Poll analysis: No, the hurricane didn’t beat Romney

I know you’re bored with post-election number-crunching but (a) ruling out silly force-majeure explanations makes the “how can we reform the GOP?” debate easier and (b) I wrote on this last week and feel obliged to follow up. My working theory then was that even low-information voters aren’t easily impressed enough to let the glorified presidential photo op that is “storm crisis management” swing a national election. And the exit poll data seemed to back that up. Notwithstanding the alarming 15% chunk of voters who said O’s hurricane response was the “most important factor” in their vote, it turned out that very late deciders broke only a bit more heavily for O than earlier deciders did. The hurricane may have padded his margin slightly but it didn’t swing the outcome.

Now here’s Sam Wang, one of the statistical modelers who called the election beforehand, “unskewing” the national polls by noting that O outperformed their predictions by roughly 2.4 points on average on election day. If you add that 2.4 points to the national poll averages in October, then you realize a few things:

– President Obama led national opinion on every single day of the final two months of the campaign.

– During this period, the only event to meaningfully move national opinion was Debate #1, which led Mitt Romney to close two-thirds of a 6-point gap between him and President Obama – overnight. Some of this gain was reversed in the closing two weeks of the campaign.

– Sandy’s measurable effect on opinion was no more than 1.0%, and even this might have reversed by Election Day.

So Bickers and Berry’s claim that “the president clearly benefited from the ‘October surprise’ of Superstorm Sandy” is unsupported by data.

Wang’s graph of the unskewed national polls shows a small blip towards O over the last few days following the storm but then a slight trend back towards Romney, suggesting a small but fleeting Sandy bounce. At the Journal, though, Gerald Seib argues that Obama’s poll uptick in the election’s closing days actually began a few days before Sandy made landfall, not after:

The picture emerges from a look back at the averages of polls conducted in the nine most hotly contested battleground state polls compiled by Realclearpolitics.com. Between Oct. 2 and Oct. 26, Mr. Romney gained ground in all nine of those states, though not enough to take a clear lead in any except North Carolina.

But then, between Oct. 26 and Oct. 30—the day the hurricane hit the East Coast—the Romney move up came to a halt. In that period, his support was flat in polling in four of the battleground states, he actually lost a point in his support in three of them, and gained a single point in two.

In other words, Mr. Romney had hit a plateau just before the hurricane shook up the political system.

Seib’s conclusion didn’t seem to match up with the RCP national poll average that I looked at last week, which showed Obama surging back into the national lead right around October 29-30 — just as Sandy was making landfall on the east coast. That coincidence made me think that maybe the storm did help push O’s numbers upward. But if you look at the national poll averages at HuffPo and TPM, which are based on a slightly different sample of national polls than RCP’s average is, then it makes more sense. Here’s what HuffPo’s looks like:

The first thing worth noting: Unlike RCP’s average, HuffPo’s average never showed Romney in the lead, a result that jibes with Wang’s “unskewed” model. The second thing worth noting is that, while Romney got within two-tenths of a percentage point on October 21, Obama’s lead started expanding the very next day — fully a week before Sandy made landfall. By October 27, two days before landfall, it was up to six-tenths of a percent. TPM’s national-poll average had the race closer but if you zoom in there, you find a similar result, i.e. that Obama passed Romney on October 26 and maintained his lead to the end. Not that any of these national numbers matter; the lesson of the election is that the state polls were more trustworthy. The point is, though, that both state and national seem to show a late break towards O, presumably fueled by undecideds finally making up their minds, that started before Sandy struck and created an opportunity for him to play Concerned Leader. The million-dollar question is why. Why couldn’t Romney close the deal with fencesitters?

If you missed Karl’s post last night about the flawed national polls, now’s the time to catch up. Gallup, it seems, may have been tripped up, ironically, by the fact that it screened its sample of registered voters too rigorously to try to find likely voters. In the past that was necessary because turnout on election day tended to show a redder electorate than you’d expect from a sample of registered voters. There were more Democrats out there, but GOPers could make up the difference because they were likelier to vote. Obama’s achievement this year was to turn out enough “irregular voters” that the electorate ended up looking like a poll of registereds, not likelies. In fact, go compare Gallup’s own tracking polls of likely voters versus registered voters and see how they did. Among likelies, the final poll was Romney by one. Among registereds, the final poll was … Obama by three. Gallup had the correct answer all along. They, and we, just made the wrong assumption about turnout.

• Suffolk poll puts Markey up 17 over Gomez
• PPP: South Carolina a dead heat
• Christie: On Hurricane Sandy, Obama has “kept every promise he’s made”

Blowback