Today’s Washington Post poll results do not look good for Miit:
And the WaPo did not oversample Dems:
We’ll know soon enough. Either Romney will win and prove that the polls were false and slanted in favor of Obama, or Obama will win because the polls were false and slanted in his favor.
Serious question, here, from someone who doesn’t really understand the basic issues at play: the accusation is that the polls are weighted using 2008 demos. Is this true? Is it a proper thing to do?
What’s wrong, in principle, to what unweightedpolls.com is doing (besides adding in Republican voters based on Rassmusen; that I understand)?
No, it isn’t true. The polls ask whether you identify as a Democrat or Republican or Independent. As the political winds shift, so does this self-identification. None of the professional polls weight the sample toward political ID based on any exit poll. They do compare their sample with the exit polls to get a sense of whether they are doing something willdy wrong.
The above should answer that. But what’s wrong is that the polls aren’t weighted to begin with. They are reporting what they are learning based on random sampling. And the random sampling suggests that at this snapshot in time an unusually high number of registered voters is self-identifying as Democratic-voting.
Actually I believe Rasmussen does weight their national tracking poll by party ID: Methodology - Rasmussen Reports®
Most other polling firms do not but rather simply report what the respondents told them.
Yes, I did not mean to suggest that Rasmussen is not a professional poll. But it is a special case, since they make no attempt to call cell phone users they necessarily have to massage their party data to partially account for that significant gap.
I think there is also some trivial number of other professional pollsters that weight by party. ABC news used to do so, I think. But it is definitely the distinct minority based on the reporting I’ve seen.
Right. Because numerous studies have shown that party ID is extremely malleable, especially for swing voters (really the only kind that matters in this type of polling). They will happily fly between calling themselves Democrats, Republicans, or Independents (although more typically they go from Dem->Ind->Dem or GOP->Ind->GOP).
Currently you have a lot of folks that decided they were independents in 2010 admitting they’re Democrats again and a lot of conservatives that don’t like Romney calling themselves independent.
It’s also, I think, a fatal flaw to place any weight on mid-term party percentages when analyzing presidential polling - the electorates are just not comparable.
It’s worth reflecting on how deeply crazy and yet deeply widespread this theory is.
This is all over the conservative mediasphere. National Review, American Spectator, Hugh Hewitt, Pajamas Media, Rush Limbaugh…it’s hard to find a bastion of conservative viewpoints that doesn’t appear to subscribe to this conspiracy theory.
And it’s so self-evidently stupid. First, even if you’re too blindered to consider facts, to buy this conspiracy you have to believe thatFox News is in on the liberal media cabal. Moreover, most of the major media polls are operated by a bipartisan team including one Republican pollster. Why are they in on the conspiracy? That entirely apart from the reality that the factual premise is just demonstrably wrong. The majority of pollsters don’t weight by party ID.
I think this really shows how far down the rabbit hole the GOP has gone. Even if some of those sites are just parroting the theory because they think it’s good politics, it shows what they think of their own readership–which is not much.
There have, of course, been Democratic conspiracy theories involving rigged voting machines and such. But have any of them been anywhere as widespread and self-evidently false as this?
I think the R’s are just prepping their stolen election theories: “How could Obama have won? All the polls were for Romeny! Voter fraud!”
No it isn’t true but it is not completely based on random sampling: it is based on random sampling then adjust to attempt to reflect the current demographics (not voter ID but income, urban vs rural, racial ID) of the RV pool. The tools to make those corrections vary between houses.
I don’t think the argument is quite as bad as you make out. I think they’re asserting that the party ID ratios reported by the non-Rasmussen polls reveal that those polling firms have skewed sampling practices. That is, the conservative complaint isn’t that the polling firms are weighting for party ID, but that the party ID ratios reported by those polls indicate that they should be.
This is almost certainly still incorrect, but not quite as obviously so as you’re portraying it.
It’s probably just polling bias.
You haven’t trotted out any facts thus far, except to cry foul.
Not really. “Party identification” is generally susceptible to question wording, which is why I’ve never trotted out a poll as evidence of party identification (Gallup, Rasmussen or otherwise, and I dare you to prove me wrong) as it can vary wildly from pollster to pollster and even from individual pollster’s polls. What I’ve said is that I’m more inclined to believe Gallup/Rasmussen as they’re trends more closely match individual state trends. That’s why I focus on-- as I have been-- a state’s actual party voter registration data.
Oh, sure. If you want to throw in voters who identify with one party but don’t bother to vote. But since they don’t vote…
Anyway, we’re right back to what I said before. Your claim is that your polls (and only the ones which you agree with) are right and state registration data is wrong and outdated. Again, that’s a tall claim; one which you cannot substantiate outside of “I want it to be true”.
Now I happen to live in Florida, so I’ll just focus on that (as it would be exhausing to go through each state individually). Florida is actually pretty good at tracking voter registration. That site updates every couple of days, with the last update being on 9/24/2012. It currently has the partisan split at 39.95D/35.96R/24.08 or more or less a D+4 split, which is about normal for the state. Take this poll conducted on September 23rd; it has a D+8 sample (44D/36R/20I), more than double the Democrat’s actual advantage in the state according to one-day old voter registration statistics. How is that not a blatant case of oversampling Democrats?
(I’d do the same for the WaPo poll Typo Knig linked to, but I’ve gone through it five times and cannot find where they posted their splits in poll. Maybe I’m just blind.)
You can’t simply discount voter registration data as a “lagging indicator” unless you’re going to argue that a large amount of registrations were in the mail at the time PPP conducted their survey or that many people are simply not choosing to update their voter information (in this case Republicans) and are simply choosing to vote for the “other” guy (which you would think would be reflected by a large percentage of Republicans voting D).
Yes, I do. Now care to answer my question?
I wasn’t even looking at Gallup (or Rasmussen for that fact). I simply looked at state registration data and compared that to the splits in any given poll, which you summarily dismissed.
Oh, before I forget, that’s not even taking into account polls which have a likely voter sample of D+7 or more, which is absurd beyond absurd.
First of all, you’ve got no evidence that Dems are overrepresented in the PPP poll. As has been pointed out, party ID is fluid, and on account of that, party registration statistics are a lagging indicator: if I go back and forth between thinking of myself as a Dem or as an independent, the state files where my party registration is kept don’t receive a text message from my brain each time I change how I think of myself.
So it’s quite possible that the registration breakdown accurately reflects what the voters identified as at the time they registered, but the PPP poll accurately reflects how Florida voters regard themselves now.
Second, you will notice that the poll has a 3.3% margin of error. What it’s saying is that there’s a 95% likelihood that if you polled everyone in Florida about their party preferences, the % of Dems would be between 40.7 and 47.3.
Which reduces your claim of blatant oversampling to a claim that no way, nohow, could the Florida voters who identified themselves as Dem have changed by 0.75% between the time they registered, and the time this poll was taken.
That’s not a claim you really want to defend.
OMG,
Let me try this again.
You have claimed that the polls are wrong because they are adjusting the polls to fit a turnout based on 2008. That is factually incorrect and cites discussing in detail where and when in the process the party ID number comes from in the polling process have been provided.
You have claimed that the polls are wrong because their party ID numbers are different than the registration rolls for party ID. You are free to believe they measure the same or even similar things; you are also free to believe in the tooth fairy. Have fun.
Party ID in polls is again a fluid item. Will it reflect on election day? Are those who change their answer to a question about self identification more likely to stay home? Or are they the swing voters? I don’t know. The pollsters likely voter screen is supposed to predict that but we’ll see how accurate it is.
But as to the issue about expecting a 2008 turn out vs a more traditional one … 538 had an interesting take on that before the GOP convention:
Oh the answer to your question is obvious to anyone who looks at the polling list and sees that the unreported in the graphic “undecided” column varies from 2 to 18% just in the last several weeks depending on the firm. In some polls those same people seem to be mostly captured as answering “independent.”
Forgot this one:
What do you mean? Do Floridians have to update their registration every year, or something?
In Maryland where I live, you register once, and it stays current as long as you keep voting, and keep living at the same place. The only reason I’d need to change my party affiliation on my registration is if I decided I wanted to vote in a different party’s primaries.
What do Republicans have to do with it? In your example, the % of self-identified Republicans in the poll is equal to that in the registration statistics, 36% in both cases. We’re talking strictly about movement from independent to Democrat, to the extent that we’re talking about movement at all, as distinct from normal sampling error.
And again, what reason do Florida voters have to update their registration, as long as they can still go and vote? None since primary season. And even then, since we’re talking about D/I slippage, was there a seriously contested Presidential or Senatorial primary on the Dem side this year? Negatory. There might’ve been such a primary on the Dem side in a few of Florida’s 27 Congressional districts, but if there was, that would only apply to a small fraction of the state’s voters, and you’d have to prove it by me anyway.
Barring that, Florida voters who go back and forth between thinking of themselves as Dem or independent have had no reason to update their party affiliation since the spring of 2010.
It’s safe to say that attitudes towards the Democratic Party are a bit better in Florida now than they have been during most of that period. So it’s quite possible that a few percent of Florida voters now think of themselves as Dems when they wouldn’t have affiliated with either side when they registered.
…So I’ll just let this one slide.
Now you guys are just making excuses. Pray tell, at what split would you begin to question whether or not a poll was adequately representing the partisan split of a state? 55 - 45? 60 - 40? 70 - 30? 99 - 1? Or are all partisan splits in a poll valid on the basis that “party identification is fluid and party registration is a lagging indicator which just hasn’t caughe up yet”? I highly doubt it. Party identification is not nearly as fluid as you are painting it. The only reason you are arguing this is because you want to assume that the polls you want to use are adequate representations of any given electorate.
Here is, essentially, what it boils down. State level data says X, Poll A says X and Poll Z says B. I say Poll A is probably correct because it mimics state level data; you say poll B is right because there are other polls like it. I find that a bit odd as, if the tables were turned, you’d be crying foul.
Highly unlikely. You do realize this is (not so) easily checkable, correct?
Interesting how people are now concerned about the MOE where all I’ve been seeing on this board is “Hey, look! Obama is leading in this poll [even though its well within the MOE]! Obama 2012! BLOWOUT!”
No, we’re only concerning ourselves with the topline, which is what’s important, in this case the percentage by which some polling firms are oversampling Democrats.
No, I claimed that some polls are oversampling Democrats (which conversely deflating GOP turnout). Have you even looked at the internals of some polls? Some polls have GOP turnout this election as less than 30% of all votes cast.
Except it’s not correct. Scroll through almost any poll of the last few weeks and you’ll notice that their numbers mimic the 2008 electorate almost to a tee (with the GOP fairing worse) with likely voter samples being far more than occurring historically. Do you seriously need proof of this?
No, what I said was, and I quote:
You’re the one who wanted to argue that Gallup and Rasmussen are wrong because other polls not named Gallup or Rasmussen agree with each other. My response was in response to that.
I disagree that this is the argument being put forward on those sites. Can you link to a major conservative website stating it that way? It doesn’t even make internal sense that way; if you concede that this is the result achieved by random sampling (weighted by objective demographics, naturally), what theory would counsel for discounting that result? In other words, in that theory, where is the error being introduced?
I didn’t say it was a good argument, just that it wasn’t quite as blatantly false as you were portraying. And it’s being made not three posts up in this very thread.
If you take the voter registration statistics as gospel like OMG is doing, then if your poll is showing a different partisan split it must be sampling wrong. Where’s the error being introduced? Who knows, but it’s clearly there since the polling results contradict the registration data.
I’m in complete agreement with the arguments being made against this position in this thread, but once you conflate party ID and registration everything else follows quite naturally.
It is a strawman, but I’ll respond anyway. Speaking for myself, if Obama were behind in the polling, I might have a lot to say about it, but I really don’t think I’d be insisting all of the polls were wrong, and just wait and see! (Maybe the obviously biased/flawed ones, like Rasmussen, but on a case by case basis, not a sweeping insistence that the polls are at odds with reality.)
I think conservatives are really arguing with reality here, and we’ll see in about 6 more weeks.