|
|
Post by James E on Jun 10, 2024 16:41:38 GMT
As we are a site about polling (some of the time at least) it is worth noting that Farage has had a whinge about polling methodologies: "Farage also said he had written to the British Polling Council to complain about what he described as the “dishonest practices” which claimed were underestimating the strength of support for his party. “There is a very misleading opinion being given to the British public as to what is going on,” he said, complaining that there was a lack of consistency in term of methodology. He went on: "One very well known polling company will say: how do you intend vote? and it will be Labour, Lib Dem Conservative and Greens - and then other. If you then take the trouble to go other you will find a list of parties I have never heard of and the Reform will be there. So if you don’t prompt for Reform you come out at 12%. One of the other practises is to re-weight the undecided voters back to the 2019 general election in which this party did not stand. "The true picture should show that Reform UK was consistently at 16 or 17%, he claimed. This seems to be a common misconception. Per Opinium etc, 'undecided' voters are allocated on the basis of the 'decided' voters with the same demographics. So when 25% of the 'decided' Con2019 voters say that they will vote for RefUK, then Refuk also gets 25% of the 'undecided' ones. [On prompting, I like the Survation approach: they do not name any parties at all , leaving it up to respondents to name the party they say they will vote for. ] |
|
|
|
Post by James E on Jun 9, 2024 18:53:39 GMT
The front page of the i newspaper has a graphic showing how wide the range is of MRP poll/models predictions for the 2024 General Election. I'll repeat it here for convenience. Survation MRP Lab 487, Tories 71, Lib Dems 43, SNP 26Electoral Calculus MRP Lab 474, Tories 67, Lib Dems 59, SNP 26Britain Elects model Lab 455, Tories 86, Lib Dems 65, SNP 20Financial Times model Lab 443, Tories 139, Lib Dems 32, SNP 14YouGov MRP Lab 422, Tories 140, Lib Dems 48, SNP 17Economist model Lab 394, Tories 182, Lib Dems 22, SNP 24More in Common MRP Lab 382, Tories 180, Lib Dems 30, SNP 35One thing is for certain, some of the people publishing these MRPs and models are going to have egg on their face on July 5th. I'm still inclined to favour the YouGov MRP on the grounds that they have the best track record for MRP polls.. Also some of the other pollsters like More in Common and Electoral Calculus/FindOutNow used small samples of 10-15,000 while YouGov's is just under 60,000. My gut feeling is that a sample size of 50,000 or thereabouts is needed to give a good MRP prediction. Thanks for that, LL. A lot of the differences in these seat figures are simply the product of the varying polling figures. For example, the Electoral Calculus/FON figures show Labour leading 46-19 in the overall vote, while YouGov's figures were 43-25, so 9 points apart. The rate at which Labour converts the swing towards them into extra seats is much the same in both at about 15 seats per 1% swing. This also means that if there has been a 1% swing to Labour since most of these MRPs were produced (and the polls suggest that), you should add 15 to the Labour total and deduct around 17 from the Tories. This is particularly relevant to MIC whose most recent polling shows Labour 5 or 6 points further ahead than their MRP of 9 Apr-29 May. For what it's worth, I think that EC/FON probably have it about right for the point at which the LDs would get more seats than the Conservatives: the Tories would need to be reduced to around 20%, 25 points behind Labour (and 8 or 9 ahead of the LDs). Also for those expecting the Tories to get around 200 seats - YouGov's figures and most others here suggest that they would need to narrow Labour's lead to around 12% to do that. |
|
|
|
Post by James E on Jun 8, 2024 22:05:03 GMT
Looking at the latest 3 polls against my preferred comparison of the pollsters' pre-election 2024 averages gives this:
Deltapoll :Lab 46% (+2) Con 21% (-5), Ref 12% (+2) LD 9% ( 0)
Savanta : Lab 46% (+2.5) Con 26% (-0.5) Ref 11% (+2) LD 10% ( 0)
Opinium : Lab 42% (+0.5) Con 24% (-1.5) Ref 12% (+1) LD 10% ( 0)
So Labour up with all three, and Reform's advance is smaller than appeared from the first polls after Farage decided to stand as a candidate.
Overall, across the ten polls we have had from regular pollsters in the past couple of days, Labour are up by an average of 0.5% on their pre-election averages, and the Conservatives down by 2.7% on theirs. I don't have an exact overall figure for Reform UK but clearly they must be up by about 2 or 3 points on average.
|
|
|
|
Post by James E on Jun 7, 2024 16:25:33 GMT
Taking the recent (post Farage's announcement) polls we have had and comparing with the same pollsters' pre-election averages from 1 Jan to 22 May 2024 shows that Labour are overall polling at the same level, and the Tories are down by 3 points.
WeThink : Lab 46% (+1) Con 21% (-3)
MIC : Lab 45% (+2.5) Con 25% (-2)
YouGov* : Lab 45% (0) Con 18% (-2.5)
BMG : Lab 42% (0) Con 23% (-3)
R & W : Lab 42% (-2) Con 19% (-3.5)
Survation : Lab 43% (-1.5) Con 23% (-3)
Techne : Lab 44% (0) Con 20% (-3)
I have not included Focaldata, as they did no polling from 1 Jan to 22 May, nor Ipsos, whose most recent poll was only very partially post-Farage (and their methodology has changed)
* for YouGov, I have used the figures they provided for 'unadjusted' polls as a like-for-like.
|
|
|
|
Post by James E on Jun 7, 2024 16:16:38 GMT
More in Common NEW @moreincommon voting intention has a Labour lead of 21. While changes within MoE it is the highest Labour lead of the campaign. 🔵 Con 25% (-2) 🔴 Lab 46% (-) 🟠 Lib Dem 9% (+1) 🟣 Ref UK 11% (+1) 🟢 Green 6% (+1) N: 2,618 Dates: 5/6-7/6 'Over 1000 of the sample are from today after the news of the PM's decision to leave the D-Day celebrations early had been reported. While the weighting efficiency isn't such I'd be happy reporting separate VIs the Labour lead today was somewhat higher than yesterday' In short the D Day issue has cut through This is also the highest Labour lead that More in Common have ever recorded, although their polling history only goes back to April 2023 (26 polls). Their average for the first 4 months of 2024 was a Labour lead of 15 points. They combine an Opinium-style adjustment of Don't Knows with relatively low Ref and green vote shares - I think the latter is probably caused by not prompting for those parties. |
|
|
|
Post by James E on Jun 7, 2024 12:53:24 GMT
jimjam - That's an interesting and informative post about internal notions within Labour regarding the EU. The 2026 renegotiation date for the TCA will mean Labour are going to forced into an element of decision making on this, whatever happens, and that's going to be the point at which Labour's stance will be tested. Up to now, I think it's fair to say many of the leaderships utterances on this have been guilty of the same 'magical thinking' as was evident on the Brexiter side, with assumptions that things we want to see happen would be easy to negotiate etc. But I do think politics has moved on now, and public opinion certainly has, so the idea of agreeing concessions in exchange for closer alignment should now be politically much easier for Starmer than it was previously. It seems to have gone unnoticed, and may well be an outlier, but we have had a first poll showing a two-thirds majority for Rejoining the EU - it was from Verian (Previously Kantar) a few days ago. The norm is around a 60/40 majority for re-joining, but with a spread of results due to pollsters house effects - possibly caused by false recall. veriangroup.com/hubfs/Britain-Barometer/Britain-Barometer-2024/5-June-2024-Britain-Barometer.pdf |
|
|
|
Post by James E on Jun 3, 2024 16:57:38 GMT
|
|
|
|
Post by James E on Jun 3, 2024 11:35:00 GMT
Re Yougov MRP, their last MRP 2 months ago had the Conservatives on 155 seats, Labour on 403 with a majority of 154 Their headline voting had Labour on 41% and the Conservatives on 24%, significantly less than their normal poll at the time The difference was because their MRP model 'probabilistically matches 'missing voters' (i.e., those indicating they do not currently have a vote intention) to similar respondents who do express a vote preference, rather than excluding them from the estimations' So a similar methodology to Opinium, it will be interesting to see if there is a similar unwinding of don't knows as Opinium There is not much sign of that in YouGov's normal polls. The three in the General Election campaign so far average 19% of Con2019s answering 'Don't Know' compared to 21% in 4 polls in April and 20% in 4 done in early May. The Labour 2019 'Don't Knows' are unchanged at an average of 8% |
|
|
|
Post by James E on Jun 3, 2024 11:13:07 GMT
He's done this to get the media outlets to cover it My guess is thar he will spout on again about the large immigration numbers Unfortunately the media outlets will fall for it and give him the attention he craves Standing His Candidates down and telling his troops to vote for Sunak! Shudder! I would not be surprised to see him do this, as he has never shown a great deal of enthusiasm for getting either himself or his parties represented in Parliament. If Refuk do stand down, we have had some polling for exactly that scenario by Survation two months ago. This made little difference to the overall Labour lead, which was 17 points (46/29) but it did significantly boost the Tories' vote share in seats where they are currently strong, improving their total seats to around 150. www.bestforbritain.org/spring-2024-polling |
|
|
|
Post by James E on Jun 3, 2024 10:42:56 GMT
The Yougov poll this evening. Is it reasonable to expect, based on punditry here, that just as the Opinium widened labour's lead, that Yougov will shrink it? So if it stays the same, that's good news for labour.... There was no change of methodology by Opinium in their most recent 45/25 poll. Their adjustment was slightly smaller (5 points) due to the lower number of Tory2019 'Don't Know' responses. For their MRPs, YouGov re-allocate 'Don't Knows' just as Opinium do, so it is reasonable to expect Labour's lead to be around 7 points lower than their normal polls. If this happens, Labour would be around 15-18% ahead in the headline VI figures. Their March 2024 MRP produced shares of Lab 41%, Con 24%, LD 12%, Ref 12%, and gave Labour 403 seats; this is the obvious starting point for a like-with-like comparison. We should expect youGov's MRP to show somewhat higher LibDem VI and lower RefUK than their normal polls, as the figures from their two previous MRPs showed. They prompt, using the previous or notional result of each constituency, which increases the LD Vi at the expense of Labour and also seems to increase the Con VI at the expense of RefUK. They then model constituency results according to the different dynamics of which party was 1st, 2nd and 3rd last time. This produces results which I think are rather generous to the LDs in those seats where they were second to the Tories in 2019 - at any rate, it produces very different results in these constituencies to Survation or Lodestone (see below). The other adjustment which YouGov are now using is what they call 'unwinding' which moves the constituency results a bit in the direction of Uniform National swing. Time will tell whether this produces more accurate results, but let's just say that this part of their methodology is not based on recent fieldwork. lodestonecommunications.com/election-hub |
|
|
|
Post by James E on Jun 2, 2024 8:00:25 GMT
We need to bin JL Partners pronto. According to Laura Kuennsberg they are the only pollster polling during this election A mere 12 point lead and the Tories would be in dreamland if that turned out to be close. But still a Labour majority, I suspect. The Dec-Jan YouGov MRP with Labour leading by 39.5% to 26% provides a good model for what would most likely happen in that scenario. A 12-point lead would be only a 0.75% swing away from that, and as YouGov show Labour gaining around 14-15 seats per 1% swing , they could be expected to take 10 fewer seats. This would reduce Labour from 385 seats (with a 13.5% lead) to 375, so a majority close to 100. The Tories would be likely to get around 185-190 seats. As well as being the closest MRP to such a result, YouGov's is also the least generous to Labour (compared to Survation and FON) in terms of how they convert increased support into seats. |
|
|
|
Post by James E on Jun 1, 2024 21:23:05 GMT
Opinium's poll really does merit a 'so what' response, given that the movement is only large when compared to an unusually low lead in their previous poll . Comparing the Labour leads in all of the latest polls from regular pollsters with their pre-election 2024 averages gives the results below. I am only including those with 4 or more previous polls this year.
30-31 May We Think Lab 25 (+4)
29-31 May Opinium Lab 20 (+4)
29-30 May Techne Lab 24 (+3)
29-30 May YouGov Lab 25 ( 0)
28-29 May BMG Lab 16 (0)
27-29 May MiC Lab 19 (+3.5)
24-28 May Ashcroft Lab 23 (+3)
25-27 May R&W Lab 23 (+1.5)
24-27 May Survation Lab 23 (+5)
24-26 May Savanta Lab 17 (0)
So the average is 2.4 points higher, with a range from 0 to +5.
|
|
|
|
Post by James E on Jun 1, 2024 19:20:27 GMT
I've just checked Opinium's tables. The Don't Knows are 14% in total, but 17% for Con2019 and 7% for Lab 2019 voters. This is a little lower than I've seen in previous polls, which is the main reason why their adjustment makes slightly less difference than it has done before. Without it, Labour would have a 25% lead - 47.5% to 22.5%. This is in line with YouGov's most recent poll (and several other non-adjusted ones such as today's Techne). www.opinium.com/polling-tables-archive/Opinium's previous poll, with a 14-point Labour lead had more Con2019 DKs (20%), and fewer Lab 2019 DKs (5%). Because of that, the adjustment made a net difference of 7 points compared to 5 in this one. |
|
|
|
Post by James E on Jun 1, 2024 18:34:02 GMT
Apparently Channel 4 News just said Tories down 2 points and Labour at highest level since Truss in Opinium poll We'll see... If that is correct then it would be: Con 25% Lab 46-49% The 'Truss-era' Opinium showed Con 23, Lab 50 on 19-21 Oct 2022. No Opinium since then has had a Labour lead greater than 18 points, or less than 10 points. |
|
|
|
Post by James E on Jun 1, 2024 9:17:39 GMT
leftieliberal " If we are genuinely seeing a reluctance to vote Labour (as opposed to an artefact of the focus group in Swindon North) then it is significant for Labour's chances in the South of England outside London. " Looking back at the details of polls in recent months, and much earlier for YouGov and Opinium, they all show a significantly larger rise to the Labour vote for the South outside London than for GB as a whole. To summarise those average Lab vote rises in the South: YouGov 16 points ( v 12 for GB) Opinium 14 points (v 8 for GB) MIC 14 points (v 10) R&W (recent large poll) 18 points (v 12) Labour Together (large April poll) 14 points (v 11) Survation ( last MRP) 14.5 points (v 12) The Tories' vote share is in each case down by significantly more than this, and the LDs are typically shown as around 1-3 points lower. However, there is good evidence that we will also see high levels of tactical voting in seats which appear to be close contests. We have had the largest-ever levels of third party squeeze in by elections on the LDs in Lab targets and on Lab in LD targets. And I would refer again to the Lord Ashcroft finding from April which showed that 56% of both the current Lab and LD VI were willing to cast their vote for the other party, if they are better placed to defeat the Conservatives.
|
|
|
|
Post by James E on May 31, 2024 17:15:58 GMT
Matt Singh (Number Cruncher) has been providing some excellent commentary on the progress of the election and today's e-mail from him is informative as ever. The final paragraph is particularly intriguing, as it links to a tweet from Calum Weir of Opinium on the poll that they have done yesterday and today. This is due for release 8pm tomorrow. On it, Weir tweets " Jesus Wept. Just doing some double checks, but Jesus Wept!". So it seems likely that we can expect something outside Opinium's normal range of Labour leads from 10-18 points. Incidentally, all nine polls released since 'that' JL Partners one show Labour's lead either steady or up ( when compared against my preferred measure of pre-election 2024 average by pollster). Here's the article: "Drilldown By Matt Singh Latest polls continue to show little movement in any direction, and Keir Starmer has said that Diane Abbott is “free to stand” as a Labour candidate, so that row appears to be over. It’s still unclear what (if any) effect it’s had. It seems that very little is actually cutting through so far. Lord Ashcroft found that no story around Starmer or Rishi Sunak had been noticed by more than 13 per cent of the public, with Starmer's message of change being the only one making it into double digits. On the campaign trailOwen Winter at the Economist has managed to back out what appear to be Labour’s targeting plans from data scraped from its website. Basically, any Conservative-held seat with a notional majority of under 28 points (and some with much more). That is equivalent to a 14 per cent swing or (uniformly) 16-point Labour lead nationally. If you think the swing is bigger in the battleground, then it equates to a lower double digit lead. In other words, exactly what you'd do if you think you're ahead by a bit more than that, and further evidence of the parties believing the polls (as is this on the Tory side from Katy Balls).
The South Asian voteIt also seems that Labour is putting resources into defending seats with a large South Asian population. Much has been said and written about how the (majority Muslim) Pakistani and Bangladeshi communities have been punishing Labour over its stance on Gaza. But then there's the Indian community, some of them also Muslims. But many more are Hindus, a group that has been trending rightwards for some time (so pre-dating any Sunak effect) and it's there that the Conservative vote may be holding up better than elsewhere. When I looked at this in 2019, it seemed the percentage of Hindus in a constituency was much more predictive of Tory swing than the percentage who are Indian, the implication being that it is indeed primarily a Hindu effect, with Indian Muslims staying left and Sikhs, Christians and others somewhere in between. Disentangling these things is difficult, particularly so at local elections, but there is evidence the trend has continued. Polling in the nations Most polling covers the whole of Great Britain (or occasionally the whole UK) but the different party structures (and in recent decades, devolved politics) make polling of Scotland, Wales and Northern Ireland valuable. Scotland is always interesting, usually complex and can be extremely psephologically volatile. A big swing from the SNP to Labour in the second half of the parliament has led to relatively narrow but consistent Labour leads (the nationalists haven’t led since February). How this plays out in Westminster seats isn’t certain. The SNP’s main vulnerability is its evenly-spread vote, which is highly efficient when the vote is high (winning almost literally everywhere on 50 per cent in 2015) but inefficient when it falls to levels that were more typical before then. Of course, efficiency is not fixed, so the question is whether outside of the SNP-Labour battleground, ex-SNP voters still go to Labour or whether they vote tactically (something that Scots are good at) to stop the Conservatives. If so, then the SNP vote could hold up better in Con-SNP seats. If not, then Tory gains in Scotland aren’t impossible. In Wales, things are typically less volatile because the Plaid Cymru vote share tends to be reasonably stable, so swings tend to correlate well with those in England. However the gap between Wales and England has been shrinking, and indeed recent Welsh Westminster polling shows Labour leads not much wider than across Great Britain as a whole. In Northern Ireland, polling is of limited frequency, however Belfast-based LucidTalk is planning three Westminster polls during the campaign. The main shift since 2019 has been gains for the hardliners, with the TUV advancing at the expense of the DUP, and Sinn Fein regaining lost ground from the SDLP. The impact on Westminster may be limited in a purely electoral sense given the low likelihood of a hung parliament, but Northern Ireland (particularly in the context of Brexit) will continue to be an important consideration politically. Over the weekendOpinium has hinted that its poll for the Observer might be worth watching. It’s usually released on Saturday evening, as are various other polls for the Sunday papers." x.com/CWP_Weir/status/1796548534181019877 |
|
|
|
Post by James E on May 31, 2024 10:58:03 GMT
Yougov Labour lead at 25 points in latest YouGov poll for The Times CON 21 (+1) LAB 46 (-1) LIB DEM 8 (-1) REF UK 15 (+3) GRN 6 (-1) Fieldwork 29 - 30 May It becoming obvious that You Gov and Opinium can't both be right. Of course both might be wrong but most likely one is likely to be significantly wrong. The obvious comparison to be made here is between YouGov's regular polls, in which 'Don't Knows' are treated as non-voters, and their MRPs where they make an adjustment similar to Opinium. These can be compared to the 4 or 5 regular YouGovs from the same period. From 11-Dec to 4 Jan, YouGov did an MRP with 14,000 respondents, which showed: Con 26% Lab 39.5% LD 12% Green 7.5%, Ref 9% This compares to their regular polling average from around the same period of : Con 22%, Lab 44.5%, LD 9.5%, Grn 7.5%, ref 10.5%. So in this case, the adjustment, plus other methodological effects which I'll explain later, reduced Labour's lead by 9 points, from 22.5 to 13.5%. The effect was similar for their more recent MRP from 7-27 March, with a sample of 18,700, which was: Con 24% Lab 41% LD 12% Green 7% Ref 12% This compares to their regular poling average for the same period of : Con 20%, Lab 44%, LD 9%, Grn 7.5%, Ref 14.5%. In this case, the adjustment (plus those other methodological factors) reduced Labour's lead by 7 points from 24% to 17%. So that's an average of 8 points difference between these two sets of YouGov data. The fact that YouGov are using 'adjusted' figures to make seat predictions suggests strongly that they they believe that they will be a more accurate predictor. It's worth looking at the net changes by party, as I think these do also reflect the other methodological factors that YouGov use in their MRPs. These are: Con + 4% Lab -4% LDs + 2.5% Ref -2% The LD and Ref changes are likely to be caused by YouGov providing details of the past result by constituency to their respondents. This is justified by the fact that Con/LD contests have different dynamics to Con/Lab contests, and they want separate data for the two. This looks to me like it has had the effect of boosting the LD Vi at Labour's expense and probably the Con VI at the expense of Ref. However, I suspect that this reminder of the 2019 result may overly effect the responses they get in those seats where the LDs were second in 2019. Many voters will not be aware of the previous/notional result in their constituency, or if they are aware of it may not pay it much attention. This goes back to the issue I mentioned in my post on Chichester a few days ago. Models which do not treat LD/Con contests as different to Lab/Con ones tend to understate the LDs in their battleground (or 2nd place) seats: example of this are Survation and Focaldata. But I also suspect that YouGov may be going too far the other way - particularly in those seats where the LDs are a distant second to the Conservatives. |
|
|
|
Post by James E on May 30, 2024 8:20:56 GMT
Many thanks to @isa , Dave , oldnat and others for all the messages about my analysis. I probably don't spend more than around 30-40 minutes a day noting and analysing polling details, so I can't help feeling that professional analysts should have access to all the information I have, plus a lot more. I retired (very) early a few years ago, but my career was largely spent in identifying anomalies, errors and fraud in data that businesses and other organisations had submitted ; and then in building up evidence and data to illustrate those errors. So I'm really just applying the same skills to polling. |
|
|
|
Post by James E on May 29, 2024 17:04:01 GMT
I have run my usual regional analysis using More In Common's figures for the first time. As others have noted, More In Common adjust their figures in respect of Don't Knows in a similar way to Opinium. Their figures do have other house effect, too: they show relatively low reform UK VI, and higher Con and Lab than Opinium (so about the same lead). These are their averaged 'regional' figures for the 9 polls since 20 March, so total respondents are about 18,000.
Headline GB
Lab 43.3% (+10.3)
Con 26.1% (18.6)
Con to Lab Swing 14.5%
London
Lab 45% (-3)
Con 27% (-5)
Con to Lab swing 1%
South East
Lab 35% (+12)
Con 31% (-23)
LD 17% (-1)
Con to Lab swing 17.5%
South West
Lab 37% (+14)
Con 27% (-26)
LD 17% (-1)
Con to Lab swing 20%
East
Lab 42% (+17)
Con 28% (-29)
Con to Lab swing 23%
West Midlands
Lab 46% (+12)
Con 28% (-25)
Con to Lab swing 18.5%
East Midlands
Lab 42% (+10)
Con 29% (-26)
Con to Lab swing 18%
North East
Lab 52% (+9)
Con 26% (-12)
Con to Lab swing 10.5%
North West
Lab 53% (+7)
Con 23% (-15)
Con to Lab swing 11%
Yorks & H
Lab 48% (+10)
Con 25% (-18)
Con to Lab swing 14%
Scotland
Lab 37% (+18)
Con 16% (-9)
SNP 29% (-16)
SNP to Lab swing 17%
Wales
Lab 45% (+4)
Con 24% (-12)
Con to Lab swing 8%
So the highest swings in the East England region, followed by the South West. This is the same with the R&W figures I analysed yesterday, and also with the Labour Together large scale poll in April. I do not have any other analysis that goes down to the level of individual English regions - most just divide by North, Midlands and South. Notice how similar the Tories share of the vote is across all of the English regions. It ranges from 23% in the North-West to 31% in the South East, with all others in between.
The 'beyond proportionate' swings are particularly large with MIC. The Conservative 'Percentage of 2019 vote retained' is as follows: remember that under UNS, this would show London and Scotland with the lowest voter-retention, and the Southern English regions as the highest.
London 84% (27/32%)
North East 68% (26/38%)
Wales 67% (24/36%)
Scotland 64% (16/25%)
North West 61% (23/38%)
Yorks 58% (25/43%)
GB Overall average 58% (26.1/44.7%)
East Midlands 53% (29/55%)
West Midlands 52% (28/53.5%)
South East 56% (31/54%)
South West 51% (27/53%)
East 49% (28/57%)
|
|
|
|
Post by James E on May 29, 2024 16:12:01 GMT
More In Common Voting intention, 27-29 May Labour's lead at 19 🔵Conservative 26 (-1) 🔴Labour 45 (+1) 🟡Liberal Democrat 9 (-) 🟢Green 5 (-1) 🟣Reform UK 11 (+1) Changes with 22-23 May n = 2008 This makes '4 in a row' polls rising, the 4 in question being the Survation, R&W, YouGov and MiC we have had since the 'National Service or Compulsory weekend volunteering' policy was announced. These are up by 5, 1.5, 2 and 3.5 points respectively on the various pollsters averages for 2024 so far - so an average rise of 3 points in the Labour lead. en.wikipedia.org/wiki/Opinion_polling_for_the_2024_United_Kingdom_general_election#Seat_predictions |
|
|
|
Post by James E on May 29, 2024 15:00:06 GMT
....... The 16 point Con to Lab swing in this (18-24) cohort is actually lower than the 22% swing in the same poll for voters over 65. This is the same as 1997, the swing to Labour was greater in traditional Tory seats than Labour seats. It opens up the whole debate on proportional swing and Uniform swing. For what it’s worth I think that, for elections where there is a significant move in public opinion proportional swing models give a better idea of how individual seats will perform. The 1997 result was 'semi-proportional', at least in terms of the Conservatives' vote losses. They lost 11.5 points overall from 1992 retaining about 73% of their share from that election. A proportionate loss would have been minus 16 points in their safe seats (with around a 60% share from 1992), but this was just short of 14 points in reality. But this time their losses are likely to be greater: they look set to retain only perhaps 60-65% of their 2019 even if they can improve to 27-29% by election day. This opens up a far wider difference between linear and proportionate losses . This is something I have looked at with a huge amount of polling data: I regularly analyse all the regional figures from YouGov, Opinium and Delta. And in recent weeks I have done the same with Survation's most recent MRP (15k sample), R&W's 12k sample poll, Labour Together's 9,400 sample poll, and the last 8 polls from MiC; to this I can also add Electoral calculus's own analysis of their FON MRP, and 3 local or constituency polls from Survation earlier this year. Every single one of these shows the Conservatives' vote-share losses as being proportionately larger in their strong regions (i.e the South and Midlands of England). And correspondingly, they lose a lower proportion of their 2019 vote share in London, Wales and Scotland. This is why we are likely to see some very high swings in Conservative safe seats, and little to no swing in many safe Labour seats. |
|
|
|
Post by James E on May 29, 2024 11:11:26 GMT
James E Agree with you on JLP and their track record doesn't seem good. On polling companies though I would say R&W are also new to the game and maybe we give them more credence because they poll regularly and seem to mostly be middle of the pack. I'm still watching yougov more carefully than any of the others. I am a little suspicious of R&W, too. Looking at their figures from early last year, they went from producing among the highest Lab leads (27,27,26) to the lowest ( 12,14,12 ) between Feb 2023 and May. While there was a downward trend at that time, the average reduction was only about 4-5 points. I believe that they may have made methodological changes, but not all pollsters are open about these. I trust YouGov, Survation and Opinium, among those who do have a track record. As I have mentioned before, Opinium's raw findings are similar to YouGov's but are then subject to their adjustment. When we get YouGov's MRP, they will be using the 'Opinium adjustment', and the headline figures will be close to those Opinium or MIC have been producing recently, rather than the normal YouGov's with labour around 25 points ahead. |
|
|
|
Post by James E on May 29, 2024 10:10:49 GMT
YouGov's latest tables are now available: d3nkl3psvxxpe9.cloudfront.net/documents/Sky_VI_240528_W.pdfThe pattern of the Conservatives vote share losses being proportionately larger in their strong areas continues: their first two polls in the campaign both show somewhat larger losses for the Tories (-55%) compared to their GB losses (-53%). Or in simple terms, YouGov show their GB share as 21% (-24), and their overall South & Midlands share as 25% (-30). Those who say they are 'certain to vote' is 56% in this one, so a couple of points lower than their last poll (58%). The comparable average figure for the early campaign figure in 2019 was 62%, so it's down by about 5 points. And the 18-24 sample of 166 voters is very much as we might have anticipated (comparisons to 2019) Lab 65% (+9) Green 20% (+16) Ref 6% (+5) LD 5% (-6) Con 4% (-17) The 16 point Con to Lab swing in this cohort is actually lower than the 22% swing in the same poll for voters over 65. |
|
|
|
Post by James E on May 29, 2024 9:07:20 GMT
I have mentioned a like-with-like comparison with pre-election polling averages in resect of JL Partners, and that in their case, this is lower. However, 7 of the 10 polls we have had since Sunak called an election show a higher lead when measured against the pollsters respective averages for 2024 (pre-election). Those averages are:
People Polling (4 polls) 26%
YouGov 25%
Ipsos (4 hugely varied polls) 24%
R&W 21.5%
Techne 21%
We Think 21%
lord Ashcroft (5 polls with non-standard VI question) 20%
Deltapoll 18%
Survation 18%
Savanta 17%
BMG (4 polls) 16%
JL Partners (2 polls, and I am using their unadjusted figures) 16%
Opinium 16%
More in Common 15.5%
This list divides quite clearly into the three approaches taken by the various pollsters which I mentioned last week, with those taking the traditional approach at the top, the Survation/Savanta/Delta group of those who don't prompt for RUK in the middle, and the 'adjusters' at the bottom.
Comparing the latest polls we have from the 9 companies who have published figures so far in this campaign, we have leads of:
28 May YouGov 27% (+2)
27 May R&W 23% (+1.5)
27 May Survation 23% (+5)
25 May JL Partners* 16% (-2)
25 May Deltapoll 22% (+4)
24 May Opinium 14% (-2)
24 May We Think 25% (+4)
23 May Techne 26% (+5)
23 May MoreinCommon 17% (+1.5)
* I am using the raw figures as JL Partners are using a variable adjustment methodology, as described earlier.
So taken as a whole, these 9 polls show Labour's lead up by 2 points when compared to the same pollsters pre-election 2024 averages.
I think that a comparison to the long-term average is often more meaningful than the standard comparison to the last poll - so will be making the same comparison as the campaign progresses.
|
|
|
|
Post by James E on May 29, 2024 8:33:11 GMT
JLP as predicted NEW: *Labour lead at 12 points* JLP/ @restispolitics 🇬🇧 election poll, 24th-25th May 2024, change on early May Labour: 40% (-1) Conservatives: 28% (+2) Reform UK: 12% (-1) Lib Dems: 10% (-1) Green: 5% (-) Without the adjustment the results of this poll would be Lab 43% Con 26% Reform 10% LD 10% Green 6% JL Partners have no track record in poling for General Elections. They have conducted only two polls so far this year (with headline leads of 18 and 15), but prior to that their only previous polls on General Election VI were three conducted in 2022; so they had a gap of 18 months before the first of their polls this year, and have only ever done 5 previous polls for General Election voting Intentions. They have done other, niche polling over the past few years, mostly for various think-tanks. They did a poll for the London Mayoral election which predicted that Khan would win by 4 points (rather than the 11 he actually won by). Polling for low turnout elections is difficult, and of course YouGov got the London Mayoral Election wrong too, but in JL Partners case, I believe that this is the only election against which their figures have been tested. As for their General Election methodology - there is a long and complicated explanation included in their tables if anyone is interested (e.g "Don't Knows are imputed 'within survey' which means that each survey has its own bespoke model...The basic principle is to build a random forest model...This implicitly assumes that those who say they don't know can be imputed from their policy stance as well as demographic factors..."). So this is a lot more complicated and experimental than Opinium of MiC, and is being 'evolved' over the series of polls which started in April. Over their three polls since April, the size of their adjustment has increased: their 7 April poll had an 'unadjusted' Labour lead of 19 points which became an 18-point lead in the headline figures (42/24); their next poll in May saw a 17 point lead reduced by 2 points to 15 (41/26), now this poll has taken a 16-point lead (39/23 before excluding the DKs, which is the same as Neil's figures above), and reduced that to 12 points. As it is always wise to compare like-with-like, the best comparison that can be made for JL Partners is that they show a Labour lead 2 points lower than their previous polling average (of 2 polls) in 2024. Or if we are comparing only to their most recent poll, it is 1 point lower. |
|
|
|
Post by James E on May 28, 2024 16:56:53 GMT
Having warned about the volatility of R&W's regional cross-breaks, I have had a look at their latest poll, and thought I'd post the figures, anyway. Overall, they conform to the pattern I have posted about ad nauseam for the past year or two, but in a highly exaggerated form. In the last R&W, the Tories' strongest 'region' was Wales; in this one, it is London. Best to treat these as entertainment rather than any kind of guide to the election. GB Lab 46% (+13) Con 23% (-22) Con to Lab swing 17.5% London (sample 1,694) Lab 45% (-3) Con 26% (-6) LD 11% (-4) Con to Lab swing 1.5% SE England (sample 1,501) Lab 40% (+17) Con 26% (-29) LD 13% (-5) Con to Lab swing 23% SW England (sample 1767) Lab 40% (+17) Con 23% (-30) LD 20% (+2) Con to Lab swing 23.5% East England (sample 776) Lab 46% (+21) Con 22% (-35) Con to Lab swing 28% West Midlands (sample 1,004) Lab 51% (+17) Con 22% (-31) Con to Lab swing 24% East Midlands (sample 559) Lab 44% (+12) Con 24% (-31) Con to Lab swing 17.5% NE England (sample 275) Lab 53% (+10) Con 24% (-14) Con to Lab swing 12% NW England (sample 1,196) Lab 58% (+12) Con 17% (-21) Con to Lab swing 16.5% Yorks & H (sample 907) Lab 51% (+13) Con 21% (-22) Con to Lab swing 17.5% Scotland (sample 679) Lab 35% (+16) Con 22% (-3) SNP 28% (-17) SNP to Lab swing 16.5% Wales (sample 394) Lab 45% (+4) Con 25% (-11) Con to Lab swing 7.5%  |
|
|
|
Post by James E on May 28, 2024 15:59:05 GMT
Survation NEW: First Westminster voting intention from the campaign period. 23 point Labour lead, highest since November 2022. LAB 47 (-1) CON 24 (-3) LD 11 (+3) GRN 3 (+1) RFM 8 (-) SNP 3 (-) OTH 4 (-) F/w 24th - 27th May. Changes vs. 22nd May 2024 The previous Survation showed an unusually high Lab VI. Compared to Survation's average in 8 polls so far in 2024, this is: LAB 47 (+2.2) CON 24 (-2.4) LD 11 (+1) GRN 3 (-1) RFM 8 (-) SNP 3 (-) By this measure, the 7 polls we have had since the election announcement show an overall increased lead of 2.5% on average - that is, when compared to the various pollsters' pre-election averages from the start of the year. |
|
|
|
Post by James E on May 28, 2024 14:00:52 GMT
Mega poll alert The Redfield Wilton poll coming out later asked 12,000 people across Britain how they will vote on 4 July. Although above a certain number I'm not sure how much benefit it is to the overall Voting Intention figures. But the regional and demographic cross breaks are likely to be more informative At one time, I used to analyse R&W's cross breaks, but the figures fluctuated so wildly that I decided they were of little value - one example I remember is the Conservative vote in the West Midlands being 5% - so down 50 points from GE2019. Their most recent poll (Lab 45%, Con 23% ) with a sample of 3,700 shows the Tories' strongest strongest 'region' as Wales where they are in the lead on 40%, and the next strongest as the North East where it is a 35/35 tie. They also show swings to Labour of over 25% in the East, South West and East Midlands. Even with a larger sample, I would put less trust in their cross break figures than YouGov or Opinium. |
|
|
|
Post by James E on May 28, 2024 9:48:15 GMT
From Numbercruncher politics ( www.ncpolitics.uk/ ) Are the polls exaggerating Labour's lead? By Matt Singh Rob Hayward, who called the 1992 polling failure (and whose analyses are always worth paying attention to) has told the Guardian he thinks there is some skew in the polls now, citing the local election results – which were less bad for the Conservatives than might have been expected based on Westminster polls – and the makeup of undecided voters in those polls.
Over the weekend I spoke to Polling UnPacked on precisely this topic, particularly with regard to the local election results, and what I see as the differences between now and 2015 when it comes to potential polling error. The whole post is worth reading, but it quotes the following summary of my thoughts:
I think the key differences [between now and 2015] are that it’s only one thing rather than everything, and in terms of “real votes”, chiefly Westminster by-elections, this looks nothing like 2015 (both Labour and Lib Dems have been getting mid-90s level swings)
I also think there are other explanations for the narrower gap in local elections than you’d expect given the polls. We know for sure that the squeezable votes are much more on the left (or at least, lean Labour over Conservative right now).....
[etc.]It's interesting to see Rob Heywood and now Matt Singh give their detailed views on which current polls are likely to be closest to the actual result. Both articles say the same as I wrote on 'seats prediction' thread, 4 days ago, albeit in rather fewer words: "I think that the likely outcome is at the lower end of the current polling range" As with so much of what is likely to happen in 5 weeks' time, I think that the 1997 precedent is relevant here. While many have been critical of Opinium's adjustment, it seems to have been forgotten that we have had 'adjusted' polls since the early 1990s. ICM adopted a 50% re-allocation of Don't Knows to the party they voted for at the previous election after 1992. And in the Labour landslides it worked well. While the polls generally overstated Labour's lead in 1997 by about 6 points, ICM were broadly accurate as their final 5 polls averaged a 12-point Labour lead. (leads of 14,17,5,15 and 10) Opinium's adjustment is actually very similar in effect to what ICM were doing in the 1990s. There is a misconception that all Don't Knows are ALL being re-allocated to the party they voted for in 2019. In fact, the adjustment allocates them according to the VI given by the 'decided' voters of the same party, so that with Con2019 DKs, around 55% go back to Con, 15% to Lab and 20% to Ref (or around those figures). So this adjustment should encompass all types of 'undecided' voter apart from those who end up not voting. I see the latter as the biggest weakness of it: the proportion of 'Don't Knows' who end up not voting is likely to be very much larger than the number who are already telling pollsters that they won't bother voting. because of this, the adjustment may be a bit overdone, but on the other hand the Conservatives are still fairly likely to recover a few points from a falling Reform UK VI. |
|
|
|
Post by James E on May 26, 2024 18:03:17 GMT
graham (re tactical voting and seats which may be Lab & LD targets) " I disagree re- Wimbledon where Labour was a very clear second place in 2017 having held the seat 1997 - 2005. Good reason to think the 2019 result was an aberration due to the Brexit and Corbyn factors.Likely to be a three-way contest." Tactical voting is almost certainly unnecessary in Wimbledon. With a notional 39% vote share from 2019, the Tories are likely to be reduced to the mid 20s - the EC, Survation and YG models show, 23,24 and 25%. That's less than half of the combined LD and Lab vote. The models all show the LDs winning, narrowly with EC and comfortably with YG and Survation. I also think that the same is true of South Cambridgeshire, which was Con 44%, LD 41%, Lab 14%. If the Tory vote falls by 14 points ( I am expecting at least that) and LD & Lab collectively pick up 6 points, then it would be impossible for the Conservatives to win (Con 30%, LD + Lab 61%). For what it's worth, I predict LD 43, Con 28, Lab 20
|
|
