Recent Posts

It is good to finally have notional 2019 results for the new boundaries, but these are really no surprise. The old boundries also showed Labour needing about a 12-13% UNS swing, and the pattern of the changes makes a very slight shift towards the Conservatives.

But the reporting of this by the BBC (and ITN) is downright misleading. There have been numerous cases in the past 30 years where the apparent, UNS-based target for either Lab or Con to achieve a majority has shifted by as much as 7 points. For example, going into GE1997, the Tories appeared to need about a 5-6% lead to get an overall majority, but after the  1997 election, this rose to around 13%. Labour's apparent target for a majority rose from 2-3% on the basis of GE2010 to 9% after the 2015 results. So a fall in the apparent Labour target from 12-13% to 4-5% would be a similar-sized shift to what has happened in the recent past.

My own model, which is consistent with MRPs from Savanta, Survation, YouGov and the Electoral Calculus models, shows that in the event of a tie, Con and Lab would probably have around 290 seats each. I believe that Con could still get a majority with a 3-4% lead in the popular vote, and Labour most likely need a 4-5% lead for 325 seats+. But if I am right, these figures are unlikely to be reliable in determining what size of lead will really be needed in 2029 or 2030. 

Re Local By-Elections:

I posted the below in December 2022, repeating what I had stated previously on the same subject and re-quoting Anthony Wells' comments on UKPR1 in 2009.

So regular readers may have read this several times before, but not for at least a year...........




"Our former host on UKPR1 wrote on this in 2009 : "Can local by elections predict general elections?"



"Sum of local by-elections Jan-May 2005: CON 33%, LAB 26%, LDEM 31%

Actual general election result 2005: CON 33%, LAB 36%, LDEM 23%



Sum of local by-elections Jan-Jun 2001: CON 32%, LAB 30%, LDEM 26%

Actual general election result 2001: CON 33%, LAB 42%, LDEM 18%



Sum of local by-elections Jan-May 1997: CON 28%, LAB 37%, LDEM 30%

Actual general election result 1997: CON 31%, LAB 44%, LDEM 17%"

So Labour underperformed by 7, 12 and 10 points in pre-GE local by elections, and the LDs overperformed by 13, 8 and 8 points. The Conservatives' share is quite close to the subsequent GE results, though.

Incidentally, the LDs overperform (and Labour tend to underperform) in all Local Elections: another UKPR article from 2012 puts the LibDem overperformance at 7 points on average.

That Telegraph/YouGov seat prediction - Some relevant details from their article:

It's an MRP of 14,000 voters, done 'over the new year'. That's a relatively low sample for an MRP - YG take about 50,000 responses for their General Election MRPs. Such a small sample means that it needs to be taken with a pinch of salt in Scotland in particuar, where they are likely to have had fewer than 1,000 positive responses.

They state that: 'It also factors in the large numbers of undecided voters and which way they are most likely to vote, known as electoral tightening'. This is really important to bear in mind. Opinium's adjustment typically has the effect of a 6-7% reduction to Labour's lead. It is not stated how YouGov are treating undecided voters, but if they are following Opinium's methodology, the adjustment with YouGov is likely to be even larger. When I did a detailed comparison of Opinium and YouGov last May/June, the average figures for Don't Know and WNV were 26% with Opinium and 31% with YouGov.

This may explain why their headline figures are so different to the normal YouGov polls, the last 4 of which average as Con 22.5%, Lab 44.5%. It looks like the VI figures here, after the adjustments, are roughly Con 27%, Lab 38%. They state that the overall Con to Lab swing is 11.5%, and that in England and Wales Con are down 18 points and Lab up by 4. I am inferring that Labour are up by a lot more than 4 points in Scotland due to their seat gains, so as to be up by 5% across GB. So an 11 point lead after adjustments compared with an average 22% lead in the most recent normal YouGovs. I can just about see how the 38/27 figures acould be within range of the closest recent YouGov of 43/24, with the re-weighting adjustment making a hefty 8 point difference, but it would be good to see the tables.

There are some interesting details in the seat results. LibDems appear to do well, the best result being winning Horsham. Labour win Mordaunt's seat of Portsmouth North, plus Banbury, the Isle of Wight seats and Welwyn. Labour just misses Mogg's seat, where their figures are Con 33%, Lab 32%, LD 17%. The weak Labour performance in safe seats is consistent with other recent detailed polling, and Local Election results. The Telegraph seems alarmed that 80% of those voters deserting the Tories were Leave voters in 2016 but this should come as no great surprise to anyone, given that that 80% of the 2019 Tory vote was made up of Leave voters.

The result conforms closely to the '15 seats for 1% swing' pattern of other MRPs. Here Labour are gaining 183 seats from the 202 they held in 2019 on an 11.5% swing, which is 15.9 seats per 1%.

[EDIT - 24 hours on, I have found that several of the figures which I took from the Telegraph article were incorrect. See the statement from YouGov themselves to this effect. The swing is in fact 12.6%, and Labour are up by more than 4 points. The actual figures were Lab 39.5%, Con 26%. With the correct swing, the 'seats per 1% swing' is in fact 14.5 per this MRP]

@jimjam

Looking at this in more detail, I can see that the adjustments could be somewhat bigger than my rough estimate.

I am allowing for Labour's DKs to return to them as well as the Tories'. So taking the most recent YouGov, the calculations are:

Con 26% DKs x 45% 2019vote /2 = 5.85
Lab 9% DKs x 33% 2019vote /2 = 1.5
So the net difference there is 4.35 points** , not 3. 

And with Reform UK on 9 points per YouGov, an adjustment of one-third to the Tories gives them another 3 points.

Of course the figures would then need to be re-based to percentages. And it is normal for the Green vote to be squeezed a bit towards an actual election, so if you factor in 1 point going to Labour for that, we get back to my rough estimate of the Tories closing Labour's lead by about 6 points.


**This is rather less than the 6-7 points difference that Opinium's methodology makes. And of course, such an adjustment should not be applied to polling figures from Opinium or MIC, whose headline figures already include their own, larger adjustment for the 'Don't Knows'.

"Reform are not standing" (in Kingswood).

This sets up an interesting race for all the other fringe parties on the right. No fewer than 8 such parties have contested one or more of the Parliamentary by-elections in the past 6 months  :

Reclaim Party
Heritage Party
UKIP
SDP
BF
CPA
English Democrats
Scottish Family Party

You could probably add the anti-Ulez candidates in Uxbridge to that, too.


Four of these stood in Uxbridge in RefUK's absence, so I'd expect similar here. Those of the above who do stand can probably expect to share 3-4% of the total vote.

Thanks for that! The polling average has been in the 15-20% range for nearly 12 months now, and this does not look like ending soon.

As for the margin needed for a Labour majority: There is close agreement between Electoral Calculus and the larger MRPs from Survation and Savanta towards the end of last year on this. The headline figures are each showing Labour taking around 400-425 seats on a lead of around 17%. The pattern in each case is that they show Labour gaining around 15 seats for each 1% swing from the Conservatives (as opposed to the 10 seats per 1% implied by UNS). I would expect the Tories to recover around one-third of the RefUK VI and up to half of their 2019 voters who 'Don't Know'* which would give them a boost of 6 points and cut the Labour lead from around 18 to around 12. With this in mind, here is a scale of how various different Labour leads might translate in seats, based on those MRPs and the '15 seats per 1% swing' pattern.

3% Labour lead- 315 seats  (10 short)
5% Labour lead- 330 seats (majority of 10)
7% Labour lead - 345 seats (majority of 40)
9% Labour lead- 360 seats (majority of 70)
11% Labour lead-375 seats (majority of 100)
13% Labour lead - 390 seats (majority of 130)
15% Labour lead - 405 seats (majority of 160)
17% Labour lead - 420 seats (majority of 190)**
19% Labour lead - 435 seats (maajority of 220)

*around a quarter of Con2019 voters answer 'Don't Know', which is 11% of everyone who voted. All precedents suggest that some recovery of this is likely (and the same applies to the 8-10% 2019Lab DKs).

** my figures are modelled around this figure of 420 seats on a 17% lead in the vote. This would be 218 Labour gains on a swing of 14%, hence about 15 seats per 1%.

The Survation and Savanta MRP figures showed the SNP still dominant in Scotland with Labour taking around 10-15 Scottish seats. If you think Labour will do significantly better/worse than this in Scotland, then adjust their seat total up or down. 

The position of Voting Intentions by social class is now rather confused, as different pollsters show very different figures. To take the average of the past 3 Deltapolls, they show:

Deltapoll ABC1s
Lab 37%
Con 30%

Deltapoll C2DEs
Lab 47%
Con 25%

However, the recent averages per YouGov are:

ABC1s
Lab 49%
Con 21%

C2DEs
Lab 42%
Con 27%

So YG and Delta are 7 points apart for C2DEs, and 21 points apart for ABC1s. If you split the difference, the two groups average almost the same result. 

 My approach to try to get a meaningful picture of changes is to compare YouGov's figures with their own previous analyses. And they are continuing to show Labour making more progress in the higher social classes, as they have been doing steadily for decades. But what is also clear is that Social Class alone is now a very poor predictor of Voting Intention. Several polls and analyses I have seen have produced similar results for their AB and DE samples. And as I have mentioned before, the Conservatives' stongest social demographic is now C2 voters, whereas Labour do best with C1s. This was already the case at GE2019 (see link below) but has become more so. 

yougov.co.uk/politics/articles/26925-how-britain-voted-2019-general-election?redirect_from=%2Ftopics%2Fpolitics%2Farticles-reports%2F2019%2F12%2F17%2Fhow-britain-voted-2019-general-election

YouGov's figures again show a very significant Labour lead (of 16%) in the South of England. And across their most recent 6 polls, the average has been a Labour lead of 14 points, ranging between 9 and 18. With each sample being around 675 people, this is based on around 4,000 responses.

This would be a swing of 23% compared to 2019, when the Tories led by 32 points. Perhaps even more remarkable is the fact that it would represent a 12% Con to Lab swing when compared to the 1997 election, in which the Tories still managed an overall 9% lead in the South.

The Kingswood by election should be very interesting. In most respects, it seems to typify the demograhics in which the Con to Lab swings (per YG) are the largest, being Southern, Leave voting and above-average ABC1. 

Yes, it's the average of the last 4 Parliaments which have run to the full 5 year. I am comparing the average of Opinion polls in the exact month 4 years on from the previous election (as we are now) with the final opinion polling average (not the actual result, because polling error is a different thing)

So...
1982-92: June 1991 polls Lab 6% ahead, final polls Lab 1% ahead.  5-point swing back
1992-97: April 1995 polls Lab 25% ahead, final polls Lab 18% ahead. 7-point swing back.
2005-10: May 2009 polls Con 16% ahead, final polls Con 8% ahead. 8-point swing back.
2010-15: May 2014 polls Lab 3% ahead, final polls Con 1% ahead. 4-point swing back.

A significant part of this is of course the recovery of the 'Don't Knows', which is factored in to some current polling. The adjusted polls, such as Opinium, point towards a similar narrowing of about 6 points for those polls which treat 'Don't know' as 'would not vote'.

But there's wider point here: Kellner seems to be saying that MRPs showing Labour on 400-430 seats with a 17% lead are going to be proved wrong if Labour were now to get (say) 375 seats, and a 100-majority on a lead of 11%. I'd disagree, because a movement of 6 points in the polls would in inevitably shift a very significant number of seats. I've seen a similar fatuous argument that an MRP taken in May 2017, when the Conservatives were 18 points ahead in the polls, was proved totally wrong by the actual result a month later, ignoring the huge shift in the polls during the month which followed. 

Actually, that should be a "fine-tooth comb".
:-)

dictionary.cambridge.org/dictionary/english/fine-tooth-comb

For a slightly more recent comparison, I think the precedent of 1996-97 is more relevant here than 1969-70.

Looking back to the 5 polls done in April 1996 - so 4 years on from the previous GE, as we are now - the polls averaged Con 28%, Lab 53%, LD 15%. The final polls showed Labour 18 points ahead, so making a like-with -like comparison, the Tories' recovery was by 7 points. The polls also overstated Labour, but polling error is a different thing to Government recovery. 

If the Conservatives can make a similar recovery now, they would cut the deficit from 18 to 11 points.  

For a couple of other comparisons to the 4-year point: In May 2009, the Tories averaged a 14% lead, which was again reduced by 7 points in the final pre-election polls. And in May 2014, Labour had around a 3% lead, which turned into a 1% Con lead in the final polls - a 4 point movement. 

As we've touched on the 1997 General Election, here's another bit of analysis. I've done my usual YouGov cross-break analysis by 'region' and class, but here the comparisons are to where things stood after the 1997 landslide. For anyone expecting the individual constituency results to look like they did in the Blair years, the short answer is that isn't going to happen.

These figures reflect the 8 YG polls from 18 Oct onwards, which currently average as a Labour lead of 22%. Some allowance for 'swing-back' really should be made - say 4% for a winning margin of 14%. 

London
Lab 55% (+6)
Con 20% (-11)
Swing Con to Lab 9% since 1997.

South of England
Lab 40% (+9)
Con 28% (-12)
LD 13.5% (-10)
Swing Con to Lab 10.5% since 1997.

Midlands
Lab 46% (-2)
Con 28% (-6)
Swing Con to Lab 2% since 1997.

North of England
Lab 53% (-2)
Con 21% (-4)
Swing Con to Lab 1% since 1997.

Scotland
Lab 32% (-13)
Con 14% (-4)
SNP 36% (+12)
Swing Lab to SNP 12.5% since 1997.

Wales
Lab 45% (-8)
Con 17% (-3)
Swing Lab to Con 2.5% since 1997.

ABC1 Voters
Lab 48% (+14)
Con 21% (-18)
Swing Con to Lab 16% since 1997

C2DE Voters
Lab 41% (-13)
Con 27% (+3)
Swing Lab to Con 8% since 1997

Some of these changes are already visible. The shift in London towards Labour has been evident in the past 3 elections, and as of 2019 the Tories were already faring better with C2DE voters than ABC1's (per YouGov's analysis, at least) . However, this has become even more pronounced in the past 4 years. Labour is likely to fare less well in Wales than in 1997 and very likely also to less dominant than it once was in the North of England. On the other hand, the South is going to look very different. Even if we take Opinium's figures as an alternative, these show a 6% Con to Lab swing from 1997 in the South, with Labour 4 points ahead where they trailed by 9 points in the 1997 landslide. And this is in the context of Opinium's re-weighting methodology which reduces Labour average lead from around 22% to 15%.

"...the outcome for the Conservatives could be bleak indeed – maybe as few as 130 seats..."

There was a little-publicised MRP from Survation just 6 weeks ago which may be relevant here. Sample was around 12,000, so not huge by MRP standards, but larger than a normal poll. The results were:

Con 29.3% (151 seats)
Lab 45.8% (435 seats)
LD 11% (15 seats)
SNP 3% (28 seats)
Grn 4% (1 seat)
RefUK 5%

So this is a poll where the Tories are getting a relatively large share (4 points above their average) and the Labour lead is a bit lower than average on 16.5%. It's worth considering what the figures would look like if the Conservatives did a bit better than this. Taking the Low/High figures from Survation's data of Can 31%, Lab 44%, this would shift around 25 seats back to the Tories - using '7 seats per point' and cutting the lead by 3.5%.

But looking at the detailed seat findings, it iseems to me that even this may be overstating the Tories. The reason is that Survation do not seem to be making allowance for the dynamics of LibDem held seats, nor where they are the obvious challengers. Survation have the LDs taking just one seat in Scotland, for example. The LDs would almost certainly profit hugely from tactical votes in their target seats if the Conservatives only get around 30%. Similarly, I can see close Con/Lab seats in England where there does not appear to be any squeeze on the projected LD share. Tactical voting is (IMO) likely to cost the Tories 30-40 seats at the next election - about half of those going to the LDs.

Putting these factors together, I would estimate that the Conservatives would win as few as 130 seats in the event that they finish around 15-16% behind Labour in the popular vote - which in my view is right at the top end of the range of plauible outcomes.

www.survation.com/archive/2023-2/
(click on 14 Nov docs)

There's a comparison to be made between polls at the start of the year and where we are now.

YouGov's first 6 polls of 2023 in Jan to early Feb averaged:
Lab 47%
Con 25%
LD 9%
Ref 6.5%
Grn 6.5%

And their most recent 6 in Nov-early Dec average:

Lab 45% (-2)
Con 23% (-2)
LD 10% (+1)
Ref 8.5% (+2)
Grn 7.5% (+1)

It's a similar story with Opinium, with Lab 44 Con 28 turning into Lab 42, Con 26.