Pollsters fear a statistical disaster

Click to follow
The Independent Online
THE NAMES Kinnock and Major haunt this presidential campaign - not only because of the parallel so comforting to Mr Bush, that here too a conservative party in power for a dozen years might yet again squeak back. For the myriad US polling organisations there is another fear: could they get it as spectacularly and embarrassingly wrong as their British counterparts in the run-up to the general election five months ago?

Superficially, a similar disaster might appear in the making. On one thing alone do polls since the Republican Convention agree, that Governor Bill Clinton still leads. But the size of that lead differs wildly: from 3 points in one 'quickie' survey immediately afterwards, to - variously - 6 points, 9 points, 11 points, 15 points, and most lately, according to the Washington Post and ABC yesterday, a gigantic 19 points.

And how could it be otherwise, the layman observes, when the samples are so small, normally between 750 and 1,000 individuals from a population of 250 million. That is the least of the pollster's worries: 'A few drops of blood are enough for an accurate blood test,' points out Frank Newport of the Gallup Organisation. Thus it is with political polling.

The technique of 'equal probability of selection' in a random sample is a computer-guided science designed to produce a cross- section of an entire country. Polling is normally spread over two to four days to remove other aberrations. 'We are only taking a snapshot of public opinion, not predicting a result two months in advance,' say the pollsters. But, they insist, the snapshot is as accurate as the obligatory 'doctors' warnings' permit.

One such 'warning' of course is statistical margin of error. In the the case of the Post/ABC survey, based on a a sample of 768 registered voters, this margin is plus or minus 4 per cent. This means that Mr Clinton's true score might be 51 per cent, not 55, and Mr Bush's 40 instead of 36 per cent. In that case the Democrat lead would be only 11 per cent, more or less in line with other recent polls.

And as the small print says, 'sampling error is only one of many potential sources of error'. These more subtle factors include the phrasing of questions, even the order in which they are asked. As an undeclared candidate for the White House, Ross Perot always fared better when people were questioned about their preference in a three-way race after being asked to choose between Mr Bush and Mr Clinton alone.

Then there are the 'don't knows'. People tend not to like admitting they do not have an opinion. Offered a blunt choice between two candidates, even undecided potential voters mostly opt for one or the other, meaning that ratings can be artificially inflated or deflated.

This year, arguably, that danger is larger than ever. Old party loyalties are frayed. Early in the campaign season, unusually high proportions of voters said they were dissatisfied with the choice on offer. The volatility of subsequent polls indicates that support for candidates may still be shallow. For that reason, as well as the British precedent, the political pros are unusually wary. 'This time we're not paying as much attention to the numbers, said one Democrat consultant, 'but going with our gut feelings.'

On balance the US presidential polls have a good track record, certainly better than their British equivalents. Not since Harry Truman's surprise win in 1948 have they suffered a serious embarrassment. The Kennedy-Nixon contest of 1960, they said, was too close to call - and it was. But even Jimmy Carter's narrow win in 1976 was correctly predicted. The trick, says Mr Newport of Gallup, is to keep polling up to the last moment.