Many people will be holding their breath as tonight's election results come in – but none more so than America's pollsters, whose near-unanimous prediction of a Barack Obama victory would, if wrong, be the biggest embarrassment since ... well, since they messed up in New Hampshire just 10 months ago.
As an underdog candidate, John McCain insists that the public polls, which give the Democrats a lead of close to double digits, are vastly overstating his opponent's advantage. What happened in the days before the New Hampshire Democratic primary on 8 January is Exhibit A in his case.
In the run-up, surveys put Mr Obama, fresh from his win in the Iowa caucuses, ahead by a steadily widening margin. But on the day Hillary Clinton beat him and kept her campaign alive. Could the polls be heading for a similar disaster now? The answer is, most probably not. For one thing, the recent track record of the final pre-general election polls is very good. When averaged out, they were spot on in 1992, 1996 and 2004. In 2000, however, most put George Bush slightly ahead, when the reality was that he lost the popular vote to Al Gore by 0.6 per cent.
Today's polling techniques are extraordinarily sophisticated, winnowing out likely voters from registered voters, and running parallel samples using only mobile phone users, to make sure that the landlines mostly used by pollsters do not distort the picture. But it is often forgotten that polls are not predictions of the future. Their function is to produce an accurate snapshot of public opinion at a given moment.
In the majority of general elections, the race tightens in the final days. New Hampshire came just five days after the caucuses in Iowa, leaving the polls no time to pick up a late swing towards Ms Clinton. But that does not mean they did not truly reflect opinion when they were carried out. Of course, polls can be wrong, even in snapshot mode. "Humility is built into our DNA," Jon Cohen, The Washington Post's director of polling, wrote on Sunday. "The mathematics of the 'random sample' on which all polling is based says that five times out of 100, we will be badly off the mark." His biggest worry is whether these samples are truly random: "What if people who pause to take a pollster's question are significantly different from those who don't?"
Pollsters call the phenomenon "differential nonresponse", and, according to Mr Cohen, it's "a live, practical concern". In a sense, it's a version of the celebrated (and much disputed) "Bradley Effect" whereby polling is said to overstate the vote a black candidate receives. The explanation advanced is that people lie to pollsters because they do not want to appear racist. Instead, it may rather be that potential anti-black voters are under-represented in the sample.
Conceivably, this might be a factor that helps Mr McCain's Republican contest. Poorer and less-educated whites, who tend to be less well disposed towards blacks, are also harder to reach by phone.
That is one reason, Mr Cohen wrote, that "although I don't think we'll get bitten, I'll be a little worried until it's all over."