John Curtice: Don't believe all you read in the polls

This election is going to be the most heavily polled ever. More than 50 polls have already been conducted since the beginning of March – as many as were published during the whole campaign in 2005. There may be another 70 or 80 to come.

But more polls do not necessarily mean greater clarity. Yesterday was an obvious example. YouGov (together with new boys Opinium) reported that the Conservatives were 10 points ahead, while ICM said the Conservative lead was just four points.

Such divergence should not surprise us. Even perfectly conducted polls of 1,000 people or so will vary one from another simply by chance. This is particularly true of the lead, which depends on getting not one figure right but two. Any two polls might be as much as 10 points apart without anything being necessarily amiss.

Since the debâcle of 1992, when the polls suggested Labour was a point ahead but the Conservatives won by eight points, pollsters have had to come to terms with the fact that the samples of people they manage to interview are rarely representative politically of those who eventually make it to the polling station. Rather, they typically exhibit a pro-Labour bias.

This problem is most obvious in the polls of the four established pollsters who conduct their polls by phone: ComRes, ICM, MORI and Populus. When they ask their respondents how they voted last time they typically find a higher proportion saying they voted Labour than actually did so. A phone pollster may have to make as many as 12 phone calls to secure one interview, and Labour voters are apparently more likely to pick up the phone and co-operate.

To counteract this, all phone pollsters other than MORI weight their data slightly in favour of the Conservatives. But they also have to bear in mind that some people may misremember how they voted last time.

Perhaps fewer people answering a poll actually voted Labour last time than say they did. In that case reducing the weight of past Labour voters would actually introduce error. The pollsters make allowance for this possibility. MORI, however, feels that it is too risky to weight the data at all.

Polling using the internet, as pioneered by YouGov, is a more radical solution. YouGov do not (because they cannot) contact people at random. Rather they rely on recruiting people into a panel of those willing to complete the company's polls.

Respondents are unlikely to be representative. But when people sign up to the panel, they are asked which party they usually support as well as how they voted last time. This relies rather less on people's memories.

YouGov aim to use this information to ensure they approach a politically representative sample in the first place. But as nobody can be sure exactly how many people in the population "usually" support each party, there is still room for uncertainty.

Pollsters must also anticipate who will actually vote and who might be particularly reluctant to declare their intentions. They do not all agree on how best to deal with these problems either. Polling nowadays is an uncertain art – so do not be surprised if there is the occasional disagreement.

John Curtice is professor of politics at Strathclyde University. He will be analysing the opinion polls throughout the campaign for 'The Independent'