This time one thing's for certain

Philip Cowley explains why the Tories are hoping in vain for a repeat of the Great 1992 Poll Disaster
Click to follow
Early last week when Ian Lang was asked about his party's standing in the polls, he delivered the standard Conservative response. "It depends," he said, "whether you think the polls are right", before going on to declare that he had been through this before: "I remember in 1992 when the polls said we would be wiped out and we went on to win."

Leaving aside his rewriting of history - the polls in 1992 did not show that the Conservatives were going to be wiped out, tending either to show the parties neck-and-neck or to place Labour just slightly in front - it seems a fair point. For, in 1992 the polls did get it seriously wrong. Two polls published on the day of the 1992 election put Labour ahead of the Conservatives, by an average of two percentage points. A third put the parties level, with a fourth, Gallup, giving the Conservatives a lead of just half a percentage point. When the votes were counted the Conservatives led Labour by a clear seven and a half points.

This was even worse than 1970, when the polls were last wrong. Then the polls were "only" six and a half points out, and one company had got within a point and a half of the actual result. In 1992 there were no saving graces: even Gallup underestimated the Conservative lead by seven points.

Yet the Conservatives should not pin their hopes on a 1992. Even if the polls are as wrong now as they were in 1992, Labour would still walk it: the current Labour lead is around 25 percentage points; in 1992 the polls were 8.6 percentage points out. The polls would need to be about three times as inaccurate as they were in 1992 to remove Labour's lead. You cannot compare now with 1992 and pretend, as Ian Lang did, that the Conservatives have "been here before". They haven't. No one has been here before.

There is a second, and more substantive, reason why the polls - or some of them - should be more accurate this time. Rather than just shrugging off 1992 as a bad year and carrying on as before, the polling companies undertook a rigorous examination of what had gone wrong. They considered the two most common explanations for the error - that there had been a late swing and that we are a nation of fibbers.

The late swing explanation is kinder to the pollsters. It suggests that the polls were accurately measuring voter intention, but that by the time real votes were cast something had changed these intentions. The more hysterical explanations have the voter undergoing a Pauline conversion while actually in the polling booth as the horrors of a Labour government hit home.

The second explanation - that people are so ashamed of supporting the Conservatives that they pretend otherwise - is also favourable to the pollsters. They may have got it wrong, but if everyone was lying to them, it couldn't really be their fault, could it?

However, the report of the Market Research Society after 1992 largely rejected both of these explanations. If there was a late swing then why did the exit polls - which measure actual vote rather than intention - also get it wrong? And, if people lie, why do methods of polling which get people to record their vote secretly not show a higher Conservative vote?

To be sure, there was a late swing and some people may well lie to opinion- polling companies (or may be deceiving themselves, which is a less depressing way of looking at it) but the bulk of the explanation for the polls' debacle in 1992 lay in two errors of a more systematic nature. First, the polls were too "downmarket": too many of those questioned were from groups likely to be Labour supporters (such as council tenants) at the expense of groups likely to be Conservative supporters (such as those from households with two cars). And second, there was a problem of disproportionate refusals: Conservative supporters seem to be more likely not to answer questions asked by people with clipboards (either by refusing point-blank or perhaps by fleeing as soon as they see the clipboard) or to claim that they don't know how they will vote.

This was an important finding. Though the "late swing" and "the fibbers" explanations largely exculpate the pollsters, if true they would also mean that there was little that could be done to improve future polls. By contrast, the problems of inaccurate sampling and disproportionate refusal rates are, potentially at least, solvable.

Unfortunately there is as yet no consensus on how to solve the problems: each of the main polling companies has introduced slightly different innovations. Some have included telephone polling, which allows for random polling as well as cutting down on the problem of unknown refusals, since you cannot run away from your own telephone. Others have made various kinds of adjustments to accommodate the don't-knows and the won't-says.

As a result the headline poll figures can vary quite dramatically. ICM's adjustment is the largest, weighting for past vote and allocating 60 per cent of those who don't express a preference to the party for whom they claim to have voted in 1992, which has the effect of depressing the Labour lead. Thus Gallup, for example, currently shows the Conservatives to be 28 points behind Labour; ICM, by contrast, puts the gap at 18 points.

The polling companies cannot, then, all get it right on the night. Even if they all predict the correct winning party, there is bound to be a degree of error in the various forecasts of the actual share of the vote received by the different parties.

But by hedging their bets in this way, it means that they are unlikely to all get it wrong either, as they did in 1992. Herein lies the problem for the Conservatives. Even with such a wide variety of polling methods, some of which are weighted to increase apparent Conservative support, the one thing - perhaps the only thing - on which all the polling companies do agree is that, short of a miracle, the Conservatives are going to lose.

The writer teaches politics at the University of Hull. He will be observing the polls for the 'Independent on Sunday' each week until the election.

Comments