Sunday, February 2, 2014

More drama as TNS-BMRB poll suggests the pro-independence campaign have closed the gap for the FIFTH time in a row

There are two new referendum polls out tonight, one from a polling company that is a veteran of this campaign, and another from a pollster that has just entered the fray for the first time. Both provide further powerful evidence that the pro-independence camp are continuing to narrow the gap (although in the case of the new pollster that evidence has been somewhat masked by a barking mad weighting procedure that has already been castigated by no less a figure than Professor John Curtice). Let's start with the familiar pollster - TNS-BMRB are suggesting that the pro-independence vote is up 2% on the equivalent December poll, which also means that it is up a full 4% since late October.

Should Scotland be an independent country?

Yes 29% (+2)
No 42% (+1)

A net one-point decrease in the No lead may not look terribly significant, but the true drama lies in the relentless long-term trend in favour of independence over the last five TNS-BMRB polls. In the late September/early October poll the No lead dropped from 22 points to 19, in the late October poll it dropped from 19 points to 18, in late November it dropped from 18 points to 16, in December it dropped from 16 points to 14, and now it has fallen from 14 points to 13. So the No campaign's advantage (with a pollster that is traditionally one of the most favourable for them) has been slashed by 9% since the late summer - almost exactly in line with the 10% drop in the No lead suggested by last week's sensational ICM poll over roughly the same time-scale.

Although the number of undecideds in this poll has dropped back a bit from the previous record high level, TNS-BMRB are still showing a significantly higher percentage for "Don't Know" than any other pollster, almost certainly due to their practice of asking people how they will vote on the actual referendum date, rather than how they would hypothetically vote in a referendum taking place right now. That makes it harder to make much sense of the headline numbers - so, for greater clarity, here is the state of play suggested by this poll when Don't Knows are excluded from the calculation...

Yes 41% (+1)
No 59% (-1)

And for the benefit of innumerate London media folk such as Mr David Dimbleby, that is not a "2-1 majority against independence". In fact, it's somewhat less than a 3-2 majority.

The pollster making its debut in this campaign is Survation, which on its raw unweighted figures shows a position of Yes 43%, No 57% (with Don't Knows excluded).  That would be almost as good for the Yes campaign as the ICM poll, and can probably be taken as a further indication that the gap is narrowing significantly.  However, Survation have distorted those figures out of all recognition by adopting a discredited weighting method that no other pollster uses, and that was finally abandoned even by YouGov a few months ago.  Basically, respondents were asked how they voted in the 2010 UK general election, resulting in figures of SNP 42%, Labour 32%.  That of course bears no resemblance to the actual 2010 result of SNP 20%, Labour 42%.  But instead of coming to the common-sense conclusion that huge numbers of people were mixing up how they voted in 2010 with their vote in the 2011 Holyrood election that produced an SNP landslide (an established phenomenon), Survation have stupidly made the assumption that everyone remembered correctly, and have accordingly adjusted their figures on an industrial scale to bring them into line with the UK general election result.  The 240 people who reported that they voted SNP in 2010 (in reality they are much more likely to represent the 45% of the electorate who voted SNP in 2011) are downweighted so that their responses to the referendum question count as the responses of just 150 people - a ridiculously low 15% of the sample.  People who recalled that they voted Labour have had their responses upweighted to a similar extent, with a significant upweighting of Liberal Democrat voters also taking place.  After all that, the published headline figures should be regarded as almost meaningless, but for what it's worth (ie. nothing) here they are...

Should Scotland be an independent country?

Yes 32%
No 52%

With Don't Knows excluded, it works out as...

Yes 38%
No 62%

It's impossible to know what the figures would be if a defensible weighting procedure had been used (most pollsters now weight by recalled 2011 Holyrood vote, although Ipsos-Mori ploughs its own furrow and doesn't weight by past vote at all).  However, it's virtually certain that the Yes vote would be significantly higher and the No vote would be significantly lower.  This is Professor Curtice's verdict -

"It looks highly likely that if Survation had followed the same practice as most other pollsters, the reported Yes vote in this poll would have been over 40% – just as it was in last weekend's ICM poll and is in this weekend's TNS-BMRB poll."

A vivid illustration of just how extreme the scaling down of SNP voters is in this poll is provided by the Holyrood and Westminster voting intention questions. On the Holyrood vote, the unweighted figures show the SNP on 45%, Labour on 29%, the Conservatives on 13%, and the Liberal Democrats on 6%. But the adjusted figures used for publication show a radically different picture (albeit one that will still give the SNP enormous heart in mid-term) -

Scottish Parliament constituency voting intention :

SNP 38%
Labour 36%
Conservatives 12%
Liberal Democrats 9%

On the Westminster vote, the unweighted figures show the SNP in the lead on 40%, Labour on 31%, the Conservatives on 15% and the Liberal Democrats on 9%. But (hey presto!) the adjusted figures for publication reverse the position and show Labour in the lead -

Westminster voting intention :

Labour 38%
SNP 30%
Conservatives 16%
Liberal Democrats 10%

Regardless of whether the adjusted or unadjusted figures are more accurate on that count (and you can probably guess my own view), those are numbers that should be scaring Labour to the core.

The poll is arguably most useful in showing the potential impact on the referendum result of a Tory surge in the UK-wide polls (the last week has seen conflicting evidence on whether that is already happening). When voters are asked to assume that the Tories will win a majority in the next general election, the No lead reduces by 6%. When they are asked to assume that the Tories will win a majority and then go on to remain in power for the next fifteen years, the No lead slumps by a full 11% - which for all we know might even be enough to put Yes ahead under the correct weighting.

There is also further proof (and we've now had so much evidence of this sort that it can't really be doubted) that lower-income voters are breaking for Yes. Even on the adjusted Survation figures, all three of the lowest income brackets - covering anyone living in a household with combined earnings and benefits of less than £20,000 - have Yes in the lead. The outright lowest income category gives Yes the nod by a whopping 2-1 margin.

* * *


It may seem utterly bizarre that such strong evidence of a narrowing gap coincides with a Poll of Polls update showing a slight increase in the No lead (finally bringing to an end a run of five successive updates showing the lead decreasing), but it's just a freakish occurrence brought about by the introduction of a seventh pollster into the sample. I was going to say that not too much should be read into it, but in fact I'd put it even more strongly than that - it's literally meaningless. If Survation had weighted their data correctly, we'd be looking yet again at an improved position for the pro-independence campaign on all three averages. Even as it is, the Yes vote has increased fractionally on the headline average.

MEAN AVERAGE (not excluding Don't Knows) :

Yes 33.9% (+0.1)
No 48.7% (+0.7)

MEAN AVERAGE (excluding Don't Knows) :

Yes 41.0% (-0.3)
No 59.0% (+0.3)

MEDIAN AVERAGE (excluding Don't Knows) :

Yes 40.8% (-0.1)
No 59.2% (+0.1)

(The Poll of Polls is calculated as a rolling average of the most recent poll from each of the pollsters that have been active in the referendum campaign, and that adhere to British Polling Council rules. Whenever a new poll is published, it replaces the previous poll by the same company in the sample. That used to mean a sample of six polls, but with the introduction of Survation it's gone up to seven.)

Regular readers will know that the Yes vote on the median average has typically lagged a bit behind the mean average, but the two figures have now virtually converged. It's hard to know which measure is likely to be more accurate - the case for looking at the median is that it isn't affected by extreme outlying figures in either direction, but on the other hand it's always possible that an outlier might be right, in which case the mean average could be slightly closer to the truth. For the time being, we no longer have to worry about that conundrum.

On the headline figures, the Yes campaign now require a 7.4% swing to draw level.

* * *

UPDATE : Some good news - Damian Lyons-Lowe of Survation has left a comment on Professor Curtice's blog indicating that in future his company won't be weighting by recalled Westminster vote, in order to make comparisons with other polls more meaningful.

On a similar theme, I was contacted this morning by Roger Scully, Professor of Political Science at Cardiff University and author of the Elections in Wales blog. Roger felt that I shouldn't have included Survation in the Poll of Polls update, because it isn't comparing like with like, ie. it makes the trend figures less meaningful. Instinctively I want to agree, not least because if I'd only included the TNS-BMRB figures in this update it would have produced a further swing to Yes across all three averages, which is likely to reflect the reality of what is actually going on. But I just don't see how that approach will work in future - it's highly likely that other BPC-affiliated pollsters will conduct their first referendum polls between now and September (Populus and ComRes are the most probable candidates), each potentially bringing a different methodology to the table, and if I ignore all of them it will leave us with a very partial picture.


  1. Can you provide a comparison on the poll of polls excluding the newcomer... Just so the previous trend can be seen more clearly?

  2. Yes, without Survation this update would have been -

    Mean Average (not excluding Don't Knows) :

    Yes 34.2% (+0.4)
    No 48.2% (+0.2)

    Mean Average (excluding Don't Knows) :

    Yes 41.5% (+0.2)
    No 58.5% (-0.2)

    Median Average (excluding Don't Knows) :

    Yes 41.4% (+0.5)
    No 58.6% (-0.5)

  3. James, as for this:

    "Regular readers will know that the Yes vote on the median average has typically lagged a bit behind the mean average, but the two figures have now virtually converged. It's hard to know which measure is likely to be more accurate - the case for looking at the median is that it isn't affected by extreme outlying figures in either direction, but on the other hand it's always possible that an outlier might be right, in which case the mean average could be slightly closer to the truth."

    In order to follow the swedish election polls, in a pretty much similar effort to yours, I calculate an average excluding the highest and the lowest poll. Maybe it would interest as something in between the mean and the median average.