Tuesday, November 06, 2012

Why the polls are all way off this year

Last Friday, Nate Silver of the NYTimes led with an article, "For Romney to Win, State Polls Must Be Statistically Biased".

No, the polls are not "statistically biased". But the pollsters almost universally are making a bad assumption about voter registration and likely turnout: they are basing their information on the assumption that the voter turnout will be exactly the same as it was in 2008.

Today, Silver is saying, Late Poll Gains for Obama Leave Romney With Longer Odds:

Instead, it is President Obama who is making gains. Among 12 national polls published on Monday, Mr. Obama led by an average of 1.6 percentage points. Perhaps more important is the trend in the surveys. On average, Mr. Obama gained 1.5 percentage points from the prior edition of the same polls, improving his standing in nine of the surveys while losing ground in just one.

But there is a major problem with many polls, and Nate Silver, as a statistician, ought to know this.

John Nolte writes for Breitbart about some recent, intensive surveys of the voting population this year. In short, the Democrats are staying home (see the related image) and the Republicans are coming out in force.

The Polls Are Wrong: Electorate is R+6:

While every single poll on the planet predicts Democrats will enjoy a turnout advantage of three to eleven points, the latest Rasmussen survey of party affiliation taken throughout October shows that Republicans enjoy a huge 5.8% Party ID advantage going into the 2012 election.

According to Twitter's invaluable NumbersMuncher, Rasmussen correctly predicted 2008 would be a D+7 election and incorrectly predicted 2010 would be D+3. (2010 was D+0, or even, so Rasmussen gave an edge to Democrats they didn't have.)

If the 2012 race is D+2, Romney's probably going to win. If the race is R+6, Romney's going to enjoy a landslide.

The most important piece of information in this poll, though, is that Rasmussen's sample is a monstrous 15,000 likely voters. Moreover, Rasmussen's Party ID results match similar results found by Gallup and Pew.

Again, if Rasmussen is anywhere near as correct as he's been in the past, all of these polls showing Obama holding small leads, are dead wrong. They're under-sampling Republicans in a major way; predicting an electorate that looks like 2008.

ADDED: Here are the specifics on the Gallup partisan breakdown. Though they buried the lede, this is all you need to know:

Now Gallup is in the game, and the numbers are brutal. In 2008, the Democrats had a 39-29 (D+10) advantage in hard party ID, and a 54-42 (D+12) advantage with leaners. In 2012 though, we’re in the post-TEA party era. Republicans now show a 36-35 (R+1) hard party ID advantage, and a 49-46 (R+3) lead with leaners. This gives us a range of party ID swings from 2008, from R+11 to R+15.

If the electorate is 49-43% as Gallup predicts -- or R+3 -- Romney wins in a landslide.


  1. Yep, that's why I'm predicting Romney 320+.

    1. Hi Ron, that's what a couple of guys (Michael Barone, George Will) are projecting. I think Rove is predicting something a bit smaller. But I think one of the big stories tomorrow will be the navel-gazing from the media on "how could the polls be so wrong?"

  2. Not so wrong after all.