The full tabs for the ComRes European election poll are now online here. The poll also asked Westminister voting intention and produced topline figures of CON 26%, LAB 37%, LDEM 8%, UKIP 20%. It did not use ComRes’s normal method, excluding the reallocation of don’t knows and their normal squeeze question, so it is not directly comparable.

The strangest thing about the poll though is the relationship between Westminster and European voting intention. I could be reading it wrongly, but it looks as though there is hardly any relationship at all. The overwhelmingly majority of people who said they would vote UKIP in a Westminster election said they would vote UKIP in a European election, as one might expect, but the other figures look very odd. Amongst Conservative Westminster voters, 12% would vote UKIP in the European elections, 39% Labour and 22% Lib Dem. Amongst Westminster Labour voters, 6% would vote UKIP in the European elections, 30% Conservative, 31% Lib Dem. Amongst Westminister Liberal Democrat voters, 2% would vote UKIP in Europe, 35% Labour, 33% Conservative.

People do, of course, vote differently at Westminster and European elections, but not to this extent. If we compare it to the Survation poll at the weekend, the vast majority of people who said they’d vote Labour, Conservative and Lib Dem for Westminster said they vote the same way in a European election. YouGov found the same when they asked back in January. ComRes’s previous European Election poll in January didn’t give cross-breaks for current Westminster voting intention, but the 2010 cross-break did at last suggest the majority of Labour and Conservative 2010 voters were sticking with their party in the European elections. I can’t work out exactly what has gone wrong in this latest poll, but it certainly looks very strange.

66 Responses to “More on that ComRes European poll”

1 2
  1. Amateur hour!

  2. Perhaps the electorate, after the ludicruous police commissioner polls, just think it is all a joke and will vote at random, if at all.

  3. So ComRes found that when it comes to the EU elections the Westminster VI breaks down as follows…

    Con: Con 27%, Lab 39%, LD 22%
    Lab: Con 30%, Lab 31%, LD 30%
    LD: Con 33%, Lab 35%, LD 26%.


  4. I don’ think this is that complicated. The discrepancy is probably something to do with the fact that the poll is (i) reporting 10/10 certainty for Euros but 5 to 10 for Westminster (ii) not weighted in the same way. Would be interesting to see the comparison with those two amendments though.

  5. This result shows the wisdom of not rushing to judgement on the basis of a single poll. I would want to see a likely explanation in “events” for any significant change.

  6. @ Anthony Wells

    I ran a full set of correlations (I know the problem with doing it like this) – anyway, the link between the two sets of tables is essentially random.

  7. Repeated from last thread as Anthony put up the new one when I was writing it:

    ComRes have indeed only used likelihood to vote values of 10 for inclusion, which means they only use 40% of the sample. In contrast for General Election voting they use LTV of 5-10 inclusive, which means 88% of the sample was used.

    If you remember they did the same for their local elections poll, presumably with the same rationale that people are so much less likely to vote. But using such a tight criterion is wrong for two reasons. One is that it ignores the fact that people who respond to on-line surveys (say they) are more likely to vote than average. So using only 10s will exclude some people who will vote. Compare that 88% using 5-10 to the last general election turnout of 65%. But if you only use LTV=10, you would only use 53% of the sample – less than expected turnout even before you take the increased involvement of on-line panellists into account.

    But there’s another specific reason for using LTV=10 being wrong in this case. Next year’s Euros will be (probably) be held on the same day as the locals (on 22 May), so some of those uninterested in the Euros may come out to vote in them (and vice versa). ComRes’s figure for LTV=10 for the locals was 56%, so this 40% is looking very low indeed.

    This would be bad enough if ComRes then showed all the data in their tables, but they don’t show the complete LTV breakdown for each EU Party choice – just the number of 10’s (in contract they do give the breakdown for those those don’t include in the GE figures). Without that it’s impossible to tell what is going on.

    Even from what we can see the figures still look very odd. For example only only 47 people said they would certainly vote Lib Dem in a GE but 126 in the Euros. UKIP also show an increase (157 to 192) but it’s much smaller. Yet as Anthony points out Lib Dems always do worse in the Euros than other elections and UKIP much better.

  8. ComRes shouldn’t have published their European poll. I’m surprised they allowed it to be.

  9. @ Andy JS

    Probably OpenEurope wouldn’t have been happy (I guess it is the standard practice to include a point on making the research to public domain).

  10. “Include a point in the contract…”

    Apologies – but with all my restrain I still post from my mobile.

  11. Looking at the data the randomness problem only seems to arise with Con/Lab/LD (the UKIP and Green lines seem plausible). Are we looking at a situation where the 3 ‘main’ Parties have been displayed in random order, but the results being picked up as if they weren’t? It has to be something technical because no real sample would be this much out of skew.

  12. Has there been a virus going around ComRes, leaving the work experience student to complie the poll.

    Billy Bob, bizarre indeed.

    On the westminster VI, UKIP VI close to what Survation had, not near the 13-15% YouGov get. The plot thickens.

  13. Yes, because YouGov polls are never wrong, are they. ;p

  14. Perhaps the stupendous momentum is finally working

  15. Actually 10/10 is not a sensible choice. Studies have found that when asked to choose on a scale people will avoid the extremes – so even if certain to vote the person is likely to choose 9. People will also avoid 0.

  16. Roger

    There are only locals in some areas, so I guess turnout will rise from 35%(2009) to no higher than 45%.

  17. Using 10/10 is crazy. I always vote, yet I would only rank myself as a 9/10 to vote, becuase you never know, I could be ill or be in hospital, so I can never say 10/10 I will vote.

    Can’t they publish results using different senario.s? But state their prefered method? ie use a 6/7/8/9/10 aswell as a 5/10

  18. @ Roger Mexico

    Yes they do look more plausible (though it is not a criteria), but the paired correlations are still all over the place.

    I don’t know if they use spreadsheet or SPSS (and if the former is transferred to the latter) – but a rogue macro can easily cause such a havoc.

  19. Criteria is meant to be criterion

  20. None of the recent polling looks likely to make much sense until after Whitsun.

  21. @ Couper2802

    I’m not sure about your views about Liker-scale. it depends, I suppose.

    Anyway a 10-scale measure would disperse the results, so you would reduce it to high-low-medium in practice.

  22. ComRes need to issue a statement on these polls, because it is inconceivable that the cross tabs between the Euro and GE polls are correct. It’s possible that the headline figures in the Euro poll might still be correct, but as matters stand that is a heck of an assumption to make.

    Clearly ComRes are capable of publishing incorrect data. In this case it is as clear as day that they’ve messed up. The question is, have they messed up in other past polling, but less spectacularly such that the anomalies were put down to random variation?

  23. @Phil

    Just take ComRes results with a pinch of salt in the future.

    What was it Com Res = Comedy Results.

    Not that I suppose Nick Clegg or Nigel Farage will complain about the results.

  24. Couper – people do indeed tend to avoid the extremes… but not on that question, people are very willing to tick the extreme. And while there isn’t much difference between them, 9/10 voters are marginally less likely to actually end up voting than 10/10 voters.

  25. Good work colleagues. I have full confidence in tonight’s YG (assuming there will be one). I know people have theories about BH fieldwork but online contributors can do so from their deckchair on the lawn if they wish. Also they have all today up to – about now, I think.

  26. @LAZLO

    I learnt about respondents avoiding the extremes in Psychology; it is called Central Tendency Bias.

    I am suprised that ComRes or any polling company use 10/10 because I thought that Central Tendency Bias was well accepted.

    But because it is so long since I studied Psychology there may have been new research.

  27. @AW

    Sorry our posts crossed. OK you are the expert I am sure you are right.

  28. @ Howard

    It’s raining here in Liverpool… Heating is on… (Partly because the dog successfully caught some ugly cough).

  29. Couper – if anything people overestimate their likelihood to vote, so dissuading them from ticking 10/10 may even be a good thing!

    (More seriously, unlike voting intention which we can never check individual survey responses with individual votes, we can check people’s responses against the marked electoral register to see if they actually voted… and the 0/10 scale actually gives a pretty fair correlation against how likely respondents actually are to vote – see the graph in the page I linked to above)

  30. Laszlo

    Yes, here too (no dogs though, but Mrs H is not well in that area).

    I suppose lots of YG respondents do so from work, otherwise I can’t see when they find the time.

  31. Howard. I am pretty certain they do. We seem to get rather a lot of responses first thing in the morning when people would be getting into the office, or when people would be having lunch.

  32. AW
    Ah, the marked sheets. I wish I had a fiver for every time someone said ‘I don’t vote’ and then turned up on them!

  33. @ Couper2802

    I know and it is pretty strong as you said (since the problems with the research of the Dutch social psychologist, it’s more dubious anyway).

    When I try to understand organisations and ask these 1-10, 1-5 questions, people do give socially expected ones even if they are extreme (5 or 10). Then a supplementary question or a differently worded opposite scaled question actually show that the original answer was improbable.

    Avoidance of extremes is stronger when it’s visual (eg you give a 10 cm line and ask people to mark their position on it.

    So, it’s a big mix. I tend to be contended with high-medium-low in my own field, but even then I have problems when the respondent gives “considerable” or “limited”

  34. Sorry Anthony, I normally use qualitative research techniques and then turn them to dichotomised quantitative analysis (recombining them eventually). And I know it is completely unacceptable in pure quantitative analysis like polling at the scale of 1000+.

  35. I’ve always assumed (perhaps wrongly) that the number of people who change their vote is overstated. More to do with people switching between voting/not voting than switching between parties. Seems very weird that so many would actively support different parties like that, with the exception of UKIP who are a sink for protest votes in Euro elections that few people bother to vote in or take seriously.

  36. Laszlo – thinking about what you said it may have been a very deliberate decision of polling companies to switch from the old 5 point verbal scale to a 0-10 numerical scale to try and take advantage of people’s reluctance to pick the extreme.

    The old 5 point verbal scale used to invariably show far more people definite to vote than actually did, so pollsters switched to 0-10 and the 10/10 figure does at least not look absurd.

    Essentially its a battle between social desirability people bias (voting is seen as the socially responsible thing to do) and people’s reluctance to give an extreme answer. I suspect there are all sorts of underlying problems with the approach, but empirically it seems to work fairly well.

  37. dan the man

    There are only locals in some areas, so I guess turnout will rise from 35%(2009) to no higher than 45%.

    Oh quite. There’s none in Scotland and Wales for example. Actually even 40% might be a bit high (the previous best was 38.5% in 2004) and there were locals in 2009 as well, However if UKIP are still as prominent as now, it will be bound to boost things as well, so we could see 40%+.

    I was more trying to make the point that ComRes respondents wouldn’t know about the locals and some might have been more willing to say “10” if they did.

    The more important point is that although the 40% of those polled being 10’s might look plausible in the context of a normal EU turnout, it actually isn’t because people in opinion polls are always more biased towards involvement. So the 56% who said “10” for the Locals to ComRes became about 33% at the polling station[1]. So if you’re expecting 35-40% at the ballot box, you should be including say 50-60% of those polled in your figures.

    [1] I can’t find a reliable turnout figure for this May. Can anyone do better?

  38. @ Phil Haines

    The question is, have they [ComRes] messed up in other past polling.
    We strongly suspect they have. They did a dramatic press release one time & we questioned it when the tables were released. IIRC, Anthony contacted them & asked about it. They said it was not an error, it was a method change but they were not keen to share what the method change was. IMO, they simply made a mistake & refused to own up to it.

  39. @Roger Mexico

    Only local figures are available – it seems that it varied between 21 and 37% – but I gave up looking at all of the counties and within them constituencies.

  40. Anthony – are the Conservatives on 28% or 26% with this poll?

  41. @ Anthony Wells

    Yes, I can see the point and I can also see how it could work on large samples (though I still think it would make sense to trace the 10-point scale to 5. There’s probably a discernable difference between 9 and 10, but I would be surprised if on a sample of 1,003 it would make a significant difference).

    Still I doubt if this is the core issue with this strange poll.

  42. @ Howard

    “I wish I had a fiver for every time someone said ‘I don’t vote’ and then turned up on them!”

    It is a better tactic than telling everyone who canvases you that you are voting for them. One person I was ‘knocking up’ when they hadn’t voted must have done that and appeared to have been pestered by every single party on the night- mistake number 2 was probably not having voted in the first place!

  43. Andy js – 26% my mistake.

  44. Is there a typo in the first para? Previous article on this poll gave LDs on 18 not 8.

  45. I emailed Andrew Hawkins earlier to see if he had any light to shed but simply got this:

    “I am on leave until Monday 3rd June. In my absence please contact the ComRes switchboard on +44(0) 20 7871 8660.”

  46. sp

    “Yes, because YouGov polls are never wrong, are they. ;p”

    Do you have an uncontrollable urge to criticise everything and everyone? I’m sure you could join a different polling site and Anthony would cope okay without your witty repartee.

  47. @ Croftee

    Kippers seem not to like YG because it shows more muted support for UKIP than most of the other firms. And I think SEAL PUrPle is a kipper. ;-)

  48. Andy JS

    According to the tables:

    the Westminster VI figures (f/w 22-24 May) are:

    Labour 37% (35)

    Conservative 26% (29)

    UKIP 20% (19)

    Lib Dem 9% (8)

    SNP 3% (3)

    Green 3% (4)

    BNP 1% (1)

    Plaid Cymru 1% (*)

    Other 1% (*)

    Figure in brackets are last ComRes on-line poll (15-16 May), though not strictly comparable as they don’t seem to have asked the squeeze questions – though it doesn’t make much difference.

  49. Looks like Cameron is doing a great job of ruining the Conservative party. 26%, that is dire.

    Some back benchers must be plotting to get rid of him.

    Will am MP jump ship if their polling VI falls any lower?

  50. Amberyee

    Ah well, being rude to the host is ticketyboo in that case.

    [He’s not as nice as my pups though.]

1 2