On the final Saturday night of the campaign I expect we’ll see plenty of polls. We already have figures from Opinium and ComRes. I know were are definitely due a new YouGov poll for the Sunday Times, ICM for the Sun on Sunday and Survation (presumably for the Mail on Sunday), and I’d expect a new ORB poll for the Sunday Telegraph too.

So far Opinium for the Observer shows topline figures of CON 43%(-2), LAB 37%(+2), LDEM 6%(-1), UKIP 5%(nc). Fieldwork was on Tuesday and Wednesday and changes are from a week ago. Again, we see the Tory lead continuing to drop down into single figures. Full tabs are here.

ComRes for the Sunday Mirror and Independent has topline figures of CON 47%(+1), LAB 35(+1), LDEM 8%(nc), UKIP 4%(nc). Fieldwork was Wednesday to Friday and changes are from a week ago. The Conservative lead is static at twelve points (currently the equal largest any company is showing). Some of the other questions in the poll are less positive for the Conservatives – asked if they have a favourable or unfavourable opinion of politicians and parties Theresa May now has a negative score. 39% have a favourable view, 42% unfavourable, giving her a net score of minus 2 compared to plus 9 in February. Jeremy Corbyn’s net score is now minus 15 – worse, but significantly up from minus 33 in February. Full tabs are here.

I’m out tonight, so will update on the other four polls tomorrow morning – though feel free to discuss them here as they come in.

UPDATE: A brief update on the other polls, not going to post anything else tonight:

  • ICM‘s poll for the Sun on Sunday has topline figures of CON 45%(nc), LAB 34%(+1), LDEM 9%(+1), UKIP 5%(nc). Changes are from the ICM poll for the Guardian at the start of the week.
  • YouGov in the Sunday Times has topline figures of CON 42%, LAB 38%, LDEM 9%, UKIP 4%. Fieldwork was Thursday to Friday
  • ORB for the Sunday Telegraph have figures of CON 45%(+1), LAB 36%(-2), LDEM 8%(+1), UKIP 4%(-1). Fieldwork was Wed-Thurs with changes from a week ago.
  • Survation for the Mail on Sunday has topline figures of CON 40%(-6), LAB 39%(+5), LDEM 8%(nc), UKIP 5%(+2). Fieldwork was conducted wholly this morning, and changes are from Survations previous online poll in the Mail on Sunday. Tabs are here.

2,192 Responses to “Saturday night polls”

1 39 40 41 42 43 44
  1. @ Canada

    At the end of the US Presidential campaign I could see how Trump had won over the blue collar workers in the former US industrial heartland, but can someone on this site please explain to me what in Ms May’s and the Conservative Party’s campaign would win over blue collar workers in the same way Maggie Thatcher won over voters?

    Partly, I believe, because the electoral system in the UK is currently in favour of the Conservative party. Labour have to do much better (in terms of the vote received) to gain as many seats as the Conservatives.

    And that is set to get worse (for Labour) during the next 5 years.

    But also there is a predominantly right wing media (the Daily Mail, Express, Times, Telegraphs, Star, Sun) and very little left wing media.

    And there are substantial echoes from Brexit relating to issues such as nationalism, patriotism, racism, Europhiles, Europhobes etc.. And they cut very markedly across traditional Labour support. It is hard to say how they are going to affect things.

    Then there is age. Older voters tend to vote Tory. And they tend to come out and vote. Younger voters tend not to bother so much.

    So overall, in terms of policies, I think most would agree, on both sides, Labour have offered more in terms of appeal to more voters than the Conservatives.

    But all those other factors are playing a role………

  2. CambridgeRachel
    There is enough ambiguity in his tweet he may be able to argue that and avoid any action. My interest was more in the assumed closeness of the race….

  3. Formister76, well that is just unsubstantiated waffle, and a line I have often seen repeated btl.

    I think any debate is best based on facts not here say. I just told you my experiences.

  4. Statement from Ealing Central and Acton. To me, this suggests that the envelopes are opened and the ballot papers could be looked at before they are placed face down.

    “A spokesperson for the Ealing Central and Acton Returning Officer, said: “We have robust processes in place to ensure that postal votes remain confidential.

    “When envelopes are opened the ballot paper is placed face down so that the vote is not seen and ballots are placed in a pile and stored securely. They will only be counted after polling closes on Thursday, 8 June.

    “All allegations of misconduct are treated extremely seriously. Everyone who attends these opening sessions is given a copy of the legal secrecy requirements and we are reminding them of their duty to maintain the secrecy of the ballot.”


  5. Ian Gillibrand :southall is in Ealing south constituency

  6. I know how the postal vote info might be out.

    If the electoral numbers are taken (if they are on the envelope) that could be compared to voter ID lists of known supporters.

    You could guess at how many people have used their postal votes from the list, and you can buy the info to tell you who has previously used a postal voted from the marked registers (I think).

  7. Mike88, seen what results of what?

  8. Charles

    Very difficult. Ultimately you are in the world of “All the old rules no longer apply and we have no data to create a new set of rules”.

    You might notice trends in the data which seem to indicate something is very different is happening (this is another area of data science which is based on unsupervised learning or cluster analysis). However it won’t tell you how to fix the model and manual “adjustments” are no more than the equivalent of a pollsters bandaid.

    It’d allow for a “pinch of salt warning” to be given and be useful in a post mortem but there isn’t realy a way of producing a predictive model without a representative labeled set of data.

    There is the possibility that you might notice that “Hey the current clusters of respondents look a lot like that which we had xx years ago, maybe we should exhume that model for this election”

  9. @ Formister76

    Sounds like bulls*** to me.

  10. Philotes, that questionable bbc story is doing the rounds in the usual places.

  11. @CARFREW

    Yes, the concept and the name were taken from a conversation between others about 2,300 posts back hence the vague attribution.

    And I agree it’s not wholly true to say DNV = Apathy, a previous run through the 2015 data suggested in excess of 1,000,000 non-voters lived in constituencies where their voting could not possibly have impacted the outcome.

    Interesting that Con exceed DNV in about two-thirds of their seats, whilst Lab do in so few of theirs. Traditional fragmentation with other non-right parties perhaps?

  12. Alan, I have been reading about this modelling and learning. Complex stuff.

    Am i correct that the accuracy of the model will go up each election?

  13. I have a question regarding the divergence in the polls. I understand the difference is largely based around turnout, given all the major companies are displaying broadly similar raw figures.

    Therefore why in the first week of May did:
    ICM = Con+17
    Survation = Con+18
    YouGov = Con+19

    At this point, they all seemed to show a level of consistency.

    Given ICM weights turnout on previous turnout rates,and others on likelihood to vote, is the divergence explained by a more energised youth sample as a result of the election campaign?

    Thoughts, or am I missing something?

  14. Catmanjeff

    Would be surprised if that had happened on the scale required to crosscheck and get a vote share. Councils are really scared of electoral fraud and take a lot of precautions with postal votes – eg supervising the people opening them. Would be a very lax council which allowed anyone to see cast votes or electoral numbers (and write them down or photograph them)

  15. MarkW

    yes and your experiences don’t necessarily make it fact, just a small part of an overall picture.

    And as regards to your unsubstantiated waffle comment you should read some of your posts back to your self occasionally before critiquing others .

  16. @ The Sheep; Sam S; Edge of Reason

    I agree with all your points to a degree, yet at the same time disagree that it wouldn’t be possible to try – after all, that’s exactly what must be done if a model is ever to be tested or proven (eg YouGov).

    We can see from Silver the large number of variables he considers. There’s no reason some of these couldn’t be employed and modeled into a forecast if we’re going to attempt to play around with aggregating:


  17. @ Marmite

    Good question, that’s been confusing me too.

  18. formister76

    Maybe there are a plethora of right wing business men applying and not many train drivers? Narratively it might make sense that all businessmen would like their chance to “influence” matters and we do regularly see them on QT.

    The guy who asked Corbyn, “I’m totally loaded and send my kids to private school, what are you going to do for me?” was the last one I can recall.

    If there are 10x the applications for the part of “right wing businessman” than they have allocated spaces to ensure balance then people applying for that part will find it tough going.

  19. Formister76, no, but my experiences challenged your uninformed notion, and you have strongly rejected that.

    As for my posts, I can only apologize if they fail to meet your expectations.

  20. @Candy

    Obviously the richer still can even lobby to secure the legal arrangements that suit them, making it even easier to afford your M&S sarnies without doing anything illegal…

  21. Back in the 2015 election there was a post on the Labour Uncut website indicating that Miliband’s actions in the few days before polling day were in response to Labour agents seeing a disproportionate number of Tory postal votes coming through.
    Has there been any similar leak of the postal votes this time?

  22. @robbiealive

    I did some similar analysis to yourself as since the the start of ukip collapse it has been one of the most interesting aspect. I also came to similar conclusions which is (coincidently) supported by my experience of talking to 2015 ukip voters – there are a lot in my part of the world:
    Majority changed to Lab post manifesto and the rest back to Con which was previous their affiliation.

    It will be fascinating to see how this pans out regionally in the post mortem of the 2017 polling debacle!

  23. Does anyone have the polls published since Saturday with changes shown?

  24. Mohi
    Thank you for that clarification. I used to live in Ealing off the A406 so thought Southall was the same seat. My mistake. I was really just flagging up the story as yet another seat to keep an eye on for Friday morning to see if Catmanjeff’s knowledge on the unreliability of early postal vote returns as predictions turns out to be correct.
    Overall surprise if Labour are having problems remains though.

  25. In the last general election a report in labour uncut that the postal votes had been seen and was bad news for labour was dismissed in exactly the same terms.

    Indeed i posted some weeks ago about this and initially got exactly the same response but then other posters began to say well actually there is a way you can get a sense of the postal vote and party officials to see the piles.

  26. Oooookay, regarding increased registrations.

    Peeps saying not seeing such an effect in marginals.

    Well you might not, if marginals already see a higher number registered because the votes “count for more” to cite that oft-used, not entirely apposite phrase.

    Is there any evidence registration tends to be higher already in marginals?

  27. “do” see the piles.

    All my posts now await moderation whether i give the weather forecast or not

  28. Markw

    IF the data remains representative and so more data can be used to “train” the model. If there is a permanent shift in “people who answered polls then” to “people who answer polls now” then pretty much you have to throw away previous data and work with data that represents “now”.

    If this election model is correct (in terms that the model parameters are not significantly different) you could then retrain the model using this election in addition to previous data in order to improve the model. This new model will be closer to “The true model” and have a smaller prediction interval. (One election won’t matter much, but as long as people who answer polls are similar now to when we collected the data used to train the model, this effect will be grow over time)

    Given an unlimited amount of representative data, the MOE for each party would be about 0.4% due to sampling on this sampling size. The rest of the error is due to uncertainty in the model which would be reduced with more data to train the model. I have now idea how many elections worth of data was used to construct this model (and how many were used to select one particular model over all of the possible models)

    How good this model gets is ultimately down to how stable the population who answer polls are. If there is a big change, then ultimately the model will have to either start from zero or hope that the one off election was an anomaly.

  29. CatmanJeff. I would have thought the Tories would rack up the postal votes as the elderly would make up a large percentage of those voting.

  30. Just watched May’s interview from earlier today on the police cuts. Its things like this which explains why labour and Corbyn are confident enough to take her on on the conservative home ground of security. Perhaps Maybot needs a software update……


    These things also eventually impact VI, each one doesn’t do much damage, but when its a consistent trend over 5 weeks it can’t help but add up.

  31. @Alan Many thanks for answering my question. I can see that a formal test is not possible but how about the kind of point that has just been made by Marmite? If at the start of the campaign one found that a model based on likelihood to vote actually predicted the proportions who had voted last time would this be evidence in its favour? (It seems to me that a model that pays no attention to likelihood to vote but simply to the proportions voting last time would make it impossible to pick up a ‘base firming up’.) So I am wondering whether a formal test is indeed impossible but this does not mean that there could be no rational debate on the pros and cons of a particular model before the fat lady sings.

  32. Formister

    “Im not suggesting that everyone is a member of Momentum or an activist.
    But it is not a secret that they stage and manage support for these rallies with there hard core activists.
    It is no secret that the Unite and Momentum galvanised their support to become labour members to elect Corbyn, in fact they went legal to extend the joining period eligible to vote for him and they then tried their hardest during his re-election to stop people joining and be eligible to vote against him.
    And my voting intention had nothing to do with my comment.”

    Ok, what’s wrong about being a member of momentum or an activist. There are hundreds of political or quasi political organizations in the UK, that what democracy is about. I’m a member of momentum, neither I nor the other momentum members I know correspond to your ideas about us.

    Mobilising support is what political movements do, e.g. The taxpayers Alliance. All organisations have hard core supporters just look at william Hague aged 16

    Momentum didn’t go to court, that was a crowd funded action. Do you think people shouldn’t have legal redress? What happened to the rule of law?

    Your last point is completely bonkers, momentum did not try to prevent people joining the labour party, it was the party itself that wanted to stop new Corbyn supporting members joining.

  33. Thanks Alan.


    If your boyfriend is angry at Corbyn over his comments on policing go show him this…..


  35. Alexw

    These things also eventually impact VI, each one doesn’t do much damage, but when its a consistent trend over 5 weeks it can’t help but add up.

    I said it earlier and I will say it again and I believe this is why the Con stats have been steady at + 40%, the vote for May is not going away ?

    The Labour manifesto is very good, I feel that regardless of leader it would have been taken well by the public, I also believe that if Labour currently had a different leader they would have won this election, who’s saying they still wont ?

    If I had the Manifesto that JC was given I believe v Tory Manifesto I could have won the election for Labour…

    Someone posted earlier that JC appears a little “Soft” and its that which may cost him the keys to NO 10….

    We will see…

  36. @Alan,

    You could well be correct, but i imagine there isn’t a shortage or left wing transport worker types applying – my opinion.

    I found it more interesting that he never received a reply, not a no thanks but we already have enough of your demographics but we will keep you on the list for future.

    But as a left winger he got an immediate reply and an invitation to appear on two radio shows.

    That said, as you imply , it could be something about nothing and they just needed that demographic

  37. @formister76

    “Interesting story today of the middle aged conservative business man who applied several times to go on BBC Question time and never received a reply…..
    ………. Fuel to the fire of BBC Bias? ”

    Probably not seeing as how Question Time isn’t made by the BBC !

  38. @Canada
    “…can someone on this site please explain to me what in Ms May’s and the Conservative Party’s campaign would win over blue collar workers…”

    I think it’s not so much a positive feeling for the Conservatives, but that they are less dangerous than the alternative, in the eyes of many voters.


    Your speculation on QT and BBC Bias has nothing to do with polling.

  40. @Cambridge Rach

    “Hearing a lot on the radio about the brave police and dedicated nurses etc, all I can think is “isn’t it about time they had a decent pay rise?”. Wonder how many other people are thinking the same and wonder also if labour will push that point.”


    Indeed. Question is, has there been polling on this point or have pollsters avoided it? How many peeps are aware of what’s been happening to public sector pay, and do they think it’s a good thing?

  41. @IAN

    There was a period of plateauing of the conservative vote, but its been slowly and steadily falling for the last few weeks if you look at the poll of polls.

    What May’s bottom for poll support is I have no idea. I think we need some more polls from today and tomorrow to see if its still falling and not hit bottom yet.

  42. @MarkW

    And i hope my opinion challenged your uniformed notion, which you strongly rejected.

    No need to apologise for your posts not meeting my expectations, I never apologised to you in the first place for mine not meeting yours.

  43. Any polls out this evening?

  44. Now, now.

    In a civilised democracy, we shouldn’t be talking about May’s bottom.

  45. As a retired (early) nurse I approve of this survey.


    People’s expectations of salaries compared to reality.

  46. @Judah

    “All very saying spend and give, the Labour’s policies would murder the economy. Amazing no one mentions this.”


    It gets mentioned really rather a lot on here. It’s the counter arguments that seem to upset the modding. But ultimately the question is, what does polling say? How many of the public buy the household economy angle vs other interpretations?

    Is there polling on this? Is the question. It’d surely be useful to track it regularly too…

  47. @Turncoat – Hello.

    Yes, I think you are getting what I am tentatively suggesting.

    The emprirical evidence from multiple general elections, going way back to 1945, clearly seems to suggest that there is not so much a ‘shy Tory’ syndrome, but a ‘shy loser’ syndrome. Whichever party is behind in the opinion polls nearly always tends to do better than the polls suggest.

    This may be to do with people being happier to publicly identify with a ‘popular’ party, but it might also be around issues of motivation to vote, if your supporters think it’s in the bag.

    It’s also fair to say that the effect is much more noticeable in scale on the occasions it benefits Cons, although here, the 1992 result stands out as a really large factor. Strip this out and the average Con error is significantly less pronounced.

    It’s such a striking feature of the polls and results that I’m surprised it hasn’t been analysed previously. The conventional wisdom is that polls always underestimate Cons, but in fact, polls nearly always underestimate the party in second place.


    I think your mistaking my comment as criticism

    I in fact was implying they were very good at mobilising their support , whether I like it or not, that is a compliment

    I did imply that because they mobilise their support so effectively it doesnt mean that the crowds you see at Corby’s rallies are typical of your average labour supporter.

    I also didn’t criticise or say they shouldn’t have legal redress , all my comments were raised about the effcetiveness of their ability to mobilise support in the right places and they managed to orchastrate Corby’s election as leader very well.

    Im fully aware how politics works thank you, so I didn’t need the lecture.

    If you could point me to where I was criticising and not just stating what I believed to be a fact?

  49. Charles

    Your question is on model selection. Giving a pool of data you can make any number of models which can be used to train the data (as you say this can include and exclude turnout filters)

    One problem with training model to fit data is the more variables you feed it, the better the fit can get (you can include literally random variables and the model with spot patterns within this randomness and “improve” the model, if you include as many variables as you have data points, you will fit the data perfectly)

    One way to ensure that the model you create actually works is to remove part of the data (in this case, a subset of the elections), train many different models on the remaining data and then check to see how the model performs with the data you removed from the pool. This process is called validation.

    If a model perform better on the validation data, that is indicative that it will perform better in the real world. (Depending on the amount of data you removed as there will be some random sampling variation in this data)

    In this way, it is possible to make informed decisions about the impact of including vs excluding a turnout filter in the model for example.

    This is the area which gives me greatest concern about the yougov model, due to a lack of information about the secret sauce. If model validation has taken place with independent data (I have no reason to suspect it hasn’t as I’m sure YouGov know their stuff but equally I have no evidence of their model selection process) then I’d expect it to perform well.

    The other concern (which is shared by all predictors of the election) is that this election will be “weird” and not follow the pattern of the previous elections.

  50. Catmanjeff
    ‘The second inner envelope should not be opened at this stage.’

    Not so. The ballot paper is removed from the envelope and then placed downwards. In practice the scrutineers can normallyeasily ascertain how the vote has been cast. I attended the opening of postal votes for the Norwich North by- election in July 2009 and on the basis of my observations came up with a very accurate forecast of the final result. In particular, I was able to predict that UKIP would beat the Green – something which surprised most people.

1 39 40 41 42 43 44