The British — Poll Convergence

Three of the "Big 4" pollsters have now pronounced for this week, and they have converged. All are near to the 50% Labour, 30% Conservative, 14% Liberal Democrat share of the vote that has been the result of nearly every poll conducted since 1st November last year. For the past six months, the state of the parties has been described as 'flat lining'. So it is today.

There's not, as they say, a cigarette paper between them, at first glance.

MORI's figures have pulled into line with the others, whereas before this poll MORI had the Labour Party share higher, at 54/55% over previous weeks, the Liberal democrats lower, at 11/12% but the Tories the same.

A change was made in the question methodology with the closing of nominations, as in the final poll in 1997, MORI showed respondents in each constituency the names of all the candidates standing in that individual constituency rather than as before asking an open-ended question of "How do you intend to vote in the General Election on June 7".

Of course it is nearly impossible to do this over the telephone, but MORI maintains its tradition of face-to-face interviewing, in people's homes.

The methodologies of the three pollsters differ considerably.

ICM and Gallup interview on the telephone, which gives a potential random sample, excluding of course the fewer than 10% who are not on the telephone, which ICM and Gallup use weighting to account for.

All three now "prompt" the respondents with the names of the parties, in the case of ICM and Gallup, with just the names of the three main parties, with MORI, the names of all the candidates and their parties standing in that constituency.

MORI reports its "top line" on all those interviewed, after reallocating those who say they would not vote (8%), are undecided (10%) and refuse to say how they would vote (2%).

This is so that comparison can be made between the actual vote and the last election and this survey, and the swing since the general election calculated.

The details of the methodology and the change in showing the names on the ballot was helpfully explained in Peter Riddell's column in the Times today.

From time to time MORI reports in the Times and elsewhere on the voting intentions of varying subsets of the electorate, such as those "certain to vote", those that "care very much who wins", etc., and take them into account in our final and only "predictive" poll, the poll that will be published next Thursday morning in the Times, on election day.

On May's 13-14 ICM, the unweighted voting intentions of the entire sample of 1,004 people were Conservative 29%, Labour 51% and Liberal Democrats 14%. But after applying weighting which includes the ICM "adjustment" of setting a target weight for a reported vote, the figures shift to Conservative 30% (up one), Labour 47% (down four!), and Liberal Democrat 16% (up two).

In the published headline figures, the lead was reduced by a further two points by taking the opinions only of those saying they were most likely to vote, 553 of their original sample of over 1,000.

We have searched for the "raw", unweighted figures on the ICM poll conduced 26-28 May, and the line in their tables has mysteriously disappeared. Shurly shom misteak?

ICM publishes an "adjusted" voting intention figure, which with the Conservatives on 28% must mean that their unadjusted figure from the last weekend's interviewing must be even lower. And so it appears.

The tables as published on the web site this week differ from last week, and do not show the unweighted bases, and neither do they see to match the published figures: 28%/47%/17%.

They show 27%/49%/16% based on all respondents expressing an intention to vote and saying likely to vote (7-10 at Q 1); filtered; 25%/50%/17% based on all of 1000 or so.

Nick Sparrow, the MD of ICM has sent me an email saying that if I want advice on polling methodology I should ring him.

Yes Nick, please explain your methodology so we can all see what it is that you are doing with these figures and how you choose which you publish, and please confirm that it is consistent.

The ICM "Variometer" seat projection on these latest figures is Conservatives 128, Labour 463, Liberal Democrats 38 and others 12, for an overall majority for Labour of 267, according to the ICM web site. We tried to run their figures through the 'Variometer' on their ICM web site and it refused to take a Tory figure under 30%. Shurly shom misteak?

When run through our "swingometer", we get somewhat different figures, their published figures of 28%/47%/17% works out to an overall Labour majority of 229. Before their calculation was lower than ours; now it is higher. Shurly shom misteak?

Gallup has helpfully provided a detailed description of their methodology on their American web site which may be found in the area under the British ensign.

Their figures are based on the subset of eligible voters identified as most likely to actually cast a ballot. The screen for registered voters, and eliminate less than 5% who say they are not on the electoral register. What they do with the people that say they are registered, but not in that constituency, they do say. (We register them, and eliminate them in the filtered analysis, not at the interviewing stage).

They then use 'the views, opinions and self-perceptions of the voters themselves as a mechanism for isolating likely voters', using a bundle of answers including whether they voted last time (notoriously inaccurate in the British polling history!), how closely the person is following the election, how often they vote in elections, "and so forth".

From this they reduce the total sample down to a percentage that best estimates the percentage of the voting age population that they think will vote on Election Day. Thus they say if the turnout will be 70% in their estimation, they throw out the 30% of the sample least likely to vote, and that's what their voting intention figures is based upon.

Thus Gallup has already taken likely turnout into account. ICM may have. I think so, but perhaps Nick will enlighten us as to just how he does it?

Seven days and counting.