As you may have seen, Sky News made an exasperating attempt last night to distort the referendum debate with a wildly unreliable online "poll", conducted by themselves and using their own paying customers as respondents, which purported to show that Theresa May is significantly more popular in Scotland than Nicola Sturgeon - a finding that is light-years out of line with recent properly-conducted polls from firms affiliated to the British Polling Council.
The numbers were so patently absurd that I didn't think they even warranted the effort of a blogpost, although I did post a tweet pointing out that it was a junk poll and should be disregarded. That prompted a reaction this morning from Harry Carr, head of "Sky Data", who insisted that it was a nationally representative poll. He also produced datasets which showed that some weighting had been done - but there was no obvious sign of many of the political weightings that are standard in Scottish online polls, such as by recalled indyref vote, recalled EU ref vote, or recalled Holyrood vote.
As the poll bears all the hallmarks of having far too many Leave voters in the sample, I asked Harry a simple question - had he weighted by EU ref vote? His answer : No. He had weighted by recalled 2015 general election vote, but not by EU ref vote. I strongly suspect that also means he didn't bother weighting by recalled Holyrood vote or by recalled indyref vote.
There's your explanation for the poll's nonsense results right there. A telephone or face-to-face poll can possibly get away without doing proper political weighting, and just relying on demographic weighting. But an online poll - no chance. Online polls are different because you know in advance that you're drawing from an unrepresentative pool of potential respondents - Sky customers, for example, may well have political leanings that are different from non-Sky customers, and you have to carefully weight to correct for that. I literally can't think of a single British Polling Council firm that would have attempted to conduct an online poll in the way that Harry has done. It also looks like he didn't bother weighting by country of birth - a failing that he has in common with only BMG.
I stand by my original assessment - junk poll, ignore.
By the way - here's a challenge for Harry. Conduct a Scottish voting intention poll, and publish it even if it puts the Tories in the lead. Then try to keep a straight face when you defend your methodology. Go on, I dare you.