John Curtice: What the electorate should – and shouldn't – read into opinion polls

OPINION polls play a central role nowadays in the narrative of politics.

When in February the Tory poll lead narrowed, many commentators criticised David Cameron for alleged mistakes. Now the Conservative lead has widened again, George Osborne has been receiving plaudits for his promise not to hike National Insurance as much as Labour would.

Little wonder, then, that many become discomfited when polls show apparently contradictory results. Yesterday was a prime example. An ICM poll suggested the Tory lead had narrowed to just four points, while both YouGov and Opinium stated that the lead was still as much as ten.

In truth, people often expect too much of polls. Even two perfectly well conducted polls can disagree in their estimate of a lead by as much as ten points simply by chance. And under the first-past-the-post system, such a difference could encapsulate everything from a Labour majority to a Conservative one.

Still, we should be on somewhat safer ground if, instead of focusing on individual polls, we look at the average across a reasonably large number. Chance then plays a much smaller role. Take this approach to recent polls and a clear pattern emerges. Immediately before the Budget, the polls, on average, put the Conservatives ahead by six points. After the Budget, the average lead increased to seven.

Now, in the week since Mr Osborne's announcement about National Insurance, the average lead (including in yesterday's polls) has risen to eight points, following a one-point increase in the Tories' share.

Even so, looking at the average of the polls presumes that, collectively, they are more or less right. Yet few pollsters will forget the 1992 election when that manifestly was not the case. The industry has been trying to eliminate an apparent pro-Labour bias ever since.

To do this, pollsters use a variety of methods, including adjusting the data to take into account how people say they voted in the past and how likely they are to vote this time. But there is inevitably a degree of uncertainty about the accuracy of these methods. Meanwhile, what people most want to know about the election is what the outcome will be in seats.

Opinion polls, however, only estimate votes. Any statement about what a poll means in terms of seats is but an extrapolation that usually assumes the whole country will swing the same way.

Yet a number of recent polls in Labour-held marginals have all suggested the Tories are making rather more progress there. If they do, Mr Cameron may well do better than the national polls suggest – but that, at least, will not mean the polls were wrong!

• John Curtice is professor of politics, Strathclyde University.