Dueling Pollsters

"Assessing Poll Performance in the 2000 Campaign" by Michael W. Traugott, in Public Opinion Quarterly (Fall 2001), Annenberg Public Policy Center, Univ. of Pennsylvania, 3620 Walnut St., Philadelphia, Pa. 19104–6220.

Political prognosticators seemed to fare almost as badly as Al Gore in the protracted 2000 presidential election. Not only did red-faced TV network anchors have to retract their hasty projections on election night, but most national pollsters had to swallow their forecasts of a narrow victory in the popular vote for George W. Bush. Should the pollsters now don sackcloth and ashes? Not necessarily, argues Traugott, a professor of communication studies at the University of Michigan.

In the popular vote, Gore got a winning plurality of 48.4 percent, beating Bush by half a percentage point. Of the 19 final preelection surveys of "likely voters" by different polling organizations, 14 gave the nod to Bush, while two had Gore with a small lead, and three (including two by the same firm, using different techniques) called the race a dead heat. That may not seem a great record, but the vast majority of all the polls accurately showed the contest for the popular vote to be very close. In fact, says Traugott, the 2000 surveys "were about as accurate as the average [of such polls] since 1956."

As the 2000 election reminded us, however, the popular vote isn’t what matters most. In what may be a trend—one likely to be accelerated by the 2000 outcome—several polling firms collected state-by-state data to forecast the all-important Electoral College vote. (Not all of them surveyed all 50 states and the District of Columbia.) All told, they made 149 predictions and were wrong 17 times. Two of the pollsters, missing the final result in Florida, foresaw a narrow Gore win in the Electoral College. The third firm, which wrongly put eight states in Bush’s column, awarded him 354 electoral votes—83 more than he actually got.

Academics did worse than the commercial pollsters in predicting the winner of the 2000 election. Using historical models based on the state of the economy and presidential-approval ratings, political scientists confidently unveiled seven forecasts at the annual conference of the American Political Science Association in August 2000. All seven anticipated a Gore victory, with the Democrat beating Bush in the popular vote by between 5.6 and 20.6 percentage points.

Did the professors’ models have a Democratic bias, or did Gore muff an election that should have been his? It’s too close to call, as a pollster might say.