Monday, December 9, 2013

New Ipsos-Mori poll confirms increase in support for independence since the publication of the White Paper

Ipsos-Mori have just released the third poll on independence referendum voting intentions to be published since the launch of the Scottish Government's White Paper, and the second to be wholly conducted since then (last week's TNS-BMRB poll was partly conducted before the WP).  It confirms the trend suggested by the previous two, of a clear swing in favour of the pro-independence campaign.  Here are the full figures -

Should Scotland be an independent country?

Yes 34% (+3)
No 57% (-2)

In the STV report on the poll, Ipsos-Mori's Mark Diffey acknowledges the boost in support for independence, but goes on to note that No retain a healthy lead.  Unfortunately, what he doesn't go on to note is that, even with this shift, Ipsos-Mori remain the outlier at the No-friendly end of the polling spectrum, showing a bigger lead for the No campaign than any of the other five pollsters that adhere to British Polling Council rules.  They also remain one of only two BPC pollsters (the other is YouGov) to be showing a raw No vote higher than 50%. In all likelihood, therefore, the true position is somewhat rosier for Yes than the raw figures of this poll would imply.

A couple of interesting titbits from the poll's datasets - Yes have a slim lead (47% to 45%) in the country's most deprived communities, while the No lead among Labour supporters (73% to 18%) is now several points lower than the Yes lead among SNP supporters (74% to 15%).

* * *


And now for the second update of this blog's Poll of Polls, which is based on a rolling average of six polls - the most recent from each of the six BPC pollsters to have been active during the referendum campaign (Ipsos-Mori, Panelbase, ICM, YouGov, Angus Reid and TNS-BMRB). This update simply replaces the last Ipsos-Mori poll with the new one, and therefore unsurprisingly sees the pro-independence campaign moving in the right direction.

MEAN AVERAGE (not excluding Don't Knows) :

Yes 32.7% (+0.5)
No 49.0% (-0.3)

MEAN AVERAGE (excluding Don't Knows) :

Yes 40.0% (+0.5)
No 60.0% (-0.5)

MEDIAN AVERAGE (excluding Don't Knows) :

Yes 38.9% (-)
No 61.1% (-)

As you can see, the Yes camp have broken through the psychological 40% threshold in the middle batch of figures. The median average is unchanged for the simple reason that Ipsos-Mori remain one of the outliers, and are therefore irrelevant to the calculation.

When Don't Knows are taken into account, the Yes side now need just an 8.15% swing to draw level. With Don't Knows excluded from the equation, the required swing is down to 10%.


  1. Ipsos don't weigh using last vote (their results tables have more Labour voters than SNP ones!) and they ask respondents how they would vote if the referendum were now. Both methods probably inflate the no figure.

  2. According to a twitter comment on the Wings twitter ( I cannot access WoS)

    Weighting used in poll Con15% Lab36% Lib8% SNP35% Other6%, yet 2011 list was Con12% Lab26% Lib5% SNP44% Other10%

  3. Reweight using the 2011 List ratios and the poll result is Yes 42% No 48% and Don't Know 9%

  4. It's not yet clear whether the 'party support' figures in the datasets refer to recalled vote from 2011 (in which case obviously the credibility of the headline figures is severely in doubt) or to current voting intentions. My guess is that it's the latter, which would mean that Ipsos-Mori are claiming (or are about to claim over the coming days) that Labour have drawn level with the SNP on the constituency vote, or perhaps even taken a slight lead. In a perverse way I would still draw encouragement from that, though, because it leaves open the most logical explanation for why Ipsos-Mori are producing bigger No leads than any other BPC pollster - namely that their sample under-represents natural nationalists. And it could be that even in their previous polls that showed an apparently healthier SNP vote, they were over-representing the 'wrong types of nationalist' in the sample - for instance there's some limited evidence to suggest that SNP supporters in the party's heartlands are proportionately less likely to be Yes voters than SNP supporters elsewhere.

    One thing that seems to be beyond dispute, though, is that Ipsos-Mori are significantly under-representing respondents with a predominantly Scottish national identity. Over at Wings, Scottish Skier has reweighted the data from the poll using national identity percentages from the census and the Scottish Social Attitudes Survey, and come up with independence voting intention figures of Yes 37%, No 47%, which are almost identical to Panelbase.

  5. I think that identity is key to the wide divergence in poll results (i.e. Panelbase and rest). There is a clear difference between the 'Moreno' national identity results produced by the Social Attitudes Survey ( and that of the 2011 Census ( Although Curtice has attempted to rubbish the Census data ( as, in effect, little more than a push poll, we have to remember that he has ‘skin in the game’ and his dismissing the Census data as simply wrong seems little more than an unconvincing and self-serving excuse for a catastrophic failure in the methodology of his own Social Attitudes Survey.

    That said, and at the risk approbating and reprobating the same body of polling data, it is interesting to look at the correlation between identity and Indy voting intention. Perhaps unsurprisingly, the Social Attitudes Survey data ( found clear correlations between identity and referendum voting intention. Whilst, this data shows the correlation falling short of one-to-one relationships it does demonstrate clearly that the more Scottish we feel our identity to be, the more likely we are to vote Yes (and vice-versa). Granted we have to take this data under caution, but given that the Social Attitudes Survey also asks a forced choice identity question, we might expect McCrone to report/explain any divergence in the correlations attributable to how the question is asked.

    Before looking at the effect of identity on the Indy poll results we have consider what (if any) role identity plays in polling methodology. The latest Panelbase poll ( also asked an identity question and produced a result almost identical to the Census findings. Contrastingly, the latest poll Ipsos Mori which used the 'Moreno' question and produced results closely in line with the Social Attitudes Survey identity data. That identity is playing a crucial role in the polling results seems beyond doubt, the question is what is happening?

    On one hand, if polling companies are actively incorporating identity data in developing their quota samples and/or weighting their results then (given the identity/voting intention correlations) we may fairly say that the polling company’s decisions on identity are driving the results. On the other hand, if different quota sampling and/or weighting methodologies simply result in these divergent identity outcomes then we may fairly say that at least one (or perhaps both) of these polling methodologies is flawed.

    Playing with the Ipsos Mori data in such manner as to re-weight/re-balance it to the Census results is little more than an interesting exercise, but we may fairly say identity will be critical to the actual referendum result and that we needn’t be too worried about polls that do not treat this criterion accurately.

  6. Ipsos don't use past votes because the believe that false recall will distort the poll results. I've made an attempt to weigh the results to 2011 using the data in the tables. I got Yes: 29% No: 52% DK: 19%. More in line with what the other polls show.

  7. Calum : That's more in line with other pollsters in the sense that the No vote is at a more realistic level and there are more Don't Knows. But the overall No lead is still exactly the same : 23%, compared to an average No lead across all BPC pollsters of just 16.3%.

    In any case, I can't see any information in the tables that would even make it possible to weight by 2011 recalled vote. The more I've thought about it, the more convinced I am that the 'party support' figures refer to current voting intention, not 2011 vote.