CNN gave us a fine example of why the polls this year are largely worthless. The poll shows a tied national race at 49-49. This is a 3 point improvement for Romney over the last national poll they took. Then you dig into the internals of the poll. They sampled 41% Democrats and 30% Republicans. Yep, in a year when almost all the evidence points to a parity in party turnout or a slight Republican advantage, the best they could manage after giving Obama an eleven point advantage was a tie for the Southside Messiah!
The Czar at Gormogons has a straight forward explanation as to why most of the polls this year are showing a huge oversampling of Dems:
Here is how they are doing it.
If you look at history, which poll takers do, they consider what percentage of the population votes Democratic and what percentage votes Republican. For many years, Democrats typically make up 36-38% of the election; the remainder is composed of Republicans and independents. The Democrat percentage is so consistent that the election is won or lost by how many Republicans vote plus whatever the independents do.
For example, in 2002, 38% of voters went for Democrats, 40% went for Republicans, and 22% were independents—and the GOP did smashingly well even though many independents broke for Democrats.
2004, 37% of voters went for Democrats, 37% of voters went for Republicans, and 26% claimed to be independent. But enough independents leaned Republican for GOP candidates to win a few elections that year.
In other words, year after year, the election is decided by how many independents break for or against Republican candidates. Republicans numbers also fluctuate greatly, but Democrats tend to hover around 36-38% pretty consistently.
In 2008, voter turnout was pretty unusual: 39% of voters went Democratic, and 32% went Republican. The remainder (29%) claimed to be independent, but many were borderline Republicans who simply voted for Democrats that year. In fact, Democrats hadn’t enjoyed turnout that high since 2000—and Republicans had not seen participation that low in decades.
Got that? Okay. Let us say you are a pollster. You conduct a poll in which you call a few thousand numbers around the US, and hope that you get about a thousand people to participate all the way through.
You ask them if they are registered voters because you don’t need participants who cannot or will not vote. Sometimes, you ask if they are likely voters. This last part is very essential, the Czar believes, because only a small percentage of registered voters actually bother to act like mature adults and vote.
But many organizations don’t care. They only want to know if you were to vote, for whom would you vote if the election were held today? Then they ask if you are a Democrat, Republican, or Independent.
You then wind up with a mish-mash of information. You run some common statistical methods on it, including the margin of error (which you can look up if you don’t already know what this is), and you might wind up with the following simplified data for 1,000 participants:
As you know, it tends to be a bit more complicated than this, because the polls factor in your gender, age, ethnicity, income level, and so on in order to figure out how to better categorize this information. But stick with us.
Looks good. But this tells you nothing, as you might have magically called the only 620 Democrats in the US somehow. So the pollster normalizes the numbers to make them more reflective of the total population. For this, you can use oversampling—you might focus more on one group than another. Or, you can just look at historical results.
If we know that Democrats routinely make up 37% of voters, and then we can assume that 620 voters is probably too high. 620 out of 1000 is of course 62%, not 37%. So if we got 620 voters to indicate they are Democrats, than the actual number of Republican voters out of every 1,000 is probably closer to 493…all we do is multiply 304 actual Republican participants by 1.62 to normalize the number. And we see that the proportions are maintained pretty well. We can then figure out the probable number of independents from there. The results wind up:
Of course, this seems very scientific but it is of course absolute horseshit. It assumes that the fact Democrats average 36-38% of the vote makes a safe baseline for you to beef up the other numbers. You still have no idea whether or not your sample was good. For example, in 1980, the pollsters must have called nearly every Democrat household in the US and very few Republican ones.
Nevertheless, if you look at the voter identification numbers dating back decades, it lines up really, really well with the end result. So pollsters laugh and point out that the methods justify the results.
And there’s that little asterisk you hardly ever see that tells you, frankly, how many more Democrats participated in the polls than did Republicans:
This year is quite interesting: despite all the evidence that Mitt Romney is dominating, those darn polls keep showing President Obama tied or ahead.
Gallup and Rasmussen have started using a different assumption: what if voters turn out in the numbers they did in 2010? In other words, what would happen if the voters were evened out at 35%?
Bang—the numbers show that Romney is ahead or tied. And “a few points” translates to a landslide in voting measurements. “Oh sure,” squawk Democrats, “These polls are rigged because you’re choosing an outlier year.” You see, in 2010, Democratic participation in the mid-term elections was at an all time low.
But, these pollsters argue back, 2010 has much more recent data; and Republican participation in 2008 was at an all time low. If it’s wrong to use 2010 to represent likely Democratic participation, then it’s equally wrong to use 2008 to show likely Republican participation. Touché.
Many folks are not interested in these byzantine and occult weighting systems that are based, ultimately, on desired results. Ever give a kid the answer to a math problem, and then ask him or her to solve the problem knowing the answer? You get handed the right answer, but sometimes some hysterical math to get there. In political science, this is called sophistry. And it is how much of the polling numbers are computed.
Go here to read the rest. In short most pollsters are making the classic error of fighting the last war and assuming that the electoral turnout is going to be a mirror of 2008. This defies both logic and history. That Romney has done as well as he has in such weighted polls points to a Romney landslide in the making. Other signs are the early voting in most states where the Democrats are down and the Republicans are up, the enthusiasm gap between Democrats and Republicans and the comparative size of the crowds turning out to see Romney and Obama. 2012 simply isn’t 2008.
A word of caution in regard to the latest polls. Hurricane Sandy has tossed off most polls. First, by interrupting polling operations for a few days, and second by giving Obama a few days of the “rally round the President in a national disaster” bump of a few points. Such bumps are very ephemeral, normally lasting only a day or two, especially as news of the inevitable problems with disaster relief begin to filter out, as they are now. (Language advisory for the below video.)
Hurricane Sandy helped Obama, but for him it came a few days too early to do him good on election day.
Update: Ed Morrissey at Hot Air tears into the CNN poll:
Only this poll turns out to be bad news for Barack Obama. Why? According to CNN’s internals of this likely-voter survey, Obama doesn’t come close to tying Romney in one key indicator — independent voters. According to the sample data, Romney leads independents by a whopping twenty-two points, 59/37. How can Romney have a 22-point lead among independents but still only get to a 49-all tie with Obama?
Yes, that’s a D/R/I of 41/30/29. In 2008, a big turnout election for Democrats, the D/R/I was 39/32/29, and Obama won independents by eight points. Despite Republicans having a five-point edge in this very poll among those “extremely enthusiastic” about voting, the CNN poll has added four points to the Democrats’ advantage in this sample.
Let’s also take a look at the gender gap. In 2008, Obama got a +14 in the gender gap, with a +13 among women and a +1 among men. In this poll, Romney wins men by nine (53/44) and Obama wins women by eight (53/45) for a gender gap advantage of +1 for Romney.
So we are expected to believe that since 2008, (a) Obama has lost thirty points in the gap with independents, (b) Obama has lost fifteen points in the gender gap, and (c) Obama is still just four points below his 2008 share of the electorate? Only in a world where 41% of the voters will be Democrats and only 30% Republicans, and that world won’t be what we see tomorrow.