A possible response to the "Skewed Polls!" arguments.

Politics & Elections
1

*tl;dr: Read the **bolded *text for summary.
Ok, I think I might get it now.

So, all these “Polls Are Skewed!” folks (e.g. here) predicting a Romney win are pointing to the polls’ internals, specifically that the pollsters’ likely-voter models keep producing samples with an larger-than-usual percentage of (D)-identified voters than (R)-identified ones (despite there seemingly being more Republican enthusiasm in '12 than in '08), AND they show Romney tied or losing almost everywhere despite also giving him a big edge in “independent” voters (as opposed to previous years, when independents were even or broke toward the winner). The poll-skeptics (even the more thoughtful ones) say that these internals make no sense in the context of a top-line of Obama +1 or +2, so there’s really good reason to think that everyone’s likely voter models and weightings actually are producing a bias in Obama’s favor.

In response, the best (and for that matter only) counter-argument I’ve seen (or made) is an argument from authority: when, as people who basically know nothing about how to conduct and weight scientific polls, we’re evaluating the relative strengths of these competing claims, it just seems super-unlikely that almost ALL of the people who do this professionally and have solid track records are wrong, while the only people who HAVE figured out the truth are partisan Republicans with no particular expertise in polling.

I actually find this argument to be quite convincing, but it’s not very satisfying, right? It DOES seem a little weird that (R)-identified voters are coming in kinda light, and it seems a lot weirder that Obama could be ahead overall despite losing decisively among independents; Democrats are really gonna crush the Republicans in base turnout? That’s what we’re expecting?

Anyway, I’ve got a hypothesis that would explain this. Of course, I’m basically trading in the same kind of amateur analysis as the Republican poll-debunkers, but at least in my case I’m butting heads with dilettantes like myself instead of, you know, every decent professional polling firm.

It’s pretty simple. If it’s correct that this is in fact the scenario that the polls are describing, then the safest assumption is that BOTH parts – the supposed Romney lead among independents AND the supposed Obama lead overall – are real. How? A modest percentage of those who, in past elections, have or would have considered themselves Republicans are, in 2012, self-identifying as “Independents.” Or, to be even more succinct: it’s the Tea Party.

Finding up-to-date Tea Party demographic/self-report numbers is surprisingly difficult, but back in 2010, per Gallup, 43% of Tea Party supporters considered themselves Independents (and 8% called themselves Democrats), even though in my opinion it’s safe to assume that the Tea Party-Independents mostly represented traditionally conservative/Republican voters. There’s no doubt that that 51% combined Dem/Ind number has come to down to earth somewhat in the past two years as the Tea Party movement has shrunk and settled into it’s role as a de facto subset of the Republican electorate, but I still get the impression – and I may well be completely off base here – that a large minority of Tea Party folks prefer to self-identify as Independents instead of Republicans, even though they’d sooner shit a hubcap than vote for Obama. Basically, in this particular election, the self-identified “Independent” group is siphoning off numbers from the self-identified “Republican” group in way – or to a degree – that has not been true in past.

The problem with this theory is, of course, a lack of evidence. It’s just a loose hypothesis which depends on my assumption that a meaningful percentage of Tea Party supporters prefer the “Independent” label to the “Republican” one despite being reliably conservative in their voting. And of course I haven’t crunched any numbers to see if the math would shake out.

On the other hand, the hypothesis kills two weird-looking birds with one stone, and I think the whole thing is at least plausible. Anyway, it struck me as kind of a neat idea that I haven’t seen anyone else hint at. Given the lack of evidence I really have no confidence that I’m right. However, in the (unlikely?) event that, when this is all over, the exit polls looks substantially like the surveys that some Republicans are questioning – with Obama losing independents badly but winning the popular vote – I think this idea will stand a decent chance of having explained a part of that result.
Thoughts?

2

The biggest problem with the argument is that Republican registration is up over 2008. Now it could be that many independents and even some Democrats just registered to vote in the primaries, thus creating a big gap between state stats of actual registered party members and polling data.

We’ll find out who is right. After these messages.

3

Well if the hypothesis is correct I certainly wouldn’t expect Republican registration to drop.

The idea is that, post-Tea Party, a chunk of already-Republican voters will have begun to self-identify as “Independents” when, for example, they’re asked about their affiliation by a pollster; this costs them nothing, and is a decision that can be made on the fly. I certainly would not expect anyone from this group to change their official registration, which would cost both time & effort (in the need to fill out the necessary forms) and the ability to vote in Republican primaries (which, it is clear, is an ability Tea Party members value highly, since it allows them to challenge or threaten incumbent Republicans from the right). I’m not saying anything about the proposition that previously-*Independent *voters may have changed their *registration *to vote in Republican primaries.
Anyway, just checked the final version of the Gallup tracking poll, and they actually have Obama winning independents 46-45. I don’t know off hand whether this represents a shift in Obama’s favor, or it’s just that Gallup uses a different methodology to reach those numbers than other firms, or what.

4

The skewed polls argument essentially boils down to an argument from incredulity. ‘There’s no way Democrats will have that large of an advantage over Republicans, therefore the polls are biased!’ The only evidence that has been presented is the registration advantage that Republicans have held over the past four years. Ignore for a moment the fact that this is in aggregate, and is likely to vary greatly by state. Also ignore the same thing for the independent advantage that Romney appears to hold in national tracking polls, which certainly is greater in say Texas than California (or Ohio). Also ignore the huge amount of uncertainty in boiling down the poll results into groups that are approximately 1/3 the size of the entire polled population.

Imagine you were a Republican upset with what you perceived to be a lack of conservatism in the party, such that you are to the right of what you consider to be the party’s positions. You may call yourself an independent, but are you going to go out to change your registration to become a (capital I) Independent? What would be the point? The only outcome would be to potentially miss out on the R primary, as several states have closed primaries. If you are interested you can switch registration to Republican to vote in their competitive primay even if you consider yourself a life-long Democrat. If not, why bother to ever switch your registration, even if as a former Democrat you became a born-again Christian and literally think Obama is the Antichrist? Symbolic reasons? I can guarantee you that this is not the case with the general population.

The point is that voting is more correlated and accurately predicted with party self-identification rather than party registration. I have seen no evidence to the contrary, and wouldn’t make any prima facia logical sense. However, this is what pollsters measure, self-identification. They do not weight or adjust for it, but only report the self-identification of the population they are polling, which is further adjusted for demographic factors. The skewed polls argument against this is that you need to weigh by party registration to more accurately reflect the registration of the voting population, but polls do not generally report this result. All the D+7 splits or whatever that you see are based on self-identification, not registration.

Anyway, we are likely to know by tonight or tomorrow, so one side will be eating crow no matter what.

5

I have seen this argument presented at other places pretty much exactly as you describe it, VarlosZ. Usually it’s accompanied by a graph of polls about party ID over time, like this one from September 2008 to the present. It certainly does seem like the line for “Independent” and the line for “Republican” has been inversely correlated, at least for the last couple years or so.