With the election barely a week away, some polls show the gap closing between Sens. John McCain and Barack Obama. That fits with the conventional wisdom that presidential elections tend to tighten up in the days before an election. Is endgame narrowing in the polls a real phenomenon?
Yes. In 10 of the 15 presidential elections from 1944 to 2000, the candidate who was leading in the polls on Labor Day saw his margin shrink by the time of the final poll. (This includes Thomas Dewey, who managed to lose to Harry Truman in 1948 despite never trailing in the polls.) If you average together all 15 of those contests, the Labor Day spread was cut in half by Election Day - although the early leader won the popular vote in every case except Dewey-Truman. In other words, while last-minute poll tightening is far from death and taxes, it is a real phenomenon.
Researchers offer differing explanations for why this might happen. While some point to buyer's remorse or cold feet, there is no statistical evidence to support these claims. Others point to the decreasing margins, as well as a reduced variation among endgame polls, to suggest that voters are drifting back toward their initial biases and preferences. In this model, voters are more likely to think independently in August than in November. If a candidate makes a newsworthy gaffe in August, a large number of uncommitted or weakly committed voters move to his opponent, resulting in a surge in the polls.
But months of appeals from the candidates to underlying voter allegiances have a real effect: When a voter's inner Democrat or Republican is awakened, they come home to their party's candidate. So the same gaffe in November would sway substantially fewer voters than it did before, and tightening poll margins reflect the number of committed partisans on either side.
Another theory attributes poll tightening to simple mathematics. Let's say that 10 percent of each candidate's supporters decided to switch sides in the final weeks of the campaign. That same percentage would reflect a larger exodus from the candidate who started with more voters - leading to a tightening of the race. Similarly, if undecided voters broke evenly in the final days, they'd add proportionally more support to the losing candidate - and again the poll margin would narrow. But few observers believe this can account for all of the observed tightening.
A related phenomenon is that the final poll, on average, overstates the actual margin of victory. From 1944 to 2000, the final polls predicted a margin 2.2 percent larger than the eventual outcome in the national vote. Given the short period between the final poll and the election, this is not likely the result of changing voter preferences. Rather, many believe that voters hesitate to declare their support for a losing candidate to a pollster, a tendency known as the "spiral of silence."