A related question I'd be curious about is the likely voter issue. In the US, left-wingers are less likely to actually go vote than right-wingers. Pollsters try to correct for this with a "likely voter" model (which varies by pollster, but tries to discover how likely a respondent actually is to vote).
If you poll everyone eligible to vote, you get left-wing results. If you poll registered voters, it pulls a bit right. If you poll voters that match a reasonable "likely voter" model, you'll get results that are even further right and tend to be closer to the reality of votes actually cast.
Does this dichotomy exist in the UK and/or Scotland specifically, and if so, do pollsters there try to correct for it?
I read that a problem the pollsters were having was simply that they didn't have any precedents. There was a referendum last century asking a different question but that wasn't that helpful. Essentially you can't make up a model without observations and opinion polls don't count
This is my understanding of the issue too. Polls don't just ask the question. Each cycle they take their input, the actual results, and refine their model. Over iterations the model should get more and more accurate.
This event is a one-off. No cycles, no iterations, leaving a model which is essentially a best-guess.
Some pollsters ask how likely one is to turn out to vote.
Amongst those that do, they weight responses differently.
Some will only report those who are 10/10, 9/10 and 8/10 likely to vote in their topline figures.
Some will multiply the likeliness with the voting intention, so that a 10/10 likely voter would count as a whole vote, a 9/10 likely voter as 9/10ths of a vote and so on.
The UK Polling Report[0] - run by the guy in charge of YouGov, if I recall correctly - is a very useful site for explaining UK polling oddities!
