John Curtice: Pity the pollsters. They got everything right on the night - except the result

'Does Tuesday not demonstrate that we are fools ever to put so much trust in opinion polls?'

Who would ever become a pollster? Before Tuesday's election, the polls correctly told us that the United States was heading for its closest ever race in modern presidential history. Within an hour or less of the polls closing, they accurately forecast the winner in most of the 50 states. But, thanks to just one mistake, one reason why this year's US presidential election looks set for a prominent place in the history books is because the exit polls got it "wrong".

Who would ever become a pollster? Before Tuesday's election, the polls correctly told us that the United States was heading for its closest ever race in modern presidential history. Within an hour or less of the polls closing, they accurately forecast the winner in most of the 50 states. But, thanks to just one mistake, one reason why this year's US presidential election looks set for a prominent place in the history books is because the exit polls got it "wrong".

Their singular but crucial mistake was Florida, the fourth biggest state with 25 crucial electoral college votes. Shortly after the polls closed, the exit poll commissioned jointly by all the major American networks declared that Gore had won the state. Alongside predictions that Gore would pick up Pennsylvania and Michigan, too, it looked as though Gore was on the road to a rather unexpected victory.

One person just simply did not believe them. And that person was none other than George W Bush, who promptly let the television cameras into his Governor's mansion to let the world know just what he thought of the prediction. And then, as the votes started to be tallied up in Florida, Bush was indeed in the lead.

The broadcasters were forced into an embarrassing retraction. Florida was now "too close to call" and the race to the White House suddenly took on a different complexion. Alas for the pollsters and the media, their troubles did not end there. Just as many people in Britain were waking up to hear about the night's events, they made yet another mistake. As much as 97 per cent of Florida's vote had been counted and George Bush was still ahead, albeit only narrowly. As Florida was all that George Bush needed to enter the White House, the American media duly crowned him the 43rd President of the United States.

But state officials in Florida had other ideas. They duly announced that there would have to be a recount. The media was forced into its second retraction. Florida was, once again, "too close to call". In just a few short hours, the reputation of all the pollsters and their media clients was torn to shreds.

We have, of course, been here before. In the 1992 British general election, two exit polls commissioned separately by BBC and ITN both suggested that there would be a hung parliament, with neither of the two main parties even close to winning. In the event, John Major secured an overall majority of 21 on the back of a quite comfortable eight-point voting lead.

How can such mistakes possibly be made? Does Tuesday not demonstrate that we are fools ever to put such trust in polls? Is there perhaps even a warning here to Labour not to put too much confidence in its opinion poll leads?

First, we must remember that exit polls and opinion polls are different creatures. Opinion polls must find a mirror image of the population by seeking people out in their homes. Exit polls simply try to interview those who have made the effort to go to the polling station and cast their vote. Exit polls can thus go wrong for different reasons than opinion polls.

The key to a successful exit poll is to find an example of polling stations that will be representative of what happens across the country as a whole. Here, in fact, American exit pollsters have an easier task than their British counterparts. In the United States, election results are declared polling station by polling station.

Exit pollsters can put together a set of polling stations that between them have voted in exactly the same way as the country as a whole in previous elections. In Britain, results are only declared constituency by constituency. Exit pollsters can never be entirely sure that they are interviewing voters in the right places.

So how could things have still gone so wrong in Florida? Of course, a sample of polling stations that mirrored how the state as a whole voted at the last election may not necessarily do so again four years later. Meanwhile, not everyone goes to the polling station to vote. Some vote by post, and thus become invisible to the exit pollster. And the position in Florida appears to be so close that the outcome could eventually turn on how those absentee voters voted.

But the potentially big enemy of the exit pollster is those who refuse to say how they have voted. How someone votes is, of course, their own business; nobody is under any obligation to tell an exit pollster how they voted. But if those who choose one candidate are less willing to tell the exit pollsters how they voted than are supporters of another candidate, the pollster can soon be led astray.

All of this might account for the pollsters' first mistake in Florida. But what of the attempt then to give the state prematurely to George Bush? As soon as results start coming in, pollsters can update their forecasts, comparing the expectations of their poll against the early results.

In effect, when on Tuesday night the pollsters said that a state was "too close to call", they were saying they wanted to see some real results before trusting the results of their poll.

But after 97 per cent of the results are in? Alas, even then, if the remaining three per cent prove atypical, and the result is very close, a forecast can still be wrong. After all, in the Welsh referendum in 1997, the BBC forecast the "no" side would win, with just one result to come. That result then duly delivered an unusually large yes-vote. When an election is very close, the forecasters' black arts will always be tested beyond their limits. And these are the occasions when mistakes are long remembered.

John Curtice is Professor of Politics at Strathclyde University

j.curtice@strath.ac.uk