Monday polls

It’s Monday, so we have the usual rush of polls – the daily YouGov, twice-weekly Populus, weekly Ashcroft and monthly ICM. I’ll update this post as they come in, and do a round up at the end of the day.

Populus have topline figures of CON 31%, LAB 33%, LDEM 10%, UKIP 15%, GRN 5%. Tables are here.

ICM for the Guardian have figures of CON 36%(+6), LAB 32%(-1), LDEM 10%(-1), UKIP 9%(-2), GRN 7%(-2). The six point jump for the Conservatives looks particularly sharp but usual caveats apply. While it tends to be the polls that show unusual results or big changes that get all the attention, they are actually the ones we should be most dubious about. If there has been a genuine surge in Conservative support, then we’ll see it across all the pollsters, and other polls so far this month have shown things pretty stable.

The weekly Ashcroft poll has topline figures of CON 30%(-4), LAB 31%(nc), LDEM 9%(nc), UKIP 16%(+2), GRN 8%(+2) (full details here). A one point Labour lead, and changes in the opposite direction to those in the ICM poll – I think we can be fairly confident that what we are seeing with these two is just random noise, the back and forth of normal sample error.

UPDATE: The last of today’s four polls, YouGov’s daily poll for the Sun has topline figures of CON 32%, LAB 32%, LD 6%, UKIP 16%, GRN 8%. After a couple of three point leads at the tail end of last week, it’s back to neck-and-neck from YouGov. Putting all four polls together I think it’s pretty much business as usual. One neck-and-neck, one Tory lead, two Labour leads. There is nothing here that’s incompatible with the steady picture we’ve had for the last six weeks: an extremely close race between Labour and Conservative, with Labour holding onto a tiny lead.

552 Responses to “Monday polls”

1 2 3 4 5 12
  1. @ Spearmint

    The interesting thing from Ashcroft was that some of his focus group from Sheffield Hallam got Christmas cards from Nick Clegg!

    (I’ve got my net ready and Mr Nameless should be here any minute!).

  2. @Unicorn – very interested in your 3.30pm post. I’m particularly interested in the use of Euclidean distances for all 5 UK parties. In particular, you are obviously coming out with a single ED score for the actual versus the target for 5 parties.

    I don’t know enough about the methodology to be sure of this, but my assumption would be that an equal variation for any one of the 5 parties would effectively contribute 20% worth of the total ED?

    You do discuss the Con/Lab gap in terms that suggest a lower ED for just these two, but I’m wondering how much the smaller parties have contributed to the final result, in particular Greens and UKIP, both of which seem to be adding quite a bit of variance to your findings.

  3. “ICM for the Guardian have figures of CON 36%(+6), LAB 32%(-1), LDEM 10%(-1), UKIP 9%(-2), GRN 7%(-2)”

    It looks good on paper but what else can I say?

  4. Postage Included

    And the weather! :-)

  5. @PAUL

    Perfectly true. Large sample size compensates but cannot negate that the GE sample is self-selected and not necessarily representative.

  6. @Hawthorn

    No need to apologise. Perhaps I should apologise for sounding rather sharper than intended – mock outrage has become a bit of a mannerism with me, perhaps due to having a 14 year old.

  7. I have not had so much fun since the 2004 Canadian federal election. Averaging the last 4 polls YouGov/Sunday Times, ICM, Populos and Ashcroft you can actually determine anti-swing back to the LD from Labour and Conservative.

    Ashcroft has LD on 9% last week and now there are two polls finding they are at 10%. Meanwhile Ashcroft has Conservative down 4 points, but UKIP and Green up 2 each.

    There you have it anti-swing back in play. :)

  8. Ashcroft Poll has a big swing in England to Labour, and this seems to be quite a consistent story.

  9. On the media “narrative”, poll spinning by the Republican press in the USA did little good for President Romney.

    At the time, I had some work dealings with Americans. I remember one right-wing Republican express disbelief at the eventual result to me and a colleague when we were having a beer.

    Partisans only fool themselves with this stuff.

  10. @ Colin

    I am not really inclined to go back through the various UKPR posts to see which contributors used the various different definitions at different times.

    As I say, it is easy enough to do comparable analyses with the definition of your choice.

    However, as I say, the Electioforecast model uses something close to my definition here and also on p.1 of the Working Paper introducing his model Stephen Fisher writes:

    ..if a party is doing unusually well in the polls some way before the election their support is likely to drop and vice versa.

    This, too, seems pretty close to the definition I am using here.

  11. Unicorn

    I was putting swingback in the context of historical data, this election with it’s massive swings may well be a poor predictor for those parties.

    I’ve looked at your analysis and I suspect you are mostly capturing information about the random shocks which are greater than the drift term. I really should go away and estimate the scale of these parameters from the electionforecast site as people seem to be taking swingback to mean “people will get what they got at the last election” when it’s a much more subtle term than that.

    Looking at your analysis it seems to show that in the past year things have moved further away from the previous election than closer. Is it by enough to declare a model that has worked previously dead? I agree the massive shifts in the minor parties seem to demand a convex term in the auto regression function.

    I’d certainly suggest it didn’t suggest antiswingback as that would demand ukips vote being pushed higher and higher the higher they went until we became a one party nation. It’s unstable as a long term model.

    I’d prefer a moving average model to something that predicts ukips vote increasing between now and the election.

    If you want to do some hypothesis testing I’d run a regression of the form VI_{t+1} -V_0 = gamma * (VI_t – VI_0). Where VI_0 is the 2010 figures. Essentially you use last months VI as a predictor for the following month.

    I suspect you’ll get a gamma something very close to 1, the trouble is electionforecasts model will be I suspect a gamma of the order 0.99 so it’ll be hard to statistically distinguish them on this elections data alone.

    Ultimately with the data we have, deciding between gamma = 0.99 and gamma = 1 produces very different results over a long term but not so much over a year.

    Going back to your analysis, essentially what is happening is you are fitting a linear fit to last years data and projecting the final VI forward. Choosing the final VIs is the same as choosing the line.

    While it produces very good in sample data, (it’s been designed that way) that method produces typically bad out of sample predictions. I suspect the optimal final VI for the Green party would be about 12% in terms of minimum distances. One way to test your model is to run these same tests but leaving several months of data out of your sample you fit to, to see how well it predicts those months which are now out of sample. Compare to a “Same as last month” fit and check for significance.

  12. @Chrislane 1945

    No such swing in Scotland, however.

  13. @Unicorn


    The numbers you’re using (36.05% for the Tories, 26.99% for Labour, 23.03% for LD and so on) are the UK results. Don’t forget the polls are for GB only. If you’re calculating the distance for a GB poll, it should be to the GB result. According to Anthony’s post of May 7 2010 (, the final GB results were CON 37%, LAB 30%, LDEM 24%, Others 10%.

  14. So one poll shows Con 4 points clear whilst another has Lab 3 clear.
    Ashcroft’s latest poll has a combined figure of just 61% for the two main parties which is as low as I can remember.

  15. @ Alec

    You are right that a good deal of the variance will come from the smaller parties.

    Well – correction – not all that much comes from the Greens. In more recent months their VIs are 6 point above 2010, contributing 36 units to the sum of squares. In contrast, Ukip has often been 12-13 points up on 2010, contributing 144-169 to the total (prior to taking the root of course).

  16. At the 2005 election Labour got 35.4% in England and the Tories got 35.7% but Labour got 286 English seats and the Tories got 194. Will we see a similar disparity in seats if the vote share in England is roughly equal again? Or have there been changes which mean it will be a lot more even?

  17. ‘”But having come closest to predicting three of the last four general elections, ICM’s regular phone poll for the Guardian is seen as the gold standard, and so Monday’s result will bring the Tories particular cheer.”

    In terms of predicting party lead that claim is not accurate. In 1997 Gallup’s final poll predicted a Labour lead of 13% which matched the actual outcome – ICM had given Labour a 10% lead. YouGov in 2001 predicted a 10% Labour lead compared with the outcome of 9% with ICM saying 11%..In 2005 NOP predicted the actual outcome of a 3% Labour lead compared with ICM’s 6% lead.Finally in 2010 YouGov and MORI predicted a 7% Tory lead compared with the 7.3% outcome – ICM had predicted 8%.

  18. Swingback continues on the Ashcroft poll.

    Labour remain in the 36-40% range, barring outliers.

    The return to the two-party system continues, with the main two parties now looking at getting over 60% (with a good campaign) in the GE.

  19. @ Spearmint

    What happens if you just do the calculation for Labour and the Tories?

    I’ll take a look. Because the Tories have been flatlining and Labour going down steadily I would expect to see modest moves in the swingback direction.

    To take this as support for swingback theory, you would have to posit a very restrictive definition of swingback.

    Specifically, It would have to be restricted to a (senior-if-coalition) governing party strictly in comparison with its main opposition. Swingback hasn’t worked at all for the junior coalition party and – as I have just shown – doesn’t work for the other minor parties. Nor has it worked for the Tory party alone.

  20. @ Martyn

    Fair point. I’ll recalculate using the GB figures and report back again.

    I don’t expect big differences but we’ll see…

  21. @Chrislane

    So few England results that’s hard to say anything much. Ascroft gave Labour a 1 point lead in England 3 months ago, 4 or 5 point Tory leads since. More MOE stuff really.

  22. LASZLO
    “the HoC once in every 5 Yeats”.
    As far as I know there was only one Yeats. (During my army service training we had an Army Educational Corps lecturer who can down to talk to us about Yeats. The day before our RSM told us we should be polite and listen carefully since, “I know that many of you fellows have never heard of a Yeat.”

  23. @ John Pilgrim

    Oh the beauties of predictive typing …

    As to the army and education. Time: 1981, Hungary. There use to be a cigarette brand called Symphony in Hungary. When on ground exercise in the army smokers had a 10 minute break every 2 hours (actually non-smokers didn’t). Seargant: “who has a Symphony?” Someone: “Beethoven”. Seargant: “private Beethoven (pronunciation Hungarianised)! One step forward!”

  24. SYZYGY et al
    “Perfectly true. Large sample size compensates but cannot negate that the GE sample is self-selected and not necessarily representative.”

    As Monty Python’s Colonel said: “Now, I’ve noticed a tendency for this programme to get rather silly.”
    The voters at the General Election are not a sample, but the whole electorate.
    That includes the 30 or so percent who choose not to vote, or fail to vote. If they wished the result had been different, they should have voted. If you wish they had voted to produce a different result, you should have campaigned, or campaigned harder or more effectively.

  25. @John Pilgrim

    There was more than one Yeats. His father was a painter of note. The famous lines from WB are topical:

    “And what rough beast his hour come round at last
    Slouches towards Bethlehem to be born”

    Well Ed sought of slides, and Nigel saunters. Boris fits the bill.

  26. ?
    “sought” should be “sort”. A mistake of extreme weirdness.

  27. @ Dave

    The whole thing was just a series of jokes at the expense of methodological perfection as a solution to better polls.

    However, you would be surprised how many people can’t vote, although they would like to. There was a paper on this in JBS sometime in the 1970s. I know that the situation has improved since then but…

  28. I think this all comes down to the UKIP score.

    Are they at 15/16% as per Populus/YouGov/Ashcroft or 9% as per ICM.

    Intuitively it feels like UKIP should be dropping given that the press is no longer banging on about immigration (and ICM reported that people are less concerned about it). But the gap between ICM and the rest is too big to be explained by that alone.

  29. TOH
    ‘what larks!’
    Do you have great expectations, Howard?


    If they are topical, their topicality lies across the Channel I think.
    The Second Coming was written in 1919, and contemplated the state of Europe after WW1.

    If you wanted a topical quote from it for UK at present I would choose :-

    ““The best lack all conviction, while the worst/Are full of passionate intensity.”

  31. John Pilgrim,
    Although Great Expectations is one of my favourite novels,to my shame I had
    to google to find out who said it.Did you know that there is actually a web site
    called,what larks pip?We learn something every day .

  32. ICM, the firm which gave Yes an 8 point lead in the referendum within days of the actual vote!

  33. @Martyn and Unicorn

    Very interesting post by Anthony Wells on overstatement of LD support in 2010, which continued for some pollsters in the 2014 European election:

    May 22 Actual 6.9%
    19-21 Opinium 6%
    20-21 YouGov 9%
    19-20 Survation 9%
    15-19 TNS 7%
    16-18 Comres 7%

    Is not part of the problem sampling errors when you get down to these low percentages?

  34. Yes, but ICM have been working hard to overcome the mess they inherited.

  35. No deal tonight in Brussels.

  36. Andy Shadrack

    No issue with sampling errors as p reduces as the sampling error reduces as sqrt(p*(1-p)). The off quoted 3% only is relevant for parties 30-35%.

    Around that level sampling error is about 2%

  37. John

    As you know I do!

    A phrase from that great novel that both my wife and i use when we are having fun with the grandchildren

  38. Does anyone know what effect the ICM reallocation of DKs by previous vote filter had on the poll?

  39. The weekly Ashcroft poll has topline figures of CON 30%(-4), LAB 31%(nc), LDEM 9%(nc), UKIP 16%(+2), GRN 8%(+2

    61% for the Tories and Labour…………Horrific but it is only one poll.

    Cross break..SNP 44% LAB 24% which appears to be quite consistent with other polls.

    Back to the UK polls. Fink Pink HSBC, any impact they had on the polls last week will probably evaporate this week.

  40. Very interesting to see the ICM/Ashcroft variation – it may well be just sample variation, but after such a bad week for the Tories the ICM surge was certainly unexpected. The ICM polls have also been reportedly the most accurate in election outcome prediction.


  41. god the guardian reporting on this is infuriating.
    I assume that when the ICM polling reverts to the mean in a few weeks time it will be “Labour surge/Tory’s sink” and put it down to ed m successfully eating a kebab whilst cameron’s trousers were too tight at a press launch.

  42. Greece had rejected the bailout offer as ‘absurd’. 75% of Greeks support them, according to polls.

    Crunch time coming, one feels.

  43. I never cease to be amazed at how many posts on here seem to suggest that 2, 3 or 4% moves between polls are the result of current news events.

    Given that the majority of voters never or almost never change their votes and that of the remaining minority most change their votes but rarely it is presumably being suggested that a few percent perhaps one in twenty are changing their vote every week or even every day, back and fore.

    This is of course absurd, this welter of people who change VI every day do not exist.

    Far too much of the time so called movements are simply changes caused by the random chances of sampling.

    I would love some polling to tell us if anyone changed their vote recently and if so the reason why.

    I find it near incredible that even 0.1% changed their vote on the single issue or event of HSBC tax avoidance/evasion, a pink vehicle, the mayor of London’s nationality, keeping cash receipts or any of the other recent events.

    Before you assume some have changed their vote “because” of one of these issues, try to think of a real flesh and blood person actually doing that and you will realise how unlikely it really is.

  44. @Reggieside

    Couldn’t agree more. I’m a loyal Guardian reader but their analysis of their own commisioned polls is myopic rubbish.

  45. @ Jonathan

    you make very strong statements for which you don’t have evidence. You maybe (even likely to be right) that none of the events created a significant change in VI, but there is absolutely no evidence at the level of the individual – simply because the individuals’ predisposition to filter information and then change the predisposition (when they had enough) varies widely. The data in the tables is much higher level in abstraction to pick up such changes (certainly in the short term). In addition, the relationship between the stimulus (the news, not to mention how and where and how it is picked up) and the behaviour is stochastic and interdependent with other stimuli.

    What we can say is that in the framework of the model applies to the general population, the distribution of votes among party choices (DK is excluded!) hasn’t changed for some time. There is no causality is implied in any of the polling models, even if there is a likeliness.

  46. jonathan

    You may be party correct but, over time, the drop in LD VI, the rise in UKIP and so on, are clearly the result of people making different decisions than before.

    The point of most localised incident polls is, I agree, hard to fathom. Almost all committed voters will automatically say “x is a rubbish idea”, without needing to hear what it is provided their least favourite party suggested it.

    So all results are skewed by that and largely irrelevant.

    Anthony [who is very clever] has details of “false” polling where policies receive different responses depending which party they are told has come up with them.

    I, of course, look at everything on their merits.

  47. @Reggie

    The political commentators at the Guardian have made no secret of their dislike for EM, so it’s not surprising. Remember this is a newspaper that backed the LDs last time.

    In fact Labour at 32 is a bit low but not that far off their current polling average of 33. The Tories being at 36 is almost certainly an outlier, but their previous poll had the Tories at just 30, which was also an outlier.

  48. Pollsters including Peter Kellner have had their moment of fame as a question on Only Connect. Sadly, the collective brains on the show thought they were the names of horse races!

  49. MS has picked up from the ICM tables that women have broken 38/32 to the Tories, which is most unusual. Have ICM made a mistake?

  50. Sorry to link to this paper, but it looks like the Express has had a complaint made to IPSO upheld for misleading headlines based on a poll:

    That’s a small victory for psephologocal reason.

1 2 3 4 5 12