Cracking ComRes poll of all Labour-held seats in Scotland shows gigantic 19% swing to the SNP

The headline pretty much contains all the information I have at the moment, but I'll update this post when more details are made available.  A 19% swing is (unbelievably) a touch lower than the Ashcroft constituency polls were generally showing, although that may simply be because the swing isn't quite as huge in affluent Labour-held areas that Ashcroft hasn't covered yet.  It would still be enough to cause absolute carnage.

If there was a uniform swing of 19% across the whole country, that would imply a national SNP lead of approximately 16% - exactly the same as suggested by the recent ICM poll.  However, you'd expect the Labour-SNP swing to be somewhat lower in seats that Labour aren't competitive in, so the fact that ComRes have only surveyed Labour-held seats would - on the face of it - indicate that they're picking up a slightly lower SNP lead than most other pollsters.

There's a big caveat here, though.  Most Scotland-wide polls are sensibly weighted by recalled vote from the last Holyrood election, and from the referendum.  One thing we'll have to look out for is whether ComRes have made the same mistake that Ashcroft did, and weighted their results to recalled vote from 2010 - a procedure that we know is wildly unreliable.  They may have felt they had no option but to do that, because they wouldn't have been able to match 2011 Holyrood vote recall to the correct boundaries.  But the reality is that if 2011 weighting isn't possible, it would be much better to simply dispense with past vote weighting altogether, because otherwise there's a severe risk of underestimating the SNP's lead.

UPDATE : According to the Daily Mail, these are the voting intention numbers in the ComRes poll.  The percentage changes are from the 2010 results in Labour-held seats.

SNP 43% (+24)
Labour 37% (-14)
Conservatives 13% (-1)
Greens 2% (+1)
UKIP 2% (+1)
Liberal Democrats 2% (-12)

Don't be startled by the seemingly 'narrow' lead - you'd see much the same thing if other polls were restricted to Labour-held seats.  The wonders of the first-past-the-post electoral system can easily translate these sorts of figures into something approaching a clean sweep of seats if the votes for the leading party are sufficiently evenly spread - and that's exactly what we suspect may be happening.  The recent ICM poll suggested that the swing was highest in the Labour heartland seats which are toughest for the SNP to win, and lower in No-voting seats where Labour are starting from a more vulnerable position, due to a split unionist vote.

ITV, who commissioned the poll, are suggesting that the SNP would take 28 Labour seats, while Labour would hold the remaining 12.  But that's based on a uniform swing, and almost certainly underestimates the SNP's potential gains.  Labour's hopes of achieving respectability in defeat depend on cutting the SNP's lead, not on getting their prayer-mat out and hoping for a uniform swing that simply isn't going to happen.

A crude look at the percentage change figures would suggest that the SNP are hoovering up almost all of the lost Labour votes, and almost all of the lost Liberal Democrat votes - ie. the SNP vote is up by far more than the Labour vote is down.  But it may not be quite as simple as that - the datasets will hopefully tell us if underlying movement from the Liberal Democrats to Labour is being disguised by the gargantuan swing from Labour to the SNP.  That's what appeared to happen in the 2011 election.

UPDATE II : I've now had a chance to look at the datasets, and they are truly bizarre.  ComRes have indeed committed the cardinal sin of weighting by recalled 2010 vote - but when they asked people how they voted in 2010, they didn't even give the SNP as an option!  The only options were Labour, the Conservatives, the Liberal Democrats, or "some other party".  That approach will take quite a bit of justifying, given that the SNP outpolled both the Tories and the Lib Dems in 2010.  (Even in 2010.)

I can think of two possible explanations for what ComRes have done - either a) stupidity, or b) it's a cunning plan to try to make people less confused over how they voted in 2010, ie. if you don't even offer the SNP as an option, people who voted Labour in 2010 but then switched to the SNP in 2011 might be more likely to answer the question accurately.  You'd have to be ultra-charitable to think that was the plan, but even if it was, it doesn't appear to have worked to any great extent.  Respondents who say they voted for "some other party" have been downweighted from 264 to 207, meaning that the SNP's position may well have been underestimated due to false recall.

At the very least, the use of 2010 weighting means that this poll is not directly comparable with any full-scale Scottish poll, and so the superficial appearance that it may be marginally less awful for Labour than the results we've seen from other firms is fairly meaningless.

The point I made above about where the Liberal Democrats' lost votes are going is borne out by the datasets - fractionally more Lib Dem voters from 2010 are planning to vote Labour than for the SNP, but that is being offset by a much bigger shift from Labour to the SNP than the headline numbers would suggest.

Perhaps the most devastating detail of the poll from Labour's point of view is that just 33% of respondents say that Labour is the party they most closely identify with, irrespective of their current voting intention.   The SNP are only just behind on that measure, on 31%.  Remember these numbers are only from seats that Labour won in 2010, in most cases by a massive margin.  It really does look as if a large number of ex-Labour voters are not merely "on holiday", but have undergone their own personal revolution, and as a result are not coming back any time soon.