New EU referendum polls from ICM and YouGov show a virtual dead heat

The weekly ICM referendum poll is out, and on the face of it, the results look pretty routine, slotting in more or less in the middle of ICM's 'normal range' -

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

Remain 44% (-1)
Leave 43% (n/c)

However, for the second poll in a row, ICM have made a methodological change, so the headline numbers aren't directly comparable with anything that has gone before. In the last poll, it was possible to see that Remain would have been roughly four points ahead (without a distortion caused by rounding) under the old methodology, leaving open the possibility that there had been a genuine swing to Remain in the immediate wake of the Brussels attacks. This time, it's slightly tricky to work out the unadjusted trend, because the format of the datasets has changed to take account of the methodological shift, and there are one or two things there that don't really make sense. For example, we're told that the headline results are filtered by voter registration - and yet as far as I can see respondents aren't asked at any point whether they are registered to vote. Possibly ICM already have that information from questions previously asked to their polling panel - but in that case, why is that particular filter necessary? Why not simply invite only people who are registered to vote? I'm sure there's a straightforward explanation, but it's not immediately apparent. We're also told that two sets of results are filtered to only include those who are "certain to vote" - but a) that makes a nonsense of the whole concept of "turnout weighting", and b) the numbers don't tally up anyway. Only 1255 respondents say they are absolutely certain to vote, and yet 1819 respondents are left in the first filtered sample, and 1625 in the second.

For what it's worth, though, it's hard to see how Remain would be further than 2-3% ahead if the changes hadn't been introduced, so whichever way you cut it, this poll is slightly better for Leave than the last one from ICM.  The improvement is not statistically significant, but it somewhat reduces the chances that the apparent swing to Remain last time around was genuine.

The disparity between Scotland and the rest of the UK is particularly gigantic this week - the Scottish subsample figures are Remain 60%, Leave 28%.

*  *  *

Sod's law - I was just about to hit 'publish', and then I suddenly noticed there's a new YouGov poll out as well...

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

Remain 39% (-1)
Leave 38% (+1)

This one does look potentially significant, because the razor-thin Remain lead is lower than the 2-5% range we saw in the last four YouGov polls.  Even if the slight progress for Leave is just an illusion caused by sampling variation, that could mean that YouGov's current 'normal range' isn't quite as good for Remain as we had thought until today.

The Scottish subsample figures are : Remain 51%, Leave 32%.

*  *  *

SCOT GOES POP POLL OF POLLS

I've been thinking for some time that I should reduce the time-frame covered by each update of the Poll of Polls, because one month seems a bit too long at this stage of the campaign.  Ideally, I'd like to cut it to a fortnight, but at the moment that would leave us with just one telephone poll in the sample.  So I'll compromise and go with three weeks for now, although at some point it'll hopefully be possible to bring it down to two, as polls are likely to become much more frequent as referendum day approaches.

50/50 ONLINE/TELEPHONE AVERAGE :

Remain 45.2% (+0.9)
Leave 41.4% (+0.5)

ONLINE AVERAGE :

Remain 41.9% (+0.9)
Leave 42.4% (+2.2)

TELEPHONE AVERAGE :

Remain 48.5% (+1.0)
Leave 40.3% (-1.2)

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