15 Jun 2024 : LAB: 43% / CON: 19% / REF: 16% / LDEM: 11%

[pjw1961]

Jun 16, 2024 9:03:30 GMT alec said:

Daily Mail trying to link Starmer to communist spying claims - x.com/jdpoc/status/1802062270828106022

They are really desperate. 

[crossbat11]

shevii

If and when my post gets to 10 "likes", I promise to undo mine in order to get it back to single figures again.

🤣

[thylacine]

I'm not sure what Sunak is up to in his interview with the Times where he is bringing his faith into the political arena. Isn't this something that politicians in the UK usually avoid ?
Perhaps it's preparation for life after No 10.

[pjw1961]

Jun 16, 2024 8:50:02 GMT shevii said:

Jun 15, 2024 22:19:24 GMT pjw1961 said:

The statistical study of the MRP models that was linked to on here a day or two back (Focaldata?) pointed to the YouGov as appearing to be the most reliable. It will be interesting to compare the final predictions of each MRP model to the actual results. 

Having comprehensively denounced the reliability of Survation's individual seat forecasts, I still can't resist quoting the Braintree one: Labour 37.1%, Conservative 34.4%

Surely you should be mildly upbeat about Braintree? I looked at the candidates and only one independent, so unless that person has some local factor which will attract voters then Braintree is one of the least complicated ones as a straight Lab v Con fight and exactly the type of seat the polls are saying will go Labour?

I would be surprised if Labour could win Braintree on current boundaries, but it is not entirely impossible. I would guess RefUK will win at least 15-20% (UKIP got 18% here in 2015). 

[Old Southendian]

Jun 16, 2024 8:35:47 GMT soph said:

The other type is broader in scope but more specific in task, usually referred to as "deep learning models". This includes AI like AlphaGo and AlphaZero - very specifically designed to learn Go, and Chess respectively. But nightmarishly good at them. So in principle, this type of AI could be built specifically for the purpose of predicting elections, and be good at it, the main roadblock is figuring out what data to use to train the AI.

For example, if you wanted to use historical polls vs actual results, making sure to feed in all understanding we have of the methodology and biases of those polls, then you'd have maybe still less than 100 data points. Compared to the billions of games of Go that AlphaGo was trained on, it's essentially not anywhere near enough to get accurate predictive results from.

Another idea could be to use the news and social media that existed in the run up to historical elections. There's a lot more of this data available, but only really useful for elections since 2010; which my hunch would be isn't varied enough to be able to predict an election that is different from the ones we've seen since 2010. It's possible that you can't predict elections with this type of data anyway, e.g. social media is flooded with bots who don't have any votes but might imply to an AI that there's a deluge of support for the party the bots are touting.

In either case, because the LLMs can't be used and we'd instead need specialised models, I would expect the pollsters to adopt an accurate AI method (if it ever exists) themselves and continue to develop it, rather than that they'd be made obsolete by AI.

Well said! I was just about to chip in and say something similar (honest!) having recent got into Machine Learning techniques for data selection.  The problem with most Machine Learning techniques is the availability of large numbers of 'correct' results (in this case election outcomes) in order to train the algorithms.  We just don't have enough, and I suspect the variation in circumstances in each election would throw off any predictive ability even if we had more.  If you're not careful, the algorithm picks out some coincidental number that bears no relation to the outcome, but just happens, by chance, to correlate well.  Not unlike human logic.  "The old man's guide is chance". 

[Deleted]

Jun 15, 2024 22:16:17 GMT norbold said:

I came across an interesting wall in Clacton today in quite a prominent position on one of the main roads leading into (or out of) the town. 

Bit worrying that someone local doesn’t know whether the main road leads in or out of your own home town.

[graham]

Jun 16, 2024 7:24:57 GMT hireton said:

It seems that if Labour win the election with an absolute majority it will be the first time in the UK that a government with an absolute majority has been replaced by an opposition party winning an absolute majority since 1970.

That could also be said to be true of 1979 in that the October 1974 election had produced a small overall Labour majority.

[graham]

Jun 16, 2024 7:31:05 GMT nickpoole said:

Jun 16, 2024 7:23:16 GMT steve said:

Might be worth also considering that Labour's upcoming historic victory might well be achieved by a lower vote share than Theresa May receive in 2017, no one accused her of being popular.

But a hell of a lot more popular than the LDs.

And more than the 36% polled by Blair in 2005.

[steve]

Donald Trump has made a point in recent months of deriding his rival Joe Biden as being cognitively impaired, mocking the 81-year-old US president for his verbal stumbles and accusing him of falling both up and down stairs.


On Saturday night Trump, who turned 78 on Friday, returned to the theme during a speech in Detroit, Michigan to the rightwing group, Turning Point Action. He sarcastically quipped that Biden “doesn’t even know what the word ‘inflation’ means”, and challenged his rival in the 2024 election to take a cognitive test just as he had done when he was in the White House.

Trump told his audience that he had “aced” the cognitive test following advice from the then presidential physician, a Republican member of Congress whom he named as Ronny Johnson. “Has anyone heard of Ronny Johnson, congressman from Texas?” he asked the crowd. "


I very much doubt anyone has heard of Ronny Johnson they might however have heard of Ronny Jackson, who was trump's physician as president and is now a congressman for Texas.


About that dementia Donald.

[chrisc]

Jun 16, 2024 8:35:47 GMT soph said:

Hello! I'm a serial lurker that enjoys reading these posts in election times. I posted a little under another username back in Brexit ref/2017 elec times; but as a programmer and general techie the AI mention got me itching to say something again!

Jun 15, 2024 22:13:37 GMT mercian said:

1) I wonder whether anyone's tried to predict the election result using AI? Even if they haven't, as AI improves will it make pollsters redundant?

There's two major types of AI that are being researched at the moment. The ChatGPT, Copilot, Bard, etc. type are called "large language models", and these ones are designed almost exclusively as knowledge resources. So they're really good at, for examples, filtering down essays into key points, collecting headlines into an overview, and sharing information that exists from the vast vast collection of data they've been trained on. What they're NOT any good at is 'thinking' - they can't play games, solve maths problems, make science hypotheses or, importantly, make election predictions. It's very unlikely that this type of AI on its own could predict elections, though it could form a part of a larger construction.

The other type is broader in scope but more specific in task, usually referred to as "deep learning models". This includes AI like AlphaGo and AlphaZero - very specifically designed to learn Go, and Chess respectively. But nightmarishly good at them. So in principle, this type of AI could be built specifically for the purpose of predicting elections, and be good at it, the main roadblock is figuring out what data to use to train the AI.

For example, if you wanted to use historical polls vs actual results, making sure to feed in all understanding we have of the methodology and biases of those polls, then you'd have maybe still less than 100 data points. Compared to the billions of games of Go that AlphaGo was trained on, it's essentially not anywhere near enough to get accurate predictive results from.

Another idea could be to use the news and social media that existed in the run up to historical elections. There's a lot more of this data available, but only really useful for elections since 2010; which my hunch would be isn't varied enough to be able to predict an election that is different from the ones we've seen since 2010. It's possible that you can't predict elections with this type of data anyway, e.g. social media is flooded with bots who don't have any votes but might imply to an AI that there's a deluge of support for the party the bots are touting.

In either case, because the LLMs can't be used and we'd instead need specialised models, I would expect the pollsters to adopt an accurate AI method (if it ever exists) themselves and continue to develop it, rather than that they'd be made obsolete by AI.

As I’m sure soph knows, but perhaps worth mentioning here, deep learning AI models don’t just play games. They have already revolutionised a number of fields, especially scientific research. For example AlphaFold has predicted  thousands of new protein structures and is a boon to academic biochemists and the drug discovery industry.  I would say these methods are one of the biggest changes to the field of molecular biology since the discovery of the DNA double helix. And, by and large, all open source. Almost makes me regret retiring…..

[Mark]

Jun 16, 2024 8:35:47 GMT soph said:

Another idea could be to use the news and social media that existed in the run up to historical elections. There's a lot more of this data available, but only really useful for elections since 2010; which my hunch would be isn't varied enough to be able to predict an election that is different from the ones we've seen since 2010. It's possible that you can't predict elections with this type of data anyway, e.g. social media is flooded with bots who don't have any votes but might imply to an AI that there's a deluge of support for the party the bots are touting.

There are other variables that, I think, would make it hard for AI to parse through the data...

* Is there a correlation between political views and using ad-blockers? Those that use them won't even see the targeted political ads, but, as it's done user side, it would be hard to pick up on this. Market research type polling shows that there is a strong correlation between choice of browser and whether a blocker is used (Firefox 42%, Chrome 18%, Edge 6% - Opera blocks ads by default), but is there such a correlation with political views?

* Is there a correlation between political views and social media accounts that are private/friends only...accounts that an AI system simply won't be able to see.

In both instances, is there a demographic correlation outside of party support/affiliation?

* How much chatter is in FB groups, where it would largely be seen by like minded people, in other words, the already converted?

[Deleted]

Not a fan of criticising people’s looks but……………………. ole Wes Streeting has a rather odd, slab face doesn’t he? (Plus an odd hair cut.)

A lot of energy though so it all balances out…. I hope.

[Mark]

Jun 15, 2024 23:07:19 GMT mercian said:

Latest from the Muslim vote site recommending candidates:
Independent 24
Workers 14
Green 10
LibDem 6
No Party 2
Kubo 1

The site seems to be very volatile. There are now 33 seats with big Muslim population where there is no recommended candidate, and there is clearly no consensus nationwide. Notable that there are no Labour, though 31 of the 33 are Labour seats. I have no idea what Kubo is.
Anyway, it will be interesting to see how much effect these recommendations have.

I suspect that this site will have very little effect indeed.

Firstly, the site doesn't reflect the Muslim viewpoint, but one Muslim viewpoint.

Then you have to consider who is behind it. Is it set up by one person/a small group of people - and what are their political leanings? Such a site set up by someone in the Socialist Workers Party will show different recommendations to those put forward by someone that is a LibDem.

Does it take tactical voting into account at all, or, is it simply which candidates are most 'on message'?

If it is a group, rather than a single person/small group that is behind it, who are they? Are they even Muslim? Could it even be the work of a foreign actor/state (EG. someone like Putin).

Whatever the answers to the above, how much reach does it have? How many of the Muslim population are even aware of it?

[jib]

I was musing over the MRP offering from Survation. Plausible indeed and it looks like the entry of Farage into the campaign has transformed things for the Lib Dems. Sunak must be despondent that his advisers didn't foresee this, but it appears they were too busy hot tailing it to the bookies for a "cheeky* flutter".

The SNP are still on course for majority of MPs in Scotland. I failed to understand those who wrote off their chances, and their travails under Sturgeon and Yousaf look behind them now. Trouble in store for Starmer as we can expect the constitutional settlement to be a big issue in the next Parliament and a weakened mandate for him north of the border. 

* illegal

[chrisc]

On the thoughts of AI and elections/polling, I am reminded of “psychohistory” , Isaac Asimov’s suggestion in his Foundation SF books that here would ultimately be a mathematical theory of human behaviour, just as there was of thermodynamics. You just needed a large enough sample size. In later books he narrowed this down to just needing a variety of different representative input cultures, rather than needing billions of humans. To get back to polling, do we have to restrict AI models to UK elections? Or is there a way of including the much larger datasets of all the elections taking place elsewhere in the world. I guess every day there is a national election somewhere, certainly every week.

The problem is that the data may be too heterogenous to be a useful input and there would be a danger of GIGO, garbage in, garbage out. But even if we restricted it to similar election systems and political systems, it could increase the sample size. There is much talk recently of a Tory meltdown like happened in Canada. That electoral system is so similar to ours (and has extensive polling data) that it would at least be one useful additional set of data points.

[steve]

The mrp figures for the SNP are inconsistent with nearly all other polls.
The electoral calculus latest poll of polls has the SNP on 20.

A 1% increase for the lib dems and 1% fall for the Tories sees the liberal democrats in second place.

It's not an unrealistic possibility the bookies currently offer just 11/4 on that result.

[James E]

jib

"The SNP are still on course for majority of MPs in Scotland. I failed to understand those who wrote off their chances, and their travails under Sturgeon and Yousaf look behind them now."

Survation's MRP figures (SNP 37%, Lab 30%) are out of line with other polling in Scotland. Not one of the past 15 polls shows the SNP ahead and the overall average is for Labour to be 4 points in front. YouGov's cross breaks show Labour an average of around 6-7 points in front, too. It seems to me that where MRPs ask a 'constituency question', the SNP may be picking up some tactical support from the Scottish Greens, so perhaps this would explain some of the difference, but certainly not all of it. Or this may be a house effect of Survation's MRPs, as they have consistently shown higher SNP seats than others such as YouGov.

en.wikipedia.org/wiki/Opinion_polling_for_the_2024_United_Kingdom_general_election#Scotland


Of course it is possible that Survation's MRP is entirely correct, and most of the polls listed above are wildly wrong. But I think that the more likely scenario is that this one is generous to the SNP, and the reality is somewhere near to a tie. This makes a accurate prediction of seat totals in Scotland almost impossible.

[steve]

Im not an expert in the Hindu faith but I'm pretty sure it doesnt promote the constant lying, immoral greed, cronyism and wrecking of people's essential services that Sunak has been advocating for the past 5 years.

[bardin1]

Jun 16, 2024 10:59:28 GMT James E said:

jib

"The SNP are still on course for majority of MPs in Scotland. I failed to understand those who wrote off their chances, and their travails under Sturgeon and Yousaf look behind them now."

Survation's MRP figures (SNP 37, Lab 30) are out of line with other polling in Scotland. Not one of the past 15 polls shows the SNP ahead and the overall average is for Labour to be 4 points in front. YouGov's cross breaks show Labour an average of around 6-7 points in front, too. It seems to me that where MRPs ask a 'constituency question', the SNP may be picking up some tactical support from the Scottish Greens, so perhaps this would explain some of the difference, but certainly not all of it. Or this may be a house effect of Survation's MRPs, as they have consistently shown higher SNP seats than others such as YouGov.

en.wikipedia.org/wiki/Opinion_polling_for_the_2024_United_Kingdom_general_election#Scotland


Of course it is possible that Survation's MRP is entirely correct, and most of the polls listed above are wildly wrong. But I think that the more likely scenario is that this one is generous to the SNP, and the reality is more likely to be somewhere near to a tie. This makes a accurate prediction of seat totals in Scotland almost impossible.

I agree that Survation is out of line with recent polls although there has been a narrowing - Ipsos/ STV [03 Jun 2024 - 09 Jun 2024] had 36% each (with marginally more SNP in the tables) recently so there may now be crossover, but I would accept unlikely to be anything like a 7% lead for the SNP.  

The Ipsos/ STV had support for independence at 51%. Their narrative, which I  give for balance below, suggested they thought Labour had more chance to shift voters than the SNP, but my thinking is that supporters of independence who were unhappy about the Sturgeon/ competence issues may decide on the day to hold their noses and vote SNP, while the 'may shift  to labour' voters may be  more influenced by the likely media negativity/ scaremongering  in the run up to election day. . That is not scientific and could quite possibly be wishful thinking on my part 

Here's the Ipsos narative

Scottish Independence

The poll shows a small lead for Yes, down 2 percentage points compared with Ipsos’ previous poll in January. Among those likely to vote either Yes or No in an immediate referendum, 51% say they would vote Yes and 49% No.



Emily Gray, Managing Director of Ipsos in Scotland, commented:



This election campaign in Scotland is about the persuadables, with 42% of likely voters saying they may change their mind by polling day. Although it currently looks a very close race between the SNP and Labour, there are signs that Labour may be in a stronger position than the SNP to win further voters over during the campaign. Of those who may change their minds, Labour is likely to be the main beneficiary, with 24% of this group saying they may switch to Labour, compared with 12% for the SNP. The Conservative vote looks soft, with 55% of those intending to vote Conservative saying they may change their mind – and those voters would be most likely to switch to Labour. Given the profile of marginal seats in Scotland, even small changes in vote share (and remember that polls have a margin of error too) can make a big difference to the final result – which means the parties still have a huge amount to play for in the remaining weeks of the campaign.

[alec]

This reasonably sums it all up for Conservatives - x.com/LBC/status/1802273265244574183

"Dishonest, Feckless and Useless" would make for a handy Labour poster, but to think that this is being routinely applied to what was once considered 'the natural party of government' is really rather extraordinary.

We've become inured to the state the Conservative Party has got itself into, but stepping back and looking at this from a more distant perspective, we really are living through some extraordinary times.

[pjw1961]

Jun 16, 2024 10:03:28 GMT graham said:

Jun 16, 2024 7:24:57 GMT hireton said:

It seems that if Labour win the election with an absolute majority it will be the first time in the UK that a government with an absolute majority has been replaced by an opposition party winning an absolute majority since 1970.

That could also be said to be true of 1979 in that the October 1974 election had produced a small overall Labour majority.

The point being made is that in 1979 and 1997 Governments that had enjoyed a majority at the previous election had lost that majority by the time of the subsequent election. If that were not the case hireton's observation wouldn't hold up, which it just about does. 

[graham]

Jun 16, 2024 10:44:15 GMT jib said:

I was musing over the MRP offering from Survation. Plausible indeed and it looks like the entry of Farage into the campaign has transformed things for the Lib Dems. Sunak must be despondent that his advisers didn't foresee this, but it appears they were too busy hot tailing it to the bookies for a "cheeky* flutter".

The SNP are still on course for majority of MPs in Scotland. I failed to understand those who wrote off their chances, and their travails under Sturgeon and Yousaf look behind them now. Trouble in store for Starmer as we can expect the constitutional settlement to be a big issue in the next Parliament and a weakened mandate for him north of the border. 

* illegal

No - this MRP survey is gibberish re- Scotland. Implies SNP sweep Glasgow but lose Falkirk.  They will do well to win 20 seats.

[Deleted]

Jun 16, 2024 10:37:18 GMT Mark said:

Jun 15, 2024 23:07:19 GMT mercian said:

Latest from the Muslim vote site recommending candidates:
Independent 24
Workers 14
Green 10
LibDem 6
No Party 2
Kubo 1

The site seems to be very volatile. There are now 33 seats with big Muslim population where there is no recommended candidate, and there is clearly no consensus nationwide. Notable that there are no Labour, though 31 of the 33 are Labour seats. I have no idea what Kubo is.
Anyway, it will be interesting to see how much effect these recommendations have.

I suspect that this site will have very little effect indeed.

Firstly, the site doesn't reflect the Muslim viewpoint, but one Muslim viewpoint.

Then you have to consider who is behind it. Is it set up by one person/a small group of people - and what are their political leanings? Such a site set up by someone in the Socialist Workers Party will show different recommendations to those put forward by someone that is a LibDem.

Does it take tactical voting into account at all, or, is it simply which candidates are most 'on message'?

If it is a group, rather than a single person/small group that is behind it, who are they? Are they even Muslim? Could it even be the work of a foreign actor/state (EG. someone like Putin).

Whatever the answers to the above, how much reach does it have? How many of the Muslim population are even aware of it?

But it makes ole mercy man happy thinking it matters - so that’s alright.👍 

[leftieliberal]

Jun 16, 2024 8:54:51 GMT pjw1961 said:

Jun 15, 2024 23:06:00 GMT RAF said:

Well that depends. STV can only work in multi member constituencies and each party will have a slate of candidates for that constituency equivalent to the seats available. So whilst you can rank the candidates you wish to select in order of preference, if you want to cast all three of your votes for a single party you will have to vote for the candidates the party has chosen to contest the seats.

I guess you could have an open list/STV hybrid which would allow you to rank your three candidates from a wider party slate (of say 10 candidates).

No, its simpler than that. The voter ranks all candidates on the ballot paper (at least those they choose to vote for at all) in order of preference. At the count the candidate with the fewest first preferences (i.e. marked as "1") is eliminated and their second preferences redistributed until eventually you are reduced to the correct number of finally elected people. Parties cannot control how voters rank their offering and the favoured party HQ 'parachute' candidate may well end up eliminated by some popular local one that the party just allowed on as a makeweight. 

The only downside I can see is that there are going to be some very long ballot papers. 

That is actually only true for AV. For STV in a multiple-member constituency, before anyone is eliminated the surplus votes of those who are above quota are redistributed first. This is normally done by redistributing all their votes, but giving each a weight that corresponds to the surplus (it's much easier using a computer program for this). In practice, STV elections would be quite like London Mayor elections when the Supplementary Vote method was used and all the ballot papers were machine read. 

[oldnat]

Jun 16, 2024 8:44:42 GMT leftieliberal said:

Jun 15, 2024 22:49:34 GMT pjw1961 said:

It is not proportional though. In fact it could easily be less proportional than FPTP!

I'm an STV man myself, as it weakens the power of the party machines. The voters choose which candidates from each party they wish to prefer rather than the party under a list based system. 

There is only one real downside to STV and that is the size of the constituencies. For reasonably good proportionality you need constituencies that return around 5-7 MPs (because the quota to win a seat is 1/6+1 (just under 17%) to 1/8+1 (just over 12.5%)). For big cities like London this is about the size of London Assembly constituencies. But in rural areas the constituencies would become very big, say Cornwall as a single constituency, and it is even worse in Scotland. As we already allow four island constituencies to be excluded from the population rules for defining constituency boundaries, I think any STV constituencies should include a few AV (single-member constituencies with preferential voting) in the Scottish Highlands.

Since 2021, the Scottish local government wards are in the process of doing precisely that - starting with councils including populated islands - (a single member ward is effectively AV, as in by elections). In my council, Arran is a single member ward.

[Deleted]

I wonder if ole Colin has finalised his decision to vote Tory again yet?

[neilj]

Jun 16, 2024 12:04:36 GMT @fecklessmiser said:

I wonder if ole Colin has finalised his decision to vote Tory again yet?

I was hoping he'd come back

[laszlo4new]

Jun 16, 2024 11:05:58 GMT steve said:

Im not an expert in the Hindu faith but I'm pretty sure it doesnt promote the constant lying, immoral greed, cronyism and wrecking of people's essential services that Sunak has been advocating for the past 5 years.

Well, in the 2019 election in India, 43% of Modi's winning candidates were under criminal investigation. Half of these were persecuted for murder, attempted murder, rape. They were elected because the Indian justice system had 30 million backlog cases. Just for balance: Reporting other candidates at election times to police is common.

So, basically the criminals stopped (they didn't) financing the political parties for advantages, but they became candidates.

Police found huge amount of cash, alcohol, jewellery in party offices that were meant for buying votes.

Sunak didn't talk about it (not even when he embraced Modi).

[leftieliberal]

Jun 16, 2024 9:10:47 GMT James E said:

shevii

But by my preferred like-with-like comparison to the longer-term average of the first 20 weeks of 2024, the Labour VI has fallen by far less. Per the most recent polls we have from Opinium, Techne, Savanta, YG, R&W, WeThink, Survation, BMG and MIC, Labour are down by an average of just 1.2 points from where they stood in the pre-election period (from 1 Jan to 22 May).

My index based on those five (in bold) plus Deltapoll (I'm just waiting for them to complete my Week 4 set) has the Labour lead provisionally down by 0.4% against Week 3, but still up 2.8% on their lead before the election. Some of the stuff on the Wikipedia page is misleading because they are using all the polls published for their LOESS and that is affected by pollsters who publish during the election but not regularly outside elections, or publish more often during election campaigns. Unfortunately I decided that I had to remove YouGov from my set because of their methodology change that made their election polls not strictly comparable with their pre-election polling. 

[oldnat]

Jun 16, 2024 10:59:28 GMT James E said:

jib

"The SNP are still on course for majority of MPs in Scotland. I failed to understand those who wrote off their chances, and their travails under Sturgeon and Yousaf look behind them now."

Survation's MRP figures (SNP 37%, Lab 30%) are out of line with other polling in Scotland. Not one of the past 15 polls shows the SNP ahead and the overall average is for Labour to be 4 points in front. YouGov's cross breaks show Labour an average of around 6-7 points in front, too. It seems to me that where MRPs ask a 'constituency question', the SNP may be picking up some tactical support from the Scottish Greens, so perhaps this would explain some of the difference, but certainly not all of it. Or this may be a house effect of Survation's MRPs, as they have consistently shown higher SNP seats than others such as YouGov.

en.wikipedia.org/wiki/Opinion_polling_for_the_2024_United_Kingdom_general_election#Scotland


Of course it is possible that Survation's MRP is entirely correct, and most of the polls listed above are wildly wrong. But I think that the more likely scenario is that this one is generous to the SNP, and the reality is somewhere near to a tie. This makes a accurate prediction of seat totals in Scotland almost impossible.

Both YG and Survation's MRP reports highlight the large number of marginals in Scotland. I'll "predict" the party tallies here sometime on 5th July - if all the recounts have finished!