You might remember that when the sensational Ipsos-Mori poll came out last week showing a Yes majority, I suggested that what we really needed now was an independence poll from TNS - the only other firm that uses a 'real world' data collection method (ie. not volunteer online panels) for polling on Scottish affairs. Right on cue, here is the first TNS independence poll since the referendum. The timing is just a coincidence - TNS polls take a long time to conduct, so it was set in train well before the Ipsos-Mori result emerged.
Should Scotland be an independent country?
This is an absolutely crucial development. It never seemed likely that the Ipsos-Mori poll was an 'outlier' in the sense that the Yes lead was an illusion caused purely by sampling issues, but if TNS had replicated the small No lead reported recently by the online firms, we might have concluded that Ipsos-Mori's distinctive methodology was producing questionable results. As it is, it looks much more like a straight split has opened up between online and 'real world' firms - with the former mostly showing a No lead and the latter showing a Yes lead.
There are a couple of caveats, however. Firstly, both the Ipsos-Mori and TNS polls were conducted more recently than any of the online independence polls, so it's possible there has been a sudden swing in favour of independence over the last few weeks, and that the next batch of online polls will show much the same thing. I think that's highly unlikely, but the possibility can't be entirely discounted, because neither Ipsos-Mori nor TNS have previously produced post-referendum independence polls, meaning we have no baseline to work from.
Secondly, it's unfortunately the case that it isn't just the 'real world' status of TNS and Ipsos-Mori that distinguish them from the online firms. Neither of them weight their results by recalled referendum vote, which could theoretically explain why their results are so different. However, TNS do (unlike Ipsos-Mori) use a form of past vote weighting, based on the election that ought to be freshest in people's minds - this year's general election. So there's no reason to automatically assume that recalled referendum vote weighting would make a massive difference to the headline figures.
If the divergence really is being caused primarily by data collection method, it could be hugely significant. Intuitively, most of us would probably prefer to put our trust in polls that use a completely fresh sample, rather than relying on respondents who have volunteered to be polled over and over again, and who are more likely to be politically committed. In which case, instinct might lead us to conclude that Yes are more likely to be ahead at the moment.
Although Ipsos-Mori and TNS are showing a very similar state of play, the historical significance of each of the two polls is very different. Ipsos-Mori have never shown support for independence at anything like the current level, whereas TNS are returning to familiar territory in showing a Yes lead - it's hard to believe given their reputation as a No-friendly pollster during the referendum campaign, but if you go back a few years there was a long spell when they were the only firm that occasionally showed a majority for independence, before a sudden and catastrophic drop in Yes support that wasn't replicated (not in full, anyway) by other pollsters. The TNS methodology has evolved over the years, though, so the long-term trend they've shown may be somewhat less meaningful than the one shown by Ipsos-Mori.
TNS have also provided voting intention figures for next year's Scottish Parliament election...
Constituency ballot :
SNP 58% (-4)
Labour 23% (+3)
Conservatives 12% (n/c)
Liberal Democrats 5% (+2)
Regional list ballot :
SNP 51% (-3)
Labour 24% (+4)
Conservatives 11% (-1)
Greens 6% (-2)
Liberal Democrats 6% (+2)
The drop in SNP support may look slightly alarming, but in fact on the list vote it's not significant at all. 51% is identical to the party's share on the list in the TNS poll of two months ago. It looks like the jump to 54% last month was just a quirk caused by sampling variation. On the constituency vote it's harder to say - 58% is (unbelievably) the SNP's worst showing in a TNS poll since the general election, but they're still only 2% lower than in the two polls prior to last month. So we could simply be looking at margin of error 'noise', or the SNP may have slipped slightly - but probably not by as much as the 4% drop suggested by the headline figures.
It's more clear-cut in Labour's case - their share of both the constituency and list vote is significantly higher than in any previous post-election TNS poll, so they probably have genuinely recovered to some extent. There have been some suggestions that this is a 'Kezia bounce', but I'm extremely sceptical about that - I think it's more likely to be the Corbyn effect finally making itself felt. That may come to an abrupt end on Saturday if Andy Burnham or Yvette Cooper upset the odds - if not, the next few months could be highly unpredictable.
It's interesting that the Holyrood figures from Ipsos-Mori last week were interpreted as wonderful for the SNP and dreadful for Labour, because the TNS trend suggests that Ipsos-Mori might have shown the SNP even higher and Labour even lower if they had conducted polls in June or July.
Last but not least, there's severe cause for concern in the TNS poll for the Tories. After Ruth Davidson led her party to a disastrous all-time low of 14.9% at the general election, you might have thought the only way was up - but instead they've slipped further. On the constituency ballot, they've dropped from 15% to 12% since the first post-election TNS poll, and on the list ballot they've slipped from 14% to 11% over the same period. Unless she can turn things around, Davidson may end up looking like the most over-rated political leader since...oooh, Jim Murphy, probably.