People often ask me why polls swing so much. Why did everyone miss the Republican wave in 1994? Why don't we have a better idea what's going to happen this cycle? The answer is simple---we have no idea who is going to vote.
Statistical sampling is very accurate. If, say, Walgreen processes 10 million prescriptions a month I can sample a thousand of them and give you a really accurate estimate of the number of mistakes in the entire 10 million.
But that doesn't work with voters--and it's not because they lie, or because we can't call the ones without land phones. The problem is that when I design a voter poll, the first question I have to ask myself is "who is going to turnout." I start my sample with "Adults" but that will tell me nothing about the election. Then I narrow the sample to "Registered voters"--half of whom won't vote, so I still have nothing. Then I try to identify "likely" voters. In normal years, I can get pretty close by polling people who always vote, and I can add in those people who assure me that they are definitely going to vote (even though a lot of them are lying). In a normal year, that will get me pretty close.
But that polling method makes it impossible for me to spot an abnormal year. That's becasue my sample assumes a normal year. Sure, I can make some adjustments. In 2008, we had a rough idea that younger voters would have higher turnout because of the hopey changey thing. This year, we think Republcans are going to win big because the likely voters are swinging Republican and we think the young Hopey Changey guys are going stay home.
Those two factors tell me that it's going to be a pretty big Republican year. But that's all we know. So is it going to be a HUGE Republican year where all the toss ups go Republican, and the Dems lose 90 seats? Or is it going to be kind of a normal Republican year where Democrats lose 30 or 40 seats? We have no idea.
That's because there are millions of ticked off people who almost never vote. It's easy to construct a polling sample in which they all vote--and Republicans have huge wins. And it's easy to construct a polling sample in which turnout among this group is higher than normal, but still within traditional ranges--and Republicans have moderate wins.
The bottom line is that we simply have to guess at turnout and that guess will generate substantial swings in the outcome. Here's an interesting article in the Wall Street Journal about Gallup's struggle with the sample.
The difference depended on who the nation's oldest and most venerable polling firm decided was likely to actually go to the polls. In one scenario, Gallup saw the GOP with a slim three-point lead nationally; in another, the margin was 17 percentage points
Here's a great example. We have plenty of polls in the Reid vs Angle race in Nevada--most of which give Reid the slight edge--but we are starting to get a glimpse of actual turnout and it doesn't look good for Reid.
In Reno’s Washoe County and Las Vegas’s Clark County, Republican turnout was disproportionately high over the first three voting days, according to local election officials. The two counties together make up 86 percent of the state’s voter population.
Rumor is that we are seeing the same thing in Arizona--Republican ballots are being returned at a higher rate than normal, while Democratic ballots are being returned at a lower rate that normal--but may be that's a timing difference. Democratic ballots may simply be skewed later in the cycle.
Is November 2nd going to be a ripple? wave? tsunami? Despite the most sophisticated polling in history, we simply have no idea.
Recent Comments