Panelbase's final poll of the campaign :
Should Scotland be an independent country?
Yes 48% (-1)
No 52% (+1)
(UPDATE : The 5% No lead that you might have seen reported from the figures that include Don't Knows has been exaggerated by the effect of rounding. On the unrounded figures, it's...
The apparent swing to No is not statistically significant, and because this takes us back to the position in the last-but-one Panelbase poll (which was conducted at roughly the same time as the famous YouGov poll that had Yes in the lead), the likelihood - but not certainty - is that we're looking at margin of error noise.
Again, when it's so close we're entitled to ask the question - is there any systemic bias in Panelbase's approach that might be understating the Yes vote? Unlike the online pollsters that reported last night, they do correct for the over-representation of English-born people in their raw sample, so we can't use that as an alibi. But at exactly the same time as they introduced country of birth weighting, they also introduced a controversial procedure which reduces the reported Yes vote, and which no other firm uses - namely weighting by recalled European Parliament vote. 252 people who recall voting SNP in May (94% of whom are Yes voters) have been downweighted to count as just 214 people. The logic for doing this is that the Panelbase sample always has a disproportionately high number of people who voted in May, meaning that it's therefore prudent to make sure those people are weighted in line with the result of that election. But the problem is that if you have far too few non-voters from May in your sample, the logical conclusion to draw is either that the sample is hopelessly unrepresentative of the population you are trying to poll, or that a lot of people are not telling you the truth (or some combination of those two factors). That being the case, you can't be sure whether weighting by vote recall is eliminating a bias, or introducing a whole new one.
The fieldwork for this poll is a bit more up-to-date than the three we saw last night - it took place between Monday and today. So it's tempting to see this as a strong clue that we shouldn't expect any significant shift to No in tonight's YouGov poll, not least because on the unrounded numbers the Yes vote is actually slightly higher than it was in the last-but-one Panelbase poll. The problem is, though, that Panelbase's trend has become completely decoupled from the YouGov trend - Yes first reached 48% with Panelbase months ago, at a time when YouGov were still showing them on 39-42%. So it's anyone's guess what the "Kellner Correction" has in store for us this evening.
More details and a Poll of Polls update to follow...