Making Sense of Political Polls and Predictions for Election Day 2012

A conversation with Harvard Kennedy School Senior Lecturer David King

November 1, 2012


Much of the media coverage of the 2012 presidential campaign has been saturated by polling and predictions. But not all polls are alike, and many are unreliable, according to David King, senior lecturer in public policy.  King studies political campaigns, polling, elections, and the U.S. Congress. We spoke with him about the value and credibility of political polling.

What should voters keep in mind when they assess the value of political polls that are handicapping next Tuesday's presidential election?

King: Most polls are poorly done and should be read with skepticism. Pay attention to how a polling outfit reports their findings. Do they explain where the sample came from? Were they calling people on landlines, reaching them by cell phones, using robo-polls, or asking people to answer questions on the Internet? Each source brings biases, and those should be disclosed up-front. Polling firms should also report the response rate for their surveys, the number of people who completed their surveys, and the actual questions that were asked. And since polls are often paid for by political parties or candidates, those possible conflicts of interest should be reported up-front.

What are the most reliable/credible polls & polling methods?


King: Twenty years ago, most polls were relatively credible and done with a sense of rigor and professionalism. The number of polls has risen dramatically since then, while the art of polling has become more difficult since it is harder to draw a representative sample. Among the better polls that have been maintaining high standards: Pew, Gallup, and the Marist College polls that are now being jointly sponsored by NBC news and the Wall Street Journal.

Because polls have been becoming less reliable, it has become popular to average a set of polls and hope that errors on the high-side and those on the low-side balance out. The “poll of polls” at RealClearPolitics.com is a good example.

An even better way of handling polling is found at Nate Silver’s site through the New York Times. Silver assigns weights to the quality of the polls as he averages them, and he pays particular attention to polls in states (instead of fixating on national polls). Since the presidential election comes down to races in a half-dozen swing states, this approach makes a lot of sense.

Nate Silver adds an additional twist by estimating the probability that a candidate will win a race, based on the distribution of votes shown in these polls. For example, an average poll result of 51 percent for Governor Romney and 49 percent for President Obama does not yield a 51 percent chance that Romney would win the state. That 2-point lead could mean a 70 percent likelihood that Romney would have won the state if the election were held on the day that the polls were done. Nate Silver has his critics -- but his innovative work is currently state of the art.

It appears that the presidential election could be very close. Is there a possibility of an Electoral College tie? What happens if that is the outcome??

King: I do not expect a tie. I do not even think that the Electoral College outcome will be close. To the extent that this election is a referendum on President Obama, the President’s current momentum makes it likely that he will win with a strong Electoral College majority. Still, there are several (implausible) scenarios by which the Electoral College becomes tied at 269 to 269. In that case, the election would be thrown into the House of Representatives, where state delegations would each have one vote. Two previous elections were decided by the House of Representatives, first in 1801 and again in 1825. With the Republican Party currently controlling the majority of delegations in the House, Mitt Romney would almost certainly become President. Again, though, a tie this year is exceptionally unlikely.

Print print | Email email
Photograph of David King

David King, senior lecturer in public policy

"Most polls are poorly done and should be read with skepticism. Pay attention to how a polling outfit reports their findings. Each source brings biases, and those should be disclosed up-front. Polling firms should also report the response rate for their surveys, the number of people who completed their surveys, and the actual questions that were asked. And since polls are often paid for by political parties or candidates, those possible conflicts of interest should be reported up-front"