Like fans and coaches working the refs, partisans and politicians' paid staff are quick to pick apart polls that show their guy or gal is losing. But the polls are probably more accurate than the partisan whining.
Today’s Quinnipiac University poll showing Mitt Romney ahead of President Barack Obama by 6 points is probably no different.
Democrats were quick to zero in on the demographic breakdown of the 1,722-person poll, which showed respondents broke down this way by party identification:
Republican (34%), Democrat (31%), Independent 29%
And in March, when Obama led Romney by 7 (and Sen. Bill Nelson led Connie Mack by 8), this was the party breakdown of the poll:
Republican (29%), Democrat (33%), Independent (33%)
But here’s the breakdown of active registered voters:
Republican (36%), Democrat (40%) and independent (23.4 %).
To recap: When more Democrats account for the overall sample, the Democrat tends to win. And when more Republicans are sampled, the Republican wins. And the loser whines that the poll isn’t accurate because the demographic breakdown doesn’t mirror registration or performance in a state where Democrats cast anywhere from 37 percent to 42 percent of the ballots and Republicans cast anywhere from 38 percent to 41 percent of the ballots in recent Florida presidential elections.
However, pollsters say the surveys (and many others like it) are actually more accurate than they appear because they’re technically not gauging actual voter registration. They’re gauging self-described party ID of voters. They’re getting a sense of the voter writ large. Although, there is considerable overlap between the self-described party ID and self-described party registration.
The specific question Quinnipiac says it asks: “Generally speaking, do you consider yourself a Republican, a Democrat, an Independent, or what?” It’s not asking: “What is your official party registration?” (The latter is a variant of the type of question asked by The Miami Herald/Tampa Bay Times’ pollster, Mason Dixon Polling & Research).
Though the two questions are alike (and rely on people to tell the truth), the responses can produce different demographic bottom-line results because the questions ultimately differ. That is, one question is: How do you feel? The other: What are you?
Consider how many felt in March, when Romney’s campaign was reeling from a brutal Republican primary capped with the now-infamous Etch-a-Sketch gaffe. People just weren’t feeling so Republican.
Now consider how many voters feel now that Obama’s campaign is producing mixed message after mixed message while Super PACs and Romney pound him on TV.
The mood of the electorate has likely shifted in response. So which type of polling method is right? Who knows?
Technically, Quinnipiac could scale its poll responses (called weighting) to make the bottom-line numbers look more like the electorate. It could boost the Democratic numbers and decrease the independent numbers in this poll. But that could interfere with what generally makes a poll valid: randomness.
“That’s a conundrum in polling,” said Charles Franklin, a University of Wisconsin polling expert who's visiting professor of law & public policy at Marquette University.
“Weighting may underestimate the real shifts in the psychology of the electorate when it comes to registration and party self-identification,” Franklin said. Also, manipulating results can leave a pollster open to charges of, well, political manipulation if it's not done with extreme care.
On the other hand, Quinnipiac and virtually all other pollsters weight for age, sex and race/ethnicity. Why is party ID different? Because it changes. Race, age and sex (except for operations, of course) are immutable.
As a result, the response of each African-American respondent, for instance, can be assigned greater weight to ensure black sentiment is adequately represented. White and older voter responses can be assigned relatively less weight because they wind up being disproportionately polled.
Weighting for party affiliation can be much more common when it comes to polls that specifically ask people how they’re registered or targets them off a list that clearly identifies their party registration, pollsters say.
Still, there are disagreements. Mason-Dixon’s Brad Coker, for instance, said the relative party ID for the Quinnipiac poll when it came to Democrats was a little too low for his comfort level. He said he probably would have adjusted the calls he made to ensure the proportion of telephoned Democrats was higher if they appeared to lag so far behind.
“Romney probably has a lead that’s really about 3 or 4 right now,” Coker said. “He’s still winning.”
The reaction by Democrats was much more vehement in January, when a Quinnipiac poll showed Romney beating Obama. Sen. Nelson's pollster, Dave Beattie, and Quinnipiac pollster Doug Schwartz sparred over the demographic issue. Beattie’s salvo is here and Schwartz is here. (Democrats were silent in March when Quinnipiac showed Nelson and Obama winning. Mack's campaign complained then. Today, they're advertising the Quinnipiac results because they show Mack up by 1, essentially tied with Nelson.
The day after Quinnnipiac announced its presidential numbers, Marist College produced a poll showing Obama and Nelson up by 4 in Florida. And, yup, you guessed it. Marist asks party registration. Its poll had Democrats at 43 percent, a about 3 points higher than current registration. Marist pollster Lee Miringoff said he prefers to use registered voters in Florida because the state keeps such solid voter-registration data. Beattie offers the same argument.
But the folks at Quinnipiac say they prefer their method, which involves random-digit-dialing Floridians instead of phoning off the voter-registration lists.
"We are currently using the same methodology that we used when we accurately predicted the Florida presidential results in 2004 and 2008," Schwartz wrote in January. "Respected New York Times polling analyst Nate Silver found that we were the most accurate poll in predicting the 2010 elections."
Polling is as much art as science. So is conducting surveys of likely voters as opposed to registered voters. Good pollsters have a feel and knowledge of how the electorate will look and try to adjust their models accordingly.
Despite all the variables, there are constants in polling.
Pollsters can often rip on each others work like catty artists. Partisans pot shot polls that don’t help their cause and over-elevate polls that do. And, within days of a poll being released, it’ll probably be out of date in the ever-changing climate of an election.
Don’t like the latest poll numbers? Stick around. They’ll change.
But the complaining won’t stop until Election Day.
(This post has been updated)