair - it’s not magic.
Their methods for unskewing are…secret. Totally different thing.
Anyway, it shows there’s clearly a lot of money to be made betting on Romney at Intrade.
Not right. You have to believe that the methodology used to correct the demographics to the identified true demographic standard is faulty, or that the data coming in somehow systematically undersamples a portion of a demographic group that would lean a particular way but then uses that subgroup as representative of the whole demographic group.
A hypothesis for how that has occurred for Rasmussen and Gallup has been cited based on what is known about their methodologies; none has been offered for why everyone else would be getting it wrong.
(Also for the FL poll one can go into the demographic crosstabs: the party ID discrepancy is mostly there in the unweighted data as well 35D 29R)