@insto "@calmac "There just isn't the evidence for it."
There is. I spelled out the evidence earlier. I think gembo's point above, with which you couldn't agree more, is also mistaken. What you need to do now, rather than just reasserting what you previously said, is explain why I'm wrong about Panelbase and TNS. Please feel free to do that with alternative evidence - numbers and stuff. The data is all in the public domain."
You haven't provided evidence for why you think Panelbase and TNS are outliers. You really haven't. It's just an assertion.
I don't have to give any numbers - my whole point is that you have insuffiecent evidence to claim that it is "clear" that they are both outliers. So, with numbers, can you show in what way the methodology of each is inflating the support of one side and suppressing the vote of the other?
Without that, there's no way you can say either are an outlier.
My point is that it's not even possible to make such a claim until after the vote itself, because there is no way of having a benchmark for the different methodologies. My evidence for that is the history of political polling.
"The one further piece of evidence I can give you is the detail of Yes Scotland's reanalysis of the polls before they decided to push the point at which they think they will overtake Yes away from June to September."
That wsort of analysis of the polls is complete drivel, and I think you give it more credence than it deserves. You can look back at political polling and see that, over a period as long as this one, it hardly ever moves in a straight line, or can be projected forward from this far out on some sort of linear regression model. More common in UK polling is that the governing party is behind and narrows the gap, often to the point of retaking the lead, anywhere in the last year. The movements are lumpy and inconsistent, and this general trend often doesn't happen at all.
This sort of linear modelling is great with some data sets, but is frequently terrible when dealing with human behaviour. How can you ever factor in a Sheffield rally, Jennifer's ear, the sopabox, I agree with Nick, Who's Ian Gray?, unpardonable folly, bigotted old woman... I don't think the No side could blow this with one almighty gaff, but the Yes side could, and no model can allow for that.
"More generally, your assertion that "there is no way you can get a genuinely representative sample. There will always be biases" is also wrong."
Ah. I'm going to have to ask where you work. Because surely you appreciate that there is no perfect sampling method in opinion polling?
Look, at least some of the pollsters are going to get this referendum very wrong, unless they change their methodology. That's axiomatic when they are reporting such different results. Each one has the same confidence you do, but some are going to be very far out. If the result is close, some are going to preduict the wrong outcome. All pollsters should be preparing for that and not making claims they won't be able to defend later.
"If surveys were always biased they would be useless."
Absolutely not. They'd be good approximations. Which is exactly what they are. And you can improve the methodology by comparing polls to final results.
And just to finish, if things are going to be kept civil then the snide digs should stop, and people should apologise when they make them.