Wednesday, February 17, 2016

Bizarre split opens up between two telephone pollsters on the EU referendum trend

Did you imagine the EU referendum polling couldn't possibly get any stranger than the enormous gulf we've already seen between telephone and online firms? Think again. At least all firms have until now generally shown the same basic trend - when online polls were showing a steady state of play with the two sides roughly level-pegging, telephone polls were showing a steady state of play with Remain in a big lead of around 20 points. Recently, the two regular telephone pollsters (Ipsos-Mori and ComRes) have been showing almost identical figures, so when ComRes agreed with the online firms that there had been a significant shift to Leave in the wake of Cameron's failure to secure a credible deal, I would have bet quite a lot of money on the next Ipsos-Mori poll showing at least some decline in the Remain lead. But, remarkably, it hasn't - or not a statistically significant one at any rate.

Should the United Kingdom remain a member of the European Union or leave the European Union?

Remain 54% (-1)
Leave 36% (n/c)

Barring a methodological tweak (which there doesn't seem to have been) it's very difficult to reconcile a no change poll from Ipsos-Mori with a poll from ComRes showing the Remain lead collapsing by 10%. The simplest logical possibility is that there was some sort of bounce-back for Remain after the ComRes poll was conducted, but in fact there was only a two day gap between the completion of the fieldwork for the two polls, and it's hard to think of anything that dramatically changed in that short period.

The most important methodological difference between the two firms is that ComRes weight by past vote recall, and Ipsos-Mori don't. You'd expect that to lead to Ipsos-Mori having a little 'house bias' in favour of Remain, because there might be too many Labour-voting Remain supporters in the sample, but it shouldn't really distort the trend. (It might affect the trend to a small extent if, for example, most of the swing to Leave was concentrated among Tory voters who are under-represented in the Ipsos-Mori sample, but the impact would be fractional.)

Intuitively, the most plausible explanation is that both firms have produced extreme results due to normal sampling variation - ie. ComRes have Leave too high, Ipsos-Mori have Leave too low, and if they both polled again they'd be likely to get results somewhere in the middle. If that's correct, it would mean there has indeed been a meaningful swing to Leave, but not quite as dramatic as we thought the other day.

Looking at the datasets, the 'surplus' Remain supporters that Ipsos-Mori are picking up seem to be very heavily concentrated among under-45s.  In the older age groups, they're actually showing a very similar position to ComRes, or in some cases a slightly better position for Leave.  But when you get down to the 35-44 age group, suddenly Ipsos-Mori have a gargantuan 67% to 25% split in favour of Remain, which compares to a tiny 45% to 43% Remain lead in the ComRes sample for the same group.  The somewhat less extreme generational gap that ComRes are showing 'feels' more realistic somehow (which would be good news for Leave), but who knows?

By contrast, political allegiance doesn't offer any clue at all to the divergence between the two firms - Ipsos-Mori are reporting better figures for Remain across the board among Conservative, Labour and Liberal Democrat voters.

So the plot has just thickened even further.  Depending on who you believe, Remain could have a huge lead (Ipsos-Mori), a modest lead (ComRes), they could be locked in a statistical tie (ICM), or they could be several points behind (YouGov).  Take your pick.

*  *  *


Given that the most recent online polls have shown movement towards Leave, it might seem odd that there has been a tiny net swing back to Remain in the online average of the new Poll of Polls - albeit not enough to knock out the Leave lead on that measure.  It's a slightly artificial effect caused by two relatively good polls for Leave from mid-January dropping out of the sample.

Should the United Kingdom remain a member of the European Union or leave the European Union?


Remain 47.5% (+0.3)
Leave 39.9% (-0.1)


Remain 42.0% (+0.4)
Leave 42.4% (+0.1)


Remain 53.0% (+0.3)
Leave 37.3% (-0.4)

(The Poll of Polls takes account of all polls that were conducted at least partly within the last month. The online average is based on eight polls - four from ICM, two from YouGov, one from BMG and one from ORB. The telephone average is based on four polls - two from ComRes and two from Ipsos-Mori.)


  1. Worth a read as an unbiased view

  2. Any chance of seeing your polls in graphical form James? There are a lot of free tools now that would make that easy to set up and maintain.