Weight your votes, ladies and gentlemen.

If you're not into the bizarre maths of polling, please don't read this, just come back later.

Anyway, ICM asked how Glasgow East constituents voted in the 2005 General Election, and, like many other pollsters, they compared that data against the actual 2005 results to see how representative their sample was. 

They expected thirty-one of their sample to report having voted Liberal, but only got six. I can imagine why, incidentally. So each of those six people miraculously become five and a sixth people. Except that one of them felt it was unlikely they'd vote, so got discounted, and, taking into account the overall weighting for likelihood to vote, only four of these 2005 Liberals remained. However, each now counted for six people. Super-Liberals, if you will.

In a final round of magic, one of these four Super-Liberals plans to vote SNP, so six votes got added to their column (see the second column from the right on page three of the ICM report). That's a 25% swing between these two parties, except for the one 2005 Nat who's gone the other way. He or she is much less important, as we'll see.

In sharp contrast, the fifty-five people who said they'd voted SNP in 2005 were weighted down to thirty-six. Of them, 94% will still vote SNP, but their opinions only count as two-thirds of a person, unlike our rogue Super-Liberal, who, you'll remember, is now six people. So if you told ICM you voted Liberal in 2005, your opinions are nine times more influential on the poll outcome than if you said you'd voted for the Nats. 

(The rows don't add up properly on that page, something I intend to ask Nick Sparrow about. Alternatively, anyone who gets how the 2005 figures or the 2008 intentions add up to the total on the left, please let me know. Paging Mark Ballard!)

Seeing as we've come this far down the rabbit-hole of psephology, there's another quirk here beyond the one Mike noted. When ICM asked about actual 2008 voting intentions, only ten people said they were voting for the Liberal candidate, but this number was too low, so got weighted up to twenty-four. 

On the other hand, twenty-three respondents planned to vote Tory, but ICM weighted them down because the sample had slightly more 2005 Tories than expected, and they eventually counted for just nineteen.

So twenty-three people in this poll said they'd vote Tory, and just ten that they'd vote Liberal. Yet the final numbers, generated by a massive amount of hand-waving and pure woo! show the Liberals on 9% and the Tories on 7%. 

Thank you for your patience if you're still here. Here's my prediction, based on the raw numbers, despite a small sample - the Tories will come out ahead of the Liberals. 

Update: the next poll has the Tories on 7% and the Liberals on 3%. This is exactly in line with the raw numbers from ICM.