The election was called on May 8. From the outset the opinion polls were unanimous — Labour was going to win, and by a wide margin.
On May 9 the Daily Express declared on its front page “Blair by 250” (NOP). As the campaign progressed, Labour’s lead was inexorable. No event occurred to change public attitudes, nor did any issue seize the imagination enough to alter polls.
The pollsters had something to prove. They were ridiculed in 1992 when they predicted a Labour victory. They overestimated Labour in 1997. In 2001 they repeated this, but now there are more reasons to excuse a small statistical error.
Advertisement
It became evident during the course of the election that there would, again, not be a national uniform swing and that the turnout was likely to decline. Although these features cannot be reflected in overall figures they were correctly identified in the commentaries accompanying many polls, especially later in the campaign.
Fewer people said they were “very interested” or “fairly interested” in the election than in 1997, but nevertheless the proportion saying that they would definitely vote was always substantially higher than the 59 per cent who did turn out.
The national polls were dominated by MORI, NOP, ICM, Gallup and Rasmussen. Their findings suggested dramatically different Labour majorities, but this may have had more to do with their different sampling methods than with fluctuations in opinion.
Most polls were conducted by telephone, only MORI collecting its data face to face. Differences emerged between polls which asked the blunt question of their interviewees “Who do you intend to vote for?” and those which, before asking the question, made mention of the parties contesting the election.
Questions could be made more precise after the nominations closed and the list of candidates was known. MORI changed its methodology to reflect this and its polls began to suggest a slight narrowing of Labour’s lead. Polls which asked questions about attitudes and issues also found slightly different outcomes according to whether interviewees were presented with a menu of issues or whether they were asked to name issues themselves.
Advertisement
The latter method suggested that the Conservatives’ emphasis on Europe and the future of the pound had failed to ignite the electorate, and this was a contributory factor in the shift of emphasis of their campaign in its closing stages.
Opinion polls can only represent a snapshot in time, reflecting views at the actual time of sampling. Over the timescale of a general election they can identify trends, but not until the last few days of an election can they be said to have a predictive capacity.
Viewed in this way the polls were relatively consistent. Indeed, ever since 1997 Labour had enjoyed a lead, usually in double figures, over the Conservatives — something not achieved by any other government of modern times. It was an open secret that the election had been pencilled-in for May 3, only to be delayed because of foot-and-mouth disease. The delay did nothing to harm Labour.
Just before the delay became apparent, MORI in The Sunday Telegraph suggested Labour leading by 48 per cent to 32 per cent (April 1) and ICM in The Guardian agreed (49 per cent to 34 per cent).
By the beginning of the campaign proper Labour’s lead, if anything, was even bigger. Throughout the month of May 8 to June 7 the Conservative percentage seldom varied. Seven polls put them at under 30 per cent. Only one put them as high as 34 per cent. Averaging all polls (itself slightly risky because of different sample sizes), the Conservatives stood at 30.3 per cent.
Advertisement
It has become a common observation that Conservative supporters are more likely to decline to disclose their voting intention than supporters of other parties, and some polls were weighted to try to take account of this. An equally plausible claim in an election with such a low turnout is that Conservatives were more likely to vote than their Labour opposite numbers and that this marginally increased the actual Conservative percentage on polling day.
This claim is strengthened by pollsters’ findings during the campaign that Conservative support, although lagging far behind Labour, was also firmer.
Labour’s vote appeared to be more fluid. From a high point of 55 per cent to a low of 43 per cent in the polls, the party’s alleged support shifted not only more than the Conservatives’ but also in a patternless way (except that all the final polls put Labour at 47 per cent or less). During the campaign Labour lost support overall to the Liberal Democrats, but this disguises a triangular movement.
Labour did lose support to the Liberal Democrats, partly because of the willingness of Labour voters to shift to the Liberal Democrats in constituencies in which Labour entered the contest in third place — again something very difficult for polls to identify precisely because of the late decisions which such voters often make.
But Labour also lost a small amount directly to the Conservatives, and even more from those who declared themselves Labour supporters who ultimately did not vote. Even from the fieldwork conducted on June 2 - 3, with the resulting polls published on June 5, 6, or 7, Labour’s final support was overestimated by between 2 per cent and 6 per cent.
Advertisement
The final polls for the Liberal Democrats and the smaller third parties were accurate predictions, even though support outside the two main parties is notoriously softer. The Liberal Democrats gained support modestly but firmly during the course of the election, and their leader Charles Kennedy was recognised by nearly twice as many respondents by June 7 as on May 8.
As coverage of the Liberal Democrats increased in all the media, so did their poll ratings—a feature which was also true of the 1997 election. The politics of Scotland and Wales involve a different pattern, (and Northern Ireland even more so, which is omitted from the table of polls published opposite), and in each country Labour maintained a bigger lead over the Conservatives than in England.
The polls published exclusively for Scotland slightly overestimated SNP support and failed to detect the rise in support for the Liberal Democrats, while those in Wales also overestimated Plaid Cymru. The Labour lead in both was, however, so great that such a difference mattered little.
Even the last polls showed a Labour lead over the Conservatives of between 11 per cent and 17 per cent. In the event it was just short of 9 per cent. The different effects of regional and sub-regional voting make predicting a translation of percentages into seats hazardous.
The closest estimate was ICM, published in The Guardian on June 6, which predicted a vote of 43 per cent to Labour, 32 per cent to the Conservatives and 19 per cent to the Liberal Democrats. On this basis ICM predicted an overall majority of 165. In fact it was 167.
Advertisement
Peter Kellner, in the London Evening Standard as early as May 29, predicted Labour on 412 (it was 413), Conservatives on 163 (actually 166) and Liberal Democrats on 53 (actually 52), an overall majority of 165.
It would be impossible in the world of polling, in changing electoral geography, to expect anything more accurate than these two forecasts. The polls emerged relatively unscathed from the election. The errors they reported were not of their own making.
In military history it is often remarked that generals prepare to fight the last war, not the next one. With such a disparate, changing and unpredictably local voting pattern in Britain today, pollsters can hardly be blamed for anything different.
This article first appeared in The Times Guide to the House of Commons 2001