I voted for his not surviving, in this thread. Just sayin'....
In fact the problem with the opinion poll that we got a good look at wasn't the poll itself, which as far as I remember brought the SNP in about 4% behind Labour - well within the possibility of a late swing producing the eventual result, and in fact probably within the confidence limits in respect of the actual vote.
The problem was with the weighting statistics which were subsequently applied to the raw figures, which transformed a Labour lead of only 4% to a lead of 15%. At the same time they looked at a very low LibDem score, decided that must be wrong and "weghted" it up a lot. The LibDems actually lost their deposit.
The actual raw data from the questioning weren't all that far out from the eventual result. The problem was the the papers headlined the weighted figures, and the raw data were buried in a pdf that wasn't even put online until several days later.
As far as I could see, there were two main sources of error in the weighting. The first was that half of all respondents who were undecided or refused to answer, were allocated in accordance with the 2005 voting split. The polling organisation themselves accepted that this tended to favour the incumbent party. In my view it's a completely invalid assumption to make when a very large swing is underway, particularly as the group you're talking about, the undecideds and refuse-to-answers, are probably the very people who are on the verge of moving away from the incumbent party.
The second was that a lot of the weighting was done using data collected when the respondents were asked which party they voted for at the last general election. It was assumed that this question would almost always be answered truthfully. The answers seemed to indicate that the survey had identified a lot more SNP fans (and a lot fewer LibDem fans) than random chance would have predicted. Results were therefore "weighted" to compensate. However, no consideration seems to have been given to the possibility that in this case, many voters might have been confused and actually responded by saying which party they had voted for in the last Holyrood election, which was more recent and more memorable. And which produced a far higher vote for the SNP.
Presumably these weighting methods have proved valid in other situations. My point is that there were particular effects at work in this by-election which should have suggestted to anyone who was actually thinking about it that the assumptions might well be invalid on this occasion.
Does it matter? Maybe. When that poll was first published, my immediate reaction was to wonder whether it was actually worthwhile for me to drive further 100-mile round trips to help with the campaigning, if we were really on a hiding to nothing. I only decided to keep at it after looking at the raw data for myself, discussing the flaws online, and coming to the conclusion that the poll was probably wrong.
If I'd only had the newspaper reports to go on, I might not have changed my mind. And maybe a significant number of other activists might have made the same decision. The eventual majority was quite slim, and it's very probable that having a lot of activists on the ground in the last few days made all the difference between a narrow win and a narrow defeat. I myself managed to get three voters who had forgotten all about it or not made the effort, to go out and vote a couple of hours before the polls closed. If adverse publicity makes activists lose heart in the last week of campaigning, I think it can be a real influence.
Rolfe.