Why we should maintain a healthy scepticism about tomorrow's Ashcroft constituency polling

11am tomorrow (Wednesday) will herald what in some people's minds is the biggest polling event of the year so far, namely the release of the endlessly-hyped Scottish constituency polls from Ashcroft.  In particular, Mike Smithson over at Stormfront Lite has been investing almost millennial significance in them for the last few months.  I think that's slightly silly - as far as I'm concerned, Scotland-wide polls are more important.  I'll certainly be taking a keen interest in the Ashcroft results, but the following big caveats should be borne in mind...

1) Constituency polling has a very patchy track record.  There's no mystery over why that should be the case - the methodology for national polls can be honed through trial and error, but most of the constituencies we'll be hearing about tomorrow have never been polled in this way before.

2) Ashcroft may weight his results by 2010 recalled vote.  We don't know that for a fact, but it seems conceivable that he'll go down that road in order to keep his methodology consistent with his English constituency polling.  In Scotland-wide polls, it's become accepted good practice to weight by 2011 recalled vote, which we know is much more accurate than 2010 recall.  However, it's not really possible to do that in constituency polls, because nobody knows exactly what the 2011 result was in individual Westminster constituencies (with the exception of the Western Isles and Orkney & Shetland).  That leaves Ashcroft with two basic options - to weight by 2010 recall, or to follow the Ipsos-Mori practice of not weighting by recalled vote at all.  He should do the latter - and if he doesn't, the results will be unreliable, and in all probability will underestimate the SNP.  For the avoidance of doubt, that point will hold true even if the results appear to be very favourable for the SNP.

3) Ashcroft unwisely asks two voting intention questions, and headlines the results of the second question.  In a constituency poll, it's a very good idea to ask respondents to think about how they will vote in the context of the local seat and the local candidates - but that question should be asked upfront.  By asking it second, and immediately after a differently-worded voting intention question, you're inviting people to think they're "supposed" to give two different answers.  If there are several constituencies in tomorrow's polling that produce completely different results on the two questions, then alarm bells should be ringing.  The media will dutifully follow Ashcroft's edict that only the second question matters - but they'll be wrong.  The true position may well be somewhere in between the two results.

I should make clear that I have absolutely no idea at this stage what the polling will show - it may be good, bad, or wonderful for the SNP.  Ashcroft has been having a whale of a time dropping hints, but a lot of them have probably been red herrings.  So I just thought it might be worth airing my concerns well in advance, so there's no question of raising convenient objections to a disappointing result, or falling silent about the flaws in an encouraging poll.

UPDATE : Literally within seconds of me saying that I had no idea what the results were, I was pointed in the direction of what appear to be the results (the Gordon result is bang in line with what Cochrane led us to expect).  The person who sent me them suggested that I shouldn't report the numbers in case they turn out to be wrong.  But suffice to say that Danny Alexander's career prospects are not looking too hot at the moment, even allowing for a huge margin of error.

UPDATE II : As the apparent Ashcroft results have been posted and discussed in the comments section below, I've just gone ahead and published the full numbers in a fresh post HERE.