Dealing with the ex-Labour supporters who said they “didn’t know”

At last! We now have the full data from the ICM Crewe and Nantwich by-election poll and I have reproduced the most interesting feature above – how the original numbers produced a 12% lead with the Tories on 47% and what happened when they went through the “spiral of silence adjuster”.

For what has happened here is that ICM has followed its standard practice and has allocated half of the 2005 Labour voters now saying “don’t know” to the Tasmin Dunwoody figure.

Looking at the finer detail less than half of Labour’s general election voters last time say for certain that they will be sticking with the party next week. A significant proportion are voting Tory with a large number, the biggest proportion I have ever seen, simply saying “don’t know” or they refused to answer.

Like all ICM voting intention polls this was past voted weighted and only the preferences of those saying they were 7/10 or more on the certainty to vote scale were included.

I find it very hard looking at these numbers to say that it will be anything other than a Tory win.

UPDATE: ICM has changed its standard turnout questioning for this poll – perhaps a reaction to what happened in the London mayoral race. As well as the normal rate your chances of voting on a scale of 1-10 respondents were also asked about their attitudes to this by-election. “They were asked whether they felt that it is not really worth voting, or whether people should only vote if they care who wins or, alternatively, whether they felt that it is everyone’s duty to vote.

Weights were applied to each cell of a 30 cell matrix based on the above two questions. At one extreme a respondent replying that they are 10 out of 10 certain to vote and who thinks it is everyone’s duty to do so received a weight of 1, but a person who said they were only 5 out of 10 certain to vote and thought it only worth voting if people care who wins received a weight of 0.3.

The weights have been devised so that the effective sample size is reduced from 1,004 to 570 thereby assuming a real turnout figure among registered voters of approximately 57%. This may be higher than the actual turnout in the by election. But 100% turnout could not be achieved because of faults in the electoral register. Evidence also suggests that polls tend to interview slightly more voters than exist in the whole population.

ICM has thus changed slightly our method for calculating which people are most certain to vote. By using a two part question we hope, more accurately, to weight our vote intention calculations towards those who bother to vote, and away from those who declare party support but do not bother, in the event, to register that preference.

Mike Smithson