We have three new polls so far today. TNS have put out a new GB poll, which has topline figures of CON 33%(nc), LAB 32%(nc), LDEM 8%(+1), UKIP 16%(-1), GRN 5%(+1) – clearly no significant change since their previous poll (tabs are here).

ComRes have a new poll of the 40 Labour held constituencies in Scotland (that is excluding Falkirk, where Eric Joyce sat out his term as an independent). In 2010 the share of vote in these seats was CON 14%, LAB 51%, SNP 19%, LDEM 14%. The new ComRes poll found support standing at CON 13%(-1), LAB 37%(-14), SNP 43%(+24), LDEM 2%(-12). The seven point SNP lead represents a swing of 19 points from Lab to SNP, the equivalent of a sixteen point SNP lead in a national Scottish poll (tabs are here).

Finally YouGov have a new London poll for the Evening Standard, which has topline figures of CON 34%(+2), LAB 45%(+1), LDEM 8%(+1), UKIP 8%(-2), GRN 4%(-1) – changes are from YouGov’s previous London poll a month ago. The eleven point Labour lead represents a swing of 4.5 points from Con to Lab since the general election, the equivalent of a two point Labour lead in a national GB poll (tabs are here).


444 Responses to “ComRes in Scotland, YouGov in London and latest TNS poll”

1 7 8 9
  1. @ omnishambles

    Thanks for posting the link to the new YouGov projections.

    I can’t find any description of the methods they are using. Judging by a comparison between the Nowcast and Forecast seat tallies, it is yet another Swingback model.

  2. How well did Election Forecast, May2015 and Electoral Calculus do in predicting today’s Ashcroft constituency polls?

    The short answer is that EF did better than the other two, but that this show some consistent biases that might be worth bearing in mind when using their projections In future.

    May2015 only gives the projected winner’s margin over the runner-up and so the only way of assessing the performance of this model is to compare this margin with that shown in the Ashcroft polls, and then I turn see how the discrepancies compare with those shown by EF and EC. In the analyses reported below, I have concentrated on the CVI measure as this is the one that everyone seems to take seriously. (I still think there is a question about whether this will prove to be the best predictor of the actual election results).

    For May2015 its margins were on average 6.96% different from Ashcroft’s with some larger and some smaller and particularly big discrepancies in the cases of Camborne and Redruth and Sheffield Hallam. For EC the figure for the mean absolute discrepancy was slightly larger (7.56%), with particular problems arising in the same two constituencies. In contrast with this the EF margins were much closer to Ashcroft’s (mean absolute discrepancy: 3.75%).

    As is often the case The Election Forecast projections are more accurate than those of the other two models. It is not at all surprising that it outperforms EC with this batch of seats. Unlike, EC it makes use of Ashcroft constituency polls, and all eight seats were polled last September. I have to admit that I don’t really understand why May2015 performs so badly in these benchmarking tests. Like EF is uses Ashcroft data, but for some reason it doesn’t seem to make good use of this information.

    By using the Euclidean Distance metric , it is possible to compare the accuracy with which EF and EC predict the VIs for the five GB-wide parties. In line with the analysis above, the EF projections were closer to the CVI scores (mean Euclidean Distance: 7.28) than were those generated by the EC model (mean Euclidean Distance: 11.53). A score of 7.28 represents a greater level of accuracy than the EF model routinely achieves in previously unpolled constituencies. The reason for this is that the September polling data was incorporated into the database and the model now benefits from the use of that information.

    So, as usual we find that Election Forecast is better than its competitors at predicting Ashcroft constituency results. It remains to see whether this will also make it a better predictor of the GE results.

    In the past I have looked at systematic biases shown by the different models. In previously unpolled seats, all have shown a bias in favour of Labour. But in the last batch (all revisited by Ashcroft) this pattern was reversed in the case of EF. With the current batch there was a small and non-significant bias in Favour of Labour. I think there is reason to suspect that the model’s biases are different in previously polled seats, and I am reluctant to try to draw any generalizations from the current batch of results.

    That said, EF reliably overstated the VIs of both the Greens and Ukip in the current batch. In the case of Ukip all eight projected Ukip VIs were higher than the CVI figures the emerged in today’s polls. This may in part be a quirk of the model’s behaviour in repolled seats, though we have seen reliable Green overestimation in the past (e.g., in the December batch of polls). Another possibility is that support for the smaller parties is being squeezed in these tightly contested marginal seats. Given the distortions introduced by looking at repeatedly polled seats, I don’t think it is safe to conclude that EF more generally overstates Green and Ukip support.

  3. @unicorn

    I found this vague description

    “In his forecast – which will be updated every week until election day – YouGov President Peter Kellner looks at the latest polling data, as well as the state of the campaign and observations from past elections, to share his best guess of what he thinks the result will actually be when all the votes are counted.”

    “observations from past elections” = “we think swingback will happen” so yeah

  4. @OMNISHAMBLES

    I find the the difference between YouGov nowcast and forecast surprisingly large.

    Nowcast has Lab (285) + SNP (52) at 337
    Forecast has Lab (262) + SNP (35) at 297

    A drop of 40 seats which must be almost all in E&W.

  5. The Nowcast actually looks more interesting, because it is based partly on how people are responding to polls in each seat. So its a semi-marginal poll for every constituency?

    https://yougov.co.uk/#/constituency/360/nowcast/

    How is the Nowcast calculated?

    Our Nowcast is based on a sophisticated statistical model that combines the respondents we have in each seat with modelled observations from similar types of people across the country. In maths-speak, it is a hierarchichal Bayesian model using multilevel regression post-stratification, and was originally developed by Professor Andrew Gelman of Columbia University and was extended and applied by YouGov’s Chief Scientist Professor Doug Rivers of Stanford University

  6. @Exilein Yorks

    I find the the difference between YouGov nowcast and forecast surprisingly large.

    Agreed. An enormous amount of change to get done in just five weeks.

  7. @UNICORN

    It’s a forecast in the same way that I can forecast tomorrow’s weather by saying it will be the same as the average for the same day over the last 10 years.

  8. Linking together some of the above posts on Yougov, Peter Kellner and forecasts, we can reach the conclusion;

    forecast = best guess

    As we can assume that the forecast will be the best forecast, using simple algebra we can see that the ‘best’ on each side of the equation cancel each other out, leaving us with the simplified proof;

    forecast = guess

    At this point, this is all we need to know.

  9. @unicorn @exileinyorks

    “large difference”

    Not really. It’s only large in numbers of seats. In terms of VI, Keller’s forecast is just CON on ~37% and LAB on 32 or 33. Which is in MOE of quite a few polls we’re seeing now. It’s not an “enormous amount of change”.

    Maybe it is more far-fetched with the Lib Dem prediction of 30 seats, I’m not sure what VI that requires

  10. Got my first leaflet today – I now feel like a real part of the election. It recommends I re-elect Paul Blomfield.

  11. I am assuming that Kelner is ‘guessing’ that Labour will recover in Scotland that is why the SNP seat count is down but crash and burn in England.

    The Labour campaign will have to do something that will simultaneously make them more popular in Scotland but less popular in England. So is Kelner guessing that Labour is going to do something to help their position in Scotland to the detriment of their position in England seems unlikely?

  12. Note: this is not an excuse to discuss the pros and cons of Scottish independence.

    With that said…

    Interesting research on the reasons why 55.3% of Scots voted ‘no’. It’s also interesting to see that nearly a majority of Aye voters in the survey greatly overestimated the significance of The Vow-

    http://www.heraldscotland.com/politics/scottish-politics/the-vow-did-not-sway-the-referendum-result-new-research-shows.121704244

    One thing that hasn’t been noted is that if the Ayes had won just a third of 2010 Tory voters over (as opposed to about 2%) then they would have (very narrowly) won. Whether they could have done so, especially while still having such brilliant success winning over SLAB voters, is perhaps doubtful, and depends on the degree to which SLAB voters voted Aye due to promises like “No more Tory governments!” as opposed to non-partisan reasons.

  13. On Cambridge its a three way marginal

    2010 Libdem 39
    Tory 26
    Labour 24

    So tories are switching to libdem to keep labour out.

    Unscientific survey no 1

    Drove through Hastings and Rye today (2010 tory 41,lab 37,libdem 16,ukip plus bnp plus ED 6)

    No posters at all spotted in Rye the tory end of the constituency

    In Hastings one labour ,two ukip posters and an enormous ukip flag (maybe their hq) on the front .

    Prediction Labour gain

  14. @omnishambles “large difference”

    MOE has no relevance in this context. It is the change from actual VI now to actual votes on May 7th that is of interest. Given the snails-pace rates of change over the last months, 37 – 32 or 33 would be large.

  15. So I am riding on a Greyhound Bus through the mountains of British Columbia heading for a appointment with the British Consulate in Vancouver and internet connectivity is intermittent – maybe others have already raised it but there are some interesting factors:

    1. UKIP are clearly spoiling Cameron and the Conservatives chances of taking certain seats like North Cornwall, St Ives and Torbay and that hurts their chances of gaining the largest share of seats in the HoC.

    2. Given that LD came within 66 votes of taking Camborne and Redruth in 2010 I am surprised that no one has commented on the fact they have lost 65.2% of their support since. How many centre-left LD MPs are there?

    3. I still thinking weighting LD above 33% of 2010 by the pollsters is problematic.

    4. Are there seats like Camborne and Redruth where Labour could leap over LD and take it from the Conservatives. If Cambridge is atypical, which Con-LD marginals are “low hanging fruit” for Labour – particularly urban ones?

    5. I assume that these eight seats are fairly centre right and quite affluent, and therefore it is very interesting to me that even with the “squeeze” Green ends up with an average 6% to Labour’s 9%. One might wonder where the Green vote will end up on May 7th in safe centre left seats, if it can gain this much support in tightly fought centre-right ones.

    6. Comments on Redcar assume that the fact that Green are running the president of a national trade union, with 250,000 members, as their candidate, and then the fact that half the local Labour Council members quit the Party, will have not any impact on the Labour vote. I think OLD NAT made the original post about Labour councillors falling out with each other, and that it was a clip from a newspaper. So I wonder if those Councillors will be working for the Labour candidate or the Green candidate? :)

    7. When are the next ICM and Opinium polls and I wonder if Ashcroft will re-poll Brighton-Pavillion, Norwich-South and even poll Bristol West?

    PS: When I referred to the polling results, as “dogs breakfast” I was not referring to Lord A, but to the results themselves.

  16. Also updated the Lib Dems v Greens rolling averages:

    http://www.statgeek.co.uk/2014/09/here/ (set of six)

  17. I remember when North Devon switched from Conservative to Lib Dem. The new Lib Dem MP won by three votes! Lib Dem activists were carrying people into cars to drive them down to the voting centre.

    Now it looks as if they are switching back to Conservative.

  18. Re: Individual constituency polling

    Thanks for all the responses to my request for links!

    Am I correct in thinking that apart from Ashcroft, Survation also do individual constituency polling? If so, does anyone have a link for that, please?

    And is there an individual constituency polling summary section on UKPR?

  19. @0705…..

    There is a “Salmond/miliband in pocket” poster in Rye, but as it doesn’t look much like Salmond it seems a waste of money.

  20. @ StatGeek

    I cannot get access to your stats. It says I am “forbidden” to enter your site.

    I have a British passport, honest :)

  21. Am intrigued how Solihull goes this time. Libs have won on a knife edge twice in what you would think would be Con territory, largely in part due to the local popularity of their candidate. I fancy Cons this time. We shall see.

  22. @ Carfew

    “People conform to the opinions of their group even in the face of countervailing evidence’

    Which is why, pre-WW2, Gramsci wrote (to the effect that):

    We need to try and see the world as it really is, not as we want or fear it to be.

    Unfortunately though, we view the world through the prism of our own experience and that can be pretty wonky :)

  23. @Andy Shadrack

    An interesting addendum to the Cambridge poll which showed the LD’s 9% ahead.

    1. Someone in another place explained that the fieldwork was completed during the university holidays. If so, this could artificially boost the LD score as the LDs tuition fees u-turn was a big negative with student voters (of which their are many in the City of Cambridge).

    2. In the same “other place” someone pointed out that Lord A asked two separate questions in Cambridge, the second being “who would you vote for if there was a GE tomorrow?”, which showed Lab 5% ahead of the LDs.

  24. BRIAN NICHOLSON

    I hope you have polling evidence for the 2nd egging or you could be in for one yourself. :<)}

  25. @Barbazenzero

    “Close by the sturdy crate the eggs unheeded sped-”

    Consider them now heeded. ;-)

  26. BRIAN NICHOLSON
    Consider them now heeded.

    Right Oh. Jolly good effort.

  27. My home town of Cheltenham looks interesting. Prime Tory territory but LibDem since 1992. Martin Horwood is well liked and that might be enough to see him home.

  28. @Peter Crawford (9.10pm, March 31st)

    Election Forecast prediction updated today. Changes are from yesterday. 1 week ago and 15th March (when I started recording these).

    Con 285 +3 +1 -1
    Lab 279 -2 +2 +6
    SNP 37 +1 -3 -5
    LibDem 26 -1 = =
    Plaid 2 -1 = =

    AT 9.30 this morning in a comment addressed to @Rivers10 I wrote:

    News night nightly prediction for those that care
    CON 282
    LAB 280
    SNP 36
    LIB 28
    Plaid 3
    UKIP 1
    GREEN 1

    In line with what I said last night these are exactly this morning’s EF figures. What this means is that the Newsnight figures are more up to date than those on the website (which were showing different tallies last night.) I suspect they type in the 10.30pm YouGov poll results before they hand the projections to Newsnight.

    As it happens, with each of these projections showing I had downloaded a copy of the current VI profiles for all seats. (I was trying to make sure I had up-to-date copies to use for Ashcroft benchmarking.). I was surprised to find that there was not a single change in the database at these two times.

    Two possible inferences from this:

    (1) That the four seat Lab/Con margin reduction comes entirely from regression switch-off and not from polling changes or

    (2) Perhaps the current list-of-seats list is not updated every time the projection is renewed.

  29. Will still be 295 plus, Con plus Ld plus NI Unionists = > 323
    Then they sort out EVEL, Ed & Ed dead in the water.

  30. For comparison with the Newsnight prediction, here are, very roughly, the equivalent numbers from the bookies:

    Con 285
    Lab 270
    SNP 43
    Lib 26
    Plaid 2
    Ukip 4
    Green 1

    An interesting bet is for the Tories to win fewer overall votes but more seats, offered at 22/1 by Paddypower. This is in line with the prevailing thinking that the failure to agree a boundary revision with the Lib Dems has left the Tories at a big disadvantage, but I have noticed that several posts have suggested that the way things are going might make this not the case, in which case the odds are obviously attractive.

  31. David.

    I don’t think you can assume the LibDems will want to coalesce with Con again especially if Clegg is a goner. They may have a lot of centre cabal who may be more comfortable with Lab.

  32. Under the Nowcast, Labour don’t need the SNP.

  33. Mikey
    Look at Victoria Street (by Westminster Abbey) and tell me Vince Cable is not going to give up that department plus his ministerial car to hand it over to someone who has not yet even been named? VC will go with his wallet!

  34. @ Omnishambles,

    Keller’s forecast is just CON on ~37% and LAB on 32 or 33. Which is in MOE of quite a few polls we’re seeing now.

    No it isn’t. MoE is attribute of individual polls, not the average. Labour on 33% is MoE. Cons on 37% when their average is 33% and 37% is higher than they’ve had with any pollster for years means there is a massive systemic bias against them. This is possible, but if it’s true it’s not an MoE effect.

    Kellner is relying on substantial changes to occur between now and May 7th. If an election held today produced these results it would mean the pollsters- including YouGov!- were way off.

  35. @Spearmint

    It’s only been a day since PK’s last prediction. On past experience we can expect another pretty soon ;)

  36. Aha. Dear Vince may get to keep his Ministerial position in any Lab/Lib negotiations….assuming he holds onto Twickenham!

  37. @ RAF,

    The debate’s tomorrow, so based on whatever the Sunday YouGov VIs randomly happen to be I’m expecting some hard hitting “Game changer!” analysis.

  38. @ernie @spearmint

    I see what you’re saying – I was wrong about the MOE

  39. Also new thread

  40. @simon:

    constituency polls – you’re quite right that Survation (and others) also do constituency polls, but far fewer than Ashcroft

    Wikipedia has a comprehensive listing of them all:

    http://en.wikipedia.org/wiki/Opinion_polling_in_United_Kingdom_constituencies,_2010%E2%80%9315

  41. Only in Britain can the capital city show one party (Labour) with an 11 point lead, and yet the leading newspaper in that city (Evening Standard) is a die-hard supporter of the other (Tories). That just shows that the newspapers are not interested in catering to the views of their readers. They’re more interested in spreading the propaganda of their owners.

  42. Evening Standard – or, alternatively, the readers of the ES are not representative of London as a whole….

  43. Here we go folks- Eyes down for a full house….

1 7 8 9