The new ComRes poll for the Independent paints a completely upside-down picture of Scottish public opinion compared to the poll just a few days ago for the Independent on Sunday. Here are the full figures from the Scottish subsample -
SNP 32% (+20)
Conservatives 24% (+3)
Labour 22% (-7)
Liberal Democrats 10% (-19)
Others 12% (+2)
I had an exchange a few weeks ago with people who felt that it was absurd to take subsamples of such a small size even vaguely seriously. This is one instance that would appear to bear out that argument, because clearly the changes of support for the SNP and Liberal Democrats in particular are thoroughly implausible. It's also incredibly unlikely that the Labour party are really in third place at the moment (although second place would not be entirely surprising). However, when you look at these subsamples over a period of months it's actually striking how rare such freakish findings are - the SNP and Labour nearly always occupy the top two places, while the Conservatives and Liberal Democrats are nearly always in third and fourth. So while subsample figures need to be treated with extreme caution - due not only to small sample sizes but also to the fact that the figures may not be properly weighted - it's clear they're not completely meaningless, or 'just a bit of fun' as some people put it in Snow-esque fashion.
Almost nothing can be learned from looking at one subsample in isolation, given the mammoth margin of error, but it seems to me that if you look at a pattern of several of them it's possible to get a feel for what is going on in public opinion. However, others take a different view, and from the SNP's perspective it may be just as well they do - since by definition Scottish subsamples of UK-wide polls relate only to Westminster voting intention, and understate the party's likely performance at Holyrood elections.
It might also be worth mentioning why I first started looking closely at subsamples in the first place. Just under a year ago, Mike Smithson rather astonishingly devoted an entire thread at his excellent PoliticalBetting.com site to a single Scottish subsample of fewer than 100 people, which he used as supporting evidence for his theory that the Labour vote in Scotland was holding up dramatically better than elsewhere in the UK, meaning that it could be reasonably inferred that the potential swing to the Tories in English marginal seats was being significantly underestimated by UK-wide polls. Given that the posters on PB.com are overwhelmingly Tory-supporting, this theory has unsurprisingly entered the site's mythology, and it is periodically referred back to without much reference to hard figures. And, as the insufferable Dennis MacShane demonstrated on Newsnight last week, it's amazing how long a mythology based on one set of dubious figures can go completely unchallenged, if it never occurs to anyone to check up on it.
In this particular poll, we certainly see no evidence of the Labour vote holding up better in Scotland - they are seventeen points down on their 2005 performance in Scotland, but only nine points down across Great Britain as a whole.