Polling Noise: What to Make of Surveys Favoring Obama
I’m guessing most readers already know that it’s generally a mistake to put a lot of stock in specific polls at this stage of the presidential contest. Poll trends often do matter—not necessarily in terms of predicting the final outcome, but certainly by way of illustrating how the dynamics of the contest are playing out. So today’s battery of highly favorable polls for the president from Ohio, Florida and Pennsylvania conducted by the formidable combo of Quinnipiac, CBS and the New York Times, when pondered on top of last week’s NBC/WSJ and Democracy Corps surveys, could indicate that something real is going on (particularly since both the DCorps and QPac polls use likely-voter screens, which typically give GOP candidates a small bonus).
Or at least that’s you might infer from the furious efforts by conservatives to discredit the QPac/CBS/NYT survey today. One line of attack focused on how participants say they voted in 2008. Aha! Much larger Obama margins than the actual results! Bad sample! Bad sample!
But as TNR’s Nate Cohn explains, this is a very common phenomenon; voters always disproportionately “remember” supporting the winner of past elections.
A similar argument is that the composition of the electorate in the poll shows just too many Democrats to be credible. But RCP’s Sean Trende, who is hardly in the tank for Obama, explodes that theory pretty decisively.
Any one poll can be an outlier, the result of random “noise” that does not reflect actual trends in public opinion. But the “noise” of those who want to dismiss any finding that doesn’t support their candidate is worse. As Cohn notes:
We’re at a peculiar point where many of the most highly regarded polls are showing a sizable advantage for the president, including NBC/WSJ, Democracy Corps, Pew Research, and Quinnipiac. And perhaps this is just the liberal media bias, but many pollsters showing a tighter race, including Rasmussen, Gallup, and JZ Analytics, have weathered an outsized share of criticism over the last few years.
New polls may soon come out creating a different impression of the race, and for that matter, a bad jobs report on Friday or some other external event could turn the arrows around for real. But in the mean time, data is data, and political gabbers should strongly fight the temptation to skew the data by perpetually claiming the data they don’t like is itself skewed.