For poll-watchers during 2010, the “likely voter model” was the X-factor; it could turn a mere lead into a landslide or make an entrenched incumbent look vulnerable.
For poll-watchers during 2010, the “likely voter model” was the X-factor; it could turn a mere lead into a landslide or make an entrenched incumbent look vulnerable. When the election results came in, it enjoyed a mix of success and failure. For pollsters Mark Mellman of Mellman Group and Ed Goeas of Tarrance Group, though, likely voters remain the way to go.
As the election approaches, pollsters attempt to include in their poll results only those voters who will actually show up on Election Day. A “likely voter model” consists of a variety of methods for screening out voters who are less likely to show up to the polls. Gallup employs a rigorous seven-question screen while some other pollsters simply ask a respondent how interested they are in voting on a scale of one to ten. The methodology varies, but the respondents left are always counted as “likely voters.”
Gallup, a highly respected pollster and one of the oldest institutions in the industry, released its last poll the weekend before the election. Employing two different likely voter models, it predicted a Republican advantage of 10 to 15 points, which turned out to overestimate the Republican edge by at least 4 percentage points.
“People in our business develop a likely voter fetish,” says Mellman. “The problem is there is never an electorate made up entirely of likely voters. Too many people exclude voters that will vote but don’t often vote, and they make up considerable chunk [of the electorate].”
Ed Goeas, president of the Tarrance Group, a firm that polls for Republican candidates, disagrees. “We take a strong look at tendency to vote and intensity,” he says. “The two key things [for modeling] turnout are age and education.”
Goeas cites his firm’s polling in the Oklahoma governor’s race as an example of a successful likely voter screen. In Oklahoma, Tarrance Group projected Republican candidate Mary Fallin to receive 60 percent of the vote, based on its likely voter model. This was unprecedented – no Republican candidate for Oklahoma governor had ever received more than 57 percent of the vote. Given that Fallin faced a credible Democratic opponent, some observers criticized the poll, arguing that it had overscreened for likely voters and omitted too many Democratic voters. Mary Fallin ended up winning with 60 percent of the final vote. “We got raised eyebrows with our results, but they turned out to be right on the mark,” says Goeas.
In 2010, public polling firms on averageprojected stronger Republican performance in Nevada and California’s state-wide races than the final election results bore out. Some speculate that Hispanics were under-polled. Some say that cell phone-only households contributed to the level of error. Perhaps Republican-leaning voters were simply more eager to answer a survey during dinner. Whatever the cause, the results were skewed towards Republicans in a number of races.
Polling continues to adapt to a changing electorate, and there is little question that polling will remain a staple for campaigns. The reality is that a handful of “unlikely” voters will always show up at the polls, and they can help decide elections. Just ask David Plouffe.
Noah Rothman is the online editor at C&E. Email him at email@example.com