First of all, I have an analysis piece in The National today about the new Savanta ComRes poll which puts the Yes vote on 53%. You can read it HERE.
Elsewhere in the same paper, Shona Craven argues that the Yes lead in the polls might be misleading, and asks whether Yes voters are significantly more likely than No voters to join volunteer online polling panels. To which the answer is: almost certainly yes, but it probably doesn't matter that much, because that's what political weighting is there to correct for. Most polling firms start by asking respondents how they voted in the 2014 referendum, or in some cases it's already known how they voted due to responses in previous surveys. If there are too many Yes voters in the sample, they are downweighted accordingly. For example, in the recent Panelbase poll for the Sunday Times, the 494 Yes voters from 2014 were downweighted to count as only 471, and the 560 No voters were upweighted to count as 583. If that adjustment hadn't been made, the Yes lead would have been higher than the 52-48 that was reported. And in many other surveys, the correction has been a lot bigger than that.
If anything, the greater concern ought to be that an over-correction is going on, and that Yes are actually being underestimated as a result. That's exactly what happened to the SNP in the 2019 election - the polls underestimated them, because people who voted for them two years earlier were being downweighted by pollsters to a target figure that was too low, due to the disproportionate number of potential SNP voters who stayed at home in 2017.
Don't get me wrong: I think there's no room at all for complacency about the current polling situation. But that's not because I think there's any particular reason to suppose the polls are wildly overstating Yes at the moment. The problem is more that the Yes lead is relatively modest, and public opinion can change in the blink of an eye.