New Pew poll: Obama wins by 10 if Democrats lead in turnout this year by, um, 19 points
posted at 9:57 pm on August 2, 2012 by Allahpundit
No need to boot up your computer’s calculator. That’s 43.5D/24.5R/32I, or D+19. The 2008 presidential exit poll, in a big blue year, was D+7. And yet, even here, we find this:
Nate Silver notes that Pew has been polling high for Obama all year and that registered voters tend to skew Democratic vis-a-vis likely voters. If you adjust for those effects, he thinks Obama’s actual lead here is around four points. Fair enough; that’s all I’ll give you from Pew. If you want more, hit the link. Instead, let me play contrarian by quoting from RCP expert Sean Trende’s latest piece playing off yesterday’s questionable Quinnipiac numbers. Hugh Hewitt did a superb job grilling Quinnipiac’s pollster about their sample today, but Trende argues that worrying about the partisan composition of samples is way overblown:
I say this, in part, because we’ve been having this debate for a very long time, and it usually goes nowhere. In 2004, re-weighting polls to reflect the 2000 exit polls was all the rage among Democratic bloggers. The argument went that Republicans hadn’t had parity with Democrats in polling in a very long time, so we should ignore polls showing Republicans even with Democrats, or perhaps even ahead of Democrats in terms of ID. Of course, the final exits showed a tie between the parties, as Republicans managed to turn out their base at “supercharged” levels.
Since then, the same thing has occurred in every election: The losing side objects to the partisan composition of polling. The polls then proceed to get the final result roughly correct…
In all three states polled, the RCP Averages (Florida, Ohio, Pennsylvania) include at least one poll of likely voters from a nonpartisan source that is roughly consistent with the CBS/NYT/Quinnipiac result. They also include polls that are not consistent with the CBS/NYT/Quinnipiac result. Overall it looks like CBS/NYT/Quinnipiac’s system places it on the more pro-Democratic side of the “house effect scale,” but not outrageously so.
That Quinnipiac poll suddenly becomes very worrisome if a guy as sharp as Trende thinks there might be something to it. Similarly, read the e-mail Bill Kristol got from a friend “with an excellent track record of reading election trends.” He makes a simple point: While it’s true Romney’s even with Obama in the national polls, it’s not true that he’s even with him in the far more important swing-state polls. (Silver has a chart illustrating this, in fact.) His theory is that the Bain attacks are working — not nationally, where comparatively few people are seeing them, but in areas that are being bombarded with tens of millions of dollars in ads. Perhaps not coincidentally, just today Romney hired a PR person for the exclusive purpose of answering the Bain critiques. That’s not to say Romney can’t make up the difference — conservatives will be carpet-bombing the swing states with ads soon enough — but the idea that he might be behind right now by a small but significant margin where it counts isn’t necessarily liberal media bias at work.
Update: Jay Cost of the Weekly Standard e-mails to reassure me:
It’s one thing to give pollsters a wide berth, but D+8 in Ohio [in the Quinnipiac poll] is absurd. My general rule of thumb is that the historical average over the last 25 years is D+3.5. The best for the GOP was in 2004 (when it was even) and the worst was 2008 (when it was D+7). Anything over D+7 is just not defensible.
And one of the points, too, is that there is a difference between methodological soundness and inference. You can dot all the I’s and cross all the T’s as a pollster, but if you get a D+9 sample in Ohio, I’m sorry, that does not lend itself to a valid inference about November.
In other words, you don’t have to go so far as to argue that pollsters should weight by party ID, but you should be aware of what the spreads are and be very skeptical of anybody going as far as Pew (!) or even those CBS/NYT/Q polls are.