Anyone with the slightest interest in the subject of polls this year has heard about The Bradley Effect, in which polls supposedly under-represented racist voters, so that black candidates were likely to fare more poorly than the polls indicated. My favorite statistical analyst, Nate Silver, argues cogently (here) that this effect did exist in particular circumstances in the past, but is misunderstood. He calls it "a persistent myth." He points out that Obama has actually outperformed his polling this year, for one.
But here's another factor that seemingly has been under-reported, perhaps because it's even harder to assess numerically. Anyone who has ever worked as a vendor knows intuitively about The Bandwagon Effect. It's simply, really. People are influenced by other people. If they see other people flocking to buy something, they will take a look themselves.
This is why baristas salt the tip jar with their own money. It's why movie studios use any possible excuse to trumpet "#1 Movie in America!" ads, no matter how little that phrase means numerically.
Social scientists say this effect holds true with polling too. A l994 study by two sociologists at the University of Kentucky found a statistically significant correlation:
This study, which sadly appears not to be available on-line, can't be quickly boiled down to a quick percentage, especially since a well-designed poll should pick it up as election day approaches.
But one has to wonder: Could this be fueling Obama's recent rise in the Gallup poll? Or is it just statistical noise?
It's probably impossible to know, but hardened lefty Marc Cooper reports from Las Vegas, where the betting on numbers is serious business, that to put a (illicit) bet down on Obama, one has to bet a $1000 to win $100...and it's considered a slam-dunk win. One expert told Cooper that betting on anyone but Obama to win was considered "an absurd wager."
