THe polling organisations are geared more towards general elections, so they have a method of trying to pick a sample thats representative across various demographics and is weighted in the way constituencies decide elections. The problem with polling for a referendum is the pollsters have very little experience of that model of vote, it would be easier to make sure your sample was typical if independence referendums were held routinely. There have actually been relatively few polls for the referendum, compared to a general election so you can claim from the polls that there have been that been that there are trends in one direction or another but the reality is there haven’t been enough of them to make anything that clear. The pollsters won’t really know how typical their sample groups are until after the vote. Certainly in Quebec the polls were wildly wrong.
The thing is though – it doesn’t matter, the more widely inaccurate a poll is the more sale-able it is because apparent shock deviation creates lots of noise and light in the press – lots of ‘journalism’ reporters can do without having to actually leave their desk. The other polls released at the same time the yougov / sunday times one showed no real change in voting patterns, which is why non of them were talked about but the Sunday Times one was.
Something that is interesting though, watch what happens theres news of a new opinion poll on something showing a surprise outcome, some sort of shock figure – notice how it takes less than a day for news of that ‘opinion poll’ to turn into news of ‘opinion polls‘. It becomes plural somewhere between Breakfast telly and Newsnight