Last Wednesday's Guardian/ICM survey put Labour on 42 points (down three on their position a week before), the Conservatives on 37 (up six) and the Lib Dems on 14 (down five). The Labour lead, cut by nine points, was down to just five. Not only that, but the Conservatives were ahead of Labour on who was best to run the economy, and Major led Blair on who would make best Prime Minister. The Guardian became suicidal. The Telegraph went orgasmic.
It is a measure of just how far Labour has been in front that a five- point lead was taken to be quite so dreadful for them. The same Labour activists who were doom-laden this Wednesday would have sold their souls to be five points clear of the Tories on 10 April 1992.
But a cut of nine points in a week seemed bad enough for Labour. Another week like that and the Conservatives would be back in Downing Street, come 2 May.
Yet the Gallup survey in the same day's Telegraph showed the Labour lead to be a massive 21 points, as did the next day's MORI poll for the Times, which put the Tories on just 27 points, the lowest of any poll during the campaign. Friday's Harris poll for the Independent put Labour on 48 points, a full 18 points ahead of the Tories. And there is a world of difference between Wednesday's five-point gap and today's 24-point gap in the Independent on Sunday/Sunday Mirror MORI poll.
Even NOP, which last Sunday had put the Labour lead at its lowest level for four years, has today shown that lead grow by four points. As a result ICM's poll is now being dismissed out of hand as a rogue. This is a little unfair to ICM.
The first clue to what is going on lies in the different polling methods adopted after the failure of the polls in 1992. ICM has been showing lower Labour leads since the campaign began, not because it has recorded a higher Tory share of the vote, but because its figures boost the Lib Dems' share of the vote, at the expense of Labour. Its three surveys before last week had put Labour on 48, 46 and 45 points. With one exception, no other poll had put Labour on anything less than 48 points.
It is often said that the reason for this different result is that ICM adjusts its figures. With the exception of MORI, all the main pollsters adjust their figures to try to estimate the preferences of those who don't know or won't say, and both ICM and NOP also weight their figures by declared past voting.
ICM does adjust more than the others, but the adjustment does not explain why the Labour lead appeared to have shrunk so much this week. On Wednesday the unadjusted figures were: Labour 42, Conservative 36, and Lib Dems 16, giving a Labour lead of six points. The adjustment only accounts for one percentage point of the difference.
The answer may lie in the way ICM carries out its polling. It might be the way it asks its questions; listing the parties before asking about someone's intention to vote may explain why the Liberal Democrats tend to do slightly better in ICM's polls; or it may be the way it samples the population utilising random digit dialling, a method it broadly shares with Gallup but which it does slightly differently.
ICM might have hit upon the right method of polling and be accurately capturing the public's mood. The Conservatives - who also use ICM for their private polls - certainly think so, and bang on about how the mood on the doorsteps is different to the polls. Take this with a pinch of salt. After all, does anyone expect Dr Mawhinney to say: "The mood on the doorsteps is dreadful. Everyone hates us"? None the less other polling companies are a bit uneasy about the continuing large leads they are showing, and many Labour activists claim that they are not finding the population as overwhelmingly pro-Labour as the polls suggest. So it might be that ICM is more accurate than the other polling companies. We will only know that for sure on 2 May.
Of course, a different polling technique does not explain the fall shown in ICM's Labour lead. But, as the pollsters continually point out, each poll figure has an error of plus/minus three points. The difference between two parties - for example, the lead - thus has an error of plus/minus six points. And a comparison between the size of two leads has a monster- sized health warning of plus/minus 12 percentage points.
Assume that in the last two weeks' ICM polls the "real" position of the public has been Conservative 34 and Labour 45. Such figures are consistent with the reported ICM figures (31 and 45 two weeks ago; 37 and 42 this week). A "slashing" of the lead by nine points, then, is entirely within the possible error caused by sampling. But "poll lead wobbles a bit as a result of sampling error" does not make for a good headline.
What is remarkable is how much panic - and/or excitement - sampling error can engender. Earlier in the campaign when a MORI poll in the Times showed a 12-point change in the lead the headlines were equally dramatic. In 1992 there was a poll at exactly the same point of the campaign which showed Labour to have increased its lead to seven points. In 1970 one NOP poll showed the Labour lead to have suddenly jumped to 12.4 points.
But the best example of the power of such polling blips occurred on 4 June 1987, when a Gallup poll was published showing Labour just four points behind the Conservatives. Panic ensued at Conservative Central Office, with Lord Young grabbing Norman Tebbit by the lapels and hysterically shouting: "Norman, listen to me, we're going to lose this f***ing election! You're going to go! I'm going to go, the whole thing is going to go". Young only calmed down when the Conservatives received advance notice of a different poll, giving them a 10-point lead.
Yet all of these variations were within the bounds of sampling error.
This is what appears to have happened on Wednesday. All the polling companies show Labour leading. ICM shows a smaller Labour lead. But, except for a slight diminution in Labour's share of the vote, with a concomitant increase for the Lib Dems, the main message from the polls has been one of continuity. The only rogues are the people who write wild headlines about statistical blips.
The writer teaches politics at the University of HullReuse content