ICM poll : SNP on course to oust Danny Alexander from his Inverness seat next year

In his parting shot upon resigning from the Liberal Democrats, Lord Oakeshott revealed that he had conducted one final constituency poll in a Lib Dem-held seat, namely Danny Alexander's constituency of Inverness, Nairn, Badenoch and Strathspey. The results are not only a horror show for Alexander, but they're also yet another body-blow for Labour, who on paper ought to be the lead challengers in the seat.

Inverness, Nairn, Badenoch and Strathspey voting intention for Westminster general election (percentage changes are from 2010 election) :

SNP 32% (+13)
Labour 25% (+3)
Liberal Democrats 16% (-25)
Conservatives 12% (-1)
UKIP 7% (+6)
Greens 4% (+2)
Others 3% (+1)

Unlike ICM's recent referendum polls, this one was conducted by telephone, so there's no potential alibi for Labour or the Lib Dems of an unrepresentative volunteer online panel. The sample size was 500, which is high enough to be statistically meaningful, and weightings were applied. In fact, the weighting has significantly reduced the reported SNP vote.  The fieldwork is bang up to date - it was conducted between last Friday and Monday of this week.

The poll also asked for hypothetical voting intentions under a variety of possible Lib Dem leaders.  Not a single one produced anything other than a decisive SNP victory.  The only permutation that even got the Lib Dems into second place was Danny Alexander himself as leader - but even the prospect of having their constituency represented by the UK Deputy Prime Minister was not sufficient to make voters turn away from the SNP.  In that scenario, Alexander would lose his seat by 31% to 27%.  Incidentally, the question that specifically asked about the status quo scenario (Cameron, Miliband, Clegg and Salmond all being the leaders of their respective parties at the general election) produced an SNP vote that was 3% higher than in the headline numbers, and a Labour vote that was 2% lower.  It appears, therefore, that reminding voters of the existence of Salmond and Miliband has an entirely predictable effect - so much for the apparent semi-religious belief of "Better Together" that the First Minister must be a liability for the Yes campaign.

Danny Alexander's spokesman has criticised the methodology ICM used in this poll, which in one sense is fair enough because I'll shortly be doing the same thing myself.  Unfortunately, however, the specific objection raised is cretinous beyond belief -

"Only 309 people contributed to the voting intention question – that is less than half of one percent of the 72,500 in the constituency."

That's a reference to the fact that, after turnout weighting was applied, only 309 of the original sample of 500 was left. But the margin of error even for a sample of 309 is only 5.5%, which means that if the rest of the methodology is correct there is still absolutely no way that the Lib Dems are ahead in the constituency (albeit there would be a small degree of doubt over whether the SNP or Labour are in first place). As for the "less than half of one per cent" point, that's even sillier - a far, far higher percentage of the target population was interviewed than would be the case in any nationwide poll with a sample of 1000.

In a strange way, what should be most alarming for Mr Alexander is that the people interviewed were actually quite favourable towards him personally - 52% thought their local MP was doing a good job as Chief Secretary to the Treasury, so this is not a particularly hostile sample. They're not rejecting him, they just seem to have had a gutful of the Lib Dems and the coalition. There appears to be no evidence at all that a strong personal vote (the "local hero" factor) is going to save the Lib Dems' bacon across rural Scotland this time - although I suspect it might just about be enough to save Charles Kennedy in the neighbouring seat.

These findings are of course very much in line with the Ashcroft polling of Lib Dem-held seats a year or two back, which showed that it was the SNP rather than Labour who were in line to take most of the spoils, even in seats where the SNP were not starting from second place. All of these predictions are, however, complicated by the fact that the general election will not be taking place until the other side of the referendum. If there's a Yes vote, I would expect the SNP to sweep all before it across the whole of Scotland, whereas if there's a No vote it's harder to guess what the impact might be. But there's no reason to expect an SNP collapse in those circumstances - No votes in the Quebec independence referendums of both 1980 and 1995 were followed up by clear victories for the Parti Québécois in the subsequent provincial elections.

Talking of the referendum, we're also given something tantalising in this poll - referendum voting intention figures from telephone fieldwork, and produced by a company other than Ipsos-Mori (who to date are the one and only firm to have conducted telephone polling during the campaign). Unfortunately, the numbers aren't very useful given that they're only from a single constituency, for which we don't have any baseline numbers. For what it's worth, though, here they are -

In the referendum on independence for Scotland on 18th September 2014, voters will be asked, "Should Scotland be an independent country?"  Do you think you will vote “Yes” or "No"?

Yes 32.4%
No 54.8%

With Don't Knows excluded, it works out as -

Yes 37.2%
No 62.8%

There are three extremely important caveats here - a) the poll was weighted by recalled past vote from 2010 rather than 2011, which we know tends to lead to a lower reported Yes vote because people's recollection of how they voted in 2010 is much less accurate, b) the referendum question was only asked after many, many other questions (which is fair enough in this case, because it wasn't primarily a referendum poll), and c) the figures are not turnout weighted. If ICM's recent online poll for Scotland on Sunday is anything to go by, turnout weighting might be expected to reduce the No lead by anything up to 4% after Don't Knows are excluded.

So can this tell us anything at all? Well, Alexander's seat makes up approximately one-third of the Highland council area, which in the 1997 devolution referendum produced a slightly lower Yes vote on both questions than the national average. When combined with the lack of turnout weighting, that means this poll is probably consistent with a national Yes vote of about 40% or 41% - significantly higher than Ipsos-Mori have been reporting in their published telephone polls, albeit probably not higher than in their notorious hushed-up mega-poll for the UK government. In any case, this piles assumption upon assumption - it may well be that that the regional differentials are significantly different this time around, and that the Yes campaign are doing proportionately worse in the Highlands than their counterparts did in 1997.  That's only speculation, but what we have much firmer grounds for believing is that ICM's usage of 2010 vote recall for the weightings in this poll is likely to have artificially increased the reported No lead by a substantial amount, in which case the results could be consistent with a national Yes vote of well above 40%.  The fact that people who recalled voting SNP in 2010 were sharply downweighted in the published results from 120 to just 72 is highly significant - that probably only happened because a number of people who voted SNP in 2011 but not in 2010 got mixed up (which in turn means that the SNP's own lead in the constituency is likely to be even higher than the headline numbers suggest).

One interesting detail is that an intention to vote SNP next year correlates significantly more strongly with a Yes vote than a recollection of voting SNP in 2010 does.  In a sense, that's counterintuitive, because you'd expect the lower SNP vote in 2010 to represent the "core nationalist vote" - another reason for suspecting that these are really SNP voters from the 2011 landslide who have been wrongly downweighted.