The Guardian’s monthly ICM poll is out (I’m not sure why it’s turned up at lunchtime – or perhaps I just missed it last night!). Topline figures are CON 37%(nc), LAB 36%(nc), LDEM 17%(+1), Others 10%. The Conservatives retain a one point lead. ICM do tend to show some of the most positive figures for the Conservatives (for reasons I’ll address below), but last month their 1 point Conservative lead did look like something of an outlier. It seems, however, that the difference is more systemic.
Now, looking at people’s reaction to this I’ve already seem plenty of comments remarking upon the big contrast between different companies figures – it probably deserves some explanation. Let’s start by comparing the sort of figures the different companies have produced over the last three months.
In their daily polls YouGov have been showing Labour leads of around 8 points, with the Conservatives firmly on around 36%, Labour on 42-44%, the Lib Dems on 9-10%.
Populus, who in most senses have methodology very close to that of ICM, have been showing the Conservatives between 34-39% (they dropped 5 points in their last poll, so it’s a big range), Labour firmly around 39-40%, the Lib Dems around 9-11%.
Ipsos MORI again have shown some variation in Conservative support, ranging from 32%-37% in their recent polls, have Labour more steady between 39%-42% and have the Lib Dems on 9-11%.
ComRes run parallel polls – telephone ones for the Indy and online ones for the Independent on Sunday. In their phone polls the Conservatives have been between 34-37%, Labour between 37-40%, the Lib Dems 11-13%. In their online polls the Conservatives have been between 36-38%, Labour have been between 37-40%, the Lib Dems at 10-11%.
The difference between an 8 point Labour lead and a 1 point Tory lead is large, but as Bob Worcester is want to say, look at the shares, not the lead which exaggerates variation. As you can see, with some variation and a couple of outliers like that 32% from MORI, the pollsters are largely showing the same pattern with the Conservatives, everyone has them at either around or slightly below the 37% they got at the election.
The big difference is the Lib Dems, where most companies have them around 10% or 11%, YouGov marginally lower, normally on 9% and ICM quite drastically higher, on 17%. For Labour, most companies have them between 37%-40%, the exceptions being YouGov who have them in the low 40s, and ICM who have them at 36%.
Now, one major reason behind the difference is topline adjustment – what happens to people who say “don’t know” when asked how they’ll vote. Pollsters treat these people in different ways. At one end of the scale, YouGov just ignore them; YouGov figures are based solely upon people who say how they would vote in a general election. Then we get pollsters who ask a squeeze question to people who don’t answer (are you leaning towards any of them, who would you vote for if it was a legal requirement). At the other end of the scale ICM (and, to a lesser extent, Populus) essentially make educated guesses about how these people would vote, and include them in the figures. Evidence from past elections shows that when push comes to shove “don’t knows” are likely to end up voting for the party they voted for at the previous election, hence ICM reallocate 50% of people who say don’t know to the party they voted for in 2010.
Almost by definition, this tends to be helpful to parties that have lost support since the last election, and harmful to those who have gained it. In this month’s poll it had a particularly sharp effect – before adjustment ICM’s topline figures were CON 37%, LAB 39%, LDEM 15%. The effect of reallocating don’t knows was to increase Lib Dem support from 15% to 17%, and decrease Labour support from 39% to 36%, producing a Tory lead.
Of course, that doesn’t explain all the difference. ICM would still be showing the Lib Dems on 15%, significantly higher than anyone else, and there would still be variation between the companies on other factors. There are various other explanatory factors at play – for example, I’ve seen it hypothesized that ICM’s question wording, which says “an election in your area” increases Lib Dem support. I am very sceptical of this explanation, but it is potentially a factor and is worthy of investigation. Another factor is likelihood to vote – most companies weight or filter by how likely people say they are to vote, YouGov only do this during election campaigns, when it reduces their level of Labour support.
Beyond that, there are probably factors connected with weighting – both the targets that people weight towards (primarily what assumptions they make about past vote and false recall) and also when the data is collected (YouGov and ComRes’s online polls can use stored data on panellists, polls conducted by telephone collect the data at the time the survey is conducted).
I draw no conclusions about what is right or wrong. When I first started this blog I always sought to explain the reasons behind the differences and let people make their own decisions, rather than say which polls I thought were right or wrong. In many cases, it is a philosophical difference, a case of polls measuring slightly different things – a YouGov poll is showing how people say they would vote tomorrow, an ICM poll is showing how ICM think those people would vote tomorrow. In other cases, some weighting targets probably are better than others! My sad conclusion after writing about polling for 6 years, however, is that most people tend to believe the poll they like the results from is the one to trust, and subconsciously interpret arguments about methodology to back up that preconception (not, I should add, very different from how we come to our opinions about anything else in life!)