The Role Of The Polls In The American Presidential Election

For BBC Online Monday, 13 November 2000 By Sir Robert Worcester

17 November 2000

Questions have been raised about the role of the polls in the American presidential election. One thing's for sure; they said it would be close, and it certainly was! Polls rained down on the American public at the rate of over four a day during the two months and a bit since Labor Day, the traditional 'home stretch' of the presidential contest. Nobody could say there wasn't enough data, but what did they tell us?

There were three or four so-called 'tracking polls', not much respected in the three-four weeks of the British elections, but utilised in America to get, as here, daily reports of poll fluctuations 'on the cheap' for the media, combining samples no self respecting pollster would call his own, perhaps as few as 200-300 a day, but combining them and reporting them on a three, four or even five day aggregate, so as to accumulate a respectable sample size. Gallup, who reported no less that 59 of these during the election (I probably missed a few) was conspicuous in the early weeks for jumping about, as their sample size was tiny, until they boosted it in the final few weeks, and steadied their findings.

In Britain, the major pollsters have held the line on the media's inclination to buy as cheaply as they can, with the easy excuse that it was the pollster's fault if they get it wrong on the night. British polls of fewer than one thousand people are polled from time to time, on issues and attitudes mostly, but the voting intention figures are not released on sample sizes of less than 1,000. The Americans might take a leaf from our book.

Neither do we publish 'raw' figures, but reallocate the 'don't knows' to bring the reported finding up to 100%. Otherwise the reader and viewer is naturally confused. I caught one poll report on CNN that was given in such a rush that even I couldn't follow it. The American pollsters look pretty silly reporting findings of 42% for Gore and 46% for Bush, 5% for Nader and 1% for Buchanan as CBS did on the 5th of November, when the result two days later was 49% Gore, 48% Bush. The fact is that the CBS figures only added up to 94%, when, believe it or not, 100% of those who voted voted (except maybe in Florida!).

Still, the polls told us that it was going to be a close race, and it certainly was. When the 267 national polls are averaged over the campaign, they came to 48% for Bush, 47% for Gore, 4% for Nader and 1% for Buchanan. Pretty damn close to the 48%, 49%, 3% and just under half a point recorded on the night. From the data, I said on BBC Radio 4/5 on the night that I suspected Gore would just pip Bush in the popular vote from two factors, the third party squeeze on the Green's Nader, and an unusually high turnout by the Afro-Americans, both of which proved out.

The Voter News Service exit poll was as reliable as anyone could expect. Conducted nationally among 13,049 voters and in 14 key states it provided very valuable demographic, geographic and attitudinal information, probably the most interesting being the gender gap was the greatest ever, with men favouring Bush by 9 points, while women favoured Gore by 12 points. It also provided the information that Nader's votes would have elected Gore, had he stood down before Election Day. Whites favoured Bush by 53 percent to Gore's 43 percent; blacks voted 10 to 1 for Gore and were 10 percent of the voters; Hispanics were nearly two to one, 63 percent to 33 percent for Gore, despite Bush's fluency in Spanish. Education, health care, the economy and jobs and social security all were selected as important to Gore supporters, while tax was overwhelmingly of concern to Bush votes, 79 percent of whom said tax mattered most to them in deciding which candidate to support, followed at 52 percent who chose world affairs.

There will be calls to ban the publication of poll findings by Congressmen who should know better. You can't ban the publication of polls in a free society, but pollsters can and should examine their own procedures, and make their story as clear as their masters, the media, will allow. More exposition by poll specialists and less by pundits would be a good start. I watched pundit after pundit election night and afterwards report confusingly and misleadingly the results of perfectly responsible poll results, reading one so-called expert castigating the pollsters once again for doing a perfectly reasonable job. Save us from the poll pickers, the academics and critics who live off commenting on the polls, and put the boot in whenever they see the opportunity. Of the 15 polls published at the end of the campaign, 13 were within the plus or minus three percent margin of error and 12 within plus or minus 2 percent for Bush. (See table)