An update on the Lord Ashcroft constituency polling from last week. Looking at the tables for Doncaster North – Ed Miliband’s seat – the weighting appears to be seriously askew, weighted as if the seat had a Conservative majority at the previous election. I’ve spoken to Ashcroft’s office and it appears to be a human error – typing the wrong weighting target in the wrong box (it’s something I’ve done myself in the past, though thankfully I’ve not done it and published the results!).

Lord Ashcroft’s team are reweighting and republishing the results, but obviously they are likely to make a significant difference to the findings, probably knocking down the reported level of Conservative and UKIP support and bumping up Labour. I’ll update when the new figures are out.

UPDATE: The corrected version of Lord Ashcroft’s polling is now up on his site here – topline figures in Doncaster North are now CON 13%, LAB 54%, LDEM 4%, UKIP 25%. UKIP are still in second place, but they are now a far larger twenty-nine points behind Labour (and given some of the reaction to the earlier poll it’s worth noting that Labour are even well ahead of the Tory and UKIP vote combined). There is no risk to Ed Miliband here.


123 Responses to “Update on the Lord Ashcroft Doncaster poll”

1 2 3
  1. Anthony

    Thanks for contacting Ashcroft’s people. I put a query in though the site, but wasn’t sure who to get to directly.

    On the subject of Ashcroft, do you know if the VI questions in his weekly polls give UKIP as a ‘main’ option. I always assumed they didn’t and the wording on those and his constituency polls up to now just say “[…]which party would you vote for?” or “[…]which party’s candidate do you think you will vote for in your own constituency” without specify which, if any, list(s) there are.

    However in the leader’s polls the wording is:

    If there was a general election tomorrow, which party would you vote for?
    Would it be: Conservative, Labour, Liberal Democrat, UK Independence Party (UKIP) or another party? [Prompted in randomising order]

    with the second sentence also used in the constituency question.

  2. As far as I’m aware he includes UKIP in the prompt for constituency polls, but does not include them in the prompt for GB polls.

  3. I thought the combined Tory & UKIP vote looked too high for that particular seat.

  4. Anthony

    Thanks for that. It may explain why his constituency polls tend to show a higher UKIP VI than his weekly ones, even though you’d expect as marginals there might be a squeeze on other Parties. I don’t think prompting for UKIP or not makes much difference in online polls but it might do in telephone ones.

    Did you also mention my other points about the reallocation? It makes a difference to the Hallam and Thanet ‘results’, though both of course are well within MoE and in practice it’s just too close to call.

    As I said in my original comment on this you feel sorry for Lord A on this one as these mistakes are so easy yo do, especially dealing with a revised format. I suppose the rule is the usual one that if it seems too good (or too dramatic) to be true, it probably isn’t.

  5. I’ve done some back-of-the-envelope calculations to reweight it. EdM is WAY ahead, looks like more than half the vote, with UKIP and CON down in the teens. I suspect we’ll get the numbers at 4pm.

  6. typing error erm I sometimes think the theory about the illumanati holds a certain amount of credence

  7. Colin (and Unicorn fpt)
    Fascinating piece by PK. I hope someone produces a calculator along these lines as the election approaches, although he is only tinkering with UMS. I wonder whether, if the current fracturing of VI continues until May, UMS will not be a great predictor anyway.

    Number Cruncher
    Is that all we’re expecting at 4pm or is he releasing some other polls?

  8. @ Cloudspotter

    His weekly national poll as well

  9. That this mistake seems to have been spotted by the online poll-chasing community (see Roger Mex’s link). But to do this the data had to be open and available in entirety.Compare the “human error” in the infamous research from Harvard […]

  10. COLIN

    Wow-this changes things :-

    https://yougov.co.uk/news/2014/12/01/uniform-swing-rip/

    Actually Kellner’s really confusing two things[1]. One is UNS as a mathematical model (which is what Unicorn is disputing) as opposed say to a proportional or more complex one.

    The other is about using the same swing uniformly across the country when it is clear that in Scotland and in strong Lib Dem seats this clearly isn’t the case. He also mentions Clacton and Pavilion as holds, but doesn’t consider other seats where there is strong UKIP polling such as Boston.

    As previously Kellner seems to believe that there is some sort of general incumbency effect, tied to sitting MPs especially new ones, that will apply to more than just Lib Dem seats. I don’t think we’ve seen much evidence for that yet, though it may be we need to see more polls with named candidates rather than Ashcroft’s more general question, before we can see if that is true.

    [1] Not I’m sure deliberately, it’s just that both include the words ‘uniform’ and the rest of us get confused.

  11. Here you go, corrected Q1 (Q2):

    LAB 51 (54)
    UKIP 27 (25)
    CON 14 (13)
    LIB 3 (4)
    OTH 4 (5)

    Details http://lordashcroftpolls.com/2014/12/ed-miliband-doncaster-north-correction/

  12. Roger

    I think the point of the article is to point out that it will be harder for Labour to gain the most seats and an OM, a 1 point Labour leads to virtually an identical number of seats. This is against current thinking.

  13. UKIP are now 22% behind, not 29%!

    [No, they are 29% behind – I copied down the wrong numbers from the results! Now corrected. AW]

  14. Number Cruncher

    I’ve done some back-of-the-envelope calculations to reweight it. EdM is WAY ahead, looks like more than half the vote, with UKIP and CON down in the teens. I suspect we’ll get the numbers at 4pm.

    In my comment I recalculated it to:

    Con 13%

    Lab 53%

    Lib Dem 5%

    UKIP 24%

    Others 5%

    which seems more plausible. The unweighted figures were actually nearer the 2010 percentages, though showing the usual recalled biases of lower votes for the Lib Dems and presumably right-wing Parties (BNP, ED).

  15. Roger

    I would have responded to you as HH did.
    That seems to be the central point being made-unless you think PK is confused about that too?

  16. @ Roger

    Sorry, didn’t see your original comment… You were very close!

  17. I should have said those were for the second (constituency) question – so they’re pretty close to NC’s figures.

    The fuss over Miliband ‘losing’ Doncaster was always a bit fake in any case – UKIP and Conservative votes don’t transfer to each other that easily. But the difference now is that adding the two together no longer gives a figure that is higher than Labour’s.

  18. Even this might not quite stop some of the wilder posters from speculating that Ed Miliband is in some sort of trouble in his own seat.

  19. Hearty congratulations to @Roger Mexico and others for sensing that something was wrong with these figures right from the outset and then taking the trouble to confirm their suspicions and (now) getting the press release put right.

    What chance, though, of a full product recall including a withdrawal of the headline about Miliband being at risk? Thought not…

  20. I don’t really agree with PK that incumbency has only become apparent in the last 30 years. Looking back at the 1964 election there were many seats that Labour failed to win – or only won very narrowly – which had only been lost in 1959.

  21. Colin & HH

    Oh no I agree with that part of the analysis. But that’s something that there’s been general agreement on here for a while, about Labour not getting a majority the same way they would with a GB-wide uniform swing. You only have to put the separate Scottish numbers into Anthony’s Advanced Swingometer to see the difference it makes.

    However the general incumbency bonus for the new Tory MPs seems much more dubious to me for various reasons. We only have evidence for the Lib Dem for it and that mainly when against Conservative opposition. But that might change with new polling – we just haven’t seen it yet.

  22. @Anthony Wells

    Thank you for contacting Ashcroft’s people about this.

  23. I agree with the conclusion that Ed Miliband is safe now.

    Readers may recollect that I made a number of posts on the Doncaster North thread suggesting Miliband might be at risk. Indeed I think that he might have been if UKIP had got their act together in Doncaster and South Yorkshire with a strong local campaign earlier.

    But the time has passed both because it is too late now for an outstanding local campaign and also because after the Rochester and Strood result UKIP nationally appear to have reached and perhaps gone past their peak.

    There is much to be said for the adage that the time to start campaigning for the next election is the day after the last one.

    Perhpas one might comment that Labour seems to have taken a number of South Yorkshire seats for granted as an area of safe seats reasonably close in journey time to London where potential party members can be parachuted in. I think Labour both in South Yorkshire and nationally would do better if they paid more attention to grassroots constituents (all of them, not just the core Labour vote) and less to the Westminster village and, in particular, to financial backers wanting their “pound of flesh”.

  24. As far as I can tell, incumbency only works when the MP has been working hard locally and has established a local positive profile. That fits very neatly into the LDs traditional modus operandi, and certain Labour MPs too (e.g. Karen Buck), but it’s far from obvious that it will apply to new Con MPs.

    Indeed, the whole series of Ashcroft marginal polls has repeatedly shown no incumbency bonus for sitting Tory MPs at all – the swing has always been in line with national polls.

  25. Never mind apologising, the Mail are still running the original.

    Published 10.13am today.

    http://www.dailymail.co.uk/debate/article-2855461/ANDREW-PIERCE-Clegg-Red-Ed-ex-MPs.html

  26. Good to see Ashcroft is for the most part being praised for his apology on Twitter and not the usual twitter storm of abuse. I believe it was a genuine error, something we all do and he showed how to handle it with class – fess up and apologise.

  27. I cannot believe the fuss some people make about nothing. If matters in Rotherham could not burn Labour out of office there, the leader of their party is hardly in danger, just down the road.

  28. “There is no risk to Ed Miliband here.”

    Weren’t Lab on 50% (or thereabouts) in Heywood?

    I appreciate by-elections are different but Lab’s problem generally is they know that a significant chunk of their nominal support won’t turn out but they don’t know how much of it won’t.

  29. ” If matters in Rotherham could not burn Labour out of office there, the leader of their party is hardly in danger, just down the road.”

    Using 2010 as the baseline for Lab support disguises the very slow and very gradual abandonment of Lab by one segment of their core vote over the last 16 years.

  30. I’m too dubious on this Conservative incumbency boost. Maybe they will get it in some areas and not in other. Only time will tell.

  31. Unicorn (fpt)

    Perhaps even the UKPR community is sometimes a little too ready to accept the headline figures without dwelling too long on the tinkering that goes on behind the scenes.

    You’re right, we probably underestimate the capacity for human error. I missed these originally and assumed the Doncaster changes might be demographic (that may well still be a factor).

    And it shows why it’s important to get the maximum information on the tables. Without the new format I couldn’t have analysed the reallocation. For example in the old format between Table 5 and Table 6, the likelihood to vote and removing DKs etc steps are combined, so you can’t work out what post-LTV DK figures are to see if they are added back in correctly. So Ashcroft and Survation should be praised for that as well as having their spreadsheet errors pointed out (who do they think they are, Harvard economists?).

    If Q3 is universally regarded as more informative in these seats, why doesn’t it become the standard question we use in all constituencies?

    […]If Q2 VIs understate incumbency effects (as the recent Lord A polls seemed to show) are the national polls underestimating Tory prospects?

    It’s more that they’re asking about different things. Q2 is measuring public opinion as it is today; Q3 is about how how people (think they) will actually vote in May. Usually the first is taken as a proxy for the second, but it’s not really the same thing. That’s why YouGov don’t normally use a LTV filter till near the election as they say it’s not relevant to Q2-type questions.

    It has to be said it doesn’t normally make much difference except in strong Lib Dem seats. If you compare the difference from Q2 to Q3 in the combined polls of the 11 Lib-Con seats in the latest Ashcroft main batch (not Hallam or Yardley) it’s quite striking (Q2,Q3 =difference):

    Con 33 27 = -6

    Lab 17 13 = -4

    Lib Dem 22 36 = +14

    UKIP 20 17 = -3

    Green 6 5 = -1

    Note that this isn’t only a tactical vote, because the biggest losers are the main challengers, the Conservatives.

    In contrast in the last batch of 12 Con-Lab marginals that were polled:

    http://lordashcroftpolls.com/wp-content/uploads/2014/11/Full-battleground-data-tables-Nov-14.pdf

    Con 32 33 = +1

    Lab 35 36 = +1

    Lib Dem 5 6 = +1

    UKIP 23 20 = -3

    Green 4 3 = -1

    It’s a pattern showing a slight boost for the ‘old’ Parties at the expense of the ‘new’ ones. It suggests a slight unwinding of the protest element, but that no ‘old’ Party benefits more than the others and it’s not that great in any case. The figures used are the published headline ones, but in both cases there’s very little difference to those before reallocation (UKIP loses another point in some cases).

  32. @AW

    No doubt my remark would have derailed the discussion when I think about it, but I don’t think the way I referred to Reinhart and Rogoff would be disputed by many on either side of the austerity fence.

    My point was that they didn’t make their work transparent and easily available, Ashcroft does and that’s to everyone’s benefit.

  33. @MR JONES
    I have heard this comment before regarding the northern blue collar vote. However GE’s come and go and still a Tory in S Yorks, Greater Manchester, Merseyside or Tyneside, is as rare as a Jewish Pope.

  34. @Roger Mexico.

    We all do it. People who analyse facts for a living frequently fall into the same trap: that the “facts” are not true, and that what was true then is not necessarily true now. Things change, and people make mistakes: twas ever thus…

    @Pressman

    Weirdly, this *is* a site for the neutral discussion of polls. There is much partisan discussion on here (because we’re all interested in politics), but that’s just the icing on the cake

    The meat of the icy cake (ouch!) Is the poll discussion and information provision: the number crunchers (unicorn,statgeek et all) and the information providers (amber, Colin, oldnat, virgilio, socaliberal, you, et al) provide free info that’s difficult to find elsewhere.

  35. Oh dear, another crisis for Ed Miliband! His presentational problems have become so bad they’ve started infecting pollsters asking questions about him. We should ask Lord Ashcroft to eat a bacon sandwich and see what happens.

    Still, a gracious apology on his website there. (And a nice incumbency bonus for Miliband. We’d expect it for a party leader, but it’s interesting to see where it’s coming from.) Good work, Roger and Anthony.

  36. Oh yes, and that Anthony Wells fellow provides some good stuff too…

    (Face palm, looks embarrassed, goes to fetch coat…)

  37. Ashcroft National:

    LAB – 32% (-)
    CON – 30% (+3)
    UKIP – 16% (-2)
    LDEM – 7% (-)
    GRN – 6% (-1)

  38. I love this bit in Kellner’s piece on uniform swing.

    “Things may – indeed, things WILL – change between now and next May. This analysis takes no account of possible Ukip gains. Nor does it allow for the way the parties will fight the marginals: will Labour win the “ground war” and reduce the Tories’ incumbency bonus? Will the Tories manage to squeeze Ukip support in Con-Lab contests with their message, “vote Farage, get Miliband”? Will Labour recover in Scotland? Above all, will the underlying support for each party change over the next five months – and, if so, how far and in what direction?”

    I mean, just how many riders and provisos does it take before the whole argument becomes, to put it diplomatically, academic self-indulgence?

    As for the incumbency bonus argument, doesn’t it all depend on the strength of the national mood for change? I can see it coming into play in some particular constituencies, but if there is a real desire to “throw the scoundrels out”, I suspect warm feelings towards the incumbent MP count for little as a vote for the party you want to govern, or the one you believe is best placed to beat the government, trumps all else. Many a good and well regarded MP has gone west when the tide is running against his/her party.

    As for Ashcroft’s slip of the weighting pen, it could happen to anyone I suppose, although it doesn’t help with the overall credibility of his polling, does it?

  39. Does anyone understand why Peter Kellner thinks the incumbency effect, which has existed for decades and may or may not help the Tories in 2015 (but probably will based on the historical evidence), will throw off UNS in this election but not in any of the previous elections that showed a strong incumbency effect?

    What’s happening in Scotland is a genuine transformation in the political landscape that logically should break UNS (obviously there can’t be a uniform swing to the SNP if they’re not standing in England, Wales and NI, so we’ll have to use Anthony’s special national UNS calculator), but his second argument seemed odd. UNS gave valid results in England in 2005 but not now?

  40. Mr N

    @”Ashcroft National:”

    The thing I find staggering in the Summary is that those few loyal souls who still declare a VI for LDs have amongst them, no less than 72% who might”end up voting differently” !

  41. Ashcroft polling showing Greens the ONLY minor party with a net positive that people would want to see involved in a coalition- and then only by a whisker. The SNP doing especially badly in that question- I guess people (in England &Wales) are probably worried about influence from a single region that maybe put them at a disadvantage.

    Well done Phil and Roger and Anthony for the Ashcroft correction.

  42. @Roland

    Sure, the Lab vote dropping obviously isn’t enough on its own and getting those sort of people to actually switch their vote rather than just stop voting entirely is very difficult especially in the NW quarter of England. My point was that the ex core Lab voters in all the areas where it has been happening have known about the grooming gangs for many years so measuring the political effect against when it was finally reported in the papers is the wrong baseline. The effect has been ongoing for 16 years – ironically leading to an ever greater reliance on the postal votes.

    I agree it’s slow and probably too slow for the GE unless Ukip figure out some magic words that work faster. I know guilt tripping works best face to face as I’ve been doing it for years but how to translate that into a political sound bite I don’t know.

  43. @crossbat

    The horrible truth is: you can’t really predict from a poll. We all know that, but we deliberately forget it and do it anyway. We invent swingback and apply it, but that’s just saying “if the vote changes by this much, then…the vote will change by this much”. People won’t make their mind up until march-ish, and until the, we’re knitting fictional wool…:-(

  44. Well done to all regarding the Ashcroft issue.

    Interesting stuff about Peter Kelner’s comments:

    “Uniform swing is now worse than useless – it is positively misleading”

    Of course when Con + Lab is struggling to maintain two thirds of the total voting intention, the UNS will fail to provide reliable information.

    UNS can’t allow for UKIP on 16% across GB, but only 2% in Scotland, and likewise the SNP on 42% in Scotland, but on 3.5% nationally.

    If we were to look at how best to design a government system for the UK from scratch, we would almost certainly opt for a federal system, given the huge political differences from North to South.

  45. Scotland cross break for Ashcroft

    SNP 49
    Lab 25
    Con 9
    LD 5
    Green 3
    UKIP 3

    Ashcroft VI before adjusting for turnout/don’t knows (trying to get a feel for the house effects of these adjustments across polling firms)

    Cons 28
    Lab 33
    LD 7
    SNP 5
    UKIP 17
    Greens 7

    So the net impact of the Ashcroft turnout/don’t know adjustments is fairly minor (Con+2, Lab-1, UKIP-1, Green-1)

  46. Crossbat

    I think the pollsters are in a real mood of confusion. Every time they think things can’t get more complex, they do. In Ashcroft’s latest piece:

    http://lordashcroftpolls.com/2014/12/ashcroft-national-poll-con-30-lab-32-lib-dem-7-ukip-16-green-6/

    commenting on how his last week’s poll split evenly on what the public expected the next government to be, he adds “(I think I am with the 21% who said they didn’t know what to expect)”.

  47. Anthony – maybe you could recommend to Lord Ashcroft and people like him that it might be worth them paying more attention to the comments of the various nerds and anoraks (amongst whom I humbly include myself) that frequent sites like UKPR and VoteUK discussion forum because they noticed immediately that there was something wrong with this poll.

    The main thing is that it wasn’t credible for the Tory vote to be up 2 points at the same time as UKIP being as high as 28%.

  48. @Crossbat
    “As for Ashcroft’s slip of the weighting pen, it could happen to anyone I suppose, although it doesn’t help with the overall credibility of his polling, does it?”
    On the grounds that a big change with huge consequences was worth a careful check before publishing?
    As AW said ” a human error – typing the wrong weighting target in the wrong box (it’s something I’ve done myself in the past, though thankfully I’ve not done it and published the results!).”

1 2 3