Friday’s polls

I will hopefully be doing my usual weekly round up later on tonight, this is just a swift update on Friday’s polls (after all, it’s very unfair if they never get their own post!)

The daily YouGov/Sun poll this morning has topline figures of CON 33%, LAB 32%, LDEM 7%, UKIP 16%, GRN 6% (tabs). Of the four YouGov polls so far this week there have been three Tory leads, one Labour one.

There was also a YouGov Scottish poll. Full results of that are here. I may write some more about that in the weekly round up, but for now I’ll just say that the SNP continue to show a strong lead, with topline Westminster voting intentions of CON 18%, LAB 27%, LDEM 4%, SNP 46%, GRN 3%. Note that there is a methodology change in this poll, with the sample additionally weighted to match how people voted in the Independence referendum.

The twice-weekly Populus poll has figures of CON 29%, LAB 32%, LDEM 8%, UKIP 18%, GRN 6% (tabs). After a few weeks when it’s looked as if there might be a slight squeeze on UKIP support the 18% score is actually the highest Populus have recorded for them, though all the usual caveats apply about it just being one poll and probably a blip unless other polls show the same. Note that the fieldwork was Wednesday and Thursday so partly pre-dated Nigel Farage’s comments on discrimination laws.


100 Responses to “Friday’s polls”

1 2
  1. @BristolianHoward
    “I understood Bromley was almost the sole reason Ken L was not able to remain Mayor of London and the irony of that is that I understand Bromley does not want to be in London in the first place.”

    Well, it was the doughnut strategy that did for KL. But yes the biggest Boris majorities come in the Bexley and Bromley GLA ward. Bromley does not think of itself as London. Probably because it was part of Kent until 1965

  2. RAF
    That does make some sense. I recall there was a degree of acrimony in Tory ranks caused by the boundary changes that came into place before the February 1974 election. Hornsby -Smith clearly expected to be the candidate for the new Sidcup seat but lost out to Ted Heath as the sitting PM who opted for the new ultra safe Sidcup rather than the much more marginal Bexleyheath. As it happens the Tories had no problems with the latter until 1997 but it was seen as quite marginal at the time. Hornsby-Smith had to make do with Aldridge-Brownhills in the West Midlands where she was narrowly defeated.

  3. Bromley is not cosmopolitan by London standards and is too wealthy to go UKIP.

  4. More on the Labour-favouring bias of Electionforecast and May2015

    I take it as agreed that it is worth knowing about the systematic biases shown by the various models we consult for seat projections. In recent weeks I have pointed out the extent to which some models assign to Labour many of the seats that are placed firmly in the Tory column on the basis of Ashcroft constituency polling (CP) data. This is particularly true of seat-projection models that don’t make use of CP data (e.g., Electoral Calculus, and May2014 – with Ashcroft data ‘switched of’). However, I have also noted hits of bias in CP models such as the ElectionForecast (EF) model and May2015. Recently the EF has been given a vote of support by Nate Silver and so it seems sensible to look at the data in a bit more detail.

    I have kept full lists of EF snapshots downloaded the day before each of the four batches of Ashcroft constituency polls reported since December (comprising 46 seats in all). In the case of May2015, I missed their projections for the December batch but recorded the rest since then. By comparing the polling VIs with the projections, it is possible to assess the degree of agreement between the models and the data. I have summarised overall accuracy measures after each batch of polls. Here my concern is more with systematic bias.

    The evidence suggests that, in advance of incorporating Ashcroft data, EF and May2015 both show a small but reliable bias in favour of Labour. For May2015, my material is very limited. For each seat, their projection is restricted to three bits of information: Winner, Runner-up and margin (%). So, it is only in Tory vs Labour seats that they report the relative standing of the two parties.

    It turns out that there were just eight such seats in Ashcroft constituencies this year (as a lot of seats were in Scotland). Across these 8, May2015 showed a (just) reliable 2.9% Labour-favouring bias relative to Ashcroft CPs. That is Labour margins were a little higher in Labour-projected seats or a little lower in those called for the Tories.

    For the EF model the dataset was much more comprehensive. (There are Labour/Tory margins for all 46 seats). In this case, the Labour-favouring bias was 2.5% and was highly reliable. (Given that it took four months to accumulate the polling data, if the data pattern were just a chance effect based on no real bias, it would take roughly 200 years before another such chance effect would occur.) Related to this there was a marked, reliable tendency to understate Tory VIs alone and a more modest (but still reliable) tendency to overestimate Labour VIs. Over the dataset as a whole, I was unable to detect any systematic biases either way for any of the other parties. SNP and Ukip projections were often quite inaccurate, but with errors falling on either side of the polling figures.

    The density of the Labour/Tory battleground is such that a shift of 2.5% could cause as many as 15 seats to change hands (with the margin varying by 30 seats).

    There are three ways of interpreting this bias. The first is to treat the Ashcroft polls as Tory-favouring and assume that there is no residual bias in the models’ projections. This is difficult to evaluation since we don’t even know for certain which companies conduct the A-polling. The second is to treat the biases as real and to adjust the overall projections accordingly. This would mean increasing the Tory margin by about 30 seats in current projections. The third possibility is that the bias represents some kind of incumbency advantage captured by the Ashcroft CVI measure but not by the models. On this account, the EF Swingback adjustment may have the effect of eliminating the bias. It has always been possible that one element of Swingback is an incumbency effect which manifests itself only in the polling booth and is invisible prior to that. On this assumption there may be nothing slanted about the overall projections of the model.

    I have no idea which of these possibilities is closest to the truth, but if the biases are unaccounted for, then this could alter the form of future coalitions.

  5. @Graham

    Thanks for that. I have only been around since 1975 and have only lived in Bromley since 2007, so unfortunately my knowledge of local history is more acquired than experienced.

    Actually AW is from Dartford which is close to Bexleyheath/Sidcup so he may be able to assist you!

  6. Unicorn

    That analysis is much appreciated!

  7. Another extract from the link

    “Perhaps more importantly, SNP voters would prefer a Tory government and lots of SNP MPs to a Labour government and only a few SNP MPs

  8. Another extract from the link

    “Perhaps more importantly, SNP voters would prefer a Tory government and lots of SNP MPs to a Labour government and only a few SNP MPs”

    That’s something I have been saying for a long time.

  9. @AC

    Okay what about the likely outcome of a Labour government and lots of SNP MPs?

  10. @Unicorn

    By comparing the polling VIs with the projections, it is possible to assess the degree of agreement between the models and the data
    _____________________________________

    Looking at previous by elections, example Newark

    http://en.wikipedia.org/wiki/Newark_by-election,_2014

    You can see there was a small party squeeze between the polls 2 months out vs the final result.

    In this case the top 2 were Cons and UKIP, so Labour and LD votes were squeezed, and ended up going to the Tories.

    So I don’t think projections that give results that match polls 2 months out are necessarily the most accurate.

    The more accurate ones would be the ones that look at who the top 2 parties are in each seat, and then try and project who the smaller parties votes will be squeezed towards, and do an adjustment for that.

    And we can only test how well they did on election day.

  11. @Alan Christie

    The question seems to be little bit of YouGov mischief making as the only options give were 1) Conservative Government with large number of SNP members or 2) Labour Government with small number of SNP members.

    The option of Labour Government with large number of SNP members was not presented as an option despite polls showing this to be the most likely option at this time.

    For an explanation of that obvious omission, you will have to ask YouGov for an explanation.

  12. POINTER

    What constitutes a freind or an enemy of Labour? Currently 67% of Britain say they will not vote Labour []

  13. @RMJ1

    Nah, I suspect snark about the dozenth poll this year showing SNP ahead in Scotland is hardly going to cause me problems. I mean, is anyone anywhere shocked whatever their political leanings?

    I do moderate it here though and avoid calling certain politicians bad names. ;-)

  14. RAF
    @AC
    Okay what about the likely outcome of a Labour government and lots of SNP MPs?
    _______

    I’m sure if that question was asked then polling being polling and the public being the public the result would probably have most people (SNP supporters) baking that outcome.

    Mind the public can be weird at times. ;-) That’s a joke.

  15. BRIAN NICHOLSON

    I probably answered your question in my response to RAF and of course it’s all subjective nonsense from the pollsters but you know it can make fun headlines.

    I like fun headlines.

  16. #backing

  17. Further review of candidates running in Scotland reveals that thus far UKIP have only nominated 10 candidates.

    If you factor in that UKIP’s impact on this election will be negligible in Scotland and transfer those potential votes to Conservative, and presume that Green will support SNP in 27 seats where they have no candidate – the Scotland votes site projects 50 seats for SNP, 6 Labour, 2 Conservative and 1 Liberal.

    As for Conservatives, Labour and LD crossing over with each other, did that not already happen in the referendum campaign?

    It certainly did in Quebec after the 1993 federal election and referendum, where the Conservatives seemed to gain some seats around Quebec City whereas the Liberal remained strong in and around Montreal, until the 2011 federal election when the NDP (Labour) took 59 of 75 seats.

    I can also point to instances where in Alberta, again after 1993, both Liberal and NDP (and sometimes Progressive Conservative) campaign teams merged during the election period to try and prevent Reform (UKIP) winning the seat.

    The most famous instance of this was in 2000 when the Liberal and NDP campaign teams publicly collapsed their campaigns and went and worked to elect a former Progressive Conservative Prime Minister to defeat a Reform candidate in Calgary.

    So my answer is yes, if the ideological differences and personal animosity gets hot enough, desperate people will take previously thought of unusual actions.

    I, personally, would not want to be a long time LD activist in this 2015 GE anywhere in the UK. But the one thing I have learnt about politics over the last fifty two years is that change is a constant.

  18. Question :

    Will the SNP achieve quite so many seats at the GE ?
    Will Labour be wiped out ?

    Personally I doubt it .

    Won’t there be a swing back to Labour ?

    In the constituencies with huge Labour majorities , a small reduction in the SNP polling will see Labour hold these seats with reduced majorities surely ?

  19. Are the Libs and Lab still active in Bromley RAF?

  20. @Cover Drive

    The LDs are. Their candidate Sam Webber had stood a few times for various elections. He’s popular and well known locally. I haven’t heard of the Labour candidate and doubt they will be campaigning very much. However, this is such a Blue seat Lab and LDs will be better off casting their net over other parts of the capital.

  21. Are we getting no more polls tonight?

    I thought we had an ICM on the way this week. This is very disappointing… I think I’m developing an addiction to polls.

  22. @Cover Drive

    That said, Labour did win a few seats in the Council Elections last year with the LDs wiped out:

    Con 51 Lab 7 Ukip 2

  23. @Allan Christie

    “Perhaps more importantly, SNP voters would prefer a Tory government and lots of SNP MPs to a Labour government and only a few SNP MPs

    That’s something I have been saying for a long time.”

    Election Forecast’s current prediction map has in Scotland the SNP up to 46, Lab on 8, with 3 Tory and 2 LD seats. (The map is not quite the same as the table for some reason).

    East Ren remains stubbornly red however!

    On their current UK wide figures Con+LD+DUP would have enough and 46 SNP members would sit in Westminster.

    In other words the cited ”SNP voters would prefer a Tory government and lots of SNP MPs” option would come to pass.

  24. @Richard

    So I don’t think projections that give results that match polls 2 months out are necessarily the most accurate.

    As you no doubt know, the Electionforecast team publish two lists. One aims to capture how things stand today and specifically expresses an ambition to predict Ashcroft polls. The second is a projection for the general election.

    My analysis was concerned only with the former. By introducing the notions of swings linked to tactical voting you are effectively moving on to a different question: the accuracy of GE seat projections. A proper analysis of this must necessarily wait until May 8.

  25. RAF

    As far as the Bromley and Chislehurst constituency is concerned, it covers most of the more suburban parts of the borough with some rural pockets. This is a Conservative bastion with very little Ukip presence, excepting perhaps Mottingham and Cray Valley (the latter of which is more Orpington than Bromley anyway).

    It’s one of those odd ‘left-overs’ constituencies which you get when more coherent ones (Beckenham, Orpington) are chopped out of a borough. It’s also changed a fair bit over the years from demographic change and mergers.

    As it happens there are actually two UKIP councillors in the constituency (the only ones in Bromley) in Cray Valley West:

    https://londondatastore-upload.s3.amazonaws.com/London-Borough-Council-Elections-2014.pdf#page=80

    (The current wards are: Bickley, Bromley Town, Chislehurst, Cray Valley West, Mottingham and Chislehurst North, and Plaistow and Sundridge).

    UKIP gained the seats on a fairly low percentage in what is a traditional Con-Lab marginal ward with an Independent ex-councillor taking votes from both (the two UKIP candidates elected probably did well from his second and third votes). (If UKIP get MPs most will be elected similarly). The seats they won in London were in similar sorts of wards in metropolitan Kent or Essex (in Bexley and Havering respectively)

    However most of the constituency is either too mixed (Bromley Town) or too wealthy (Chislehurst), though UKIP did well further out on the Kent borders in Bromley because those were safe Tory wards, 30% gives you a distant second rather than a win.

    The old Chislehurst constituency consisted of the old Chislehurst and Sidcup boroughs of Kent and so was more marginal, going Labour in 1966 and indeed 1945.

  26. @Roger Mexico

    Thanks for the coherent analysis :)

  27. Hmmmm.

    Anthony has changed his polling averages to 33 for the Tories and 32 for Labour. This is probably the first switch for a very long time.

    My own calculations using his weight give the Tory average as 32.522 and Labour 32.503.

    I can see why the Tory figure is rounded up to 33%. But why is the Labout figure rounded down?

    The unweighted average of all the polls included in the list give the Tories 33.7 and Labour 33.9.

    The fact is that the VIs are almost identical at the moment.

  28. Labour have rather let the side down by choosing a new candidate (though he seems to be a long way from your average SpAd fighting the token hopeless seat before receiving a parachute). However the candidate for the other four main Parties in Bromley and Chislehurst all seem to be the same as in 2010:

    http://en.wikipedia.org/wiki/Bromley_and_Chislehurst_(UK_Parliament_constituency)

    which must be fairly unusual.

  29. @ Unicorn

    There is an assumption among some that voters who are considering opting for someone other than Conservative and Labour will somehow miraculously be persuaded to return to the fold sometime before May 7th.

    My experiential observations are that during a transition phase, especially when it is unclear which parties are in fact in the lead, some voters who have already cut themselves loose from the mainstream political parties are not that easy to coral.

    In Canada when the Reform (UKIP) party emerged and Progressive Conservatives collapsed it took about a decade and three more elections for things to settle out.

    And in fact it took thirteen years and a merger before the centre-right won power again in Canada, and another two elections, five years, before they achieved a majority of seats in the HoC.

    What I keep seeing, in this UK election, is a basic reluctance of more than 65% of the electorate to even consider voting for the two “naturally governing” parties, something that has not happened since the Liberals first disappeared as a governing party in the 1920’s.

  30. Unicorn’s analysis of the various election forecast models, and their seeming to have a marginal bias to Labour compared to Ashcroft constituency polling is absolutely fascinating – thank you Unicorn!

    It does add further weight to what I have suspected is a key issue which I highlighted a couple of posts ago. Namely, Conservative incumbents in marginal constituencies, where there are few public sector workers, may well resist the general swing to Labour in seats where public sector worker density is higher.

    If this happens it will make the application of UNS defective as a guide to the ranking of vulnerability to Labour capture.

    So, not all marginals with seemingly near identical majorities, are in reality as marginal as each other?

    Does anyone have data for the proportion of public sectors workers by constituency in the Con/Lab marginals – or point me in the direction of where this information is available?

    I have a hunch it may well be a major determining factor in the outcome!

  31. I know pretty much nothing about NI politics and have seen the possibility of a Tory/DUP coalition raised elsewhere, but recent comments from them seem to be less that enthusiastic. Of course, that doesn’t make it impossible but they seem to be leaning towards Labour from some reports.

    “DUP leader raises prospect of Lab coalition”

    http://www.theweek.co.uk/politics/general-election-2015/62049/general-election-2015-labour-must-do-more-for-gifted-children-says-Hunt

  32. UNICORN

    @” But why is the Labout figure rounded down?”

    Because the weighted average from AW’s current “Polls currently included in the average” Table is 32.42721.

    His Conservative weighted average is 32.5758-which he has rounded up.

  33. @Unicorn

    As you no doubt know, the Electionforecast team publish two lists. One aims to capture how things stand today and specifically expresses an ambition to predict Ashcroft polls. The second is a projection for the general election.
    ___________________________________________

    Thanks, no I had not realised this

    But it is useful as now we can look at a seat by seat level what they expect to happen between now and election day.

    I’ll pick one of Ashcroft’s recent marginals, Colne Valley

    Their current projection is, pretty much matching the Ashcroft poll
    Con 33, Lab 32, LD 11, Green 9, UKIP 14

    Their final projection is
    Con 36, Lab 32, LD 14, Green 8, UKIP 8

    So they seem to be using 2010 vote share to project where we will be on election day.

    Just looking at recent by elections, in a 2 horse race between Lab and Cons, that does not make sense.

    Around half or more of the LD’s will probably move to Lab, the Greens will probably move a bit to Lab, and some of the UKIP may move to Cons, or vice versa, that UKIP– Tory swing seems to do its own thing.

    The LD’s will definitely not increase their share in a seat where they are no longer in contention when they are so unpopular. Even they have said they only expect to hold on in seats they hold.

    I guess I am starting to understand what Andy has been going on about….

  34. @statgeek

    The Lib Dems are still counted as a major party in Scotland by Ofcom because of their 2010 GE result. If they are reduced to ~5% of the vote and only 1-2 MPs they would be in some danger of losing it.

    I don’t think that status will be an issue in the 2016 Holyrood election because I think the distinction will be fudged by having five way debates. The current composition of Holyrood and the polling (Greens do better in Holyrood polls) would point towards that.

  35. J R Tomlin

    There’s a good chance that you know as much about NI politics as many in GB.

    Political parties there, as everywhere else, will make their calculations as to whether to support either of the big parties or sit on the sidelines, as best suits their interests.

    Who is Secretary of State for Communities and Local Government is of as little concern in NI as it is in Scotland.

  36. @Andy S

    This is a very tardy response to a query you posted a couple of threads ago.

    You asked why no one was giving serious consideraton to the possibility that Ukip may take significant numbers of seats in middle England. You identified a number of patterns that may help their prospects and argued that their eventual seat counts may be much higher than most are expecting.

    My own reason for taking this view is that in most of their target seats they are in competition with a Tory party that – on today’s polling averages – is only 3-4 points down in their 2010 vote share. In that election the best Ukip showing was less than 9% and even assuming they scoop up all BNP support, in most seats they have en enormous mountain to climb to dislodge the Tories.

    Take an example of one target which may be just about reachable: Louth and Horncastle. Here the 2010 results were: CON (50%, rounded), LAB (17%), LD (22%), Ukip (4.3%), BNP (4.4%) and Greens (No candidate). Using regional crossbreaks, assume the Tories fall to the low forties. Allow Ukip their 12 point national boost, all the ex-BNP voters, a good percentage of walkabout LibDems, and even some Labour tactical voters. Even with all this help it is difficult to see them garnering enough support to out-poll the Tories’ starting position. The same kind of pattern repeats itself over and over again.

    Add to this the fact that the Ukip VI is now below its late-2014 peaks (on at least some measures), and it is difficult to see why anyone should be more optmistic about their propects than the 2-4 seats allocated to them by most of the models.

  37. I didn’t realise that Bromley extended as far east as the Cray Valley. I could indeed see UKIP doing something in the St Mary Cray area.

  38. Two questions perhaps to ponder while you’re waiting for the next set of poll results to arrive.

    1) How many of the present set of mainstream party leaders will still be in office in twelve months’ time?

    2) Will the Greens be the first political party in this country to elect a hamster as leader?

  39. @Richard

    So they seem to be using 2010 vote share to project where we will be on election day.

    Correct. Some weeks ago they were ‘giving back’ to parties about 50% of the amount they had dropped since 2010, and taking away an equivalent amount from parties currently in the ascendent. These adjustments are now being steadily scaled back as the election approaches.

    There is no provision for tactical voting and to the extent that this become important it is possible that quite a lot will change. But my own concern was with systematic biases that might tilt several seats one way or the other.

  40. @ Unicorn

    Unicorn’s analysis of the various election forecast models, and their seeming to have a marginal bias to Labour compared to Ashcroft constituency polling is absolutely fascinating – thank you Unicorn!

    For your next instalment, maybe you could give some consideration to Ashcroft’s polls having a ‘marginal bias’ against Labour. ;-)

  41. @ Hawthorn

    Bromley is wealthy as you said.

    However, in my grandson’s class in one of the state schools there, there were 7 nationalities (and in spite of being my grandson he was counted as British). To me, it’s cosmopolitan.

  42. @Colin

    Because the weighted average from AW’s current “Polls currently included in the average” Table is 32.42721.

    That’s not the figure that comes out of my spreadsheet. I multiply the Populus 32 by 0.95 to get 30.4 and work through the next 18 eventually getting to Opinium’s 33 time 0.01 = 0.33. These products sum to 209.97. Dividing this figure by the sum of the weights (= 6.46) gives 32.5031.

    I have no idea where your slightly lower figure comes from.

  43. @ Unicorn

    As usual thank you. What I am wondering about is whether UKIP is going to get a boost once the writ is dropped because Farage is back in the limelight and thus when I look at a seat like Cannock Chase, Portsmouth South, St Austell and Newquay, etc I think that 3 seats is a bit low and maybe 10 might be a better number to project.

    The bloom may have come off UKIP, I still have them down .6% on the beginning of February but surely so is LD down from October 2014.

    The Midlands appears to be rock solid for UKIP between high teens to possibly low 20’s and you may indeed be very right that it will not translate into any seats.

    My nephew, who is in England, thinks they will do well along the the East Coast of England.

  44. @Amber

    I’m not sure of the methodology for Lord A’s constituency polls, but his national polls seem to be heavily affected by likelihood to vote. Take that away and his polls are consistent with the others.

  45. The pollsters appear to be showing somewhat different trends at present. YouGov provides evidence of slight swingback to the Tories recently, whilst Populus records a better position for Labour than the end of August last year which was the last occasion the Tories were ahead. MORI is also better for Labour than in the last quarter of 2014.

  46. @Amber

    Ashcroft polling bias?

    Quite possible..and included in the summary at the end of my analysis.

    But prior to the election itself, what evidence is there we could use to test this possibility? All will become clear(er) in mid-May.

    But for now, I would adjudicate in favour of polling data rather than ‘VI reallocation algorithms’. The least that can be said is that polling methodology has been through years of refinement to deal with biases that have emerged in the past. With a longer track record there are stronger grounds for placing more weight on polls.

    This is certainly the explicit position taken by both Electionforecast and by May2015. In the absence of polling data they try to calculate how things might stand in each constituency. But when someone does a constituency poll, the calculations are jettisoned abruptly and replaced by the new polling data.

  47. Raf – I haven’t checked to see what the average effect of turnout weighting is on Ashcroft’s polls (I did think week, but I’ve no idea if that’s typical), but there is no obvious reason why his methodology would produce a large impact.

    Ashcroft’s turnout weighting is relatively mild – it’s MORI who have very strict turnout weighting. ComRes and ICM are also stricter than Ashcroft’s approach in their own ways.

  48. Weekly summary on New Thread!

  49. UNICORN

    From AW’s current table.

  50. @AW

    MS said (well tweeted) earlier this week that Ashcroft’s recent Con 34 Lab 30 poll (1st March) would have been level without LTV; and Ashcroft’s latest national VI poll – Con 34 Lab 30 (8 March) would have had the Tories just ahead without LTV.

1 2