8 October 2020

Will Biden win? US Election 2020

Four years on from 2016, and we're back looking at polls, swings, gains, losses, etc. Well, in a way, it's only one year on from the last time we did this for the 2019 UK General Election.

First, let's get the disclaimer out of the way: I am not a predictor; I am merely attempting to interpret the polls in terms of seats electoral college votes. In 2016, this is something that I had great success with in a way. The takeaway point in 2016 was that the swing in the marginal states was higher than the national swing, which was enough to push Donald Trump into the White House. https://rhysbenjamin.blogspot.com/2016/11/how-did-trump-win.html

However, Trump has a very difficult task if he wants to stay there.

Theoretically, the swing that Biden needs to win is just 0.4%. That's 4 out of every 1,000 Trump 2016 voters going to Biden in 2020. You can also see this in his "easiest" path to the White House: gain Michigan, Pennsylvania, and Wisconsin. 



What we actually find is that, as with 2016, the current swing in marginals appears to be bigger than the national swing. If we use the swingometer we see that if we place the arrow on what the polls say is the national swing, 2.8% (using the average of the different "poll of polls" models, so basically the poll of poll of polls). On a uniform national swing Joe Biden will get a landslide, looking at 350 electoral college votes to 188. 



But what are the "marginal" states in this election? It's time to have a look at...

THE BATTLEGROUND.

This is where the election will be won and lost. These are the states in their 2016 colours, with the white line representing, to an extent, Joe Biden's "winning" post. Starting on 232 electoral votes, he needs to gain Michigan (+16), Pennsylvania (+20), and Wisconsin (+10), if he does it in order of difficulty, to get over the 270 line, because, of course, 232 + 16 + 20 + 10 = 278. 

If Biden gains Florida, that means, however, he can afford to lose any two of those three states and still win the election. Even Florida + Wisconsin gives Biden 271. If Biden loses all three of Michigan, Pennsylvania, and Wisconsin, however, or indeed if Trump makes gains (not currently predicted, it has to be said), in any of these seven targets below, then Biden's task becomes harder.

Looking at the polls across these states, however, the average swing on the current statewide polls is 3.5%, and if you weight these polls by electoral college vote, then the swing in marginals is 3.0%. This indicates that Trump is doing slightly better in the safe seats, but they're no good to him as they're unlikely to be unseated. Sorry, safe "states".

So on election night, here is your cut-out-and-keep guide to the battleground. Mark these off as they are declared, and you will have an idea of what is happening. Trump's easier states to win are at the bottom, and Biden's are at the top, with the white line representing the winning point on either side. As these come in, you should be able to tell who's winning.



Here's one more pic dump, Trump's target list:


Not paid for by either campaign. Just some independent research.

8 May 2020

2019 General Election under Proportional Representation?

This is something I always do, and I always use the only form of PR which we can use (since others require a "second choice" vote anyway, and everybody will have different #2 preferences).

That's Regional Party List. In each region, you vote for a party, not a person, and we apply that.

Now, this election was difficult to do because of the fact that not everyone stood in every seat, but we'll ignore this for our purposes. Just a bit of fun, just a bit of fun...

2017 results, first, for reference: CON 280, LAB 269, LD 43, SNP 22, UKIP 7, DUP 7, SF 6, GRN 6, PC 4, SDLP 2, UUP 2, APNI 1

2019 results, per region:
Scotland: SNP 28, CON 15, LAB 11, LD 5
North East: LAB 13, CON 11, BRX 2, LD 2
North West: LAB 36, CON 29, LD 6, BRX 3, GRN 1 (excluding Chorley)
Yorkshire: CON 24, LAB 22, LD 4, BRX 3, GRN 1
East Midlands: CON 27, LAB 15, LD 3, GRN 1
West Midlands: CON 33, LAB 21, LD 4, GRN 1
Wales: LAB 17, CON 15, PC 4, LD 2, BRX 2
London: LAB 36, CON 23, LD 11, GRN 2, BRX 1
East of England: CON 35, LAB 14, LD 8, GRN 1
South East: CON 46, LAB 19, LD 15, GRN 3
South West: CON 31, LAB 13, LD 10, GRN 2
Northern Ireland: DUP 6, SF 4, SDLP 3, APNI 3, UUP 2

TOTAL:
Conservatives 289
Labour 217
Liberal Democrats 70
Scottish National Party 28
Green Party 12
Brexit Party 11
Democratic Unionist Party 6
Plaid Cymru 4
Sinn Fein 4
Social Democratic Labour Party 3
Alliance Party of Northern Ireland 3
Ulster Unionist Party 2
(Speaker 1)
CON SHORT BY 37

In this situation potential coalitions would be as follows:
CON (289) + BRX (11) + DUP (6) + UUP (2) = 308
LAB (217) + SNP (28) + GRN (12) + PC (4) + SDLP (3) = 264

It would be the Liberal Democrats who would act as kingmaker, in effect. And with the Lib Dems having ruled out any coalition with either party during the campaign, who knows what would have happened next?

CHANGES ON 2017 UNDER RPL:
Conservatives +9
Labour -52
Liberal Democrats +27
Scottish National Party +6
Green Party +6
Brexit Party +11
Democratic Unionist Party -1
Plaid Cymru +/-0
Sinn Fein -2
Social Democratic Labour Party +1
Alliance Party of Northern Ireland +2
Ulster Unionist Party +/-0

CHANGES ON 2019 ACTUAL RESULT:
Conservatives -76
Labour +15
Liberal Democrats +59
Scottish National Party -20
Green Party +11
Brexit Party +11
Democratic Unionist Party -2
Plaid Cymru +/-0
Sinn Fein -3
Social Democratic Labour Party +1
Alliance Party of Northern Ireland +2
Ulster Unionist Party +2

25 April 2020

What if F1 had the Eurovision points structure?

Formula One has had various points systems over the years, but for this thought experiment we're going to apply the following points structure to each year:

1st place: 12 points
2nd place: 10 points
3rd place: 8 points
4th place: 7 points
5th place: 6 points
6th place: 5 points
7th place: 4 points
8th place: 3 points
9th place: 2 points
10th place: 1 point
(No points for fastest laps, pole positions, or sprint races)

Between 1950 and 1990, not all races counted towards the championship, but for our purposes, every race will count. Any driver not classified won't get points. The final round of 2014 had double points; again, we are ignoring this (it doesn't make a difference in any case).

The top three every season, therefore, would have been as follows, with different champions in bold.

1950: L Fagioli 48; G Farina 43; J Fangio 36 (1st title)
1951: J Fangio 56 (3 wins); A Ascari 56 (2 wins); J Gonzalez 50 (1st title)
1952: A Ascari 72; G Farina 52; P Taruffi 41 (1st title)
1953: A Ascari 67; M Hawthorn 60; G Farina 56 (2nd title)
1954: J Fangio 87; J Gonzalez 55; M Hawthorn 49 (2nd title)
1955: J Fangio 58; S Moss 39; E Castelloti 29 (3rd title)
1956: J Fangio 63; P Collins 54; S Moss 48 (4th title)
1957: J Fangio 68; S Moss 45; M Hawthorn 34 (5th title)
1958: M Hawthorn 76; S Moss 58; H Schell 44 (1st title)
1959: J Brabham 57; T Brooks 44; M Trintignant 43 (1st title)
1960: J Brabham 67; B McLaren 65; S Moss 42 (2nd title)
1961: P Hill 62; W Von Trips 51; D Gurney 50 (1st title)
1962: G Hill 80; B McLaren 59; J Clark 43 (1st title)
1963: J Clark 102; R Ginther 66; G Hill 55 (1st title; first time a driver scores 100 points)
1964: G Hill 67; J Surtees 62; R Ginther 57 (2nd title)
1965: G Hill 81; J Clark 72; J Stewart 61 (3rd title)
1966: J Brabham 65; J Rindt 48; J Surtees 42 (3rd title)
1967: D Hulme 85; J Brabham 82; J Clark 61 (1st title)
1968: G Hill 75; J Stewart 65; D Hulme 63 (4th title)
1969: J Stewart 89; J Ickx 63; B McLaren 58 (1st title)
1970: J Ickx 64; J Rindt 60; D Hulme 59 (1st title)
1971: J Stewart 88; R Peterson 65; F Cevert 50 (2nd title)
1972: E Fittipaldi 88; D Hulme 74; J Stewart 69 (1st title)
1973: J Stewart 115; E Fittipaldi 92; F Cevert 88 (3rd title; new highest ever score)
1974: E Fittipaldi 97; C Regazzoni 95; J Scheckter 82 (2nd title)
1975: N Lauda 108.5; E Fittipaldi 79; C Reutemann 71 (1st title)
1976: N Lauda 106; J Hunt 103; J Scheckter 96 (2nd title)
1977: N Lauda 117; J Scheckter 89; C Reutemann 88 (3rd title; new highest ever score)
1978: M Andretti 99; R Peterson 85; C Reutemann 84 (1st title)
1979: J Scheckter 109; G Villeneuve 89; A Jones 67 (1st title)
1980: A Jones 106; C Reutemann 92; N Piquet 91 (1st title)
1981: C Reutemann 88; N Piquet 87; A Jones 80 (1st title)
1982: K Rosberg 86; J Watson 71; D Pironi 67 (1st title)
1983: A Prost 96 (4 wins); N Piquet 96 (3 wins); R Arnoux 85 (1st title)
1984: A Prost 107 (7 wins); N Lauda 107 (5 wins); E De Angelis 80 (2nd title)
1985: A Prost 119; M Alboreto 87; E De Angelis 76 (3rd title; new highest ever score)
1986: A Prost 122; N Mansell 115; N Piquet 113 (4th title; new highest ever score)
1987: N Piquet 121; A Senna 104; N Mansell 91 (1st title)
1988: A Prost 154; A Senna 139; G Berger 82 (5th title; new highest ever score)
1989: A Prost 129; A Senna 86; R Patrese 76 (6th title)
1990: A Senna 116 (6 wins); A Prost 116 (5 wins); N Piquet 94 (1st title)
1991: A Senna 141; N Mansell 100; R Patrese 91 (2nd title)
1992: N Mansell 138; M Schumacher 99; R Patrese 97 (1st title)
1993: A Prost 137; A Senna 107 (5 wins); D Hill 107 (3 wins) (7th title)
1994: D Hill 130; M Schumacher 116; G Berger 71 (1st title)
1995: M Schumacher 132; D Hill 101; J Herbert 93 (1st title)
1996: D Hill 129; J Villeneuve 118; M Schumacher 91 (3 wins) (2nd title)
1997: M Schumacher 120; J Villeneuve 111; H Frentzen 84 (2nd title)*
1998: M Hakkinen 136; M Schumacher 123; D Coulthard 98 (1st title)
1999: E Irvine 126; M Hakkinen 110; H Frentzen 98 (1st title)
2000: M Schumacher 142; M Hakkinen 137; D Coulthard 123 (3rd title)*
2001: M Schumacher 165; D Coulthard 113; R Barrichello 108 (4th title; new highest ever score)*
2002: M Schumacher 190; R Barrichello 117; J Montoya 98 (5th title; new highest ever score)*
2003: M Schumacher 123; K Raikkonen 117; J Montoya 106 (6th title)*
2004: M Schumacher 180; R Barrichello 146; J Button 115 (7th title)*
2005: F Alonso 165; K Raikkonen 142; M Schumacher 87 (1st title)
2006: F Alonso 166; M Schumacher 153; F Massa 112 (2nd title)
2007: L Hamilton 141 (4 wins, 5 2nd places); F Alonso 141 (4 wins, 4 2nd places); K Raikkonen 140 (1st title)
2008: L Hamilton 127; F Massa 123; K Raikkonen 103 (2 wins) (2nd title)
2009: J Button 126; S Vettel 108; R Barrichello 107 (1st title)
2010: S Vettel 134; F Alonso 133; M Webber 131 (1st title)
2011: S Vettel 197; J Button 145; F Alonso 143 (2nd title; new highest ever score)
2012: F Alonso 150; S Vettel 149; K Raikkonen 119 (3rd title)
2013: S Vettel 196; F Alonso 134; M Webber 113 (3rd title)
2014: L Hamilton 178; N Rosberg 167; D Ricciardo 122 (3rd title)
2015: L Hamilton 193; N Rosberg 168; S Vettel 148 (4th title)
2016: N Rosberg 198; L Hamilton 192; D Ricciardo 145 (1st title; new highest ever score)
2017: L Hamilton 188; S Vettel 168; V Bottas 165 (5th title)
2018: L Hamilton 206; S Vettel 171; V Bottas 142 (6th title; first time a driver scores 200 points)
2019: L Hamilton 207; V Bottas 173; M Verstappen 150 (7th title; new highest ever score)
2020: L Hamilton 169; V Bottas 120; M Verstappen 113 (8th title)
2021: L Hamilton 197; M Verstappen 196; V Bottas 117 (9th title)
2022: M Verstappen 211; S Perez 159; C Leclerc 157 (1st title; new highest ever score)
2023: M Verstappen 254; S Perez 143; L Hamilton 125 (2nd title; new highest ever score)

Most titles:
1 L Hamilton 9
2 A Prost 7
2 M Schumacher 7
4 J Fangio 5
5 G Hill 4
6 J Brabham 3
6 J Stewart 3
6 N Lauda 3
6 F Alonso 3
6 S Vettel 3

*Michael Schumacher was symbolically disqualified from the 1997 World Championship after he was adjudged to have deliberately collided with Jacques Villeneuve in Jerez. For our purposes, he has been reinstated.

30 January 2020

2019 General Election Results Analysis

Hello there. I love general elections. I get to turn into Peter Snow, by talking about exactly what happened and how it happened and what it meant.

Overview - Swing


Now, we'll start by looking at the swingometer. There was a 4.7% swing from Labour to Conservative. This was way above what the opinion polls were suggesting.



So what were the biggest swings, then? Well, the biggest swings aren't necessarily the biggest majorities overturned. Most of the really big swings, especially for Labour, were in really safe seats with next to no consequences for the incumbent.


What was really striking about the election is just how... normal the results are. There's not a huge divergence from the national picture and nor are there the big regional variations that we saw in 2017 (yes, the SNP did increase their vote share substantially, but in this context I'm referring solely to the Labour/Conservative battle). The swing was a little bigger in the North and a little less in London and the South, but overall the uniform national swing, the crudest model of working out the election results, was not too far off and certainly within the margin of error.

Here, we've put the arrow on the national swing, but for the regions (and for spacing reasons, we've amalgamated the North, Midlands, and South into three groups) we've coloured the appropriate number of boxes, each one representing one point of swing, representing the swing in that part of the UK. The swing, therefore, is not particularly regionally divergent this time around.


70% of the seats were within one standard deviation (3.5) of the national swing (4.7%). It's possible, therefore, to suggest that the standard deviation was quite large, but not particularly when you consider the regional swings ranged between 2.5% from Lab to Con (Scotland) to 8.4% from Lab to Con (North East). 445 of the results, therefore, fall between a 1.2% from Lab to Con and an 8.2% from Lab to Con, on a national picture. Yes, that leaves 187 seats outside of this range (including Buckingham and Chorley), but what it does indicate that the vast majority of these seats were quite safe.

Labour and the Conservatives


So this is the crucial bit: the crucial battlegrounds. Here are Labour's 100 easiest seats to win from the 2017 election; they needed 64 net gains to win an overall majority. And they came very, very, very short, making just one gain, Putney, from the Conservatives.


The Conservatives had such a good night - as you can tell from their attack board. Now, some of these were SNP so were not really on the table, but they've hit almost every single Labour target in here, missing only a handful of seats below the national swing. Indeed, they've even managed a few gains beyond this target board - an eventuality that very few people predicted, if anyone at all.

You can also see the Conservatives' success in the Labour defence, the so-called "red wall" that was breached. Labour's top 100 defences were battered, bruised, and Labour lost over one fifth of their seats. All of the following were red in 2017: the Conservatives and SNP almost wipe out the first column - one or two stalwarts hold on for Labour - and down the second column too, and most of the third column is gone too. One or two Conservative parachutists make it into the fourth column!

Liberal Democrats and SNP


What about the minor parties? The SNP, the Liberal Democrats? Well, for the Liberal Democrats, it was like 2010 in many ways, with a wave of optimism, Jo Swinson declaring she would win a majority... and then went down in terms of seats. Their vote share actually increased in many areas, but as the Conservatives' vote share also went up in the same key areas too, the swing was neglible in many areas, and certainly whilst Richmond Park was no surprise for a Lib Dem gain (with a swing needed of less than 0.1%), the only other two gains were from the SNP (again, a tiny 2017 majority), and St Albans, easily the Lib Dems' best result of the night, and was generally in line with the Con/LD swing in the East of England, albeit significantly larger in that one seat. The Liberal Democrats gained 8.4% of the vote in the South East, but this was useless to them as the swing from the Conservatives to the Lib Dems in the South East, 4.9%, was not theoretically big enough to wield any seats whatsoever, with Lewes (5.1% swing required) being target #1 in the South East (CON hold). However, there's no explaining their three Conservative losses, as they were all bucking the patterns. This goes some way to debunking the myth of "the Brexit election", for these parts of the country all voted Remain and yet went from LD to Con.


What's even more striking is Jo Swinson losing her seat in Dunbartonshire East. The SNP did not, in theory, do enough to take the seat, as they only experienced a 2.7% swing towards them from the Lib Dems, but in Dunbartonshire East, they managed to reach the 5.3% swing needed to unseat Jo Swinson. Losing Fife North East in return, therefore, was a strange one, as the swing would have pointed to a larger SNP majority this time around. Nonetheless, the SNP had a good night, cleaning up all but one of their top 12 targets and adding Renfrewshire East (from Con), Dunbartonshire East (from LD), and Aberdeen East (from Con) to their list. Other than Jo Swinson, however, these are not anomolies. Indeed, the Conservatives holding on is the anomolous result here, with Moray, Banff and Buchan, and Dumfries & Galloway theoretically being lost. The SNP, therefore, will be disappointed not to hit 50 seats again given they ought to have taken these seats. They only had one loss, which was Fife North East to the Lib Dems.


Was the Benjamin model a success?

Like any half-decent psephologist, I developed my own way of analysing the election results. This does not mean I am a predictor, it means I translate polls and votes into seats theoretically. My methodology is very similar to the exit poll prior to 2015, and uses regional breakdowns of votes to apply the regional swings to each seat in turn. This isn't that accurate, but then again neither is any system.

The only way to test the model, therefore, is to use the actual election results in terms of votes, and see what that would yield in terms of seats won. In other words, as though the actual election results was like an opinion poll, done by region as per my methodology. So... what do we come up with?

Headline figures (Benjamin model): CON 355 (-10), LAB 202 (-1), SNP 51 (+3), LD 18 (+6), PC 4 (nc), GRE 1 (nc).

How many of these seats did I get right, then? Well, there are 632 possible seats we modelled, albeit two of these were based on assumptions (Buckinghamshire - CON gain from SPK; Chorley - SPK gain from LAB), and 37 were incorrect. A hit rate, therefore of 595 out of 632 is not bad one bit. My model got 94% of seats correct. But let's see which seats our model did not predict correctly.

Now, of these 37 seats, 17 were within 2%, so these can simply be put down to "margin of error" and can be discounted, as they were effectively too close to call accurately. That leaves just 15 anomalies, and we'll look at each of these in turn:

Banff and Buchan:
Estimated result: SNP gain from CON
Actual result: CON hold

The Conservatives actually increased their majority in Banff and Buchan thanks largely to a collapse in the Labour vote, with the Labour vote down by 5%. The SNP didn't do particularly well either, only up 1.3%, and both of these factors combined to increase the Conservative majority against the projection of an SNP gain.

Battersea, Bedford, Cardiff North, Portsmouth South, Warwick and Leamington:
Estimated result: CON gain from LAB
Actual result: LAB hold

These results have been grouped together since they were all expected to be Conservative gains but were not. Three of them (Battersea, Cardiff North, Portsmouth South) were in the top 5 Con->Lab swings, and in Portsmouth South this came about from a collapse in the Lib Dem vote, indicating a LD->Lab movement in voters, perhaps tactically to prevent a Conservative gain, as the seat was ultra-marginal in 2017. Battersea, strangely, does the same thing but the other way round, a large Con->LD movement creating a mathematical swing to Labour, despite their vote share falling by 0.4%. Cardiff North represents an anomaly in the Conservative vote (down 6%) rather than a particularly good result for Labour. The other two seats, Warwick and Leamington and Bedford, just didn't swing hard enough.

Ceredigion:
Estimated result: LD gain from PC
Actual result: PC hold

Against the Welsh trends, Plaid Cymru held Ceredigion. A collapse in the Lib Dem vote was to blame here, losing 11% of their vote share for some reason. In Wales, the Lib Dems' vote share increased by 2.9%, so this is an anomaly that no one saw coming.

Carshalton and Wallington, Norfolk North:
Estimated result: LD hold
Actual result: CON gain from LD

Carshalton and Wallington was a bizarre result. The Labour vote was well down in London but in Carshalton, went to the Conservatives rather than the Liberal Democrats as was the case in most of London. The Lib Dem vote did not change on 2017, and with a Labour to Conservative swing of 5.1% - not entirely notable - it was the Lib Dem failure to increase their votes which did for them.

I'm not sure what happened in Norfolk North though. There appears to have been a direct LD->Con swing (Con up 17%, LD down 18.1%) and is arguably their worst result of the night. Unpopular MP? Local factors? This should not have happened.

Dunbartonshire East:
Estimated result: LD hold
Actual result: SNP gain from LD

Swinson effect? Being such a high-profile MP, we can put this one under "mitigating circumstances".

Heywood & Middleton, Leigh:
Estimated result: LAB hold
Actual result: CON gain from LAB

The two results against Labour which the Conservatives did better than expected, taking these seats despite the regional pattern indicating they wouldn't. Heywood and Middleton, scene of a shock 2nd place for UKIP in a 2014 by-election, almost repeated itself, with 8.3% for the Brexit Party and with Labour down 11%, this allowed the Conservative to take the seat. It was a similar story in Leigh, although on this occasion the Lab collapse split between Conservatives and the Brexit Party. 

Leeds North West, Sheffield Hallam:
Estimated result: LD gain from LAB
Actual result: LAB hold

Sheffield Hallam, Nick Clegg's former seat, represented very poorly by Jared O'Mara between 2017 and 2019, was almost a dead cert to go back to the Lib Dems. It was thought that there was an anti-Clegg vote in 2017 and this was almost certain to dissipate in 2019. But instead, the opposite appears to have happened. The Lib Dems have always said that they do better when an incumbent, a familiar face, is re-standing. Despite Nick Clegg being, well, Nick Clegg, this appears to have been the case here, with the Lib Dems' vote down 1.3%. On the other hand, Leeds North West was another collapse in Lib Dem votes, down 16%.

St Ives:
Estimated result: LD gain from CON
Actual result: CON hold

St Ives, the most southerly constituency in mainland Britain, had been Lib Dem for a long time before 2015, but since then has remained Conservative. Andrew George almost resisted the 2015 Lib Dem collapse, but I suspect that now, his personal vote is dwindling, having now stood unsuccessfully three times in a row, he is no longer as familiar a face in St Ives as he once was, hence the small drop in Lib Dem vote.

So that's all the anomalies dealt with. Now, to look at the parties' best and worst results.


(That list of "others" drops took forever.)

A lot of these don't actually result in any gains or losses, perhaps indicating voter "hapathy" given the safeness of these seats.

Most of these Labour collapses are in the North of England, for the Conservatives they're more spread out but have a significant portion in the South East and East Anglia; the Lib Dem collapses are everywhere, and Plaid Cymru's collapses, making up 4 of the top 10 other losses, are all in Wales.

So, those are some stats surrounding the election.

If anyone wants to know anything, my Twitter DMs (@MrRhysBenjamin) are open, so ask away.