Warning Signs on Traditional Polling
The failure of Mitt Romney’s 2012 presidential campaign pollsters to forecast the race — they truly believed he was in position to win — demonstrated just how easily pollsters can miss an election’s result by misjudging the turnout. But at least in that instance, most of the big national polls had President Barack Obama in the lead. There was no systematic failure.
Since then, however, there have been more warning signs. A prominent one came when Ed Gillespie, a Republican, lost the 2014 Virginia Senate race to incumbent Democrat Mark Warner by less than 18,000 votes out of 2.2 million cast. Not one of the nine organizations to release their polling on the race predicted anything close. Most showed Warner with a double-digit lead. The most sympathetic to Gillespie, a poll conducted by the conservative American Action Network, said he’d lose by 6 percentage points.
Because of the polling, Republicans focused their resources on other races. If they’d spent more to help Gillespie, it’s conceivable he could have won. “I really believe our volunteers and activists and donors believed we were in position to win,” says Christopher Leavitt, Gillespie’s campaign manager. “The problem was we didn’t get that extra help and that was affected by the polling.”
The polling news website FiveThirtyEight called it “a once-or-twice-a-decade error” and blamed it on turnout models that failed to account for a big dropoff in Democratic-leaning counties.
But what if such errors are becoming more common? Indeed, it was only last November when Republican Matt Bevin defeated Democrat Jack Conway in the Kentucky governor’s race by 9 percentage points. Only one of 13 publicly released polls showed Bevin leading, and that one barely. All the rest showed Conway ahead, winning by margins ranging from 2 points to 11.
It won’t take many more such mistakes for candidates and parties to start thinking more seriously about Internet polling.