• The Guardian October Poll

    At a lead of 17-points, this is the joint second highest EVER recorded in the dating back to 1992, only beaten by a Gordon Brown crisis of June 2008. The Tories 43% is two-points adrift of its highest share ever recorded by ICM (three times after the 1992 election, Jun-08 and Nov-08 plus 44% on three occasions straddling 08-09). We did have them on 43% back in July this year.

    As for Labour, their 26% has only been ‘beaten’ by 25% in Jun 08 and Aug 09, and the last Guardian/ICM at this level was Sep 09 (the only other time). However, ICM did also put them on 26% three weeks’ ago for a Sunday newspaper.

    In fact though, Labour’s share has only been saved from a record low by ICM’s standard post-fieldwork adjustment techniques, which ordinarily help the Tories. The reason for this is that this week, there are a high number of people who say they voted Labour in 2015 but DK/refuse to say what they will do next time, and our reallocation of them back to the party for they voted for ends up adding two-points back to Labour.

    Labour’s struggles are especially prevalent among women (Tory split 39% vs 52% in favour of women).

    We also see further decline in the UKIP share, possibly prompted by recent fisticuffs and the return of Nigel Farage. At 11% they dip below their 2015 election share for the first time with ICM and it’s the lowest we’ve ever had them on an online poll (although phone polls pre-2015 had them lower very often).

  • Guardian/ICM voting intention poll

    ICM interviewed a nationally representative sample of 2,001 GB adults aged 18+. Fieldwork was conducted online on 24-26 June 2016.

  • EU referendum tracker

    This week, ICM’s regular online tracker sees a slight shift towards Leave, with 47% in favour of leaving vs 44% in support of remaining in the EU.

    However, our telephone poll conducted for the Guardian is also showing Leave ahead – for the first time – with 45% in favour of leaving vs 42% in support of remaining.

    For both polls, the overall result stands at 52% Leave vs 48% Remain.

    Online poll – weekly tracker:

    Overall Excluding DKs
    Remain 44% 48%
    Leave 47% 52%
    Don’t know 9%


    Telephone poll:

    Overall Excluding DKs
    Remain 42% 48%
    Leave 45% 52%
    Don’t know 13%


  • Guardian EU referendum poll

    The disparity in findings between online and phone polling on the EU referendum continues apace. In a second mode test, with exactly the same questions and exactly the same weighting adjustments being applied to both sets of data, a differing picture; indeed, a differing referendum outcome emerges.

    Phone Online
    Remain In 47% 43%
    Leave 39% 47%
    DK 14% 10%


    At some point, it’s fair to ask whether there will be a narrowing of the gap as we approach the June 23rd or whether this undoubted mode effect is set in stone? If polling history is to be believed (or indeed the recent British Polling Inquiry into the 2015 debacle, which identified a minor, but helpful herding effect) the former will be the case.

    I’m not so sure. The features that differentiate the modes on the EU question are persistent and consistent, and their effects are tending to work in opposite directions. Polling has often depended on hidden error cancelling itself out, but it seems increasingly unlikely that pollsters can depend on that on this occasion. So you pay your money and you take your chance on what you believe.

    If you want to ask me, which is unlikely, the answer you’d get is “I just don’t know”. I can see reasons why phone polls overstate Remain shares, and reasons why online polls overstate Leave shares. That inevitably leads to a conclusion that reality lies somewhere in the middle, but just hold that thought. More aggressive weighting schemes (privately) employed on these very data sets – schemes intended to correct for observed Westminster vote intention skews the like of which have previously consumed us – are not reducing the gap on the EU referendum but increasing it.

    The narrative that phone polls are more likely to be right ignores some fundamental flaws in phone methods. Labour supporters are continually oversampled by phone, and that may matter more than those same phone polls missing out on supposedly pro-Remain types, who are disproportionately less likely to turn out to vote. Similarly, what’s lurking under online covers could be equally nasty, and we should not ignore that the fact the UKIP voters are again, as they have long since been, higher in online polls than phone (or indeed at recent elections).

    Bemused? You have every right to be.

  • EU referendum tracker

    Keen followers of all things EU Referendum related will be aware of the divergence between outcomes derived from phone and online polls. To date, ICM has only published weekly online polls, but today we compare our usual weekend poll with a phone alternative, conducted on behalf of The Guardian for their monthly poll.

    Outside of the obvious difference in data collection mode, both polls employed analytical techniques as close to each other as possible. Both contained an identical suite of vote intention questions, were demographically and past vote weighted (PV targets set separately in the usual way for phone polls (80% to GE 2015 result and 20% to the average recall of that result) but all the way back to the General Election result for the online – which also employed 2015 vote quotas at interviewing stage). More on the online methodology will be released at a later date, with still more ICM analytical interventions (yet) to be applied. On this occasion, however, we limited those interventions to quotas and harder PV weighting in order to be as consistent as possible. The only other difference on the online poll was to exclude anyone from EU referendum voting intentions who claimed they were not registered to vote (2%).

    Turnout will be critical in this referendum, and both polls show great consistency in outcome. If we loosely point to the 10/10 certainty score as being more likely to reflect turnout than anything else, the consistency between the two polls (Phone: 65%; Online: 63%) is an advantage. On the assumption that all polls still overstate though, one might read that actual turnout will be in the high 50’s, although this particular pollster continues to be innately sceptical about levels this high.

    In line with noted differences between poll methods, the average likelihood to vote of an online respondent is slightly higher than a phone respondent.

    Turning to the referendum vote assessment, the previously established gap between phone polls and online polls is sustained. Remain enjoy a 7-point lead over the phone (48% vs 41% translating to 54% vs 46%) while Leave (this week) are neck-and-neck in the online equivalent (Remain 43% vs Leave 44%, translating to 50:50).


    Phone Online
    Remain 48% (54%) 43% (50%)
    Leave 41% (46%) 44% (50%)
    DK 11% (-) 13% (-)


    Previous mode comparisons have focussed on differing levels of “Don’t Know”, with phone polls finding fewer. This was the case in our test, but at 11% and 13% respectively there is little of note to compare. In our experience of online polls, the option to answer DK was linked to the (previous) inability to respond that they would not vote, and the addition of the turnout question has very much reduced the gap.

    But a gap does still exist, and in the expected direction. Although small comparative impurities might be present, it is difficult to attribute this gap to anything other than mode effect (the method of data collection itself causing different data to be collected).

    The question of ‘why’ is obviously one to speculate over, and much thought has already emerged, likely of some merit. However, the immediate need for us is to ponder over accuracy rather than theoretical abstraction. It’s hard not to agree that the ‘real’ answer lies somewhere between these two scores. I have long argued that phone polls contain too many Labour voters in their raw samples, and the demographically weighted base sample on this poll underlines that. Given that Labour voters split in the general ratio of 2:1 in favour of Remain, the view here is that the phone poll slightly overstate its share. Conversely, online polls over-stated UKIP before and during 2015, and (at least this one) continues to do so. If we accept that online polls probably contain too many UKIP supporters (even after quota controlling for them, as well as for supporters of all other parties) this will fairly obviously translate into help for Leave.

    So the answer does probably lie somewhere between the two. However, given that after the exclusion of DKs we have 54% vs 46% by phone, and 50% vs 50% online, that gap is not generous, making this referendum very much in play for both sides.

    What else can we take from the test? Well, there is enough consistency across demographics to take some comfort. Men are slightly more pro-Leave than women, and the long observed hardening of anti-EU sentiment does increase with age. Indeed, there is more clarity on this online, with older online respondents more likely to say Leave than their phone counterparts (there is some evidence to suggest that these older online respondents are also more UKIP inclined, which is worth a more detailed methodological poke for wider reasons). Social grade is clearly also correlated, with greater affluence linked to higher support for Remain. This is especially the case over the phone, where, it should be noted, raw samples traditionally strongly over-sample the more affluent AB group, as does this poll.

    Both retain the standard form of turnout modelling based on the 10-point turnout scale, with its net effect being negligible. This is likely because of largely offsetting effects. For example, the least affluent DE group are up-weighted for lack of raw numbers, but will be turnout down-weighted given long established disinclination to vote. Age is more powerful here, with more elderly people more likely to turnout, and vote Leave (which makes the Remain strategy for young people to ‘get granny out’ a highly risky one).

    Some people might also be interested in how the vote intention section compares. The headline phone VI vs headline online VI are follows:


    Phone Online
    Conservative 38% 36%
    Labour 33% 31%
    LibDem 7% 7%
    SNP 5% 4%
    PC 1% 1%
    Green 3% 4%
    UKIP 13% 16%
    Other *% 1%
    Total 100% 100%


    Last month’s Guardian poll, of course, was the first to spot a sudden change in favour of Labour. We doubted the truth of that and pointed to strong methodological causes that might undermine it. However, since then a number of other polls have revealed a similar outcome. Yet our poll reverts more to mean, with a 5-point Conservative lead (last month neck-and-neck) pointing to March being an outlier, were it not for other Labour lead polling emerging. It’s all in a state of flux, really.

  • Guardian Poll – December 2015

    The final Guardian/ICM poll of 2015 is pretty much as you were, with the Conservatives remaining static on 39% (but still a point ahead of their General Election showing) and Labour up one to now stand on 34% – equally their post-election high.

    Top line numbers are:

    Conservative 39%
    Labour 34%
    Liberal Democrats 7%
    UKIP 10%
    Green 3%
    SNP 4%
    Plaid Cymru 1%
    Other 2%

    That Labour stand above their election share is possibly less the result of a commanding leadership performance that has invigorated the Labour masses, but more likely the continued over-statement of Labour in phone polls for sampling reasons. It should be noted that nearly all online polling has Labour hovering around the 30% floor, and although this is only a relatively small variation between methods, it is sufficient to exercise the mind as pollsters (especially us here at ICM) await the findings of the BPC Inquiry in January.

    For our part, it is clear that phone polls steadfastly continue to collect too many Labour voters in the raw sample, and the challenge for phone polling is to find a way to overcome the systematic reasons for doing so. The methodological tweaks that we have introduced since the election in part help mitigate this phenomenon by proxy, but have not overcome the core challenge. In our view, attempting to fully solve sampling bias via post-survey adjustment methods is a step too far and lures the unsuspecting pollster into (further) blase confidence.

    We will have more to say on our methods in the coming months.

  • Guardian Poll – October 2015

    The latest Guardian/ICM poll for October reveals shows that far from causing Labour to haemorrhage voters, the new leader Jeremy Corbin has a little electoral traction. Gaining 2-points compared to last month, Labour now stands at 34%, a figure which matches that post-election high point. The Conservatives hold firm on 38% with UKIP on 11% and the Liberal Democrats on 7%.

    In effect the conference season has played out to a nil-all draw, with standings exactly as they were in mid-Summer.

    ICM Unlimited interviewed a random sample of 1,002 adults aged 18+ by telephone on 9-11th October 2015. Interviews were conducted across the country and the results have been weighted to the profile of all adults. ICM is a member of the British Polling Council and abides by its means.

  • Guardian Poll – September 2015

    The latest ICM/Guardian poll shows a slight drop in the Conservative (38%) share of the vote, with newly elected Labour leader Jeremy Corbyn enjoying an immediate honeymoon boost of 1-point (32%), which of course, is most likely to be a function of normal sampling variation.

    Full poll numbers are:

    Conservative 38% (-2)
    Labour 32% (+1)
    UKIP 13% (+3)
    Liberal Democrat 8% (+1)
    Green 3%
    SNP 5%
    PC 1%
    Other 1%

  • Guardian Poll – June 2015

    The first post-2015 national voting intention poll from ICM almost exactly mirrors the actual result of the election, with the Conservatives enjoying a 6-point lead over Labour. UKIP re-establish their 13% share, while the Liberal Democrats remain in the same lowly territory they found themselves in the real poll last month.

    It should be noted that all ICM methodological techniques are under review, and while this poll reflects pre-2015 approaches (with the exception of past vote weighting to the 2015 General Election), we will likely to be refining and/or introducing new measures once we have the full evidence base to support them.

  • BBC Question Time Election Leaders’ Special Poll

    A flash poll conducted by ICM for The Guardian after the BBC Question Time Election Leaders’ Special shows that David Cameron was judged to have ‘won’ the contest. Among a sample of over 1,000 people who watched the show, 44% said the Conservative prime minister performed best on the night, ahead of the Labour leader, Ed Miliband, on 38% and Nick Clegg, the leader of the Liberal Democrats, on 19%.

    When asked how they think each leader performed, Cameron – by a small margin – is adjudged to have been the most accomplished: 65% state that he did very or fairly well compared to 62% for Clegg and 61% for Miliband.

    ICM interviewed c.3,500 adults aged 18+ online on 28-29 April. All agreed to watch the BBC Question Time Election Leaders’ Special, and to complete a second interview immediately after it finished, which 1,288 did in the first few minutes. The data are weighted to the profile of all GB adults, including to recall of 2010 General Election voting. In essence, the post-wave data is ICM’s best guess on what a representative sample of the voting population would say had they all watched the programme.

    View original article