There are two Scottish polls in today’s papers. Panelbase in the Sunday Times has topline figures of YES 40%(nc), NO 47%(+2). In an increase for NO after three Panelbase polls in a row showing a lead of 5 points, but nothing beyond the normal margin of error.
ICM in the Scotland on Sunday have figures of YES 34%(-5), NO 46%(+4). This looks like a sudden big shift to NO, but I suspect a lot of that is a reversion to the mean. ICM’s last Scottish poll was the one showing the NO lead shrinking to just 3 points… I suspect that one was just a bit of an outlier and this is a return to normality. Even so, in John Curtice’s analysis he suggests that ICM have changed their approach to turnout in this poll, and that it would have been even worse for YES on their old method.
There is a big contrast between what different pollsters in Scotland are showing, and many people trying to read narratives and trends into the polls that aren’t really there. Even attempts at “polls of polls” are tricky because of the difference between pollsters and their uneven pattern of publication (so you can either have an average that leans towards polls that are better for YES or better for NO, or have an average than includes all pollsters but has polls that are way out of day). My view is that the best way of seeing what is happening is still the rather laborious and imprecise process of looking at trends in individual pollsters:
Taking them one at a time, and excluding don’t knows so they are comparable, ICM had YES on 40% last September, then 46% in January, then 43% in February, 46% in March, 48% in April… now 43%. ICM have been more erratic than other companies (and have messed about with their methods more) so it’s quite difficult to differentiate trend from volatility or method change.
Ipsos MORI we had YES on 34% last September, 37% in December, 36% in February. We haven’t had anything since.
Survation we had YES on 38% in January, then after a methodological change they’ve been extremely steady showing YES on a consistent 44% or 45%.
TNS-BMRB had YES on 36% in October, 38% in November, 40% in December and January, 41% in February, 40% in March, 41% in April, 42% now.
YouGov appears to show a similar steady but slow trend – 38% in September, 39% in December and January, 40% in February, 42% in March and April.
Panelbase have consistently shown better scores for YES than other companies. 44% in September, 45% in October and November, 43% and 44% in February. 47% in polls in March and April, 46% today.
My perception is still that there was a tightening in the polls at the tail end of last year after the white paper, and a very slow trend towards YES since then. The trend may well have slowed or stopped completely in recent weeks, but the single ICM poll or the normal variation in Panelbase is not enough to conclude it has reversed.