One of my pet bugbears is that "opinion" polls, because of their designation, are taken to be representative of the state of public opinion.
In fact they aren't. The headline figures quoted by the newspapers for voting intention aren't the raw data showing what the public currently think about the parties - i.e. the "opinion" of the nation. In fact they are filtered to take into account propensity to vote so what we are actually given is the "opinion" of the people who the pollsters think will vote - they are therefore a prediction of the likely outcome of a General Election (useful in itself), not a measure of the popularity of the parties.
The February Ipsos MORI poll illustrates this. (http://www.ipsos-mori.com/polls/2008/mpm080226.shtml)
Looking at the headline figure quoted in the press, you would think that the Tories were very marginally more popular than Labour (Con 39%, Lab 37%, LD 16%).
But that's the figure after they've stripped out all the low propensity to turnout voters. The actual "opinion" of the nation is listed below that: Lab 42%, Con 34%, LD 15% i.e. an 8% Labour lead and a government that with the population, as opposed to the high propensity to vote population, is actually rather popular. Of course the very people Labour policies on health, education, regeneration, improving social housing, minimum wage, workplace rights, tax credits, full employment etc. have done most to help are by definition the least well off in society, who because of social exclusion are the least likely to vote. It's bad enough that their opinions don't get registered at the ballot box in proportion to the rest of the population, but nowadays they don't even get picked up by opinion polls.
Imagine how little political momentum David Cameron would have picked up if it had been made clear that for almost the entire time he has been leader, the Tories hadn't been ahead of Labour in terms of public support - only ahead amongst those most likely to vote.
Polling companies and the media ought to stop saying "opinion poll" and start saying "election predictive model". Even the utility of it for predicting elections breaks down if we reject the assumption that election turnout will always be as low as it was in 2001 and 2005. Surely if we had a 1992-style election which people thought was highly competitive again, turnout will go up again and people other than those "certain to vote" will get excited enough by the closeness of the contest to turn out?
Labour needs to focus on how we get the people who constitute our 8% lead in public opinion turned into people who will actually vote. I think we should revisit Tom Watson and Mark Tami's Fabian pamphlet from a few years back that recommended we introduce compulsory voting on the Australian model.