US Elections
October 21, 2008
American voter surveys have often been off-the-mark in 2008 -- most notoriously when pollsters predicted a big Obama win in the New Hampshire primary, which then promptly went to Hillary Clinton.
Andreas Graefe of the Karlsruhe Institute of Technology (KIT) and the University of Pennsylvania's Scott Armstrong think they have come up with a way of being more accurate.
Graefe and Armstrong's method is based around collating likely voters' stated preferences for a given candidate with their responses to questions about key issues.
"What's new about our model is that we analyze all the problems for the country raised by current opinion surveys together with voters' opinions on who can better solve those problems," the pair of researchers stated on KIT Web site.
Their methodology, called "PollyIssues" and "PollyPolicies," predicts a clear, if hardly overwhelming victory for Obama with 52.2 percent of the vote over John McCain. The two researchers are convinced that that will, indeed, be the result.
"PollyIssues would have correctly picked the winner for seven out of nine elections from 1972 to 2004, along with providing an idea of the margin of victory," Graefe and Armstrong wrote in a June 2008 academic paper. The system failed to predict the winners in 1976 and 1980 due, the report said, to the small number of polls available for evaluation.
Their current prognostication is in line with the "poll of polls" averages used by many popular political Web sites in America. They show Obama maintaining a lead of up to 6 percent.
Political curveballs?
But there is considerable reason for skepticism given pollsters' relatively poor recent track records.
To try and combat inaccuracies within surveying systems, Americans Nate Silver and Sean Quinn have begun using methods from baseball, arguable the most statistic-obsessed and American of sports.
One core of their idea is distinguishing between the various established national polls.
"We assign each poll a weighting based on that pollster's historical track record, the poll's sample size, and the recentness of the poll," the pair state on their Web site. "More reliable polls are weighted more heavily in our averages."
Silver and Quinn's site, FiveThirtyEight.com, is named after the number of delegates to the American Electoral College, and their predictions focus on the likely number of those delegates for each of the candidates.
Delegates, not percentage of the popular vote, elect the American president.
Silver and Quinn had Obama up on Wednesday, Oct. 22, by a margin of 5.2 percent and winning the Electoral College by a whopping 150 delegates. They also calculated the chances of him winning the election at 88.4 percent.
X-factors
Still, life isn't easy for political prognosticators. For a variety of reasons, this could be the hardest election to forecast in US history.
Most pollsters still concentrate on calling people on landlines around dinner time. But increasing numbers of Americans work flexible hours and communicate using mobile phones. That might change the candidates' amount of support from forecasted indications.
What's more, first-time voters usually don't show up on pollsters' calling lists, something many observers see as aiding Obama if turn-out on election day is high.
Pollsters, however, have no guarantee that respondents are telling the truth, and this year's election features the first African-American presidential candidate from a major party.
That's led to considerable speculation about whether latent racism might cause some Americans to vote differently than their stated intentions when they get in the voting booth.
Election day will show whether the pollsters have it right this time round or have committed another colossal blunder. But whatever the result, they'll be consolation, if needed, in numbers.
All the polls currently put Obama ahead by 5 percent to 6 points. If that doesn't change in the next two weeks, all the pollsters will either get it right or wrong -- together.