|
|
Post by James E on Jun 3, 2023 10:23:12 GMT
PJW, Apropos ''I strongly recommend Omnisus against giving up the day job and taking up comedy. Stick to the polling chaps.'' Not sure they are that good at the day job mind; or else all the rest are pants; as with the disappearance of Peoples Polling Omnisis are usual the most regular significant outlier. Omnesis's 27-point Lab lead on 12 May was a real outlier, but otherwise they are not so much higher than other pollsters. Taking their 9 polls since the start of April, they have exactly the same average Conservative VI as YouGov, but have Labour 4 points higher at 47% to YouGov's 43%. Omnesis have each of LDs, Greens, and RefUK a point or two lower than YouGov. And it looks to me like Omnesis's recent figures are within about 1 point of Opinium's 'Old Methodology', which would be producing Labour leads of around 20% without the major changes they made in early 2022. |
|
|
|
Post by James E on Jun 2, 2023 17:24:08 GMT
Re Watford.
The seat has a long history of electing LibDem Councillors while at the same time being a Con/Lab bellweather. In the 2004 Council elections, the constituency voted LibDem 47%, Con 27%, Lab 17%. This did not prevent it being a Lab hold at GE2005, with 35% of the vote to the LDs' 31%.So the result diverged by 34 points from the LEs the previous year, despite the polls generally showing similar figures in May 2004 and May 2005.
The LDs will go into the next election 22 points behind Labour, so almost as far as the 28 points which they needed to make up in 2005 (and failed). Labour's 38% vote share in 2019 was 6 points higher than the best that the LDs have ever managed in Watford (32% in 2010).
As it is Lab No 38 target seat (from Con), needing a 4% swing, it sits at roughly the point where if the Tories hold it they might expect to have a (small) majority.
|
|
|
|
Post by James E on Jun 1, 2023 19:16:45 GMT
James E Isn't it normal that when a party already has a very large lead the swings are smaller? Conversely when it has no lead the swings are normally bigger? All other things being equal of course This is the issue that we discussed just a few days ago. See page 134 onwards of this thread. I analyse a lot of polls and my own view is that current polling strongly supports the idea of 'Proportionate Swing'. The Tories appear to be suffering higher swings against them where they were strongest - so voters in the South and Midlands of England, property owners, rural voters, Leave voters. However, no less an authority than Peter Kellner thinks otherwise. This is what he wrote a week ago: kellnerpolitics.com/2023/05/24/how-grumblers-and-defectors-have-laid-a-trap-for-our-pollsters/While the article purports be be about MRP, the underlying point is a defence of Uniform National Swing, and he questions (without, it seems, examing any recent polling data) whether the extrapolations showing a proportionate swing can be right. I think he is right to claim that most previous UK General Elections have conformed fairly closely to UNS, although this is unsurprising when all bar 1997 have seen overall swings of 5% or lower. UNS copes less well with larger swings, though. For example, in my view, the 1997 GE saw a "semi-proportionate swing", ranging from around 7-8% in safe Labour seats to around 13% in the Tory-held seats. I am actually in the process of re-analysing the 1997 election, but suffice to say that I don't think that Tactical Voting can serve as the main reason for the higher swings in the Con-held seats. I have given a lot of examples and data of what looks to me like 'Proportionate Swing' in recent polling over the past week or more, so won't repeat it, but it should all be accessible here. ukpollingreport2.proboards.com/user/48/recent |
|
|
|
Post by James E on Jun 1, 2023 18:30:57 GMT
Re "Labour leads Conservatives by 33% among voters who live in rented accomodation." As @trevor says, this is less than many people would have expected. Per Ipsos Mori's analysis, they held about a 12-point lead (about 45/32) at the last General Election, so using that as the comparitor, it's only a 10 point Con>Lab swing in this demographic compared to 13% per R&W's last GB poll. Lab 53% (+8) Con 20% (-12) www.ipsos.com/en-uk/how-britain-voted-2019-electionThe same poll shows Labour leading 36/35 among property owners (as opposed to mortgage holders). The Conservatives led by 57 to 22 in this demographic in 2019, so that's an 18% Con>Lab Swing in this previously very Tory demographic... [TO ADD: There is actually a simple proportionate formula which can explain both these swings. The Tories are down by 38% of their 2019 support, so their current VI is 0.62 of the GE2019 figures of 32% for renters and 57% for home owners respectively. Labour overall are picking up 65% of the votes that the Tories are losing in each case. The same formula works on the overall GE2019 result, turning Con 44.7%, Lab 33% into Con 27.7%, Lab 44%.] |
|
|
|
Post by James E on May 31, 2023 15:26:46 GMT
You've got to love the positive spin being put on inflation data on the bottom of the screen here. BREAKING NEWS - Food inflation drops to 19% - from March high of 19.1% |
|
|
|
Post by James E on May 31, 2023 11:35:38 GMT
It is far from clear in many seats which opposition party is in second place today. It makes no more sense to rely on 2019 results to predict the outcome in them than it would have done to rely on the 2017 results to predict the 2019 outcome - indeed less so because of the much bigger time gap involved between 2019 and 2024. It is ,however, reasonable to suggest that given that LD support at 9% - 12% is no higher than its national 2019 vote share whilst Labour has surged from 33% in 2019 to circa 45% today , that Labour will have overtaken the LDs in many seats where it came third in 2019. This is all true, but in reality there are surprisingly few seats where the tactical position is unclear AND the seat looks like a viable target for both LDs and Labour. And by a 'viable target' in mean the top 50 LD targets and the top 250 for Labour, which is those needing swings of up to 14.5% for the LDs and 19% for Lab. I am working on the basis that any seat where Labour is second in 2019, they will be the better tactical choice. Current polling shows Lab as being 12-16 points better off relative to the LDs compared to 2019, and taking 20-30% of LD2019 voters. So the seats which are in doubt are those where the LDs were 2nd, but could still be Lab targets. These can be divided into 4 groups: those which are short-medium range, long-shots, and very long shots. I have listed them with the swings needed by LDs and Lab respectively. Wimbledon 0.6%, 13% Cities of Ldn and Wmnstr 4.6%, 6.3% Hitchin and Harpenden 6%, 10% Finchley and GG 6%, 10% Sutton and Cheam 8.3%, 19% Cambs SE 9%, 17% Woking 9%, 16.2% Wantage 9.4%, 18% Chelsea & Fulham 12%, 13% Totnes 12%, 18% Mid Sussex 14.5%, 18% These are based on the old boundaries, so this can change. But it looks to me like the principle of backing the better placed of LD and Lab from 2019 (or the notional result) is going to hold true in all but a small number of seats. |
|
|
|
Post by James E on May 31, 2023 10:20:58 GMT
And re "with 18 months to go things might change...." (@trevor)
All the precedents point to narrowing of the polls over the last 18 months of a 5 year Parliament - the question is, how much will they narrow, and what has happened in previous long parliaments?
To take the most recent example, Labour were still 5 points ahead in the polls around Oct 2013, which changed to a 1% average Con lead prior to GE2015 - so around 6 points movement, or a 3% swing-back. The polls were of course also wrong by around 6%, but that is a separate phenonemon.
18 months before GE2010, the Conservatives were averaging around a 12-point lead, which contracted to 7-8 points in the final polls. A 2% swing-back there. And the polls were accurate that time, as they were in 2019.
GE1997 is the most obvious precedent for the current position, and here the historic polls do give some hope to the Tories. There were fewer polls then, and their results were more varied, but Labour were averaging a 27% lead in the autumn of 1995. This shrank to 18% in the final polls, so a 9% cut or 4.5% swing-back. As in 2015, the 1997 polls were flattering to Labour by around 5-6 points. But of course polls are prepared rather differently now. One comparison which I think is valid is to compare ICM's polls from auturmn 1995 to their final polls - which were by some way the most accurate, as their final 5 averaged a 12-point Lab lead. ICM's 5 polls in Autumn 1995 averaged a 19-point Lab lead, so they recorded a 3.5% swing-back in the final 18 months.
|
|
|
|
Post by James E on May 31, 2023 9:35:54 GMT
..... Looking at cross-breaks for the South of England (outside London), there still seem to be larger than average swings. YouGov's last 4 polls show Labour an average of 5 points ahead in the South, and Opinium's last 4 average a 1-2 point lead. With R&W's last 4 the average is 6 points. So these are swings of 17-19% in the context of polls which are clustered around a 14% 'national' swing. If you press the quote button then your reply will register in my notifications. Anyway, to wrap this up then I'm quite happy to say R&Ws seat selection for their 'Blue Wall' is somewhat arbitrary and they should have left out London seats, given a lot of London specific issues (eg the expansion of the ULEZ). Onward's recent analysis based on age demographics is possibly a better approach, at least for party HQs looking for how they can motivate different groups to vote for them. Although that is possibly why I highlighted GE'17 which was where weighing by age groups became the big polling 'error' issues (which saw Dr.Mibbles become a 'one hit' wonder on UKPR but hopefully a lot of us made a few ££ out of that) However, polling companies don't weight their regional x-breaks. Whilst you are clearly looking for evidence of PNS then there isn't much difference between 17-19% and 14% IMO anyway. So, perhaps PNS is "better" mid-term if you exclude certain polls or certain GEs/regions (nations in the case of Scotland). With 18mths to a GE then things might change but by all means keep plugging away with the regional analysis - I'm certainly interested to see if Rishi-CON are recovering better in some parts of the country than elsewhere (something that you would see as PNS starting to look less accurate - as per my original reply). I have been looking at the detailed figures from pollsters from some time. It isn't so much that I am 'looking for evidence of PNS' but that I am looking to establish what the detailed figures show. And the persistent pattern is of differential English regional swings - higher in the South and Midlands, lower in the North and (especially) London. Proportional swing seems to be a good explanation for this, as the tories seem to be losing the same proportion of their support in the South and the North of England. And of course I am aware that polling companies don't weight their cross-breaks, so it is necessary to look at the figures over a long period, and to see whether the patterns shown by one pollster correspond to those of others. And in that regional pattern I have described ad nauseam on UKPR2, that is what has been happening for some time - longer than I have data, although of course it is possible to go back and check. What I can confirm is that these regional differences have been there for at least the past 18 months. So while we can't be certain that this pattern will still be there in 18 months time, it is not a recent phenonemon, nor a product of Labour's poll lead since Dec 2021. Taking the 5 YouGovs from November 2021, when the Conservatives still had a 1-2 point lead (so a 5% Con2019>Lab swing), the regional swings were: London 2% South 9% Midlands 7% North 3% And the Tories vote retention was 80% for GB as a whole, 80% in the South, 74% in the Midlands, and 82% in the North. |
|
|
|
Post by James E on May 30, 2023 21:54:15 GMT
@ Trevor
"You were quite excited by PNS as the 'explanation' for CON seeming to have a higher swing against in seats where CON had a higher% (eg Blue Wall / Southern regions).. until they didn't now that all bricks in all walls have started to look pretty much the same."
It is only the most recent R&W Blue Wall poll in which the swing is the same as for their full GB polls. This may just be one poll on the low side. And it should be remembered that the sampled 42 seats include 10 in London, which has seen lower swings in all the pollsters I have looked at.
Looking at cross-breaks for the South of England (outside London), there still seem to be larger than average swings. YouGov's last 4 polls show Labour an average of 5 points ahead in the South, and Opinium's last 4 average a 1-2 point lead. With R&W's last 4 the average is 6 points. So these are swings of 17-19% in the context of polls which are clustered around a 14% 'national' swing.
|
|
|
|
Post by James E on May 30, 2023 19:54:24 GMT
"I'm not sure why Kellner's piece seemed to ignore GE'17 ..."
It is only when there is a big swing that there is any noticable difference between Uniform swing and proportional swing, and the overall swing in 2017 was just 2%.
To give an example that I've been looking at, 1997 saw a 10% UNS, with the Tories down 11.2% and Lab up 8.8%. If you then treat those figures as fractions, the Tories held on to 73.3% of their 1992 vote, and Lab overall gained 78% of their losses.
If you apply these proportions to a safe Conservative seat where they had 60% of the vote, it predicts that they would be down by 16 points, and Lab up by 12.6, so a 14.3% swing instead of 10%. In reality, in 1997 safe Conservative seats showed a swing of 13.4%
|
|
|
|
Post by James E on May 30, 2023 19:13:27 GMT
One wall looks very much like another. R&W Blue Wall 22/5/23 Con lead by 1. Con>Lab swing 14% R&W GB poll 28/5/23 Lab lead by 15. Con>Lab swing 13% R&W Red Wall 28/5/23 Lab lead by 17. Con>Lab swing 13% Are you making the argument for UNS now? I thought you were arguing for proportional swing a couple of days ago ukpollingreport2.proboards.com/post/85346/threadNot at all. I just thought it was interesting that these two samples just happen to be showing a similar swing to each other and yesterday's GB R&W poll - although generally the southern 'Blue wall' has shown a noticably higher swing since they started these polls a little under a year ago. Re proportional swing - I have posted numerous details as to how the very variant regional swings per YouGov, Opinium, Deltapoll and R&W are fuelled by the same proportional Conservative vote losses. And in a proportionate swing, you would expect a seat close to the norm (so Con 45%, Lab 33%) to produce an rather average swing rather than a large one. Many of the 'Red Wall' seats are around those figures. It's the safer Tory seats which would produce larger movement in a proportional swing. |
|
|
|
Post by James E on May 30, 2023 17:09:46 GMT
One wall looks very much like another.
R&W Blue Wall 22/5/23 Con lead by 1. Con>Lab swing 14%
R&W GB poll 28/5/23 Lab lead by 15. Con>Lab swing 13%
R&W Red Wall 28/5/23 Lab lead by 17. Con>Lab swing 13%
|
|
|
|
Post by James E on May 29, 2023 18:01:57 GMT
The cross-breaks of latest R&W provides ample reason to doubt their sampling, even if the overall result looks close to the norm. A few highlights: The Tories lead in London by 39/28, their strongest region. But they are reduced to just 6% of the vote in the West Midlands, down 48 points from 54% in GE2019. Meanwhile, Labour lead in all age groups, even the over 65s, with whom they trailed by nearly 50 points at GE2019. However, there's a small swing to the Tories among 18-24s. The LDs surge to around 25% of the vote for 25-45 year olds, but slump to around 5% in all other age groups. And a quarter of Plaid Cymru voters switch to Reform UK. Sub-samples aren't demographically balanced and the statistical errors on these small sub-samples are massive (10% s.d. for a sample of 100). It's why people like oldnat combine a number (typically 7) of polls together when analysing the Scottish sub-sample of GB/UK polls. At least it reduces the statistical errors even if it does nothing for the demographics. Talking in terms of percentages only serves to hide just how small some of these samples are. I do the same with several pollsters including R&W, for whom I have noted the latest figures alongside their previous ten. So I am well aware of their normal variability. But the latest tables are the most freakish and unlikely of any poll that I have seen. To add (21:35) - It's clear from the data that something has gone wrong in the interviewing for this poll. While there is always a need for weighting within a sample to match the demographics of the population, the geographical sampling should not be too difficult to target to get the numbers right (or close). But in this poll, they have just 114 respondents in the East, who need to be upweighted to 200, and 419 in London who need to be downweighted to 280. This isn't normal. It makes me wonder whether they mistakenly got respondents in parts of (historic) Essex like Romford or Upminster for their 'East' sample, without realising that these places are in Greater London. |
|
|
|
Post by James E on May 29, 2023 17:21:40 GMT
The cross-breaks of latest R&W provides ample reason to doubt their sampling, even if the overall result looks close to the norm. A few highlights:
The Tories lead in London by 39/28, their strongest region. But they are reduced to just 6% of the vote in the West Midlands, down 48 points from 54% in GE2019.
Meanwhile, Labour lead in all age groups, even the over 65s, with whom they trailed by nearly 50 points at GE2019. However, there's a small swing to the Tories among 18-24s. The LDs surge to around 25% of the vote for 25-45 year olds, but slump to around 5% in all other age groups. And a quarter of Plaid Cymru voters switch to Reform UK.
|
|
|
|
Post by James E on May 29, 2023 15:42:14 GMT
Following Peter Kellner's data-free defence of UNS as a means of forecasting the next General Election yesterday, I have taken a look at the most recent YouGov GB polls by age and 2016 referendum vote. This all needs to be viewed in the context of a set of seven polls averaging Lab 43%, Con 26.5%, so an overall swing on GE2019 of 14%. And that Con VI shows them retaining 59% of their 2019 vote.
Age 18-24
Lab 61% (+5)
Con 10% (-11)
Swing 8%
(and Con vote share retention of 48%)
Age 25-49
Lab 54% (+10)
Con 16% (-17)
Swing 13.5%
(and Con vote share retention of 48%)
Age 50-64
Lab 40% (+13)
Con 28% (-23)
Swing 18%
(and Con vote share retention of 55%)
Age 65+
Lab 25% (+9)
Con 45% (-19)
Swing 14%
(and a Con vote share retention of 70%)
So the pattern is (of course) one of distinctly non-uniform swings, but with the Tories losing a similar proportion of voters in all three groups up to age 64. It is only with the over-65s that the Tories fare better-than-proportionately, and that UNS would look like a better predictor than proportionate swing.
And on the Remain/Leave axis, its:
Remain 2016 voters
Lab 56% (+7)
Con 14% (-5)
Swing 6%
(and Con vote share retention of 74%)
Leave 2016 voters
Lab 26% (+12)
Con 44% (-30)
Swing 21%
(and a Con vote share retention of 59%)
So the Tories are faring very much better (relatively) with Remain voters, from whom they achieved 19% of the vote in 2019 than with Leavers, of whom they took 74%. Note that their vote share retention of 59% with Leavers matches their overall vote retention. I believe this is due to them faring very badly indeed among voters who did not, or could not vote in the 2016 Brexit referendum.
This is likely to be significant at the next election as the battleground seats in England and Wales are disproportionately Leave-voting overall. And note too that R&W's 'Blue Wall' polling of London & Southern England Remain-inclined, Con held seats also shows a higher-then-average Con>Lab swing. I suspect that this is due to other demographics outweighing the Leave/Remain ones, and in any case, the seats in question still had around a 40-45% Leave vote.
So overall, I can see just one strongly-Tory demographic - over-65 voters - who are defying the proportionate fall in the Conservative vote share, and instead moving at a lower uniform rate. Meanwhile all the geographical areas with high Con2019 votes, all age-groups under 65, as well as those leave voters who supported them so strongly at the last election appear to be deserting in a proportional or greater than proportional manner.
|
|
|
|
Post by James E on May 29, 2023 11:12:13 GMT
[ Survation's 'Rural Seats' poll with a 17% swing, precedents from 1997 of Lab winning untargeted or unexpected seats etc...] ...If Labour was about to win MPs in non-urban constituencies lacking sizable towns you would expect to see this being reflected in on the ground votes in local elections and parliamentary by-elections and that isn't happening. Instead the 'normal' pattern for England is asserting itself of anti-Tory votes in urban areas mainly going to Labour and those in more rural areas to the Liberal Democrats. Hence my view that the MRP's are significantly overstating Labour in such areas and so in turn overstating the likely size of a Labour majority. UNS may well be a more realistic indicator of the overall number of seats for each party, although quite possibly a poor way to indicate which particular seats will change hands. Surely you don't still expect that Local Elections and by-elections to be a good predictor of the LibDems' General Election performance? I have recently posted data showing that the average LD "overperformance" in recent years in Local Elections has been +9 points compared to their GE poll figures. So in recent polls, this is a an 11% GE polls rating, and then a 20% share in the LEs. Labour underperform in LEs (compared to GE polls) by an average of around 4-5 points, but generally more when they are polling high, and less when polling low. This has always happened. And as per the articles I have re-posted from UKPR1, Anthony Wells has shown that Council by-elections are an even worse predictor. Moreover, the pattern of Lab underperformance and LD overperformance is very uneven. In some places, Labour have fared somewhat better in LEs than GEs. In others they fare much worse. Meanwhile, we now have that Survation poll of the "100 most rural seats in England and Wales". Survation's most recent GB poll put the Conservatives down 16 points on GE 2019, Labour up 12, and the LDs unchanged. But in those rural seats, it's Con -18, Lab +16, and LDs -3. So this shows labour outperformaing the headline GB polling figures in rural areas, let alone the LE results. And this should not come as a surprise when English regional polling figures consistently show the highest swings in the South, especially the South West of England. And they have been doing so for at least the last 18 months. It isn't MRP that you (and Kellner) are arguing against here: it's detailed polling data. |
|
|
|
Post by James E on May 28, 2023 20:57:27 GMT
It's a very interesting article and provides a rationale for the approach that pollsters like Opinium have chosen to use, that of assuming that "Don't knows" are likely to return to the party they supported at the last GE. In Kellner's terms they are the "grumblers". That's not right, LL. If you look at Opinium's regional headline figures, which are compiled after allowing for the disporportionate Con2019 'Don't Knows', you still get the kind of 'proportionate swing' that I have highlighted. For example, the latest Opinium shows Lab 3 points ahead in the South, but this becomes a tie when they upweight the Con2019s who give a positive response. But despite this, their VI figures for England still show the kind of regional variation which Kellner expects to disappear in an actual General Election. They average an 18% Con>Lab swing in the South, 17% in the Midlands, 11% in the North and 6% in London. These regional figures reflect a pattern of (mostly) proportional vote losses by the Tories. Overall the Conservatives' VI is 63% of their 2019 GB level. It is 63% of the 2019 level in the South samples, and 64% in their North of England samples. The slight variations are a low vote retention in the Midlands (58%) and a high retention in London (75%), which is the opposite of what UNS would predict. EDIT - to add: Looking at Kellner's article, it seems to me that he has made the same mistake. to quote: " I hope they will use their large MRP samples to check for any relationship between past Tory support and current defections. As the election approaches, I expect many Tory grumblers to return to the fold, after their mid-term protests. A lot of them have been telling pollsters they “don’t know” how they would vote. The number is likely to diminish sharply when the time comes to choose the next government. Assuming that the biggest overall shift will still be away from the Conservatives, the key to success will be to estimate the correct relationship betweenf Tory loyalists and defectors in each seat, bearing in mind that the proportions are likely to be different in strong Tory areas than in marginal seats." So he hasn't actually checked polling data himself to see whether the movements look proportional or uniform. As someone who DOES check and analyse detailed polling data, I find this an extraordinary admission on Kellner's part. He doesn't know, but someone else should check all their figures to find if his guess is correct or not! And he thinks that the proportional swing pattern found by those applying MRP might have something to do with the high number of Con 2019 voters answering 'Don't Know') . It doesn't - as per data that I have shared on this board today and over the past year. |
|
|
|
Post by James E on May 28, 2023 18:59:07 GMT
@john Chanin
Thank you, that's a very interesting article, and one that I will need some time to consider carefully. But even on a first reading, I would take issue with some of the (rather limited) data which Peter Kellner cites:
1. His graph on the Con vote in by-elections does not distinguish between Con/Lab and Con/LD contests. The latter, when the Tories are in power, have always tended to produce far larger swings, and larger Tory vote losses. The only specific by-election he cites is Christchurch in 1993.
2. His use of the 2019 Election is dubious to me. The overall swing of around 4.5% was not particularly high, and much of the vote lost by Labour went to the LDs rather than Labour.
3. I don't think he can claim the 1997 election as a triumph for UNS, either. There is more detail in the article I linked in a reply to pjw1961, but a fairer description would be that the 1997 results were semi-proportionate. In Con/Lab contests in 1997, the swings ranged from typically 8% in a safe Lab seat to 13% in a safe Con seat, moving on a sliding scale for those in-between. This was roughly at the half-way point between uniform swing (of 10%) and proportionate swing, which would range from about 5-16%.
4. He does not appear to engage at all with the detailed polling findings showing the Tories losing most support where they are strongest. It looks to me like we are just asked to accept that the pattern of huge losses in the South and Midlands of England are due to the presence of a huge number of 'grumblers' who will come back, while the relatively small losses in London, or in Scotland (of around 6-7 points rather than 20+) are explained by a lack of 'grumblers' and a much higher proportion of 'defectors'
5. (Added 21:10) The issue is not really MRP itself, even though a lot of Kellner's article seems to be about MRP: it's the polling findings which it is based on. Where MRP shows high swings in the South or Midlands of England, or in rural seats, this will be because polling in these demographics shows just that. And these are long-standing features of polling showing up across multiple polling organisations, which I have been highlighting here for the best part of a year. They are a consistent feature of the past 12 months' polling, not a product of MRP.
|
|
|
|
Post by James E on May 28, 2023 16:58:38 GMT
[Rural seats poll...] results, with comparison to GE2019, it's: Con 41 (-18) Lab 36 (+16) LD 13 (-3) Con2019>Lab swing of 17% compared to current average of 14%. Survation's most recent GB poll in late April was also a 14% swing. etc... What the polling doesn't and can't allow for is the degree of effort parties put into seats and the message the voters take from that. Labour will not be targeting Central Devon (no Labour councillors, likely few members, a Lib Dem council) but the Liberal Democrats might - at least locally even if the national party demurs. Likewise South East Cambridgeshire. I will bet Labour don't even have canvass returns for Central Devon and will have no idea who or where their supporters are. This is not to say that Labour can't win seats they have never won before, as with Canterbury in 2017 - but Canterbury has a city in it capable of electing Labour councillors, so there was something to build on. Therefore, Labour winning Aldershot or Basingstoke I can believe, but not places like Central Devon. Thanks for that, and I'd agree that these seats are not likely targets - although that must apply to all of them, not just Central Devon. But that may not prevent Labour making better-than-average progress, as they have done in seats such as Central Devon, Huntingdon, or South Norfolk over recent elections. And the precedent from 1997, when Labour strictly targetted the 90 most winnable seats, showed that the outcome may not vary much regardless. From memory, there were several gains such as Harwich and Hastings which had not been on the radar at all. Looking again at the GE1997 analysis linked below (see p14 of18), the largest Tory vote losses and largest Labour vote gains of all in 1997 came in safe Conservative seats, as opposed to those which were marginal or 'possible'. www.dannydorling.org/wp-content/files/dannydorling_publication_id1318.pdfConservative held seats from 1992-97, vote change Marginal : Con -12.3, Lab +12.2 Possible: Con - 12.2, Lab +13.7 Safe: Con -13.1, Lab +13.8 The outcome was therefore noticably more than the overall 'uniform' swing of 10%, but less than a 'proportionate' swing might have given (15%?) |
|
|
|
Post by James E on May 28, 2023 13:47:14 GMT
Universal swing obviously doesn't work well at the extremes, but proportional swing doesn't work there either. For two party marginals there isn't of course any significant difference. But where seats are further away from par, and there are several parties, past experience shows that universal swing is closer to reality than proportional swing. Of course this is counter intuitive - you would expect a similar proportion of Conservative or Labour voters to shift/abstain. But that isn't what happens.MRP seeks to slice the electorate into sections which behave differently. It works quite well where there are two parties in contention, if done properly. But it can come up with nonsense where it doesn't reflect the starting position. Occasionally there is a different dynamic, where tactical voting predominates, generally to get rid of an unpopular government. Then results will be worse than a straight reading of opinion polls might suggest. I'd be interested to know the evidence for this - and in particular the 'past experience'. My evidence for a proportional swing is the most recent polling from YouGov, Opinium and R&W in which there is a clear pattern of large swings where there were most Tory voters in 2019, and small swings where there were fewest. So to take the figures from the most recent 7 YouGovs and 8 Opiniums - with an overall swing of 14% (YG) and 13% (Opin): London: 7% (YG), 6% (Op) South of England: 18.5% (YG), 18% (Op) Midlands: 16.5% (YG), 17% (Op) North of England: 12% (YG), 11% (Op) However, if you look at the proportion of their 2019 vote which the Conservatives appear to be retaining, there is a consistent pattern. YouGov's GB figures is 59% retention, and by English areas, it's 59% in the South, 57% in the Midlands, and 61% in the North. Per Opinium, the Tories GB retention is 63%, while it's 63% in the South, 58% in the Midlands, 64% in the North. (Incidentally, both show 75% Tory vote retention in London, which is the opposite of what UNS would produce). With Redfield and Wilton (10 most recent polls), the pattern is even more clear. The Con>Lab swings by region fall in inverse proportion to the GE2019 result: Large 2019 Con lead: W Mids 21.5% swing, South West 21.5% swing, E Mids 19% swing, SE 17.5% swing, East 17% swing. Small 2019 Con lead: Yorks and H: 14% swing. Small 2019 Lab lead: North West 12% swing, North East 10% swing, Wales 10% swing Large 2019 Lab lead: London: 0% swing. And the pattern by region is the same per R&W: their polls show the Tories retaining 64% of their 2019 vote overall. Aggregating their regional figures, they show 63% vote retention in the South of England and 66% in the North. The variances again come with the Midlands where they are again doing worse than proportionately with only 54% retnetion, and London, where the Tories appear to be faring better with 84% retention. For completeness, both cross-breaks and full polls show the Tories holding an above-average share in Scotland, and a slightly-above average share in Wales - but this should be apparent to anyone who simply looks at those full polls. |
|
|
|
Post by James E on May 28, 2023 10:37:03 GMT
Polling from the Telegraph… Tories face huge losses in rural areas at next election, poll suggestsJeremy Hunt and Jacob Rees-Mogg could lose their seats if polling results were replicated “ The Country Land & Business Association (CLA) last month commissioned Survation to poll 1,017 adults living in the 100 most rural constituencies in England.
The poll found that support for the Tories among residents of the seats has collapsed by 18 points since the 2019 general election, while Labour has surged by 16 points.
The reversal in voter intentions has meant the parties are almost neck-and-neck, with the Conservatives on 41 per cent and Labour just behind on 36 per cent.
The Liberal Democrats were meanwhile on 13 per cent (down three points on 2019), while the Greens were on 5 per cent (up two points). “Other” parties were on 4 per cent (down three points). Of the 100 most rural seats in England, the Tories currently hold 96.
However, according to Survation’s analysis, if the polling results were replicated at the next election, the Conservatives would lose 21 seats, with Labour taking 18 and the Lib Dems picking up three.
Among those being displaced would be Mr Hunt. The Chancellor would lose his South West Surrey seat – which has been Conservative since its creation in 1983 – to the Lib Dems.
The other Cabinet ministers predicted to lose their seats would be Work and Pensions Secretary Mel Stride in Central Devon; Transport Secretary Mark Harper in the Forest of Dean; and Culture Secretary Lucy Frazer, who would be narrowly beaten by Labour in South East Cambridgeshire.” Many thanks for this - it addresses exactly the debate that I have had occasionally with pjw1961 about seats where there is no local Labour tradition. It also tests my 'proportionate swing' argument. Setting out the headline results, with comparison to GE2019, it's: Con 41 (-18) Lab 36 (+16) LD 13 (-3) Con2019>Lab swing of 17% compared to current average of 14%. Survation's most recent GB poll in late April was also a 14% swing. Con vote share in this poll of rural seats is 69% of their 2019 vote. This is a bit higher than Survation's 8 GB polls so far this year, which show the Tories retaining an average of 66% (on a VI of 29.4). Survation generally show higher Con share, and lower for Green and RefUK, as they do not prompt with party-names. This means that the result is near the mid point between UNS and Proportionate swing. As mentioned above, Survation's headline GB figures this year typically show the Conservatives down 15 points on about 29%. But a proportionate swing (with 66% retention) would see them losing 20% in these constituencies. So the 18-point loss could be interprted as the Conservatives faring less badly than a proportionte swing in rural areas. Perhaps the most surprising thing is that nearly all of the benefit appears to have gone to Labour. Also, note the words 'last month'. This poll was carried out in April, so presumably the 'Country Land and Business Association' were in no hurry to share the good news. However, the polls are much the same now as they were a month ago. |
|
|
|
Post by James E on May 27, 2023 20:52:36 GMT
Opinium's tables show 26% answering either 'Don't Know' or 'Would not vote', with a disproportionate number of the DKs being Con2019 voters.
Of the 74% who do state a VI, it's Con 19%, Lab 34%.
Under the 'old methodology', this would be a Labour lead of 20 points, with them having 46% to the Tories' 26%.
|
|
|
|
Post by James E on May 27, 2023 17:59:19 GMT
Interesting enough article, but I am still amazed to read Kellner claim that "...if Labour wins 45 per cent of the vote, then it is likely to have somewhere in the region of 350 MPs." This seems to be based on UNS, and is consistent with the UNS model used by Polling Report, with Labour getting 343 seats on a 45% vote share to the Conservatives' 29%, as per current polling averages. pollingreport.uk/seatsBut the inadequacy of the UNS model is clear once you start to look at Polling Report's figures for individual seats - particularly those where the Tories were on a low share in 2019, and are now predicted to get zero. If UNS was a worthwhile model for seats then detailed polling would show the regions of the UK moving by an equal swing.So the Tories would be down from 32% to 16% in London, and from 54% to 38% in the South and Midlands of England. As I have highlighted before, the polls show almost the opposite. The areas where the Tories are holding a higher proportion of their vote are where they were weakest in 2019 - such as London, Scotland, and perhaps the North East of England. Across England-outside-London, they appear to be losing around two-fifths of their 2019 vote share, as a proportional loss; in the Midlands, they are losing an even greater-than-proportionate share (per YouGov, Opinium, and R&W) and possibly in the South-West, too (per R&W). If you plug the same shares used by Polling Report (Lab 45, Con 38.7, LD 11) into Electoral Calculus, the outcome is 420 Lab seats, or 482 with 50% Tactical voting between Lab, LD and Greens. Obviously, neither of these is a likely scenario, as we can still expect some further Tory recovery towards an actual election. Electoral Calculus makes use of the differential regional swings, as well as a proportionate methodology. And while it is clearly far from perfect (it understimates the LDs, and cannot allow from differential tactical voting) it is likely to be far closer to the outcome than UNS. |
|
|
|
Post by James E on May 26, 2023 21:50:11 GMT
@leftieiberal
The local elections you refer to were at a time (May 2022) when Labour were averaging a 5% lead in GB polls, as opposed to 16% now. But I would agree that it is likely that the Conservatives will fare rather better at the next GE in Harrow East then they do across GB. And all pollsters are showing lower Con>Lab swings in London than for the rest of England - especially R&W.
But if Labour are not getting high swings in London, but consistently getting higher than average swings in this 'Blue Wall' sample (in which 10 of the 42 southern seats are in Greater London) it follows that they are doing especially well in the 32 southern seats polled outside London, most of which (22) they were in 3rd place behind the LDs in 2019.
|
|
|
|
Post by James E on May 26, 2023 21:00:49 GMT
Here is my patented highly-tuned model algorithm for predicting the number of Lib Dem seats at the next election. LD_seats_2024 = LD_seats_2019 + all seats where (LD_votes_2019 + LAB_votes_2019 + CON_votes_2019*10%) > CON_votes_2019*90% i.e. 11 + 24 = 35 Lib Dem MPs. By the way I also do economic and long range weather forecasts but you'll have to subscribe to get those I'm afraid. Hmmm... If you look at the R&W 'Blue Wall' figures, which includes most LD target seats* , plus 15 southern Con/Lab marginals, their churn figures, averaged over the past 4 polls in April & May are: Lab2019>LD: 3.5% LD2019>Lab: 18% Con2019>LD : 4% LD2019 > Con: 5% These are all Conservative-held seats, but the overwhelming majority of 2019 LD voters must be in the 27 seats where they were in 2nd place rather than the 15 where they were 3rd. *LIST OF CONSTITUENCIES Bournemouth East Chelsea and Fulham Cheltenham Chingford and Woodford Green Chippenham Chipping Barnet Cities Of London and Westminster Colchester Esher and Walton Filton and Bradley Stoke Finchley and Golders Green Guildford Harrow East Hendon Henley Hitchin and Harpenden Lewes Milton Keynes North Milton Keynes South Mole Valley Reading West Romsey and Southampton North South Cambridgeshire South East Cambridgeshire South West Surrey St Ives Sutton and Cheam Taunton Deane Thornbury and Yate Totnes Truro and Falmouth Tunbridge Wells Uxbridge and South Ruislip Wantage Watford Wells West Dorset Wimbledon Winchester Woking Wokingham Wycombe Headline figures, in chronological order: Con 35, 32,34,34 LD 20,24,23,22 Lab 37,34,36,33 |
|
|
|
Post by James E on May 25, 2023 10:08:11 GMT
hireton"Sarwar has done a competent job in appealing to Tory voters" The Con>Lab movement in Scottish polls looks to me very much the same as for GB as a whole. The most recent GB YouGovs put this at a 16% & 15%, whereas it is 17% in both of their two most recent Scottish polls. Survation's most recent poll (SNP38, Lab 31) have it rather lower at 13%. There is definitely now greater SNP>Lab than Lab >SNP movement, but with the pattern of churn we are now seeing, and a SNP lead averaging around 7 points, I would currently expect around 5-10 Labour gains in Scotland.
|
|
|
|
Post by James E on May 24, 2023 21:48:45 GMT
R&W is beginning to look a bit odd- they've already had three "outliers" on national opinion polls which makes them feel less like outliers to me and have gone from one of the better ones for Labour to one of the worst. That Blue Wall today seems to have followed a similar pattern to their national polls and none of these are comparable to other pollsters who are recording something more like polldrums. I wonder if they have changed their methodology or if their methodology is causing the smaller leads where other pollsters methodologies keep things stable? I know some people don't quite trust yougov and they can certainly be very sensitive to small bits of news for some reason but I always pay more interest to them than any of the others. I'd trust youGov and Opinium more than other pollsters, too, for two reasons. One is their good track record, but another is their panel, which eliminates the issue of false recall which can affect those who ask for respondents' past vote as part of their polling questions. As others have pointed out, this latest Blue Wall poll from R&W follows the pattern of their previous polling of this sample, which was Con 50%, LD27%, Lab 21% in 2019. It's a 14% Con>Lab swing compared to around 12% in R&W's last couple of GB polls, so as usual R&W are showing Labour doing better where they are attacking. Labour would still win both Bromley & Chislehurst and Chelsea & Fulham for the first time ever on these figures, as well as comfortably taking Johnson's Uxbridge seat, so no great cause for worry in the red camp. |
|
|
|
Post by James E on May 17, 2023 18:17:21 GMT
Another low swing in Wales per R&W 7.5%, but this is consistent with other polling evidence. R&W's own recent cross-breaks put the Con>Lab swing in Wales at 11% compared to an average of 14%.
It would still be still enough to reduce the Tories to very few seats in Wales, where there are a lot of close-range labour targets. On the old boundaries, the Conservatives would retain 3 seats per UNS on this poll - or none if their vote-losses are proportionate in the 3 longer range seats of Monmouth, Camarthen West and Pembrokeshre South, and Clwyd West.
|
|
|
|
Post by James E on May 16, 2023 18:12:58 GMT
Stats for Lefties 🏳️⚧️ @leftiestats · 59m 🚨 NEW: Labour lead drop to just 11pts (-3) 🔴 LAB 42% (-2) 🔵 CON 31% (+1) 🟠 LD 13% (+3) 🟢 GRN 5% (-1) 🟣 REF 5% (-) 🟡 SNP 3% (-) Labour majority of 34 seats. Via @moreincommon_ , 12-15 May (+/- vs 6-11 Apr) I find it rather silly for this site to provide a seats projection for each and every poll. Yesterday's 45/29 Deltapoll was projected by them as a 146 seat majority, so today's Savanta (@46/29) will presumably be even higher, but for now their most recent projection is a majority of 34. Electoral Calculus - using the same 'moreincommon' figures of 42/31 shows a majority of 92, rising to 146 with tactical voting. |
|
|
|
Post by James E on May 15, 2023 21:15:05 GMT
A bit of analysis of Opinium's recent English regional cross-breaks. Overall, the 8 Opinium polls since the start of March show an average of Lab 43%, Con 28%, so a swing of about 13% compared to GE 2019. By English region the averages are: London Lab 53% (+5) Con 24% (-8) Swing 6.5% South Lab 38% (+16) Con 34% (-20) Swing 18% Midlands Lab 45% (+12) Con 32% (-22) Swing 17% North Lab 50% (+7) Con 25% (-14) Swing 10.5% These movements follow the same pattern as with my previous analysis of R&W, Deltpoll and YouGov. And as I have noted before, Opinium's detailed figures are very close indeed to YouGov's. For the past 5 YouGovs we have averages of: London: Lab 52%, Con 24%: swing 6% South: Lab 37%, Con 33% : swing 18% Midlands: Lab 43%, Con 32% : swing 16% North: Lab 52%, Con 23% : swing 12% There is a further point on which Opinium and YouGov are very similar: the Conservatives' regional vote retention. Overall, YouGov show the Tories keeping 60% of the 44.7% overall GB share they took in 2019, and Opinium 63%. Both show the Tories doing rather better than average in London, retaining 75% of their vote share, which is down from 32% to 24% with both. In the South the Tories retain only 63% with Opinium and 61% with YouGov. So the Tories' vote-losses in the South entirely proportionate to GB as a whole, but are high because they had most support there in 2019. In the Midlands, it's even worse as their retention is just 58% with Opinium and 59% with YouGov. And in the North, they retain 64% with Opinium and 60% with YouGov - very much the same proportions as the South - or GB as a whole. This differential regional pattern makes a huge difference when it comes to seats. A uniform 13% as per Opinium's headline average (with lower swings of 10% in Wales and 7% against the SNP in Scotland) would see Labour making 130 gains, so a total of 332 seats and a majority of 14. However, using the differing English regional figures (and the same as above in Wales and Scotland), we get 174 Labour gains due to all the long-range ones in the South and Midlands. This would give a majority of 102. www.electionpolling.co.uk/battleground/targets/labour |
|
