Tonight's breathtaking findings from the traditionally No-friendly pollster YouGov...
Should Scotland be an independent country?
Yes 51% (+4)
No 49% (-4)
Most media outlets are reporting the figures which take Don't Knows into account as Yes 47% (+5), No 46% (-2), although the YouGov website says Yes 47% (+5), No 45% (-3). I would guess the former is most likely to be correct, in which case the lead on the unrounded figures with Don't Knows excluded is probably only just over 1%.
Nevertheless, a Yes lead is a Yes lead, and there was certainly nothing inevitable about us seeing one of those at any point before polling day. Strictly speaking, this is actually the second time in the long campaign that a polling firm affiliated to the British Polling Council has shown a Yes lead, although it has become part of the orthodoxy to more or less disregard the previous occasion (a Panelbase poll from just over a year ago), because it's assumed that an unusual question sequence distorted the result. Apart from that, we have to go back to well before the start of the campaign to find a poll from a BPC firm that showed a lead for independence - the most recent one was a TNS-BMRB poll in the late summer of 2011.
So does tonight's poll mean that Yes are 'really' in the lead? Not necessarily. Even before you take account of methodological mistakes that a pollster might be making, and also the fact that pollsters can't legislate (or not without difficulty) for respondents sometimes lying to them, there's a standard 3% margin of error that applies to every poll because of normal sampling variation. So tonight's result is still consistent with No having a modest lead, and because of the almost unbelievable scale of the swing YouGov have reported over the last few weeks, you'd have to assume that's actually the most likely scenario.
But then again, even after the last YouGov poll on Tuesday, the significantly lower swing we were pondering at that point seemed implausible to me, and I was therefore fully expecting some kind of reversion to the mean in tonight's poll. That hasn't happened, and that's what really matters, regardless of whether Yes are being slightly flattered by sampling variation. Unless this is an out-and-out rogue poll (only a 5% chance), the worst position for Yes that the margin of error allows is Yes 48%, No 52%, which would still be better than the headline numbers in the previous poll. So we can now say with almost absolute confidence that what we saw on Tuesday was not a fluke, or a fleeting post-debate "Salmondgasm" - it was a real and very hefty swing to Yes.
There is of course an important piece of supporting evidence tonight which might lead us to suspect that Yes are more likely to be on the lower end of the range allowed by the margin of error, and that's the new Panelbase poll commissioned by Yes Scotland -
Should Scotland be an independent country?
Yes 48% (n/c)
No 52% (n/c)
With Don't Knows taken into account, the position is -
Yes 44% (+2)
No 48% (+2)
On the unrounded numbers in the last published Panelbase poll (we know there's been an unpublished poll conducted by the firm in the interim), Yes were actually on 47.6% after DKs were excluded, so it's possible that when we see the datasets we'll find that they've gained a little and have narrowed the gap to the lowest level since Panelbase changed their methodology - but even if that's the case the movement will obviously have been very slight.
If this was a normal campaign, you'd look at the two polls tonight and think they were perfectly consistent with each other, both in terms of the headline numbers and the trend - you'd imagine there has maybe been a 2% swing to Yes, and that due to sampling variation Panelbase have underestimated it and YouGov have overestimated it. But the situation looks very different when you bear in mind the huge disparity there has been between various firms throughout this campaign, and the fact that YouGov have until the last few days been firmly on the No-friendly end of the spectrum, and that Panelbase have been firmly on the Yes-friendly end. So it's an astounding paradox that YouGov have become the first pollster to put Yes in the lead in a credible poll, and that they've done it on the same night that a Panelbase poll still shows No in the lead, albeit only by a very narrow margin.
Let's try to take in the sheer extent of the gulf between the trends shown by Panelbase and YouGov that has been necessary to move us to the unlikely position where YouGov are, as of this moment, the more Yes-friendly of the two. Just over a year ago, YouGov published a notorious poll which purported to show that Yes were on 33% and No were on 67% - that means there has been an 18% swing to Yes since then. By contrast, in late August/early September of last year Panelbase were showing Yes on roughly 44% and No on roughly 56%, which would imply a far, far smaller swing of just 4% to Yes between then and tonight's poll. One month ago, in a poll that partly took place immediately after the first leaders' debate, YouGov put Yes on 39% and No on 61% - meaning there has been a 12% swing to Yes in the space of a few short weeks. The swing suggested by Panelbase over roughly the same timescale is just 2%.
To be fair, part of this disparity can be explained by methodological changes, and especially by YouGov's imperfect moves over the last year-and-a-bit to put their house in order. They ditched their Dodgy Preamble, they moved away from weighting by Westminster-centric target figures for party identification (although that dreadful practice was directly succeeded by the No-friendly "Kellner Correction"), and much more recently they introduced weighting by country of birth. All of those changes are likely to have boosted Yes somewhat. Over the same period, Panelbase have also made some modest methodological changes, the net effect of which may well have been slightly of assistance to No. But even these moves to promote a degree of convergence can't explain the bulk of the extra swing to Yes reported by YouGov.
So what in the name of Foulkes is going on? The most likely explanation is that the huge swing to Yes is something that is specific to YouGov, and that won't be fully replicated by other pollsters - or at least not by the other online pollsters. The reason is that YouGov are effectively measuring something different to the others. Specifically, there are two factors that set them apart from their three online rivals. Most obviously, there's the Kellner Correction, which in practical terms leads to a sharp upweighting of the small group of respondents who voted Labour in 2010 but switched to the SNP in 2011. In Tuesday's poll, that group was upweighted two-fold. If the pro-Yes swing is concentrated in that group, the Kellner Correction will obviously be magnifying it, and therefore nothing on the same scale will be seen in the findings of the other online pollsters. The other factor is that YouGov just seem to have far more Labour supporters on their books than the other firms do - that was one of the points Survation made in response to Peter Kellner's infamous diatribe. So, again, if the swing is particularly significant among Labour supporters (as it seems to be), it would look bigger in YouGov's results.
The fact that Panelbase and Survation have published polls since the second debate which have been very good for Yes but haven't shown anything like such a big shift is consistent with the above theory. However, there is an alternative possibility - which is that it's Panelbase which is different, and that we can therefore expect the other pollsters to be more in line with the trend shown by YouGov. On the face of it, it's much harder to make the case that Panelbase are measuring something different to the others, because their methodology and the composition of their panel is much more in line with their online rivals ICM and Survation. But in fact there is quite a bit of evidence that Panelbase have in the past produced trends that don't tally up with the overall picture. In retrospect it's absolutely clear that there was a significant swing to Yes over the winter, and yet Panelbase were literally the only BPC pollster that didn't detect that at all. Even when they did belatedly show movement in the spring, it wasn't on anything like the same scale that ICM and TNS-BMRB had picked up during the winter. OK, it could be argued that the other firms were just catching up with a strength for Yes that was factored into Panelbase's figures from the word go - but why would that have been the case? What makes Panelbase's sample so distinctive? Whatever it is, that could - I only say could - be the explanation for why Panelbase are failing to show further progress for Yes this evening,
One other point intrigues me - why did Yes Scotland actually release this poll, given that it didn't show any further clear-cut breakthrough? I can think of four possibilities -
1) They felt under pressure to do it, after the conspiracy theories started to mount about why they hadn't published last week's poll. (And unfortunately we have to take our share of the blame for that.)
2) They were keen to highlight the swing to Yes among women in the poll.
3) They had heard about the YouGov poll, and decided to use the Panelbase poll to dampen down the hype, because being seen as a slight underdog works in their favour. If there's any truth in that, it would mean Kenny Farquharson missed the point in the most deliciously ironic of ways - when he first heard about the poll results, he wrote a tweet accusing the Yes campaign of poor expectation management. It could be that the fact that he heard about it in the first place was in itself part of a clever expectation management strategy.
4) They knew that the simple act of releasing the poll would help Yes in the Curtice Poll of Polls, which is based on a crude average of the last six polls to be conducted, regardless of which firms conducted them. I haven't been able to get onto Professor Curtice's blog so far this evening, but I would expect to find that he'll have Yes at 47% in his new Poll of Polls - a higher figure than in the Scot Goes Pop Poll of Polls, which uses only one poll per firm.
* * *
Swing required for 1 out of 6 pollsters to show Yes ahead or level : 0.0%
Swing required for 2 out of 6 pollsters to show Yes ahead or level : 2.0%
Swing required for 3 out of 6 pollsters to show Yes ahead or level : 3.0%
Swing required for 5 out of 6 pollsters to show Yes ahead or level : 4.5%
Swing required for 6 out of 6 pollsters to show Yes ahead or level : 7.0%
* * *
SCOT GOES POP POLL OF POLLS
I've got a slight problem in calculating this update of the Poll of Polls, because of the two different versions I've seen of the YouGov numbers with Don't Knows taken into account. For the time being, I'll assume it's a one point lead for Yes, but obviously if that's wrong I'll have to correct the figures tomorrow. (UPDATE : It turns out the Yes lead was two points, not one, so I have indeed had to correct the figures below - the average No lead is now 0.2% lower still.)
MEAN AVERAGE (excluding Don't Knows) :
Yes 45.9% (+0.9)
No 54.1% (-0.9)
MEAN AVERAGE (not excluding Don't Knows) :
Yes 40.5% (+1.2)
No 47.8% (-0.2)
MEDIAN AVERAGE (excluding Don't Knows) :
Yes 45.7% (n/c)
No 54.3% (n/c)
(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 six - YouGov, TNS-BMRB, Survation, Panelbase, Ipsos-Mori 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.)