‘New poll, no change, polls apart’ has so far been the dominant theme of polls on the independence referendum. Today’s TNS BMRB poll illustrates it perfectly.
According to this latest measure, 25% say they will vote Yes next year, 43% No, while as many as 32% say they do not know. The Yes figure is exactly the same as in TNS’s previous poll four weeks ago, the No tally down just one point and the Don’t Knows up one. Changes of that size are well within the range of random variation to which all polls are subject.
Once we leave the Don’t Knows aside, the poll points to a 37% Yes vote, with 63% backing No, an (equally insignificant) one point increase in the Yes share. All five of the referendum polls TNS BMRB have conducted this year have put the Yes vote on this measure at between 35% and 39%, thereby presenting a picture of remarkable stability.
Yet TNS BMRB’s figures are at variance with those produced by a Panelbase poll conducted just a week earlier for the Wings over Scotland website. That recorded a 45% Yes vote once the Don’t Knows were set aside – exactly in line with the 44% or 45% Yes vote that every Panelbase poll bar one has reported this year.
In truth none of the pollsters that have been polling at all regularly during the course of this year have identified much in the way of change in the balance of referendum opinion. They simply disagree with each other about what precisely that balance is.
One (partial) clue why the difference between the two most recent polls arises can, however, be discerned by looking at the number of people they contain that say they did not vote in 2011 or cannot remember what they did.
TNS BMRB’s face to face sample contains far more such people (33%) than does Panelbase’s (19%). Moreover, TNS BMRB (unlike Panelbase) then go on to weight their data such that the total number of non-voters (including those who can’t remember) matches the 50% who did not vote according to the official 2011 result. That means every 2011 non-voter in their sample counts for as many as 1.5 people rather than just one.
These 2011 non-voters are distinctive in their referendum vote intentions – they are less likely to say they will vote Yes. In today’s TNS-BMRB poll only 33% of this group (leaving aside the Don’t Knows amongst them) say they will vote Yes, compared with 40% of those who did vote two years ago. In Panelbase’s poll, the equivalent figures are 36% and 46% respectively.
This suggests that if Panelbase had more 2011 non-voters in their sample their estimate of the Yes vote would be somewhat lower – albeit still not as low as that of TNS BMRB.
Of course TNS BMRB’s approach to weighting their data assumes that 2011 non-voters are just as likely to vote next year as are those who did cast a vote in 2011. The company’s own figures of people’s reported likelihood of voting suggests that is not likely to be the case. But even if we confine our attention to those who told TNS BMRB that they are certain to vote next year, their estimate of the Yes vote only edges up from 37% to 38%.
In a ballot in which turnout seems likely to be high, it looks as though it is going to be important for pollsters to ensure they are reaching out to those who may not have graced a polling station with their presence for some years – hardly the easiest group to persuade to participate in a political poll at the best of times.
Those who did not vote in 2011 are also distinctive in another way – they are more likely to say they do not know how they will vote. However, this does not mean that their greater presence in TNS BMRB’s poll explains why, at 32%, its proportion of Don’t Knows is much higher than in Panelbase’s (17%). (Indeed this is one figure where TNS BMRB’s estimate is at variance with that of every other pollster, including the company’s own estimate in polls it conducted earlier this year).
As we have previously noted, the explanation appears to lie in the fact that TNS BMRB ask people what how they ‘intend’ to vote next year, whereas most other polls simply ask people to indicate how they would answer the referendum question now. If that is indeed the reason then at least it should mean that at least this particular polling gap should begin to close as polling day draws near.