The polls have been largely static for the last three months, showing Conservative leads between 13 and 24 points.
Written like that, it doesn’t sound so static does it? Say the position was closer between the main parties, no one would describe a situation where polls were showing results varying between a 6 point Labour lead and a 4 point Tory lead as consistent.
The reality though is that they are. My first sentence did rather deliberately make things look volatile by focusing on the lead, which since it includes variation in both the Labour and the Conservative share of support falsely magnifies small differences. It’s more accurate to say that recent polls have shown the Conservatives at 41%-48%, Labour at 24%-29%, the Lib Dems at 15%-20%.
Over at PoliticalBetting Bob Worcester puts it all down to normal sample error. Looking at those bands of support that seems to fit, especially if you dismiss the one Populus poll that had the Tories down at 41% as an outlier. The reality is however that this is an overly simplistic explanation for the variation.
If it was all down to random variation we’d see some YouGov polls showing a low Tory lead, some Populus ones showing a high lead. It would – as you would expect – be random. We don’t see that. We see some pollsters consistently showing a lower share for the Conservatives and higher share for the Lib Dems than other pollsters.
Taking polls since the start of June:
YouGov average is CON 46.4%, LAB 25.8%, LDEM 16.6%
Ipsos MORI average is CON 46.7%, LAB 26.3%, LDEM 16%
ComRes average is CON 45.2%, LAB 25.7%, LDEM 16.8%
All very close, but…
Populus average is CON 43%, LAB 26.7%, LDEM 19%
ICM average is CON 43.8%, LAB 27.4%, LDEM 19%
If the difference was down to sample error, which is as likely to go one way as the other and affects all pollsters, then given enough polls the average from each pollster should be much the same. Clearly they are not – Populus and ICM are showing a lower level of Conservative support and a higher level of Lib Dem support. This is only three months data, so could easily be chance, but regular readers will know that these are actually long term trends and suggest that the contrast in the polls is due to methodological differences between the pollsters.
Part of this is because ICM and Populus are measuring slightly different things from the other pollsters. YouGov, MORI and ComRes’s figures are based on how people say they will vote. ICM and Populus’s also take into account how ICM and Populus think people who say don’t know will vote, and at this moment in time this cuts the Conservative lead. The rest is likely to be down to differences in their respective weightings.
That doesn’t change the central point that the polls are NOT all over the place, they have been very stable over the last couple of months. The broad picture we have is not, however, of polls all agreeing with each other with differences explained solely by sample error, they have slightly different ways of doing things that produce slightly different figures…albeit, each set pretty stable when compared to themselves.