Why is less importance attached to the view of Labour voters?

Detailed data from the December ICM and Populus polls shows that the reported shares followed the biggest scaling back of the views of those who said they voted Labour last time since the last election. If this process had not happened then, in spite of all the recent troubles, Blair-Brown’s party would have probably had poll leads.

For in each of the two polls that we have had in December more than half those who answered the “how did you vote last time” question said they had supported Labour at the last election. With both firms this vote recall proportion is the highest figure since May 2005. Normally they find an average of about 44%.

It will be recalled that the actual Labour share at the last General Election was just over 36%. So what the two leading phone pollsters do is adjust their samples, allowing for a level of misremembering, to bring it more in line with the General Election result and these are the figures that we see published. The December 2006 “adjustments” were the biggest I have ever seen.

If Populus and ICM had operated like Ipsos-Mori and Communicate Research, which don’t use such an approach, they would probably have reported significant Labour leads.

Understanding what is going on is central for those who like predicting and betting on election outcomes. I recently found this article from ICM, written after the 2001 General Election, which sets out the firm’s rationale for its methodology.

Pollsters hope that by getting the demographic profile of their samples to match the whole population, polls will give an accurate picture of voting intentions. Trouble is the days when all the toffs voted Tory and the flat capped working classes supported Labour have long gone. A demographically representative poll is no longer necessarily politically balanced. Added to which, response rates are low and falling. Some groups within the population are difficult to interview and some don’t want to reveal their voting intentions. But at present the pollsters simply replace refusers with others who share the same demographic profile, ignoring the possibility that, in doing so, it may be easier to find Labour voters…

You would have thought that most people could remember how they voted in the last election. Yet according to one poll conducted just three weeks after the 2001 election only 26% remembered having voted Conservative (7% too low) while 48% said they had voted Labour (6% too high). On the face of it, such polls simply contain too many Labour voters and too few Tories.

So why not use past voting behaviour to ensure the polls are demographically and politically representative? Some say you can’t trust past votes because some people forget how they voted and others align past votes to present intentions, but it’s the only candidate in town. Of course pollsters have to make some allowance for faulty recall, but the indications are that if the pollsters were to target recall votes closer to the outcome last time they will also get more accurate predictions.

I believe that there is a strong case for the ICM-Populus approach which is why I rate their polls ahead of Ipsos-Mori and Communicate Research. If, as happened this month, half your respondents are saying they voted Labour last time then clearly the sample is not balanced.

Latest prices on which party will win most seats at the next General Election are: CON 0.86/1: LAB 1.18/1.

Mike Smithson