New Survation poll :
Should Scotland be an independent country?
Yes 48% (+1)
No 52% (-1)
So all three of tonight's polls are showing a 'statistical tie', meaning that the margin of error leaves open the possibility that either side are in the lead. However, that only holds true if you look at each poll in isolation - if you take the three polls together, it becomes much more likely that No have a slender lead, albeit only if the polling firms are getting their methodology roughly right. The latter point has been the million dollar question all along.
Malevolent John Rentoul, he of the colonialist demands that Scotland adjusts its national identity to suit his own personal tastes, is claiming tonight to have "evidence" that the polls are systemically biased towards Yes. He says that he's spoken to someone who saw the Yes campaign's canvass returns up to Thursday of last week, and that they show Yes to be much further behind. I'm not going to completely dismiss this idea out of hand, because it's perfectly plausible that the polls are systemically biased in either direction, and although Rentoul is a thoroughly objectionable individual, he does at least tend to apply some rigour to his journalism. But there's something really odd about the claim, and perhaps someone with more knowledge of how canvassing is collated and analysed can offer a potential explanation. The one thing we know about canvassing is that people don't tell the whole truth, so Yes canvassers will get better results for Yes and No canvassers will get better results for No. It does therefore seem fantastically improbable to me that the Yes canvassing returns are worse than the results of two very recent scientifically-conducted telephone polls, including one commissioned by the No campaign itself (ie. the 46.5% Yes vote reported by Survation on Saturday). Could there be some sort of time-lag with the canvassing figures, with old data only being very gradually replaced as people go round houses again? Or would the centrally-collated figures always be bang up-to-date? Are the numbers somehow adjusted to take account of people's dishonesty, so that a best guess of the true position can be arrived at? Could it be the adjusted guesstimate that Rentoul is talking about, rather than the raw returns which surely must be better for Yes than he suggests?
Certainly on the face of it, there's a lot that doesn't quite ring true about the story, so although it's unsettling, we probably just have to work on the assumption that it's wrong (or that Rentoul is guilty of wishful thinking in the way he's interpreted the information). So let's look at it the other way around. Is there any obvious reason for supposing that tonight's three polls might be systemically biased in favour of the No campaign? Yes, there is, and it's a drum I've been banging for a very long time. As far as we know, not one of tonight's pollsters has weighted by country of birth. It's been long-established that ICM have too many English-born people on their panel, and although we don't know that the same applies to Survation and Opinium, it seems a reasonable assumption that it might do, given that it's a problem that's been uncovered by three different online polling firms. If so, weighting by country of birth would only help Yes by a small amount - but that's OK, because they're only behind by a small amount!
As already stated on the previous post, there's a particular issue with Opinium's decision to weight by recalled 2010 vote, which other pollsters used to do, but stopped after it became clear that a lot of people weren't recalling accurately. Although the effect is diluted by the fact that Opinium additionally weight by 2011 vote recall, the net result of the two weightings is a reduction in the Yes vote. If they only weighted by 2011 recall, in line with the practice of all but one of the other firms, the Yes vote would be considerably higher - and indeed there would probably be a Yes lead.
Sticking with the theme of what would happen if you applied one firm's methodology to another firm's data, Survation would be showing an even tighter race if they applied Opinium's very strict turnout filter that only includes people who are 10/10 certain to vote...
Survation (absolutely definite voters) :
There's the perennial problem with Survation that they've had to severely upweight under-25s and respondents from the South of Scotland electoral region. When trying to guess how that might have distorted the overall results of the poll, what you're really hoping to see is (paradoxically) a very low Yes vote among those groups, because that maximises the chances that the overall Yes vote has been underestimated due to sampling variation. In this case, it's a middling picture - the Yes vote among under-25s doesn't particularly look too high or too low, while the Yes vote in the south is maybe on the high side, but not necessarily by that much.
* * *
Swing required for 2 out of 7 pollsters to show Yes ahead or level : 0.5%
Swing required for 5 out of 7 pollsters to show Yes ahead or level : 2.0%
Swing required for 6 out of 7 pollsters to show Yes ahead or level : 2.5%
Swing required for 7 out of 7 pollsters to show Yes ahead or level : 7.0%
* * *
SCOT GOES POP POLL OF POLLS
Yes only slip back in this update of the Poll of Polls due to the last ICM poll (which now almost certainly looks like an extreme outlier) being replaced in the sample. But we presume that Yes are still being understated due to the fact that an ancient Ipsos-Mori poll is still being taken into account. That problem should finally be rectified tomorrow - unless of course Ipsos-Mori's No-friendly status persists and they fail to converge in the way that YouGov and TNS have.
MEAN AVERAGE (excluding Don't Knows) :
Yes 47.4% (-0.7)
No 52.6% (+0.7)
MEAN AVERAGE (not excluding Don't Knows) :
Yes 42.7% (-0.9)
No 47.4% (+0.3)
MEDIAN AVERAGE (excluding Don't Knows) :
Yes 47.8% (-0.1)
No 52.2% (+0.1)
(The Poll of Polls is based on a rolling average of the most recent poll from each of the pollsters that have been active in the referendum campaign since September 2013, and that adhere to British Polling Council rules. At present, there are seven - YouGov, TNS-BMRB, Survation, Panelbase, Ipsos-Mori, Opinium and ICM. Whenever a new poll is published, it replaces the last poll from the same company in the sample. Changes in the Poll of Polls are generally glacial in nature due to the fact that only a small portion of the sample is updated each time.)