Federal Review

Solve the world's problems. One beer at a time.

How the Polls Might Be Wrong (R337-O221)

Posted on | November 6, 2012 | 3 Comments

Everyone on the right is speculating that the polls must be wrong because many are predicting that the partisan turnout model to be similar to the 2008 election, when Democrats were energized and Republicans often stayed home.  This is an absolutely wrong assumption, they argue.  Besides, the polls – even in Ohio – regularly show a double digit Romney lead among independents.

So, I decided to look at the numbers.  Gallup had a poll showing a Romney lead of 7 (which has shrunk back to 1).  Yet Gallup predicts that turnout will be more heavily Republican than in 2008. Gallup expects the breakdown to be +3 for the GOP, yet conservatively is only modeling GOP +1:  Republicans 36%, Independents 29%, Democrats 35%.  To compare the numbers, I’ve taken the internals from the other polls.  They tell us the percentage of each party in the sample, and then how the self-identified Republicans, Independents and Democrats plan to vote.  For instance, most polls say Obama will win some Republicans and Romney some Democrats.  Where the full data isn’t given by a poll (I’m looking at you ABC/Washington Post who must have something to hide!), I’ve made some assumptions based on their reported result.

I demonstrated in a recent post how an NPR poll showing Romney with a 1 point lead was consistent with Gallup which showed a 5 pt lead.  It was all in the partisan breakdown, because NPR showed a lead for Romney among independents of 51-39, and partisan sample of Dem+6.5.  Plug those numbers into Gallup’s turnout model, and you get a Romney lead of 6.  Seems to be in the MOE of the Gallup poll (in fact, the 1 pt lead and 5 pt lead were within each others’ MOE too).  So, I took all the latest polls, worked them out with the Gallup turnout model, the 2008 turnout, the 2010 turnout and the 2004 turnout (also a very close election).  Here’s how those numbers look.

Taking an average of only those polls in November – Romney likely has somewhere between a 3 deficit and a 4 point lead.  2008 was a historic “wave” election – where GOP turnout was lower than normal and Democrat was higher.  No one in his right mind believes that same turnout will happen this year as the Romney Ryan ticket is much more acceptable to suburban voters than McCain Palin, they are drawing larger crowds than Obama Biden, and Obama and Romney have even favorable ratings (though Obama has higher negatives). In addition, the same polls getting the turnout model wrong also consistently reported higher definite and probable voters among Republicans and much more enthusiasm among Republicans. So, assume a more normal election turnout, I think it’s not unreasonable to see the 2004 model (close election, equally energized bases) or the 2010 model (recent and therefore indicative of the current voting attitudes which swept a Republican to a big win in a recall election in 2012).  Note that the CNN poll on October 28 showed a dead even race, but under the Gallup model and the 2004 model, Romney would have been in double digits – landslide territory.  There are no models even hinting at that possibility for Obama – thus giving some support to the Michael Barone’s of the world.

So, if the state polls have the same problem as the national polls and are off by as many as 4 net points, then Romney would win 337-201.  Early voting partisan ID has swung by as much as 14 pts in the direction of the Republicans, so this swing from the polling isn’t out of the question.


Related Posts:


3 Responses to “How the Polls Might Be Wrong (R337-O221)”

  1. How the Polls Might Be Wrong 2 : Federal Review
    November 6th, 2012 @ 5:22 pm

    [...] the polls are off by 2.5 pts in Obama’s favor -a very real possibility as explained here – then here’s the map – Romney wins with 285-253.  Here’s the [...]

  2. Martin
    November 6th, 2012 @ 5:26 pm

    I hope you’re right about this one!

  3. Landslide Scenario – Turnout Based : Federal Review
    November 6th, 2012 @ 5:28 pm

    [...] models means that Obama’s support in the states is overestimated by as many as 5 points as explained here. If they are wrong across the board by 4 points, then here’s the landslide scenario that [...]

Switch to our mobile site