A year since the election, we have two new GB voting intention polls (from YouGov and Survation) and a new Scottish poll (also from YouGov) today.

Looking at the YouGov/Times GB poll first, voting intentions are CON 44%(+2), LAB 37%(-2), LDEM 8%(-1). The seven point Conservative lead is the largest since the election but normal caveats apply – it is only one poll. Over the last two months YouGov have been showing a steady Conservative lead of around 4 or 5 points, so normal sample variation alone is enough to explain the occassional 7 point lead. Watch the trend, rather than getting excited over individual polls. Full tabs are here.

Survation‘s topline figures are CON 41%(nc), LAB 40%(nc), LDEM 9%(+1). Changes are since mid May. Like YouGov, Survation have shown a steady position for the last couple of months, but there’s an obvious contrast in terms of what that position is – YouGov have a steady small Tory lead, Survation are showing the parties steadily neck-and-neck. As I’ve said before, there’s not an obvious methodological reason for this (while Survation have a very distinct sampling approach to their phone polls, this is an online poll and their online polls use broadly similar methods to YouGov, ICM and other companies, so there’s no obvious reason for differing results). Full tabs for the Survation poll are here.

Meanwhile YouGov’s Scottish voting intentions are

Westminster: CON 27%(+4), LAB 23%(-5), LDEM 7%(+1), SNP 40%(+4)
Holyrood constituency: CON 27%(+1), LAB 22%(-1), LDEM 6%(-1), SNP 41%(+3)
Holyrood regional: CON 26%(+1), LAB 21%(-1), LDEM 7%(nc), SNP 32%(nc).

Changes here are since the previous Scottish YouGov poll, way back in January. There is very little movement in Holyrood support, but in Scotland the Conservatives have moved back into second place. Full tabs for the Scottish poll are here.


757 Responses to “Latest YouGov and Survation voting intentions”

1 2 3 16
  1. @ANTHONY WELLS

    There was an interesting post from @BAZINWALES in another thread about this:

    “I know it is only 1 poll so far in June but the YG averages for each month this year are showing the trend clearly (hope the formatting holds)

    Con Lab
    41.0 41.7
    41.0 41.4
    42.0 39.8
    42.0 39.3
    42.4 38.2
    44.0 37.0”

  2. … my comment was supposed to have quoted the following from Anthony Wells:

    “Watch the trend, rather than getting excited over individual polls.”

  3. Peak Corbyn?

  4. Peak May ,? My apologies , that was April 2017.

  5. @robbiealive

    “. N. Sturgeon made a mess of it, muddling the GE & a 2nd ref when she should have concentrated on preserving the SNP seats & fighting for the 2nd ref on another battle field. ”

    Well actually she didn’t. The SNP didn’t campaign on a 2nd ref in the 2017 GE, as their view was and is that they have a mandate from the 2016 GE.

    The Unionist parties did with the Scottish Labour leader actively urging Labour voters to vote Tory where they were the best placed Unionist party and the Lib Dems did not campaign actively in many seats.

  6. Scottish figures

    SLAB – pretty disastrous, I would say.
    SC&U – still failing to make the breakthrough which is needed to challenge the SNP hegemony, even after so many years of SNP government.

    What are the odds of another minority SNP government after the next election north of the border? What are the odds of a ‘Grand Coalition’ of SLABs, SC&Us and SLIBs to oust the SNP from power in Holyrood?

  7. Scottish YG

    VI retained/transferred since last year’s Westminster election is usually of interest

    Including DK’s, current VI of 2017 voters is –

    SNP – SNP 78% : DK 9% : SLab 5% : Oth 2% : SLD 1% : SCon 1%

    SCon – SCon 71% : DK 20% : SLD 3% : Oth 3% : SNP 2% : SLab 2%

    SLab – SLab 61% : DK 15% : SNP 11% : SLD 4% : SCon 3% : Oth 2%

    Of course, it would be unwise to read too much into these data. Some of those switchers might be deserting a long supported party; others “returning home”.

    Still, it may suggest that some of those former SNP voters who dallied with Corbyn last year, may have reverted to their previous VI, and there is little to suggest that Corbyn/SLab are attracting ore from the SNP camp.

    The DKs who do decide to vote in the next Westminster GE will, obviously, have an unknown effect!

    They are mainly former SCon/SLab/SLD voters. However, most Scottish constituencies are a struggle between the SNP and one other party, so the critical element will be those swithering between their previous vote and voting SNP – or voting for the most likely party to defeat the SNP.

    A significant proportion of DKs probably just haven’t thought about it at all yet!

  8. https://twitter.com/election_data/status/1005013629912088576

    Quite a swing.

    I know the MOE is bigger for small cross breaks-but there are only two Social Class groups.

  9. Colin

    Your 7:07 Tweet link: it is worth reading the responses. Some of them are very on-board.

    You 7:42 Tweet link: the story would be different depending on the acceptance of the poll as mainstream or out of line. I think it is mainly an attitude issue (after all, Labour doesn’t really offer anything, while life is probably a little bit easier for the second 25% from the bottom), so, yes, an increasing acceptance of the government, but I would be very hesitant to equate it to voting intention (on the other hand, even if they shifted to DK, it would be troubling for Labour).

  10. JohnB: “SC&U – still failing to make the breakthrough which is needed to challenge the SNP hegemony, even after so many years of SNP government.

    What are the odds of another minority SNP government after the next election north of the border?”

    To be fair, the Scottish Tories’ breakthrough was winning more than one MP for the first time in 25 years. It is almost 60 years since they topped the voting in Scotland, and 63 years since they got the most seats.

    As for the SNP, they should have learned from Brexit that they need an absolute majority of believers to carry out the independence project. Anything less and the bigger party can just turn up the campaign until the non-believers in the legislature weasel out.

  11. John B

    “What are the odds of another minority SNP government after the next election north of the border?”

    3 years might as well be 30 as far as predicting the next Holyrood election is concerned, as we await whatever Westminster imposes on us, in terms of Brexit, reduced powers etc.

    But, according to Scotland Votes, these figures would give SNP 54 (-9) : SCon 31 (+2) : SLab 26 (+2) : SLD 7 (+2).

    A little caveat would be that some respondents won’t remember that the List vote isn’t a 2nd preference, as in STV, which may explain part of the drop in the SNP List VI.

    It’s a little hard to see how the Unionists could create a common legislative programme for government, however.

    I suspect it would be another SNP minority, with co-ordinated whining from the Scottish branches of the GB parties.

  12. test

  13. Not sure that the Scottish Conservatives can be described as a branch. The most unionist of the Scottish parties also seems to have the most autonomy.

  14. I shall assume that the last thread was so worn out by AW’s long, opening post on polling [?] that the system blew up – seems loads of us were locked out and Lazslo is now Lazslp.

    Anyway @ Carfrew and tipping points [from last Fred] – this is tangential, but I wondered if you had any similar ideas on the development of new “trendy” words/expressions and similar tipping points?

    It seems to me that many [eg “From minute one”. For sure.” and many others] originate from football and the influx of foreign managers using their own idioms and translating them into English literally and then having them being picked up by others and proliferating.

    Then there are things like “absolutely”, beginning almost all responses to questions with “so”, etc etc., which just seem to explode into public use from nowhere unless [like me for example] you have been properly inoculated and didn’t even say “Far out man” when I was young and had very long hair and thought I was a hippy.

    Actually, I don’t quite know what my question is: maybe just simply – does one require a sufficient percentage of the population to join in before the herd instinct takes over?

    But that is almost a self evident truism: if only three people in the UK [Antonio Conte and two others] said “For sure” instead of the traditional “certainly” then, by definition, it wouldn’t be seen as a trendy new word.

    I guess I am intrigued by what it is about human nature in some people that means they instinctively pick up on new terminology where others are resistant – and the rest seemingly oblivious.

    Yours sincerely,

    Crofty.

    PS To all – Brexit seems to be going jolly well at last.

  15. I shall assume that the last thread was so worn out by AW’s long, opening post on polling [?] that the system blew up – seems loads of us were locked out and Lazslo is now Lazslp.

    Anyway @ Carfrew and tipping points [from last Fred] – this is tangential, but I wondered if you had any similar ideas on the development of new “trendy” words/expressions and similar tipping points?

    It seems to me that many [eg “From minute one”. For sure.” and many others] originate from football and the influx of foreign managers using their own idioms and translating them into English literally and then having them being picked up by others and proliferating.

    Then there are things like “absolutely”, beginning almost all responses to questions with “so”, etc etc., which just seem to explode into public use from nowhere unless [like me for example] you have been properly inoculated and didn’t even say “Far out man” when I was young and had very long hair and thought I was a hippy.

    Actually, I don’t quite know what my question is: maybe just simply – does one require a sufficient percentage of the population to join in before the herd instinct takes over?

    But that is almost a self evident truism: if only three people in the UK [Antonio Conte and two others] said “For sure” instead of the traditional “certainly” then, by definition, it wouldn’t be seen as a trendy new word.

    I guess I am intrigued by what it is about human nature in some people that means they instinctively pick up on new terminology where others are resistant – and the rest seemingly oblivious.

    Yours sincerely,

    Crofty.

    PS To all – Brexit seems to be going jolly well at last.

  16. Still having to fill in all me tikklers for every post.

    Very odd.

  17. This is bonkers. I posted but was told to fill in all my details. Did so and post then went through twice.

    Think I shall give up for lent.

  18. From what I am hearing from Scotland residents living in Perthshire

    – much conservative literature being sent out.
    – lib dems have a spring in their step
    – labour not happy with their local Scottish leadership
    – SNP strong

  19. It really has been an extraordinary period of stability in the polls. It doesn’t matter how much wrangling there is over Brexit, so long as nothing actually happens in “real life” the public seem fairly static in their views. I suspect they stopped listening quite a long time ago.

    There are two edges to that sword of course. If Brexit goes very badly, they will be unprepared and it will seem like an earthquake. On the other hand, if Brexit goes reasonably smoothly (by which I don’t mean all the Leave campaign’s “promises” being fulfilled – just that the average voter’s lifestyle isn’t greatly impacted) then all of the slings and arrows and alleged government “incompetence” may be so much chip paper in the bins of the politically engaged.

    But for now, the polls aren’t telling us much at all. If the public really share the opinion of HMG of most of the contributors here then they must have an even dimmer view of Corbyn and Cable.

  20. @Colin – read para 46 of the Joint Report, that May agreed – “The commitments and principles outlined in this joint report will not pre-determine the outcome of wider discussions on the future relationship between the European Union and the United Kingdom and are, as necessary, specific to the unique circumstances on the island of Ireland.
    They are made and must be upheld in all circumstances, irrespective of the nature of any future agreement between the European Union and
    United Kingdom”.

    .

  21. DW

    “The most unionist of the Scottish parties also seems to have the most autonomy.”

    Both “most unionist” and “most autonomy” would require some supporting evidence.

  22. @Alec:

    I think the UK thought the agreement included no hard border within the UK.

    That is what comes of using duress as a negotiating weapon, and that “clock is ticking routine”. And letting the other side think there is a fudge.

    You get someone to agree something that is extraordinary, something that no state can ever agree to except after badly losing a war, and let them believe they haven’t agreed to that. Agree that the alternatives are fully explored – then present a fully fledged backstop which (once agreed) will leave no incentive for the EU to agree anything else.

    And then your side is surprised that, in the cold light of day, the UK wants to stick to what it thought it agreed.

    (Or do you think Barnier’s Isle of Man and Hong Kong parallels stack up.)

    I must say that the only way the EU and Remain can lose is to overplay their hand. Thinking the EU can do no wrong is a good start.

  23. @ Colin

    “Quite a swing.”

    I’ve been watching this one for a bit, since we last discussed it. I do think there’s a trend there, and it seems to me that it could well be the driving factor behind the slow drift from Lab to Con seen this year. But the tweet is of course playing fast and loose with statistics.

    For example, in the previous YouGov, the split was 41/41. So he chose not to tweet that day, but waited for this one which is far more dramatic. It’s the classic method of being selective with starting point and end point for comparison, just like wealth management people do, choosing to compare an old stock market low point with a newer high point.

    Here’s, as best I can find them, the C2DE figures for this year from YouGov:

    08/06 48/37
    29/05 41/41
    21/05 44/36
    14/05 40/42
    09/05 43/40
    02/05 41/43
    25/04 41/40
    10/04 36/42
    05/04 41/42
    28/03 43/40
    15/03 40/40
    05/03 42/44
    20/02 36/47
    16/02 39/43
    06/02 43/40
    31/01 38/48
    17/01 38/44
    08/01 35/46

    As I said in the last thread, they bounce around a lot. So while you can identify a period (as the tweeter has done) where it seems the swing is 11% to Con, you could also, if you so wished, chose different dates, such as 28/03 to 10/04, which shows a 4-5% swing from Con to Lab over a 2 week period.

    Having said that, given that the best swing to Lab you can find is about 5%, and the best to Con is 11%, that does hint that underlying these extreme examples, the Con VI is doing better. Averaging gives a better idea. Using a 4 poll running average (from latest to the start of the year) we get:

    43/39
    42/40
    42/40
    41/41
    40/41
    40/42
    40/41
    40/41
    42/42
    40/43
    39/44
    40/44
    39/45
    40/44
    39/45

    I feel that I’d be happy to interpret that as a likely swing of 4-5 % from Lab to Con. Nothing like as dramatic as the 11% claimed, but there’s something there I think.

  24. I perhaps phrased that badly. When I say I’m “happy”, I mean I am satisfied that this would be a reasonable interpretation of the data. There’s little else about a swing to Cons that makes me happy, especially the way they’re currently behaving.

    Perhaps the public are just enjoying the entertainment value of watching them struggle to come up with a policy. Any policy. On anything. After all, the show must go on.

  25. @Triguy

    Thanks for the 2143 post.

    But the tweet is of course playing fast and loose with statistics.

    100% agree. It’s cherry picking two values from a quite varied distribution, for particular effect (in this case showing the maximum Lab to Con swing among C2DE voters).

    The Con VI is better for the following reasons:

    1) 2017 Labour voting are supporting Labour less (81% -> 70%)
    2) They have gone largely to Don’t Know (10% -> 15%)
    3) Some 2017 Lab voters have gone blue (from about 2% -> 3%)

    So the headline VI represents no real increase in the number voters supporting the Conservatives, but some 2017 soft Labour voters being a bit less sure.

  26. I think the Survation looks much the more interesting of these two polls!

    :-) :-) :-)

    You’ve got to hand it to the current clutch of VI opinion polls though, haven’t you? They offer something for just about everyone.

    Which suggests to me that they not be terribly illuminating. However, if I was a YouGov bod, I may be a little concerned that I was sticking out a bit like a sore thumb here. Sore thumb polls add to the gaiety of the nation, I accept, which is why I greatly miss the great Angus Reed contributions, but they usually end up being wildly wrong. Mr Wells in the last, albeit, short-lived thread, shed a bit of light on the reasons for past polling inaccuracies, although I fear that they may be being repeated.

    Anyway, as a bit of off-the-wall explanation as to why VI opinion polls are proving ever more difficult to get right, try this slightly maverick view. It’s by Stephen Bush and appeared in today’s I News: –

    https://inews.co.uk/opinion/comment/british-politics-david-cameron-good-performer/

  27. I went back a little bit into last years polls, and made a four week average of the C2DE for (approximately) September and November. Concatenating those with Jan, Mar, May above, we get this longer trend:

    42/43 ~Sept 17
    40/43
    39/45 ~Jan 18
    40/41
    42/40 ~May 18

    Which is even more inscrutable! Perhaps Labour had a Christmas boost.

  28. @Colin:

    I’m baffled by Faisal Islam’s tweet. Surely Barnier has only reiterated what was agreed in December and Faisal even quotes the appropriate passage?

    In December, HMG agreed a backstop: in the absence of another solution, N Ireland would essentially remain in the Customs Union and the Single Market.

    The EU are now seeking a legal agreement on that. The proposal from HMG would (as a backstop) keep the entire UK in the Customs Union, but it would leave Northern Ireland and the rest of the UK outside the Single Market. It wouldn’t commit NI (or the rest of the UK) to regulatory alignment and hence doesn’t accord with what was agreed in December.

    I note that Barnier clarified that a short while after his press conference today.

  29. Interesting to see a discussion on here on C2DEs having a Con VI.

    It’s a bit surprising that some still seem to think that the GB (predominantly English polity) party duopoly should still be based on the “class divisions” of the mid 20th century.

    For comparison, the “Social Grade” VI distribution in the latest YG Scots poll is

    ABC1 – SNP 34% : SCon 31% : SLab 24%
    C2DE – SNP 47% : SCon 22% : SLab 22%

    There’s a reasonable argument to be made that VI is affected by whether the status quo is beneficial to the individual voter or not.

    But, other than among some of us geriatrics, it seems increasingly unlikely that Labour (in any of its PR manifestations) is widely seen as being representative of an imaginary “working class” monolithic group.

  30. @ ON

    Fair point, not to mention that it’s not so easy to come up with a modern definiton of ‘class’. The discussion was really on whether the cross-break stats were significant, rather than whether the definition made any sense in the first place.

    It seems that geography and age are far more significant factors, and perhaps the ones that are worth more scrutiny.

  31. Sebastian (from previous thread)

    You are pretty much right, what the model assumes is that there are certain patterns dependant upon demographic characteristics as well as a local effect.

    Even if you don’t measure the local effect well, as long as you have good estimates for every other part of the model, you should get a measurement, which although has a large uncertainty due to the small sample size locally, is on average centred.

    Taking the distributions for votes cast, it’s possible to run the election enough times to get an estimate for the distribution of seats and hopefully any local errors cancel each other out over the 650 seats.

    There are a few issues which could lead to the model failing, which I raised at the time.

    One is measuring the factors related to the demographics well, these will all have confidence intervals and getting an estimate for a few of these far away from the true values would lead to bias in the model. Obviously the more data you have, the better these factors will be measured.

    The second issue is selecting the right model to use, it’s easy to create a complex model which when applied fits past elections very well (the more complex the model, the better it will fit existing data, even if some of the signals you use are completely worthless).

    However, this can end up training the model to fit to noise and not repeated patterns in the data (this is known as overfitting). Usually you will take a separate portion of the data, not included for fitting the model and check that the model fits this set of the data as well. Usually you will apply several competing models and select the model which performed the best on this “validation set”, which might not be the same as the model which performed the best on the data used to train the model. At the time, this was my main concern as there was no information about how the model was validated and so it didn’t quite pass academic muster.

    As the model seemed to perform well, either they did have a well validated model or they just got lucky.

  32. TrigGuy

    “The discussion was really on whether the cross-break stats were significant, rather than whether the definition made any sense in the first place.”

  33. Trigguy

    Doesn’t look statistically significant to me once you start averaging multiple polls out going back to Sept. ’17

    That’s the problem with looking at data, eventually you’ll find a subset which in isolation appears to provide a narrative, even in a randomly generated series.

    Remember that you are taking a subset of a poll which isn’t correctly weighted to that subset, only to the entire sample. Also, you are taking a subsample. The MOE for each measurement increases as the effective sample size falls.

  34. @ Alan

    Yes, I agree. That’s sort of what I meant by ‘inscrutable’. I was (at least partially) disagreeing with my original conclusion. It could well be there’s no pattern at all.

  35. TrigGuy

    Oops! I didn’t add my comment.

    Also a fair point.

    The selection of demographic crossbreaks in polls has largely been around for a long time, but I wonder if they are necessarily the most useful in the 21st century?

    Just as we have largely abandoned the concept of a Universal Swing (very useful in the 1950s, when David Butler invented it), have some of these crossbreaks lost much of their meaning?

    In (possibly) a more segmented political state, and the ability to capture much larger samples, is it time to move towards a reassessment of the critical aspects, and to report polls using aggregates of internally weighted samples?

    I was struck by an aspect of the BBC/YG poll on Englishness – that regional identity increases with distance from London. While, it is quite understandable that politicians and media who want to promote a vision of unity, wouldn’t normally want to highlight such in routine polling, I wonder whether having English geographic crossbreaks internally weighted (instead of being randomly used as part of an overall GB weighting) might be more revealing than what we get at the moment.

  36. Trigguy

    It’s unclear what you have is a pattern, what you have is a hypothesis, which can be tested with future data (that way you avoid selecting the most significant look subsample (timewise) of your data.

    One of the problems with time series data is it’s very easy to look at a large number of start and finish dates (and we spot patterns very easily) and find something which looks significant, when in effect you have just tested so many hypotheses that one appears to be significant.

    Test 20 hypotheses which are insignificant and it’s not surprising that one appears significant at a 5% confidence level.

    https://xkcd.com/882/

  37. Alan

    “Test 20 hypotheses which are insignificant and it’s not surprising that one appears significant at a 5% confidence level”

    You’re good with explaining numbers! :-)

    Even I understood that.

  38. Sorry in a bit of a hurry, but as @Alan pointed out the problem between hypothesis, data and pattern, this may help.

    https://m.youtube.com/watch?v=5OL1RqHrZQ8

  39. OldNat

    Thanks,

    I do try to tailor my posts to a non-technical level where possible (apart from explaining MRP which requires a more in depth look when technical questions are asked)

    I do recognise I have a tendency to get carried away and disappear down a rabbit hole of detail, I’ll try to pare things down to more simple explanations where possible in future

  40. The fundamental problem with cross breaks and time series in polling.

    A single poll is measuring the probability of the data given the hypothesis. The hypothesis being that all parties have an equal chance. The data shows that parties have no equal chance, this we have to reject the hypothesis.

    Obviously, it is rather obvious and boring, so we adjust for demography and whatever we have. Now the data shows differences in party preferences, but what we are really testing is the distribution of the sample relative to the population expressed as demographic and other variables.

    The trouble is that we aren’t interested in this at all – we are interested in the validity of the hypothesis (that considering demography and so on of the British population, the data verifies or rejects it). But polls don’t do this at all, I’m afraid. As a result, any analysis based on a hypothesis different from given in the poll is flawed. It could be still right though.

    All the endeavour of somehow reversing the relationship (testing the hypothesis given the data) tells us that what the polls tell us is not what we are interested.

    On the one hand, the two things are related, unfortunately, it is not possible. Yougov’s alternative model kind of reconciled them, but certainly not fully.

    Because each poll is conducted as an individual experiment to test the data, the trend is projected onto the data, it is not part of the data. It can be correct or incorrect, but the data is quite ignorant to it. Even with unchanged methodology the problem of sampling appears, not to mention the construct itself. So technically a number of polls are not time series at all, and drawing conclusions from them is reversing the purpose of polls – testing the hypothesis given the data – which is not warranted at all, even if it is done.

    Just to reiterate it (a response I have to @Charles in the previous topic) – polls are quite ignorant to our wishes. They aim at showing that given the factors of sampling the sample is representative. Pure question is – given the data whom do I think would win the elections. You can dress it up, but polls won’t ever give the information in this way. It involves two distinctly different tasks (as AW said in his previous post) – sampling and the construct.

    Obviously there are trends and so on – just don’t assume that the data prove it.

  41. Laszlo

    You’re not bad at this teaching thing either! :-)

  42. Laszlo

    https://xkcd.com/1478/

    Of course this effect can happen, which is why selecting your sample sizes is important. In your clip, 2 sample sizes of N=32 (and a small difference in the means) would be considered to have very low statistical power (probability that an existing significance will be found). if you chose N= 3200 your p values would vary a lot less and you’d have a much greater chance of finding significance where it exists. Part of experimental design is to select your sample size such that you are likely to find significance where it exists.

    The other side of the coin is by repeatedly performing an experiment and selecting the one with the “best” p-value you can “find significance” where it doesn’t exist. This is a big no-no but stuff which isn’t published can’t be scrutinised. You could also get this effect by being “lucky”. As stated 5% of experiments without significance will find significance at the 5% confidence interval. Repeatedly trying similar experiments until one appears to be significant isn’t part of best experimental practice, particularly drawing attention to an irrelevant variable which was changed.

    “When the experiment was carried out while wearing a red hat, the results were significant (p=0.499)”

    This is one of the reasons for performing replication studies, just because one experiment found results which appear to be significant until it is replicated, there is still doubt over the findings and confirmation is a powerful ally in finding “truth”.

  43. Essentially:

    A class C citizen has a 68% probability to vote Labour. Now, we have a respondent who is category C and says voting for Tories. Is he or she part of the 32% or the 68%? If the latter, it is an error – but error of the data or error of the assumptions?

    The same applies, of course, for someone in category C voting for Labour – if he or she is part of the 32%, then it is an error.

    Within the methodology of traditional polling there is no possibility of resolving both types of errors, so polling companies focus on reducing one type o of error (which works in inter-polling periods).

    The margin of error stated meant to address it, but – as it was said earlier – is arbitrary. It is arbitrary, because it assumes that the distribution of the population and the sample is similar, and indeed Scotland demonstrates the fallacy of it.

    So, even if at the level of the total sample the moe is right when we go to cross breaks, it is not simply the increasing, calculated moe, but a distribution that invalidates it. Or in the case of Scotland, the sampling principles of the GB, because of the overwhelming weight of England creates a completely wrong distribution pattern for Scotland.

  44. Lazslo (continued)

    Further down the comments someone calculated the statistical power as 52%. This means each experiment only has a 52% chance of returning a significant result. Where possible, experiments should be designed to have at least a 95% statistical power so that significance is highly likely to show up in your experiment. If half of your experiments fail due to selecting a poor sample size, I suspect you’ll find it quite hard to get future funding.

    I’d say the video was a little misleading, in that it suggests that p-values are unreliable when in reality, the experiment which he set up was designed to fail 48% of the time. The experiment performed as expected, which was to fail to find significance a lot of the time.

    It is instructive in terms of the importance of selecting a sample size which gives you at least a 95% chance of finding significance where it exists, although the video avoided mentioning this point.

    p-values fail in two ways.

    Showing significance where none exists – this is set by your threshold, so if you set a threshold of 0.05, you’d expect 5% of the time to face this type of error. Setting your threshold after you perform the experiment is bad practice.

    Not showing significance where it exists. This is determined by the experiment, and i

  45. Lazslo (cont even more after getting truncated)

    Not showing significance where it exists. This is determined by the experiment, and it is down to the researcher to design an experiment which has a low chance of this type of error. Yes, you can be unlucky but it’s the error which is under the researchers control.

  46. Alan

    Sorry, my response will only be tomorrow morning. Apologies.

  47. Joseph1832,
    “You get someone to agree something that is extraordinary, something that no state can ever agree to except after badly losing a war,”

    So you think leaving the EU is akin to badly losing a war? Probably. And it is the Uk government insisting we do this?

    The difficulty with your analysis is that it was the UK government saying the agreement did not mean all the things you object to, and the EU explaining that it did.

    Catmanjeff,
    “So the headline VI represents no real increase in the number voters supporting the Conservatives, but some 2017 soft Labour voters being a bit less sure.”

    My thought also. And this is precisely what happened before the last election causing labour to lag.

1 2 3 16

Leave a Reply

NB: Before commenting please make sure you are familiar with the Comments Policy. UKPollingReport is a site for non-partisan discussion of polls.

You are not currently logged into UKPollingReport. Registration is not compulsory, but is strongly encouraged. Either login here, or register here (commenters who have previously registered on the Constituency Guide section of the site *should* be able to use their existing login)