Tuesday, February 10, 2015

Tumultuous TNS-BMRB poll suggests the SNP are on course for a second overall majority in the Scottish Parliament

Although the new TNS poll shows a somewhat lower SNP lead in Westminster voting intentions than other firms have been reporting of late (albeit still enough to win the party two-thirds of the seats), the Scottish Parliament voting intention figures for the 2016 election have a much more familiar look to them.  Admittedly, even on those, Labour have a higher reported vote share on the constituency ballot than other firms have been suggesting, but that wouldn't be enough to stop the SNP defying the PR voting system by powering to a second successive overall majority.

Constituency ballot :

SNP 47%
Labour 31%
Conservatives 13%
Liberal Democrats 4%

Regional list ballot :

SNP 44%
Labour 26%
Conservatives 13%
Greens 9%
Liberal Democrats 6%
SSP 1%

Now that the datasets have been published, we also have partial answers to the questions I posed last night - did TNS foolishly use 2010 recalled vote weighting (which we know is far less reliable than 2011 weighting), and is that the explanation for their Westminster figures being out of step with other pollsters?  Horror of horrors, they DID use 2010 weighting, although they used 2011 weighting as well, which at least would have diluted the distortion caused.  Nevertheless, it seems likely that this is at least part of the explanation for TNS showing a lower SNP lead than other firms - it probably can't explain all of the huge disparity with the 28-point lead suggested by Ipsos-Mori, but it might well explain all of the more modest disparity with other firms.  The reason I'm having to use words like 'probably' and 'might well' is that there doesn't seem to be enough information in the datasets to make the calculation.  All we have is an explanatory note at the top which reveals that 2010 weighting was used.

And there is another puzzle.  Unlike the eccentric Panelbase poll in early October, TNS have used exactly the same weightings for both their Westminster and Holyrood voting intention questions.  So the foolishness of using 2010 weighting can't explain the failure of TNS to replicate the finding of other firms that the traditional big difference between voting patterns for the two parliaments has more or less vanished.  The explanation must lie elsewhere - perhaps in the face-to-face data collection method, or perhaps in the way TNS pose the questions.

Curiously, the Herald chose to headline the predicted turnout figures from the poll, rather than the voting intention numbers.  I don't think there's any great surprise that many referendum voters are planning to give the general election a miss - after all, recent local council by-elections haven't really shown any 'referendum effect' in their turnouts.  However, opinion polls haven't tended to be great guides in respect of turnout, so it's still perfectly possible (likely even) that Scotland will have a markedly higher turnout than the rest of the UK in May.

18 comments:

  1. TNS poll:

    "Data was also weighted to match turnout and share of vote from the
    2010 General Election"


    They weighted to 2011 also, but certainly any 2010 weighing would hit the SNP and boost Labour as usual.

    Probably explains the lower SNP share for both Westminster and Holyrood.

    ReplyDelete
  2. Annoyingly, they haven't cross tabbed 2010 or 2011 recall so gauging the extent of any 2010 issue (or even 2011 problems common to TNS) is not really possible.

    Anyway, taking the above into account and the timing of the fieldwork, TNS just look to be in agreement with other pollsters but just at the lower end due to methodology.

    ReplyDelete
    Replies
    1. Actually they've done the equivalent, giving the tables for 2010 vote (with crossbreaks for current Westminster and Holyrood votes) on page 16 and for 2011 Constituency (same crossbreaks) on page 40.

      James seem also to have misunderstood the weighting. You would expect them to use the same weighting with respect to both the nest Westminster and Holyrood elections. It's the same sample after all and if something needs correcting in it, it need correcting with respect to both.

      The only differences would then arise from differences in turnout - either a different level of turnout or different people in the sample voting in one but not the other. TNS don't use a Likelihood To Vote weighting in their polls, they just include everyone who says they they are 'certain' or 'very likely' to vote. For Westminster (p11) this is 75% (755 people) and for Holyrood (p35) it is also 75% (753 people). But this is just a coincidence that means the final tables are going to look very similar. Though the high level for Holyrood may also tell us about increased enthusiasm for Scottish politics, given the usual 50% turnout for that.

      As it happens I suspect weighting isn't responsible for the slight oddness of this poll. Down-weighting on current VI seems equivalent for SNP and Labour so it probably isn't much of a factor. TNS's face-to-face method and the high level of 'Uncertain's it produces may be more the reason.

      Delete
    2. Roger, I've responded to your comments on UK Polling Report in a new blogpost HERE. I hadn't noticed that you'd left comments on this thread, but they mostly cover the same ground anyway.

      I'm afraid it's you that's completely misunderstood my point about TNS using the same weighting for Holyrood and Westminster - I was not criticising TNS for doing that, I was criticising Panelbase for NOT doing it in their eccentric October poll. The point being that there was an obvious artificial reason in the Panelbase poll for the disparity between the Holyrood and Westminster numbers, but there is no such obvious explanation in the TNS poll.

      Delete
  3. Due to the large amount of undecided voters in this poll [for the headline figures] - the margin of error increases to between 4.5% and 5%. This could affect the results by quite a bit.

    ReplyDelete
  4. re; the choice of the Herald to focus on the lack of turnout boost, rather than the VI numbers, I think this is because there is no way to objectively cover the VI results, given the lack of preceding polls by TNS. It's only if you want to spin a narrative of Labour doing better / SNP doing worse (like David Maddox in the Scotsman), you can then compare apples with oranges to do that. Without any reference point, then any reporting of the VI can only focus on the results themselves, which suggest a bad although perhaps not disastrous result for Labour. That isn't as newsworthy it would have been if TNS had shown a much bigger SNP lead ("Labour wipeout fears confirmed") or a much smaller lead ("neck and neck" or such).

    I think the biggest reason for the difference between this poll and the other polls is the question - how do you intend to vote on date x, rather than how would you vote if there was an election today. That understandably produces a bigger don't know result, which I think is more likely to hurt a party that has recently received a significant boost in support (and help a party that has lost support recently).

    ReplyDelete
    Replies
    1. Further to this, I think the TNS question method may explain some of the difference between WM and Holyrood VI. Everyone knows in the past there has been this group of voters who voted Labour in 2010 and SNP in 2011. I think YouGov (Kellner) have said that the recent swing in WM VI they have detected is largely due to this group. I think it's plausible that they are sticking with the SNP at Holyrood (hence TNS having similar results to 2011), but have not yet fully decided to vote SNP in WM.

      Delete
    2. I think the biggest reason for the difference between this poll and the other polls is the question - how do you intend to vote on date x, rather than how would you vote if there was an election today.

      No it's not that. If you look at the two questions that Ashcroft asks the first is a 'tomorrow' one and the second a 'next General Election one'. If anything there's often a slight drop in the DKs between the two, though that may also be due to Ashcroft also asking people to think about their constituency and the likely candidates. Certainly there's not the sort of difference that we see here.

      Also TNS's GB-wide online polls ask the same question and they don't get the same level of uncertainty. The face-to-face method is probably the reason.

      Delete
  5. It seems like a lot of work has to be done to boost the turnout.

    Those who voted for independence surely want more powers for Scotland as the next step.

    SNP needs to keep the focus on the constitution.

    The "More Seats, More Powers" message is clear and simple..

    ReplyDelete
  6. The question is - would a high or low turnout really favour the SNP? During the referendum, turnouts were highest in the high NO areas - could the silent majority be persuaded to come out and save the union one more time?

    ReplyDelete
    Replies
    1. Possible, but unlikely. The boost to turnout in the referendum in the end came from the fear amongst people sympathetic to the union that Yes might win. Independence (as opposed to more powers) isn't really a live issue now, because the SNP aren't demanding another referendum.

      Delete
  7. In the TNS Westminster poll, around a quarter of people were 'undecided'. With the lib dems reduced to fringe party status and the tories still toxic in Scotland, it's fair to say they wont be going in either of those two directions. So, the real figures are, undecideds included:

    SNP: 31%

    Labour: 23%

    Could go either way: 25%

    The SNP think they have this in the bag but in reality its a toss up. I fully expect the two parties to be much, much closer in the final result than what is being predicted.

    ReplyDelete
    Replies
    1. Out of the ~1000 sample, 10% said they would not vote, 6% refused to give an answer and 26% said don't know. That means 58% gave a firm preference to one party or another. The same poll also said that 64% were 10/10 certain to vote, which is usually a pretty good indicator of turnout - some of those 10/10 certain won't vote for whatever reason, but some of the 7, 8 and 9/10 will vote.

      This all means that a large proportion of the "don't knows" simply won't vote. As I outlined above, I think the remainder (the "don't knows" who will vote) are the Labour / SNP switchers. Other polls, which ask who the subject would vote for if the election was that day, indicate they now the SNP. TNS indicates that these voters could perhaps go back to Labour, but they could just as easily (perhaps more so) be confirmed in their current preference (SNP).

      Delete
    2. Around a quarter of people are always undecided in polls.

      They're the ones that usually don't vote. They generally say this by saying they are not certain to vote, but tend not to say they definitely won't vote as this suggests they are apathetic.

      TNS got ~65% saying certain and 75% Certain or likely. Your 26% undecided largely fall into unlikely to vote group.

      If all the undeciceds voted, we'd be looking at a turnout of 90% or so. That would break all records.

      I expect 75% at most turnout which ties in with TNS's certain + likely.

      Delete
    3. "The SNP think they have this in the bag but in reality its a toss up"

      The standard of trolling on this blog is truly impressive. It takes a special kind of talent to look at a situation where the polls are showing the SNP ahead by anything between 10% and 28%, and to call that a "toss-up".

      Delete
    4. I also remember that TNS's polling during the referendum campaign produced a much higher proportion of Don't Knows, even in the final week of the campaign - probably a result of the face-to-face methodology.

      Delete
    5. TNS have tried to mitigate that by having the subjects enter their preference into a computerised system out of the sight of the interviewer. But some reluctance to show a "controversial" preference in a face-to-face interview is inevitable.

      I suspect this effect was pretty much equal in the referendum, as both opinions were "controversial" for different reasons. Whereas I think it is a little bit more likely that voting SNP would be seen that way than voting Labour. But the effect of this should not be huge; the bigger cause of the high "don't knows" is asking people how they intend to vote at some point in the future, rather than how they would vote now.

      Delete
    6. The Tories being 'toxic' does not mean no one will vote for them! they will probably pick up about 15% of the vote, but since this poll was 'face to face' admitting you are voting Tory in Scotland may be a bit difficult.

      So I'm afraid that although you may get some relief from your straw clutching beliefs that all these 'don't knows' will vote either Labour or SNP and that the bigger majority of them will plump for Labour, the facts would seem to suggest otherwise.

      Delete