Theory test

General election polling is not supposed to be like this. The standard theory is that there will be smallish changes for the two main parties, while the Liberal Democrats enjoy a slight boost thanks to broadcasting rules that give them guaranteed coverage. But movements on the scale that we have seen during this election are unprecedented. Whatever the result on 6 May, the polls will be discussed for decades to come.

The gap between the three parties, according to some polling firms, is now just a few points. The Conservatives' numbers, which went up after the second leaders' debate, have edged back down. But the Lib Dems are still in second place, with a current voting intention share of about 30 per cent. The first leader approval ratings since the second debate show Clegg polling extraordinary figures (see chart) - almost equal to Tony Blair's peak in 1997.

Bank holiday bailout

At a big pre-election gathering back in January, telephone pollsters were concerned about carrying out surveys during bank holiday weekends. An infamous 2005 Populus/Times poll - carried out over the weekend before the 5 May election - put Labour on 42 per cent, 14 points ahead of the Tories.

Three days later, Blair's party won with 36.2 per cent, while the Tories were on 33.2 per cent. The discrepancy was blamed on the difficulty of finding balanced samples - in particular, finding enough Conservative supporters - during long summer weekends. It has become accepted wisdom that surveys during such periods can produce skewed results. In 2005, the result seemed like a foregone conclusion anyway - but what effect could such poll findings have in the present, highly febrile environment?

Angus stakes

This election has seen a number of innovative polling developments, largely focused on the leaders' debates. YouGov and ComRes have both been publishing online and automated telephone polls within minutes of the debates ending. With Angus Reid, you can watch the numbers on its website change by the second as the data rolls in. Trends can be spotted even when only a few of the thousand or so responses have come in, making it the place for polling junkies to find out first who "has won".

The firm has also imported an interesting approach to online polling from its native Canada. Respondents key in their postcode, and are invited to
cast their vote on the basis of the candidates standing in their constituency. This approach, Angus Reid argues, may produce more accurate polling figures than the standard approach, which just involves choosing a party.

Mike Smithson is the editor of politicalbetting.com